Address video-audio example review feedback

Signed-off-by: Hongsheng Liu <liuhongsheng4@huawei.com>

.buildkite/cuda/test-weekly.yml

@@ -30,6 +30,14 @@ steps:
       - pytest -s -v tests/dfx/reliability/test_reliability_hunyuan_image.py -m "slow"
     mirror_hardwares: h100_4
 
+  - label: "Reliability Test - voxcpm2"
+    timeout_in_minutes: 180
+    depends_on: upload-weekly-pipeline
+    if: build.env("WEEKLY") == "1" || build.pull_request.labels includes "weekly-test"
+    commands:
+      - pytest -s -v tests/dfx/reliability/test_reliability_voxcpm2.py -m "slow"
+    mirror_hardwares: l4_1
+
   - group: "Reliability Test - Invalid parameters Test"
     depends_on: upload-weekly-pipeline
     if: build.env("WEEKLY") == "1" || build.pull_request.labels includes "weekly-test"

.buildkite/npu/test-npu-nightly.yml

@@ -69,12 +69,6 @@ steps:
           VLLM_ALLOW_LONG_MAX_MODEL_LEN: "1"
           HF_TOKEN: "${HF_TOKEN}"
         commands:
-          # Pin the Seed-TTS scorer to the second card: _get_eval_device() otherwise
-          # auto-picks npu:0, which the omni server under test already fills with its
-          # KV cache, so Whisper OOMs on `.to(device)` and every clip reports
-          # "ASR / WER failed".
-          - export SEED_TTS_WER_EVAL=1
-          - export SEED_TTS_EVAL_DEVICE=npu:1
           - |
             set +e
             pytest -s -v tests/e2e/accuracy/minicpmo_4_5/test_minicpmo_4_5.py -m "full_model and npu and A3" --run-level full_model

.buildkite/npu/test-npu-ready.yml

@@ -1,15 +1,4 @@
 steps:
-  - label: "Platforms Test"
-    key: ready-platforms-test
-    depends_on: upload-ready-pipeline
-    mirror_hardwares: a3_npu_2
-    commands:
-      - >-
-        pytest -s -v
-        tests/platforms/npu/test_minicpmo_code2wav_npugraph.py
-        -m 'core_model and npu and A3 and omni'
-        --run-level 'core_model'
-
   - group: ":card_index_dividers: Omni Model Test"
     key: ready-omni-group
     depends_on: upload-ready-pipeline

docs/.nav.yml

@@ -43,7 +43,6 @@ nav:
     - Overview: features/README.md
     - Runtime and Stage Execution:
       - Execution Modes and Streaming: user_guide/diffusion/execution_modes.md
-      - Prefill-Decode Disaggregation (experimental): features/pd_disaggregation.md
       - Sleep Mode: features/sleep_mode.md
     - Quantization:
       - Overview: user_guide/quantization/overview.md

docs/configuration/README.md

@@ -14,6 +14,10 @@ For introduction, please check [Introduction for stage config](./stage_configs.m
 
 - **[GPU Memory Calculation and Configuration](./gpu_memory_utilization.md)** - Guide on how to calculate memory requirements and set up `gpu_memory_utilization` for optimal performance
 
+## Multi-Stage Recipes
+
+- **[Prefill-Decode Disaggregation](./pd_disaggregation.md)** - How to derive a PD-aware Qwen3-Omni stage config from the default config without introducing another bundled YAML
+
 ## Optimization Features
 
 - **[Diffusion Features Overview](../user_guide/diffusion_features.md)** - Complete overview of all diffusion model features and supported models

docs/configuration/pd_disaggregation.md

@@ -0,0 +1,171 @@
+# Prefill-Decode (PD) Disaggregation
+
+PD disaggregation splits the Qwen3-Omni thinker into separate prefill and decode
+stages so prompt processing and token generation can run on different workers.
+
+This is documented as a stage-config recipe instead of a bundled YAML because the
+deployment-specific values usually change per environment:
+
+- GPU placement
+- `tensor_parallel_size`
+- connector backend and connector ports
+- connector IPs or bootstrap addresses
+
+Start from the [default Qwen3-Omni stage config](gh-file:vllm_omni/deploy/qwen3_omni_moe.yaml)
+and copy it to your own file, for example `qwen3_omni_pd.yaml`. Then apply the
+changes below.
+
+## Requirements
+
+- 3+ GPUs for a basic layout: prefill, decode, and talker+code2wav
+- A KV connector supported by vLLM, such as `MooncakeConnector`
+- Matching `tensor_parallel_size` on the prefill and decode thinker stages
+
+## 1. Split the thinker into prefill and decode stages
+
+Replace the original thinker stage with two stages:
+
+```yaml
+stage_args:
+  - stage_id: 0
+    stage_type: llm
+    is_prefill_only: true
+    runtime:
+      devices: "0"
+    engine_args:
+      max_num_seqs: 16
+      model_stage: thinker
+      model_arch: Qwen3OmniMoeForConditionalGeneration
+      worker_type: ar
+      scheduler_cls: vllm_omni.core.sched.omni_ar_scheduler.OmniARScheduler
+      gpu_memory_utilization: 0.9
+      enforce_eager: true
+      trust_remote_code: true
+      engine_output_type: latent
+      distributed_executor_backend: "mp"
+      enable_prefix_caching: false
+      max_num_batched_tokens: 32768
+      hf_config_name: thinker_config
+      tensor_parallel_size: 1
+      kv_transfer_config:
+        kv_connector: "MooncakeConnector"
+        kv_role: "kv_producer"
+        kv_rank: 0
+        kv_parallel_size: 2
+        kv_connector_extra_config:
+          mooncake_bootstrap_port: 25201
+    final_output: false
+    is_comprehension: true
+    default_sampling_params:
+      temperature: 0.4
+      top_p: 0.9
+      top_k: 1
+      max_tokens: 2048
+      seed: 42
+      detokenize: True
+      repetition_penalty: 1.05
+
+  - stage_id: 1
+    stage_type: llm
+    is_decode_only: true
+    runtime:
+      devices: "1"
+    engine_args:
+      max_num_seqs: 64
+      model_stage: thinker
+      model_arch: Qwen3OmniMoeForConditionalGeneration
+      worker_type: ar
+      scheduler_cls: vllm_omni.core.sched.omni_ar_scheduler.OmniARScheduler
+      gpu_memory_utilization: 0.9
+      enforce_eager: true
+      trust_remote_code: true
+      engine_output_type: latent
+      distributed_executor_backend: "mp"
+      enable_prefix_caching: false
+      max_num_batched_tokens: 32768
+      hf_config_name: thinker_config
+      tensor_parallel_size: 1
+      kv_transfer_config:
+        kv_connector: "MooncakeConnector"
+        kv_role: "kv_consumer"
+        kv_rank: 1
+        kv_parallel_size: 2
+        kv_connector_extra_config:
+          mooncake_bootstrap_port: 25202
+    engine_input_source: [0]
+    final_output: true
+    final_output_type: text
+    is_comprehension: true
+    default_sampling_params:
+      temperature: 0.4
+      top_p: 0.9
+      top_k: 1
+      max_tokens: 2048
+      seed: 42
+      detokenize: True
+      repetition_penalty: 1.05
+```
+
+Notes:
+
+- `is_prefill_only: true` marks the thinker stage that only saves KV.
+- `is_decode_only: true` marks the thinker stage that resumes from remote KV.
+- `kv_transfer_config` is required on both stages.
+- The orchestrator forces the prefill stage to run with `max_tokens=1`, so the
+  prefill side only processes the prompt and exports KV.
+
+## 2. Shift the downstream stages by one index
+
+After inserting the extra thinker stage, renumber the remaining stages:
+
+```yaml
+  - stage_id: 2
+    runtime:
+      devices: "2"
+    engine_input_source: [1]
+    sync_process_input_func: vllm_omni.model_executor.stage_input_processors.qwen3_omni.thinker2talker_token_only
+
+  - stage_id: 3
+    runtime:
+      devices: "2"
+    engine_args:
+      max_num_seqs: 1
+    engine_input_source: [2]
+    sync_process_input_func: vllm_omni.model_executor.stage_input_processors.qwen3_omni.talker2code2wav_token_only
+```
+
+Compared with the default Qwen3-Omni config:
+
+- the talker becomes stage `2` instead of stage `1`
+- the code2wav stage becomes stage `3` instead of stage `2`
+- the talker now reads from decode stage `1`
+
+## 3. Add runtime edges for the four-stage pipeline
+
+```yaml
+runtime:
+  enabled: true
+  edges:
+    - from: 0
+      to: 1
+    - from: 1
+      to: 2
+    - from: 2
+      to: 3
+```
+
+## 4. Launch with your custom config
+
+```bash
+vllm serve Qwen/Qwen3-Omni-30B-A3B-Instruct --omni --port 8091 \
+    --deploy-config /path/to/qwen3_omni_pd.yaml
+```
+
+## Operational Notes
+
+- `MooncakeConnector` does not support heterogeneous TP sizes across the PD
+  pair. Keep prefill and decode at the same `tensor_parallel_size`.
+- If the thinker requires TP=2, both thinker stages must use TP=2 and be given
+  separate GPU sets, for example `"0,1"` for prefill and `"2,3"` for decode.
+- Choose connector ports and addresses that match your deployment. The values
+  shown above are examples only.

docs/configuration/stage_configs.md

@@ -38,12 +38,11 @@ Each entry under `stages:` accepts any `StageDeployConfig` field directly (no ne
 |-------|------|----------|---------|-------------|
 | `stage_id` | int | required | — | Stage identity; matched against `PipelineConfig.stages[*].stage_id`. |
 | `max_num_seqs` | int | optional | `64` | Max concurrent sequences per stage. |
-| `gpu_memory_utilization` | float | optional | `0.9` | Per-stage total memory target; used for automatic KV-cache sizing. |
-| `kv_cache_memory_bytes` | int \| null | optional | `null` | Explicit per-rank KV-cache budget in bytes; overrides automatic sizing. |
+| `gpu_memory_utilization` | float | optional | `0.9` | Per-stage memory budget. |
 | `tensor_parallel_size` | int | optional | `1` | TP degree for this stage. |
 | `enforce_eager` | bool | optional | `false` | Disable CUDA graphs. |
-| `max_num_batched_tokens` | int | optional | `32768` | Per-stage prefill/token budget; also contributes to the native maximum in-flight token limit. |
-| `max_model_len` | int \| null | optional | `null` | Per-sequence context or KV length; `-1` enables native cache-capacity auto-fitting, while values above the HF default auto-set `VLLM_ALLOW_LONG_MAX_MODEL_LEN=1`. |
+| `max_num_batched_tokens` | int | optional | `32768` | Prefill budget. |
+| `max_model_len` | int \| null | optional | `null` | Per-stage context length (auto-sets `VLLM_ALLOW_LONG_MAX_MODEL_LEN=1` when larger than HF default). |
 | `async_scheduling` | bool \| null | optional | `null` | Per-stage async scheduling toggle. |
 | `devices` | str | optional | `"0"` | `CUDA_VISIBLE_DEVICES`-style device list. |
 | `output_connectors` | dict \| null | optional | `null` | Keyed by `to_stage_<n>`; values are names registered under top-level `connectors:`. |

docs/contributing/ci/test_execution_guide.md

@@ -184,6 +184,9 @@ Failed jobs: 1/2
     # Reliability: HunyuanImage DiT (H100 × 4)
     pytest -s -v dfx/reliability/test_reliability_hunyuan_image.py -m slow
 
+    # Reliability: VoxCPM2 (L4)
+    pytest -s -v dfx/reliability/test_reliability_voxcpm2.py -m slow
+
     ```
 
     The latest L5 CI jobs (reliability + invalid-parameter weekly steps) are in [test-weekly.yml](https://github.com/vllm-project/vllm-omni/blob/main/.buildkite/cuda/test-weekly.yml).

docs/contributing/ci/test_writing_guide.md

@@ -562,6 +562,7 @@ L5 level testing focuses on the performance of model services under ***long-runn
     -   `test_reliability_qwen3_omni.py` — Qwen3-Omni chat / multimodal reliability (GPU OOM, process kill, recovery, error contract under `--async-chunk` vs default).
     -   `test_reliability_wan22.py` — Wan2.2 I2V video API reliability (`/v1/videos` under OOM and process kill, recovery).
     -   `test_reliability_hunyuan_image.py` — HunyuanImage-3.0-Instruct DiT-only reliability (`/v1/images/generations` under OOM and process kill; deploy `hunyuan_image3_dit.yaml`, H100 × 4).
+    -   `test_reliability_voxcpm2.py` — VoxCPM2 TTS reliability (`/v1/audio/speech` under OOM and process kill; deploy `voxcpm2.yaml`, L4 × 1).
     -   `helpers.py` — Shared primitives used by current suites: raw HTTP probes for `/v1/chat/completions` and `/health`, OpenAI-style error parsing, GPU OOM sidecar (`inject_gpu_oom` / `stop_gpu_oom_hogs`), and `pgrep`-based process-kill injector construction (`make_process_kill_fault_injector`).
     -   `conftest.py` — `fault_injector` and `omni_server_after_fault` / `omni_server_after_fault_function` fixtures to run a callable **after** the server is ready.
     -   `README.md` — Short local run commands for this directory.
@@ -661,12 +662,13 @@ Reliability tests are **short fault-injection** integration runs (L5 **(b)** in
 | `test_reliability_qwen3_omni.py` | Qwen3-Omni: OOM vs **text vs text+audio** error contract, large multimodal chat under OOM, concurrent pressure, **SIGKILL** on `VLLM::Worker`, `/health` → 503 + fast-fail + concurrent chat; optional OOM recovery scenario (may be skipped while tracked in issues). |
 | `test_reliability_wan22.py` | Wan2.2 I2V: large `/v1/videos` under OOM, **SIGKILL** on `multiprocessing.spawn` chain, health / fast-fail / concurrent video requests; optional recovery test (may be skipped). |
 | `test_reliability_hunyuan_image.py` | HunyuanImage DiT-only: large `/v1/images/generations` under OOM, **SIGKILL** on `vLLM-Omni::` workers and serve/tree targets, health / fast-fail / concurrent image requests; some OOM/recovery cases may be skipped while tracked in issues. |
+| `test_reliability_voxcpm2.py` | VoxCPM2: `/v1/audio/speech` under OOM (error contract), **SIGKILL** on `VLLM::` workers and serve/tree targets, health / fast-fail / concurrent speech requests; some OOM cases may be skipped while tracked in issues. |
 | `README.md` | Minimal run / collect examples. |
 
 ##### Parametrization and markers
 
 - Each test module defines a **`RELIABILITY_SCENARIOS`** list (`test_name`, `server_params`: model, `stage_config_name` or diffusion `server_args`, etc.). **`create_reliability_omni_server_params()`** in `tests/dfx/conftest.py` resolves stage paths (including XPU substitutions where applicable) and builds **`OmniServerParams`** lists consumed by **`@pytest.mark.parametrize(..., indirect=True)`** on `omni_server` or `omni_server_function`.
-- Cases are tagged **`@pytest.mark.slow`** for weekly / selective CI. GPU-heavy suites use **`@hardware_test(res={"cuda": "H100"}, num_cards=...)`** (Qwen3-Omni **2**× H100; Wan2.2 **1**× H100; HunyuanImage DiT **4**× H100).
+- Cases are tagged **`@pytest.mark.slow`** for weekly / selective CI. GPU-heavy suites use **`@hardware_test(res={"cuda": "H100"}, num_cards=...)`** or **`@hardware_test(res={"cuda": "L4"}, num_cards=1)`** (Qwen3-Omni **2**× H100; Wan2.2 **1**× H100; HunyuanImage DiT **4**× H100; VoxCPM2 **1**× L4).
 - **Process-kill** tests use **`@pytest.mark.skipif(os.name == "nt", ...)`** because injection uses POSIX **`pgrep` / `kill`**.
 
 ##### CI trigger
@@ -678,11 +680,13 @@ Weekly Buildkite (`.buildkite/cuda/test-weekly.yml`) runs one step per model sui
 | Reliability Test - qwen3-omni | `test_reliability_qwen3_omni.py` | H100 × 2 (`mithril-h100-pool`) |
 | Reliability Test - wan22 | `test_reliability_wan22.py` | H100 × 2 (`mithril-h100-pool`) |
 | Reliability Test - hunyuan-image | `test_reliability_hunyuan_image.py` | H100 × 4 (`mithril-h100-pool`) |
+| Reliability Test - voxcpm2 | `test_reliability_voxcpm2.py` | L4 × 1 (`gpu_1_queue`) |
 
 ```bash
 pytest -s -v tests/dfx/reliability/test_reliability_qwen3_omni.py -m "slow"
 pytest -s -v tests/dfx/reliability/test_reliability_wan22.py -m "slow"
 pytest -s -v tests/dfx/reliability/test_reliability_hunyuan_image.py -m "slow"
+pytest -s -v tests/dfx/reliability/test_reliability_voxcpm2.py -m "slow"
 ```
 
 ##### Local commands
@@ -692,6 +696,7 @@ pytest --collect-only tests/dfx/reliability
 pytest -s -v tests/dfx/reliability/test_reliability_qwen3_omni.py -m slow
 pytest -s -v tests/dfx/reliability/test_reliability_wan22.py -m slow
 pytest -s -v tests/dfx/reliability/test_reliability_hunyuan_image.py -m slow
+pytest -s -v tests/dfx/reliability/test_reliability_voxcpm2.py -m slow
 ```
 
 ##### Adding a new model suite
@@ -704,4 +709,4 @@ pytest -s -v tests/dfx/reliability/test_reliability_hunyuan_image.py -m slow
 </details>
 
 -   -   ***Stability***: `pytest -s -v tests/dfx/stability/scripts/test_stability_qwen3_omni.py` or `pytest -s -v tests/dfx/stability/scripts/test_stability_wan22.py` (or add `test_stability_<model>.py` alongside a matching JSON config)
-    -   ***Reliability***: `pytest -s -v tests/dfx/reliability/test_reliability_<model>.py -m slow` (current suites: `qwen3_omni`, `wan22`, `hunyuan_image`; add `test_reliability_<suite>.py` and a matching step in `.buildkite/cuda/test-weekly.yml` for new models)
+    -   ***Reliability***: `pytest -s -v tests/dfx/reliability/test_reliability_<model>.py -m slow` (current suites: `qwen3_omni`, `wan22`, `hunyuan_image`, `voxcpm2`; add `test_reliability_<suite>.py` and a matching step in `.buildkite/cuda/test-weekly.yml` for new models)

docs/design/feature/offloader/distributed_layerwise_offload.md

@@ -10,30 +10,6 @@ For user-facing commands, see the
 [distributed layerwise offloading guide](../../../user_guide/diffusion/offloader/distributed_layerwise_offload.md)
 and the [Cosmos3 recipe](https://github.com/vllm-project/vllm-omni/blob/main/recipes/cosmos3/Cosmos3-DistOffload.md).
 
-## Feature compatibility
-
-Host-storage optimization and runtime compatibility are separate decisions.
-When direct checkpoint mmap is unavailable, DLO can still use tensors produced
-by the ordinary loader. "Compatibility path" below means that fallback is
-implemented but has less end-to-end coverage than the primary path.
-
-Legend: ✅ supported, ⚠️ compatibility path or limited validation, ❌ unsupported.
-
-| Feature | DLO + AllGather | DLO without AllGather |
-|---|---|---|
-| **DP** | ✅ Primary path; host weights are sharded across the DP group. | ✅ Each DP rank streams complete rank-local blocks. |
-| **SP** | ✅ When DP=1, DLO uses the SP group for weight sharding. | ✅ SP remains active without a DLO weight collective. |
-| **TP > 1** | ⚠️ Ordinary TP-aware loader only; no direct checkpoint mmap. | ⚠️ Ordinary TP-aware loader only; no direct checkpoint mmap. |
-| **HSDP** | ❌ Rejected to avoid double-sharding parameters. | ⚠️ Limited end-to-end coverage. |
-| **Per-tensor online FP8 linears** | ✅ Ordinary loader finalizes weights and scales before DLO sharding. | ✅ Ordinary loader retains complete rank-local tensors. |
-| **Other online quantization methods** | ❌ Rejected until runtime packing and scale layouts are validated. | ⚠️ Allowed through the ordinary loader; validation is method-specific. |
-| **Model-level or standard layerwise CPU offload** | ❌ Disabled because DLO takes priority. | ❌ Disabled because DLO takes priority. |
-| **Resident leading layers** | ❌ Rejected. | ✅ Requires eligible resident paths in the model's `OffloadPlan`. |
-
-See [Parallelism compatibility](#parallelism-compatibility) and
-[Request and loading constraints](#request-and-loading-constraints) for the
-detailed contracts and validation boundaries.
-
 ## Status
 
 DLO is implemented for multi-device diffusion execution. The default
@@ -205,11 +181,8 @@ requirements.
 
 Direct checkpoint mmap can back either transfer path. It is currently limited
 to proven TP1, non-HSDP, non-online-quantized layouts. Other layouts use the
-ordinary loader. Per-tensor online FP8 linears can use DLO AllGather after the
-ordinary loader finalizes their runtime weights and scales; DLO then shards and
-reconstructs those tensors with their recorded layouts. Other online methods
-must use `--dlo-no-use-allgather` or disable online quantization until their
-runtime layouts are validated.
+ordinary loader. Online quantization remains incompatible with DLO AllGather;
+use `--dlo-no-use-allgather` or disable online quantization.
 
 A normalized runtime mmap cache, built through the ordinary loader, is the
 proposed general mechanism for sharing transformed TP or quantized layouts.
@@ -224,8 +197,6 @@ Current source-level validation includes:
 - HSDP + DLO without AllGather acceptance at configuration level;
 - loader preflight fallback for TP, HSDP, online quantization, unknown custom
   loaders, missing keys, and shape/dtype mismatches;
-- ordinary-loader fallback for per-tensor online FP8 linears followed by DLO
-  sharding of finalized weights and scales;
 - exact loader-to-backend plan transfer and ordinary-loader fallback;
 - rank-local mmap source retention, bounded two-slot staging, and adapter
   transforms without parameter-side flags;

docs/design/feature/quantization.md

@@ -190,15 +190,12 @@ configured parallelism strategy:
 - `WeightsMapper` transformations must be applied before matching ignored-layer
   lists or component prefixes.
 
-Distributed layerwise offload has an additional boundary: direct checkpoint
-mmap cannot prepare online-quantized tensors because it bypasses the ordinary
-loader that creates their runtime weights and scales. Per-tensor online FP8
-linears can still use DLO AllGather through the ordinary loader: each rank first
-finalizes the FP8 tensors, then DLO shards those runtime tensors for
-reconstruction during prefetch. Other online methods remain limited to
-`--dlo-no-use-allgather` until their packing and scale layouts are validated.
-See the [DLO design document](offloader/distributed_layerwise_offload.md) for
-the full compatibility matrix.
+Distributed layerwise offload has an additional boundary: its sharded mmap
+AllGather path is not compatible with online quantization because the path
+expects already materialized checkpoint weights. Use the rank-local
+`--dlo-no-use-allgather` path or a compatible pre-quantized checkpoint when
+combining these features. See the [DLO design document](offloader/distributed_layerwise_offload.md)
+for the full compatibility matrix.
 
 ## Adding a quantization backend
 

docs/design/module/vllm_omni_config.md

@@ -18,8 +18,8 @@ validation_paths:
   - tests/config/**
 upstream_refs:
   - vllm.config
-last_reviewed: 2026-08-11
-last_verified_commit: e356708da4cd39992405686dacfc32b18fcccc7f
+last_reviewed: 2026-08-07
+last_verified_commit: 3d7fc3b9ba3cac88d579d4dc35b78b0b641675fc
 ---
 
 # vLLM-Omni configuration
@@ -39,66 +39,11 @@ The PR review proposal to capture environment-derived values in configuration
 objects at initialization is an open design question, not a current
 invariant.
 
-## Current structured-stage reuse behavior
-
-The current RFC #4021 implementation uses upstream vLLM config classes as the
-schema source for AR and generation stages:
-
-- `OmniStageLoadConfig` inherits `vllm.config.LoadConfig`.
-- `OmniStageCacheConfig` inherits `vllm.config.CacheConfig`.
-- `OmniStageSchedulerConfig` inherits `vllm.config.SchedulerConfig`.
-- `OmniStageParallelConfig` inherits `vllm.config.ParallelConfig`.
-
-This inheritance preserves the existing load, cache, scheduler, and parallel
-concern boundaries while reducing duplicate downstream declarations. It does
-not, by itself, make every inherited field an effective vLLM-Omni runtime
-option. Construction and runtime consumption are separate surfaces:
-
-| Surface | Current behavior |
-| --- | --- |
-| Structured schema | Upstream dataclass fields are inherited. Their defaults, default factories, and applicable Pydantic validation can participate when the structured object is constructed. |
-| Omni input ownership | Pipeline/deploy/CLI construction continues to accept only fields with an existing structured owner. Direct construction of an inherited sub-config can expose a wider upstream schema. |
-| Engine projection | For AR and generation stages, reusable fields are discovered by intersecting each upstream config dataclass with upstream `EngineArgs`. Only constructor-explicit values are emitted; inherited defaults remain deferred to terminal materialization. |
-| Terminal materialization | The engine-owning process constructs the final upstream `VllmConfig` and performs model-, platform-, rank-, port-, and backend-dependent initialization. |
-
-The generated projection maps include known upstream naming differences such
-as `cache_dtype` to `kv_cache_dtype`, `policy` to `scheduling_policy`, and
-`data_parallel_master_ip` to `data_parallel_address`. A field added to both an
-upstream concern config and `EngineArgs` therefore enters the AR/generation
-projection without requiring a second downstream allowlist. Explicit inherited
-fields that have no projection raise an error rather than being silently
-dropped.
-
-Ownership exclusions remain deliberate. Stage topology owns `scheduler_cls`,
-cache owns `disable_hybrid_kv_cache_manager`, and Omni runtime owns
-`distributed_executor_backend` and `worker_cls`; vLLM's private API-process
-fields are terminal internals. These exclusions prevent one input from being
-constructed or projected through two config concerns. Effective-engine-argument
-tests cover both the dynamic mapping and this exclusion boundary.
-
-`CompilationConfig` and `ProfilerConfig` are direct structured-stage fields:
-`stage.compilation_config` and `stage.profiler_config`. They accept mapping/YAML
-inputs and complete upstream value objects. Mapping inputs are validated and
-materialized as their concrete upstream types when the structured config is
-constructed, so downstream consumers see one resolved representation.
-Prebuilt upstream objects retain their type across the structured projection
-boundary.
-
-Quantization remains on its existing Omni-owned transport contract. This
-reuse step does not adopt upstream `QuantizationConfigArgs`, because the
-current engine-specific quantization split needs to be resolved separately.
-
-Diffusion keeps its existing engine-field ownership and projection surface.
-Shared Python inheritance must not be treated as evidence that an LLM-only
-parallel input is effective for a diffusion stage.
-
 ## Safe-change guide while deferred
 
 Changes should trace every affected producer to its runtime consumer and test
 the structured, legacy, CLI, and deployment paths that are actually supported.
-For inherited vLLM fields, tests should separately cover constructor behavior
-and effective `EngineArgs` projection. Do not infer a stable contract from
-this placeholder page.
+Do not infer a stable contract from this placeholder page.
 
 ## Promotion gate
 

docs/features/README.md

@@ -20,19 +20,13 @@ an implementation contract has a user-facing workflow.
 | Choose serial, batched, step-wise, or streaming diffusion execution | [Execution Modes and Streaming](../user_guide/diffusion/execution_modes.md) | [Diffusion Continuous Batching](../design/feature/diffusion_continuous_batching.md), [Async Diffusion Output](../design/feature/async_diffusion_output.md) |
 | Reclaim stage memory without restarting the server | [Sleep Mode](sleep_mode.md) | Runtime lifecycle behavior is documented in the user guide |
 
-The [Prefill-Decode Disaggregation design reference](pd_disaggregation.md)
-describes the experimental Qwen3-Omni runtime topology and its current
-integration boundaries. It is not yet a supported launch recipe.
-
 Some runtime designs are deliberately not promoted as standalone User Guide
 features yet:
 
-- Generic, OmniConnector-based
-  [Disaggregated Inference](../design/feature/disaggregated_inference.md) is not
-  yet a standalone User Guide feature. Its
+- [Disaggregated Inference](../design/feature/disaggregated_inference.md) and
   [OmniConnector implementations](../design/index.md#communication) describe
-  transport contracts, while practical connector configuration remains under
-  [Pipeline and deploy configurations](../configuration/stage_configs.md).
+  topology and transport contracts. Practical deployment configuration remains
+  under [Pipeline and deploy configurations](../configuration/stage_configs.md).
 - [Async Chunk](../design/feature/async_chunk.md) and
   [Async Omni Output Materialization](../design/feature/omni_async_output_materialization.md)
   are model- and pipeline-dependent. Use the selected model's deploy

docs/features/pd_disaggregation.md

@@ -1,140 +0,0 @@
-# Prefill-Decode (PD) Disaggregation (experimental)
-
-!!! warning "Experimental design reference"
-    PD disaggregation is only partially integrated on the current `main` branch.
-    The runtime contains PD detection and routing scaffolding, but there is no
-    supported deploy overlay or end-to-end validated launch recipe yet. Treat this
-    page as a description of the current design, not as production guidance.
-
-Prefill-decode disaggregation splits the Qwen3-Omni Thinker into two logical
-stages: a prefill worker that processes the prompt and produces KV cache, and a
-decode worker that imports that cache and generates tokens. The Talker and
-Code2Wav stages remain downstream; PD does not disaggregate the entire
-Qwen3-Omni pipeline.
-
-## Current design
-
-### Runtime topology
-
-```mermaid
-flowchart LR
-    client["Client request"] --> orchestrator["Omni Orchestrator"]
-    orchestrator -->|"original prompt<br/>max_tokens = 1"| prefill["Thinker prefill<br/>KV producer"]
-    prefill -->|"KV blocks"| connector[("vLLM KV connector")]
-    prefill -->|"KV routing metadata"| orchestrator
-    orchestrator -->|"original prompt<br/>remote KV parameters"| decode["Thinker decode<br/>KV consumer"]
-    connector -->|"remote KV load"| decode
-    decode -->|"generated text"| client
-```
-
-The orchestrator owns the request lifecycle, while the vLLM KV connector owns
-the bulk KV transfer. The original prompt is submitted to both Thinker stages:
-the decode worker still needs the request tokens and multimodal features to
-construct its request, while the imported KV cache prevents it from repeating
-the completed prefill computation.
-
-For the full Qwen3-Omni pipeline, splitting the Thinker adds one logical stage:
-
-| Stage | Role | Output or handoff |
-| --- | --- | --- |
-| 0 | Thinker prefill | Saves prompt KV and prompt-side multimodal output |
-| 1 | Thinker decode | Loads remote KV, generates text, and produces decode-side conditioning |
-| 2 | Talker | Converts combined Thinker conditioning into codec codes |
-| 3 | Code2Wav | Converts codec codes into audio |
-
-Text remains a final output of the decode stage, and audio remains a final
-output of Code2Wav.
-
-### Request lifecycle
-
-1. At startup, PD detection looks for one stage marked `is_prefill_only` and
-   one downstream stage marked `is_decode_only`. The decode stage must list the
-   prefill stage as an input source.
-2. The orchestrator clones the Thinker sampling parameters for prefill, forces
-   `max_tokens=1`, clears stop conditions, and sets producer-side
-   `kv_transfer_params`. This makes prefill finish by length after exporting KV.
-3. When prefill finishes, the orchestrator captures its KV routing metadata and
-   Qwen multimodal output under the original request ID.
-4. The orchestrator submits the original prompt to decode with consumer-side
-   `kv_transfer_params`. Those parameters identify the transfer, prefill engine,
-   bootstrap endpoint, and, when required by the connector, remote request.
-5. The decode engine imports the KV cache and continues generation. The normal
-   Omni stage path then carries Thinker conditioning through Talker and
-   Code2Wav.
-
-### Control and data paths
-
-PD uses two different connector layers. They solve different handoffs and are
-not interchangeable.
-
-| Path | Payload | Owner |
-| --- | --- | --- |
-| PD control path | Request ID, original prompt, `transfer_id`, bootstrap address, engine ID, and connector-specific request metadata | Omni Orchestrator |
-| PD data path | Thinker attention KV blocks | vLLM KV connector |
-| Downstream model path | Thinker hidden states and embeddings, Talker codec codes, and audio tensors | vLLM-Omni stage connectors and processors |
-
-### Detection and validation
-
-The current runtime accepts one logical prefill/decode pair per pipeline. Each
-stage may use tensor parallelism, but both sides of the pair must satisfy these
-invariants:
-
-- both stages define `kv_transfer_config`
-- prefill uses `kv_producer` or `kv_both`
-- decode uses `kv_consumer` or `kv_both`
-- both stages use the same KV connector
-- `kv_buffer_device` and `kv_buffer_size` agree when set on both stages
-- both stages use the same `tensor_parallel_size`
-
-These checks happen before request processing so an inconsistent PD topology
-fails during engine initialization.
-
-## What is implemented today
-
-| Area | Current state |
-| --- | --- |
-| Pair detection and topology validation | Implemented in the shared PD helper |
-| Prefill sampling preparation | Implemented for sync and async entrypoints |
-| Prefill-to-decode routing | Implemented in the current Orchestrator |
-| Deploy-based configuration | Not yet migrated; the supported `stages:` deploy schema cannot currently expand Qwen3-Omni into the four-stage PD topology |
-| Qwen3 Thinker-to-Talker handoff | Bridge state and embedding-merge helpers exist, but the merge is not wired into the live stage processor |
-| Connector compatibility | Incomplete; the Orchestrator currently requires `remote_request_id`, while the pinned Mooncake connector completes prefill without returning that metadata |
-| Validation coverage | PD entrypoint tests and the PD deploy overlay in online end-to-end tests are temporarily disabled during migration |
-| Performance | No non-regression or throughput/latency result is established for the current path |
-
-Because these pieces are still converging, the previous legacy `stage_args` YAML
-and `vllm serve` command have been removed. The serve CLI now accepts the
-deploy-based `stages:` schema, so retaining the legacy recipe would imply a
-launch path that current `main` rejects.
-
-## Configuration contract under development
-
-The intended deploy integration starts from the
-[default Qwen3-Omni deploy config](gh-file:vllm_omni/deploy/qwen3_omni_moe.yaml)
-and expands its frozen three-stage pipeline into the four stages shown above.
-A complete integration must:
-
-- split the Thinker definition without duplicating the Talker or Code2Wav model
-  contracts
-- remap downstream stage IDs, input sources, and stage connectors
-- inject producer and consumer `kv_transfer_config` values into the two Thinker
-  stages
-- allocate separate device sets for prefill and decode
-- preserve both final outputs: decode text and Code2Wav audio
-- provide connector-specific routing metadata without hard-coding one connector
-  contract
-
-The required number of GPUs depends on tensor-parallel sizes and whether Talker
-and Code2Wav share a device. A basic TP=1 layout uses one device for prefill, one
-for decode, and one for the two downstream audio stages.
-
-## Implementation map
-
-- PD detection, validation, and sampling-parameter preparation:
-  [`PDDisaggregationMixin`](gh-file:vllm_omni/entrypoints/pd_utils.py)
-- startup detection and connector metadata:
-  [`AsyncOmniEngine`](gh-file:vllm_omni/engine/async_omni_engine.py)
-- request-scoped prefill-to-decode routing:
-  [`Orchestrator`](gh-file:vllm_omni/engine/orchestrator.py)
-- Qwen3 downstream conditioning helpers:
-  [Qwen3-Omni stage input processor](gh-file:vllm_omni/model_executor/stage_input_processors/qwen3_omni.py)

docs/models/supported_models.md

@@ -39,7 +39,6 @@ th {
 | `WanImageToVideoPipeline` | Wan2.2-I2V | `Wan-AI/Wan2.2-I2V-A14B-Diffusers` | ✅︎ | ✅︎ | ✅︎ | ✅︎ | — |
 | `LingBotWorldCausalDMDPipeline` | LingBot-World 2.0 v1 (experimental) | `robbyant/lingbot-world-v2-14b-causal-fast-diffusers` | Experimental | | | | — |
 | `Cosmos3OmniDiffusersPipeline` | Cosmos3 T2I, T2V, I2V, V2V, T2V with sound, action policy | `nvidia/Cosmos3-Nano`, `nvidia/Cosmos3-Super` | ✅︎ | ✅︎ | ✅︎ | ✅︎ | — |
-| `SanaWmPipeline` | SANA-WM | `BBBBruce/SANA-WM_bidirectional-stage1-diffusers` | ✅︎ | | | | [Repository](https://github.com/vllm-project/vllm-omni/blob/main/recipes/NVIDIA/SANA-WM.md) |
 | `Wan22S2VPipeline` | Wan2.2-S2V | `Wan-AI/Wan2.2-S2V-14B` | ✅︎ | ✅︎ | ✅︎ | ✅︎ | — |
 | `Wan22VACEPipeline` | Wan2.1-VACE | `Wan-AI/Wan2.1-VACE-1.3B-diffusers`, `Wan-AI/Wan2.1-VACE-14B-diffusers` | ✅︎ | ✅︎ | ✅︎ | ✅︎ | — |
 | `Wan22VACEPipeline` | Wan2.2-VACE | `Pyros13/Wan2.2-VACE-Fun-A14B-Diffusers` | ✅︎ |   |   |   | — |
@@ -54,7 +53,6 @@ th {
 | `LingBotVideoPipeline` | LingBot-Video dense and MoE T2I, T2V, TI2V | `robbyant/lingbot-video-dense-1.3b`, `robbyant/lingbot-video-moe-30b-a3b` | ✅︎ | | | | — |
 | `MiniMaxH3Pipeline` | MiniMax H3 T2VA, first/last-keyframe FL2VA, and mixed-reference Ref2VA | `MiniMaxAI/MiniMax-H3` | ✅︎ | ✅︎ | | | [Published](https://recipes.vllm.ai/MiniMaxAI/MiniMax-H3) |
 | `DreamZeroPipeline` | DreamZero-DROID | `GEAR-Dreams/DreamZero-DROID` | ✅︎ | ✅︎ | ✅︎ | ✅︎ | — |
-| `Pi0Pipeline` | π0 (Pi-Zero) VLA | `lerobot/pi0_base` | ✅︎ | | | | [Repository](https://github.com/vllm-project/vllm-omni/blob/main/recipes/lerobot/Pi0.md) |
 | `HeliosPipeline`, `HeliosPyramidPipeline` | Helios | `BestWishYsh/Helios-Base`, `BestWishYsh/Helios-Mid`, `BestWishYsh/Helios-Distilled` | ✅︎ | ✅︎ | ✅︎ | | — |
 | `MagiHumanPipeline` | MagiHuman | `SII-GAIR/daVinci-MagiHuman-Base-1080p` | ✅︎ | ✅︎ | | | — |
 | `OvisImagePipeline` | Ovis-Image | `OvisAI/Ovis-Image` | ✅︎ | ✅︎ | | ✅︎ | — |
@@ -79,7 +77,6 @@ th {
 | `SoulXSingerPipeline` | SoulX-Singer (SVS) | `Soul-AILab/SoulX-Singer` | ✅︎ | | | | — |
 | `SoulXSingerSVCPipeline` | SoulX-Singer-SVC | `Soul-AILab/SoulX-Singer` (`model-svc.pt`) | ✅︎ | | | | — |
 | `AudioXPipeline` | AudioX | `zhangj1an/AudioX` | ✅︎ | ✅︎ | | | — |
-| `MiniMaxMusic3ForConditionalGeneration` | MiniMax Music 3 (text-to-music) | `MiniMaxAI/MiniMax-Music3` | ✅︎ | | | | — |
 | `Qwen3TTSForConditionalGeneration` | Qwen3-TTS-12Hz-1.7B-CustomVoice | `Qwen/Qwen3-TTS-12Hz-1.7B-CustomVoice` | ✅︎ | | | | [Published](https://recipes.vllm.ai/Qwen/Qwen3-TTS-12Hz-1.7B-CustomVoice) |
 | `Qwen3TTSForConditionalGeneration` | Qwen3-TTS-12Hz-1.7B-VoiceDesign | `Qwen/Qwen3-TTS-12Hz-1.7B-VoiceDesign` | ✅︎ | ✅︎ | ✅︎ | ✅︎ | — |
 | `Qwen3TTSForConditionalGeneration` | Qwen3-TTS-12Hz-1.7B-Base | `Qwen/Qwen3-TTS-12Hz-1.7B-Base` | ✅︎ | ✅︎ | ✅︎ | ✅︎ | — |

docs/serving/speech_api.md

@@ -141,20 +141,7 @@ The response shape depends on the streaming parameters:
 
 **Non-streaming (default).** With `stream=false` and no `stream_format`, returns the
 complete clip as binary audio data with an appropriate `Content-Type` header (e.g.
-`audio/wav`). Because the raw-bytes body has no JSON carrier, successful
-non-streaming responses from non-diffusion speech servers report usage through
-response headers:
-
-| Header | Description |
-| --- | --- |
-| `x-vllm-omni-input-tokens` | Total input tokens (`text_tokens` + `audio_tokens`). |
-| `x-vllm-omni-output-tokens` | Generated codec/audio tokens. |
-| `x-vllm-omni-total-tokens` | `input_tokens` + `output_tokens`. |
-| `x-vllm-omni-input-text-tokens` | Tokens from the synthesized text (`input` plus `instructions`). |
-| `x-vllm-omni-input-audio-tokens` | Reference-audio codec frames, non-zero only for in-context voice cloning. |
-
-Diffusion-mode speech servers route through a separate response path and do not
-emit these headers.
+`audio/wav`). The raw-bytes body has no JSON carrier, so no `usage` is reported.
 
 **Raw audio stream** (`stream_format="audio"`). Streams raw audio bytes (PCM or
 WAV) as they are decoded.

docs/serving/videos_api.md

@@ -167,14 +167,6 @@ transfer options such as `control_path`, `control_guidance`,
 `num_conditional_frames`, `show_control_condition`, and `show_input`; see the
 Cosmos3 recipe for complete examples.
 
-HTTP redirects for `image_reference.image_url` follow vLLM's
-`VLLM_MEDIA_URL_ALLOW_REDIRECTS` setting. Before starting the server, set it to
-`1` (the default) to allow redirects or `0` to reject them. A redirect target
-can differ from the original host, and vLLM-Omni does not yet fully implement
-upstream vLLM's media URL allowlist protection. Deployments should therefore
-treat remote media URLs as untrusted and choose this setting as part of their
-URL access policy.
-
 ### Speech-to-Video
 
 For models that support audio-driven generation (e.g., Wan2.2-S2V), pass both

docs/user_guide/diffusion/offloader/distributed_layerwise_offload.md

@@ -138,11 +138,8 @@ must enter each collective.
   but it provides no shared-mmap host-memory benefit or guarantee.
 - HSDP plus AllGather is rejected to avoid double sharding. HSDP without
   AllGather has limited end-to-end validation.
-- Per-tensor online FP8 linears use the ordinary loader and can run with either
-  DLO transfer path. With AllGather, every rank temporarily materializes the
-  complete FP8 model in host memory before DLO retains only its shard. Other
-  online quantization methods require no-AllGather until their runtime layouts
-  are validated.
+- Online quantization uses the ordinary loader with no-AllGather. It remains
+  incompatible with DLO AllGather.
 - Resident leading layers require `--dlo-no-use-allgather` and a model
   `OffloadPlan` that declares eligible `resident_dit_paths`.
 - DP concurrency requires an explicit, identical inference-step count.

docs/user_guide/diffusion/parallelism/hsdp.md

@@ -152,8 +152,7 @@ In `DiffusionParallelConfig`:
 **Constraints:**
 
 - `hsdp_replicate_size × hsdp_shard_size == world_size`
-- HSDP cannot be used with Tensor Parallelism, ordinary Data Parallelism,
-  Pipeline Parallelism, or Expert Parallelism.
+- HSDP cannot be used with Tensor Parallelism (`tensor_parallel_size` must be 1)
 
 ---
 

docs/user_guide/diffusion_features.md

@@ -168,14 +168,6 @@ The following tables show which models support each feature:
 | **Cosmos3**                  |     ❌     |     ✅      |           ✅           |       ✅        |         ✅         |         ❌         |   ✅    |             ✅             |  ✅ (encode/decode)   |       ✅        |        ❌         |
 | **LongCat-Video-Avatar-1.5** |     ❌     |     ❌      |           ❌           |       ❌        |         ❌         |         ❌         |   ❌    |             ❌             |          ❌           |       ❌        |        ❌         |
 | **MiniMax-H3**               | ✅ (FL2VA) |     ✅      |           ✅           |       ❌        |       ✅ (DiT/TE)  |         ❌         |   ✅    |             ✅             |       ✅ (tile)       |      ✅ (DiT)      |        ❌         |
-| **SANA-WM**                  |     ❌     |     ❌      |          ❌<sup>5</sup> |       ✅        |         ✅         |         ❌         |   ❌    |             ❌             |          ❌           |       ❌        |        ❌         |
-
-> Notes:
-> 5. SANA-WM cannot support sequence parallelism: its bidirectional gated delta
->    recurrence carries state across frames, so a rank cannot denoise a slice of
->    the token sequence in isolation. Doing so would need a distributed scan or
->    an all-gather before every GDN block. The remaining ❌ columns are simply
->    unvalidated on this model, not known-broken.
 
 > **Step execution note:** Helios supports single-request step execution only;
 > use `max_num_seqs=1`.
@@ -183,7 +175,7 @@ The following tables show which models support each feature:
 **Frame Interpolation Support**
 
 - **Supported**: Wan2.2 text-to-video, image-to-video, and TI2V pipelines
-- **Not supported**: Wan2.1-VACE, LTX-2, LTX-2.3, LTX-2.5, Helios, HunyuanVideo-1.5, DreamID-Omni, SANA-WM
+- **Not supported**: Wan2.1-VACE, LTX-2, LTX-2.3, LTX-2.5, Helios, HunyuanVideo-1.5, DreamID-Omni
 
 ### AudioGen
 

docs/user_guide/examples/offline_inference/x_to_video_audio.md

@@ -3,7 +3,26 @@
 Source <https://github.com/vllm-project/vllm-omni/tree/main/examples/offline_inference/x_to_video_audio>.
 
 
-The `DreamID-Omni` pipeline generates short videos from text, image and video.
+The shared runner generates video with optional audio from text, image, and audio
+inputs. It can be extended through `vllm_omni.model_extras` without adding
+another model-specific runner.
+
+## Shared request and output contract
+
+The example builds one canonical request envelope: `prompt` is text,
+`modalities` is `["video"]`, and optional `multi_modal_data` contains RGB PIL
+images plus audio `(waveform, sample_rate)` tuples. Audio files are loaded in
+full by design; the runner does not apply a time window. Prepare a trimmed input
+file explicitly when only part of a recording should condition the model.
+DreamID-Omni's audio conditioning sequence grows with the complete input
+duration, so long recordings increase denoising memory and step time.
+
+Model-specific request translation (for example, DreamID-Omni's two negative
+prompts and its 16 kHz input requirement) is declared in `model_extras`. Model
+outputs must provide `fps` and, when audio is present, `audio_sample_rate`.
+`--fps` and `--audio-sample-rate` are available as explicit metadata overrides.
+The shared encoder normalizes supported video tensor layouts, so this runner
+does not branch on a model name or assume a model-specific frame layout.
 
 ## Local CLI Usage
 ### Download the Model locally
@@ -91,7 +110,8 @@ python x_to_video_audio.py \
 Key arguments:
 - `--prompt`: text description (string).
 - `--model`: path to the model local directory.
-- `--height/--width`: output resolution (defaults 704 * 1024).
+- `--model-class-name`: optional diffusion pipeline class override.
+- `--height/--width`: output resolution (defaults 704 * 1280).
 - `--image-path`: path to the input image list.
 - `--audio-path`: path to the input audio list, indicate the timbre of the output video.
 - `--cfg-parallel-size`: number of parallel cfg parallel (defaults 1).
@@ -101,6 +121,9 @@ Key arguments:
 - `--num-inference-steps`: number of denoising steps (defaults 45).
 - `--video-negative-prompt`: negative prompt for video generation.
 - `--audio-negative-prompt`: negative prompt for audio generation.
+- `--extra-body`: JSON object containing parameters declared by the selected model.
+- `--fps`: explicit output FPS override when a model does not return that metadata.
+- `--audio-sample-rate`: explicit output audio sample-rate override.
 - `--enable-cpu-offload`: enable CPU offload (defaults False).
 
 ## Example materials

docs/user_guide/fault_injection_reliability_matrix.md

@@ -65,7 +65,7 @@ More failure modes (for example, network jitter and network interruption) will b
 
 ## Source of Conclusions
 
-The behaviors and conclusions above are summarized from current fault injection validation results on **`Qwen3-Omni`**, **`Wan2.2`**, and **`HunyuanImage-3.0-Instruct`** (DiT-only, `/v1/images/generations`). Automated coverage lives under `tests/dfx/reliability/` (`test_reliability_qwen3_omni.py`, `test_reliability_wan22.py`, `test_reliability_hunyuan_image.py`) and runs weekly via `.buildkite/cuda/test-weekly.yml`.
+The behaviors and conclusions above are summarized from current fault injection validation results on **`Qwen3-Omni`**, **`Wan2.2`**, **`HunyuanImage-3.0-Instruct`** (DiT-only, `/v1/images/generations`), and **`VoxCPM2`** (`/v1/audio/speech`). Automated coverage lives under `tests/dfx/reliability/` (`test_reliability_qwen3_omni.py`, `test_reliability_wan22.py`, `test_reliability_hunyuan_image.py`, `test_reliability_voxcpm2.py`) and runs weekly via `.buildkite/cuda/test-weekly.yml`.
 
 ### Example: `SIGTERM` fault injection log (Qwen3-Omni)
 

docs/user_guide/quantization/fp8.md

@@ -179,24 +179,6 @@ python text_to_image.py --model <your-model> --quantization fp8 --ignored-layers
 vllm serve <your-model> --omni --quantization fp8
 ```
 
-### Distributed layerwise offload
-
-Per-tensor online FP8 linears can be combined with DLO's default AllGather path:
-
-```bash
-vllm serve <your-model> \
-  --omni \
-  --quantization fp8 \
-  --enable-distributed-layerwise-offload
-```
-
-Direct checkpoint mmap is not used for this combination. Each rank runs the
-ordinary loader to create the final FP8 weights and scales before DLO shards
-them, so startup temporarily materializes the complete FP8 model in host memory
-on every rank. Runtime DLO host residency is sharded normally. Use
-`--dlo-no-use-allgather` when the synchronized AllGather request-wave contract
-does not fit the workload.
-
 ## Parameters
 
 | Parameter | Type | Default | Description |

examples/offline_inference/custom_pipeline/image_to_image/image_edit.py

@@ -53,6 +53,7 @@ from typing import Any
 import torch
 from PIL import Image
 
+from vllm_omni.diffusion.data import DiffusionParallelConfig
 from vllm_omni.entrypoints.omni import Omni
 from vllm_omni.inputs.data import OmniDiffusionSamplingParams
 from vllm_omni.outputs import OmniRequestOutput
@@ -138,6 +139,12 @@ async def main():
 
     # ---- Torch setup ----
     generator = torch.Generator(device=current_omni_platform.device_type).manual_seed(args.seed)
+    parallel_config = DiffusionParallelConfig(
+        ulysses_degree=args.ulysses_degree,
+        ring_degree=args.ring_degree,
+        cfg_parallel_size=args.cfg_parallel_size,
+        tensor_parallel_size=args.tensor_parallel_size,
+    )
 
     # ---- Cache Config ----
     if args.cache_backend == "cache_dit":
@@ -164,10 +171,7 @@ async def main():
         vae_use_tiling=args.vae_use_tiling,
         cache_backend=args.cache_backend,
         cache_config=cache_config,
-        ulysses_degree=args.ulysses_degree,
-        ring_degree=args.ring_degree,
-        cfg_parallel_size=args.cfg_parallel_size,
-        tensor_parallel_size=args.tensor_parallel_size,
+        parallel_config=parallel_config,
         enforce_eager=args.enforce_eager,
         enable_cpu_offload=args.enable_cpu_offload,
         diffusion_load_format="dummy",

examples/offline_inference/diffusion/lingbot_world_v2.py

@@ -154,6 +154,8 @@ def _positive_finite(value: float, flag: str) -> float:
 def build_omni_kwargs(
     args: argparse.Namespace,
     paths: LingBotPaths,
+    *,
+    parallel_config: Any,
 ) -> dict[str, Any]:
     """Build Omni configuration without overriding checkpoint class discovery."""
 
@@ -165,7 +167,7 @@ def build_omni_kwargs(
     return {
         "model": model,
         "flow_shift": flow_shift,
-        "tensor_parallel_size": args.tensor_parallel_size,
+        "parallel_config": parallel_config,
         "enforce_eager": args.enforce_eager,
         "model_config": {"lingbot_action_root": str(paths.action_root)},
     }
@@ -260,10 +262,12 @@ def main(argv: Sequence[str] | None = None) -> Path:
     # helper tests work on machines without the native vLLM runtime.
     from diffusers.utils import export_to_video
 
+    from vllm_omni.diffusion.data import DiffusionParallelConfig
     from vllm_omni.entrypoints.omni import Omni
     from vllm_omni.inputs.data import OmniDiffusionSamplingParams
 
-    omni_kwargs = build_omni_kwargs(args, paths)
+    parallel_config = DiffusionParallelConfig(tensor_parallel_size=args.tensor_parallel_size)
+    omni_kwargs = build_omni_kwargs(args, paths, parallel_config=parallel_config)
     omni = Omni(**omni_kwargs)
     try:
         outputs = omni.generate(

examples/offline_inference/diffusion/lingbot_world_v2_realtime.py

@@ -125,6 +125,7 @@ async def run(argv: Sequence[str] | None = None) -> Path:
     # require a CUDA-enabled vLLM installation.
     import torch
 
+    from vllm_omni.diffusion.data import DiffusionParallelConfig
     from vllm_omni.diffusion.models.lingbot_world.actions import LingBotCameraControlReducer
     from vllm_omni.entrypoints.async_omni import AsyncOmni
     from vllm_omni.experimental.ar_diffusion.consumer import ARDiffusionOmniTickConsumer
@@ -140,7 +141,7 @@ async def run(argv: Sequence[str] | None = None) -> Path:
         model=args.model,
         engine_backend="vllm_omni.experimental.ar_diffusion.engine.ARDiffusionEngine",
         enforce_eager=args.enforce_eager,
-        tensor_parallel_size=args.tensor_parallel_size,
+        parallel_config=DiffusionParallelConfig(tensor_parallel_size=args.tensor_parallel_size),
         max_num_seqs=1,
         model_config={
             "ar_diffusion_height": args.height,

examples/offline_inference/image_to_image/image_edit.py

@@ -96,6 +96,7 @@ from typing import Any
 import torch
 from PIL import Image
 
+from vllm_omni.diffusion.data import DiffusionParallelConfig
 from vllm_omni.diffusion.utils.image_output import extract_images_from_outputs
 from vllm_omni.diffusion.utils.param_utils import apply_declared_extra_args
 from vllm_omni.entrypoints.omni import Omni
@@ -451,6 +452,14 @@ def main():
 
     generator = torch.Generator(device=current_omni_platform.device_type).manual_seed(args.seed)
 
+    parallel_config = DiffusionParallelConfig(
+        ulysses_degree=args.ulysses_degree,
+        ring_degree=args.ring_degree,
+        cfg_parallel_size=args.cfg_parallel_size,
+        tensor_parallel_size=args.tensor_parallel_size,
+        enable_expert_parallel=args.enable_expert_parallel,
+    )
+
     # Configure cache based on backend type
     cache_config = None
     if args.cache_backend == "cache_dit":
@@ -481,11 +490,7 @@ def main():
         vae_use_tiling=args.vae_use_tiling,
         cache_backend=args.cache_backend,
         cache_config=cache_config,
-        ulysses_degree=args.ulysses_degree,
-        ring_degree=args.ring_degree,
-        cfg_parallel_size=args.cfg_parallel_size,
-        tensor_parallel_size=args.tensor_parallel_size,
-        enable_expert_parallel=args.enable_expert_parallel,
+        parallel_config=parallel_config,
         enable_cpu_offload=args.enable_cpu_offload,
         enable_diffusion_pipeline_profiler=args.enable_diffusion_pipeline_profiler,
         profiler_config=args.profiler_config,

examples/offline_inference/image_to_video/image_to_video.py

@@ -56,6 +56,7 @@ import numpy as np
 import PIL.Image
 import torch
 
+from vllm_omni.diffusion.data import DiffusionParallelConfig
 from vllm_omni.diffusion.utils.param_utils import apply_declared_extra_args
 from vllm_omni.entrypoints.omni import Omni
 from vllm_omni.inputs.data import OmniDiffusionSamplingParams
@@ -502,6 +503,17 @@ def main():
         }
 
     profiler_enabled = args.profiler_config is not None
+    parallel_config = DiffusionParallelConfig(
+        ulysses_degree=args.ulysses_degree,
+        ring_degree=args.ring_degree,
+        cfg_parallel_size=args.cfg_parallel_size,
+        tensor_parallel_size=args.tensor_parallel_size,
+        vae_patch_parallel_size=args.vae_patch_parallel_size,
+        use_hsdp=args.use_hsdp,
+        hsdp_shard_size=args.hsdp_shard_size,
+        hsdp_replicate_size=args.hsdp_replicate_size,
+        pipeline_parallel_size=args.pipeline_parallel_size,
+    )
     omni_kwargs = dict(
         model=args.model,
         enable_layerwise_offload=args.enable_layerwise_offload,
@@ -512,15 +524,7 @@ def main():
         diffusion_kv_cache_skip_steps=args.diffusion_kv_cache_skip_steps,
         diffusion_kv_cache_skip_layers=args.diffusion_kv_cache_skip_layers,
         enable_cpu_offload=args.enable_cpu_offload,
-        ulysses_degree=args.ulysses_degree,
-        ring_degree=args.ring_degree,
-        cfg_parallel_size=args.cfg_parallel_size,
-        tensor_parallel_size=args.tensor_parallel_size,
-        vae_patch_parallel_size=args.vae_patch_parallel_size,
-        use_hsdp=args.use_hsdp,
-        hsdp_shard_size=args.hsdp_shard_size,
-        hsdp_replicate_size=args.hsdp_replicate_size,
-        pipeline_parallel_size=args.pipeline_parallel_size,
+        parallel_config=parallel_config,
         enforce_eager=args.enforce_eager,
         model_class_name=model_class_name,
         cache_backend=args.cache_backend,

examples/offline_inference/longcat_video/end2end.py

@@ -15,6 +15,7 @@ import imageio
 import numpy as np
 import torch
 
+from vllm_omni.diffusion.data import DiffusionParallelConfig
 from vllm_omni.diffusion.models.longcat_video.pipeline_longcat_video_avatar import (
     prepare_longcat_video_avatar_model_for_omni,
 )
@@ -235,6 +236,14 @@ def main():
         dtype="bfloat16",
         enforce_eager=args.enforce_eager,
         additional_config=additional_config,
+        parallel_config=DiffusionParallelConfig(
+            ulysses_degree=1,
+            ring_degree=1,
+            cfg_parallel_size=1,
+            tensor_parallel_size=1,
+            vae_patch_parallel_size=1,
+            pipeline_parallel_size=1,
+        ),
     )
 
     audio_input: str | dict[str, str] | None = args.audio

examples/offline_inference/speech_to_video/speech_to_video.py

@@ -23,6 +23,7 @@ from typing import Any
 
 import torch
 
+from vllm_omni.diffusion.data import DiffusionParallelConfig
 from vllm_omni.entrypoints.omni import Omni
 from vllm_omni.inputs.data import OmniDiffusionSamplingParams
 from vllm_omni.platforms import current_omni_platform
@@ -249,6 +250,17 @@ def main():
             "scm_steps_policy": "dynamic",
         }
 
+    parallel_config = DiffusionParallelConfig(
+        tensor_parallel_size=args.tensor_parallel_size,
+        cfg_parallel_size=args.cfg_parallel_size,
+        vae_patch_parallel_size=args.vae_patch_parallel_size,
+        ring_degree=args.ring_degree,
+        ulysses_degree=args.ulysses_degree,
+        use_hsdp=args.use_hsdp,
+        hsdp_shard_size=args.hsdp_shard_size,
+        hsdp_replicate_size=args.hsdp_replicate_size,
+    )
+
     profiler_enabled = args.profiler_config is not None
 
     omni_kwargs = dict(
@@ -259,14 +271,7 @@ def main():
         vae_use_tiling=args.vae_use_tiling,
         enable_cpu_offload=args.enable_cpu_offload,
         enable_layerwise_offload=args.enable_layerwise_offload,
-        tensor_parallel_size=args.tensor_parallel_size,
-        cfg_parallel_size=args.cfg_parallel_size,
-        vae_patch_parallel_size=args.vae_patch_parallel_size,
-        ring_degree=args.ring_degree,
-        ulysses_degree=args.ulysses_degree,
-        use_hsdp=args.use_hsdp,
-        hsdp_shard_size=args.hsdp_shard_size,
-        hsdp_replicate_size=args.hsdp_replicate_size,
+        parallel_config=parallel_config,
         enforce_eager=args.enforce_eager,
         enable_diffusion_pipeline_profiler=args.enable_diffusion_pipeline_profiler,
         cache_backend=args.cache_backend,

examples/offline_inference/text_to_audio/text_to_audio.py

@@ -38,6 +38,7 @@ from typing import Any
 import numpy as np
 import torch
 
+from vllm_omni.diffusion.data import DiffusionParallelConfig
 from vllm_omni.diffusion.utils.param_utils import apply_declared_extra_args
 from vllm_omni.entrypoints.omni import Omni
 from vllm_omni.inputs.data import OmniDiffusionSamplingParams
@@ -305,13 +306,17 @@ def main():
     print(f"  Seed: {args.seed}")
     print(f"{'=' * 60}\n")
 
+    parallel_config = DiffusionParallelConfig(
+        use_hsdp=args.use_hsdp,
+        hsdp_shard_size=args.hsdp_shard_size,
+        hsdp_replicate_size=args.hsdp_replicate_size,
+    )
+
     # Initialize Omni. AudioX requires an explicit pipeline class; Stable Audio
     # is resolved from its config.
     omni_kwargs: dict[str, Any] = dict(
         model=args.model,
-        use_hsdp=args.use_hsdp,
-        hsdp_shard_size=args.hsdp_shard_size,
-        hsdp_replicate_size=args.hsdp_replicate_size,
+        parallel_config=parallel_config,
         cache_backend=args.cache_backend,
         cache_config=cache_config,
         enable_diffusion_pipeline_profiler=args.enable_diffusion_pipeline_profiler,

examples/offline_inference/text_to_speech/ming_tts/end2end.py

@@ -13,7 +13,6 @@ import yaml
 from transformers import AutoTokenizer
 from vllm.utils.argparse_utils import FlexibleArgumentParser
 
-from vllm_omni.model_executor.models.common.ming.speaker_extractor import SpeakerEmbeddingExtractor
 from vllm_omni.model_executor.models.ming_tts.config_ming_tts import (
     KEY_CFG,
     KEY_MAX_DECODE_STEPS,
@@ -22,6 +21,7 @@ from vllm_omni.model_executor.models.ming_tts.config_ming_tts import (
     SAMPLE_RATE,
 )
 from vllm_omni.model_executor.models.ming_tts.prompt_assembly import build_ming_dense_prompt
+from vllm_omni.model_executor.models.ming_tts.speaker_extractor import MingSpeakerEmbeddingExtractor
 
 try:
     from .runner import (
@@ -142,8 +142,7 @@ def _resolve_speaker_embedding(args, case: dict, paths: list[str]):
         return _load_speaker_embedding(args.speaker_embedding)
     if not (case.get("auto_extract_speaker_embeddings") or args.extract_speaker_embeddings) or not paths:
         return None
-    # Runs before the engine is built, so nothing has pre-fetched campplus.onnx.
-    embs = SpeakerEmbeddingExtractor(args.model, allow_download=True).extract_many(paths)
+    embs = MingSpeakerEmbeddingExtractor(args.model).extract_many(paths)
     if not embs:
         raise RuntimeError("Speaker extraction produced no embeddings")
     return embs[0] if len(embs) == 1 else torch.stack(embs, dim=0)

examples/offline_inference/text_to_video/text_to_video.py

@@ -10,6 +10,7 @@ from typing import Any
 import numpy as np
 import torch
 
+from vllm_omni.diffusion.data import DiffusionParallelConfig
 from vllm_omni.diffusion.utils.param_utils import apply_declared_extra_args
 from vllm_omni.entrypoints.omni import Omni
 from vllm_omni.inputs.data import OmniDiffusionSamplingParams
@@ -461,6 +462,17 @@ def main():
             "scm_steps_policy": "dynamic",
         }
 
+    # Configure parallel settings
+    parallel_config = DiffusionParallelConfig(
+        ulysses_degree=args.ulysses_degree,
+        ring_degree=args.ring_degree,
+        cfg_parallel_size=args.cfg_parallel_size,
+        tensor_parallel_size=args.tensor_parallel_size,
+        vae_patch_parallel_size=args.vae_patch_parallel_size,
+        pipeline_parallel_size=args.pipeline_parallel_size,
+        enable_expert_parallel=args.enable_expert_parallel,
+    )
+
     profiler_enabled = args.profiler_config is not None
 
     omni_kwargs = dict(
@@ -469,13 +481,7 @@ def main():
         vae_use_slicing=args.vae_use_slicing,
         vae_use_tiling=args.vae_use_tiling,
         enable_cpu_offload=args.enable_cpu_offload,
-        ulysses_degree=args.ulysses_degree,
-        ring_degree=args.ring_degree,
-        cfg_parallel_size=args.cfg_parallel_size,
-        tensor_parallel_size=args.tensor_parallel_size,
-        vae_patch_parallel_size=args.vae_patch_parallel_size,
-        pipeline_parallel_size=args.pipeline_parallel_size,
-        enable_expert_parallel=args.enable_expert_parallel,
+        parallel_config=parallel_config,
         enforce_eager=args.enforce_eager,
         model_class_name=model_class_name,
         cache_backend=args.cache_backend,

examples/offline_inference/x_to_video_audio/x_to_video_audio.md

@@ -1,6 +1,25 @@
 # X-To-Video-Audio
 
-The `DreamID-Omni` pipeline generates short videos from text, image and video.
+The shared runner generates video with optional audio from text, image, and audio
+inputs. It currently documents DreamID-Omni and MagiHuman, and can be extended
+through `vllm_omni.model_extras` without adding another model-specific runner.
+
+## Shared request and output contract
+
+The example builds one canonical request envelope: `prompt` is text,
+`modalities` is `["video"]`, and optional `multi_modal_data` contains RGB PIL
+images plus audio `(waveform, sample_rate)` tuples. Audio files are loaded in
+full by design; the runner does not apply a time window. Prepare a trimmed input
+file explicitly when only part of a recording should condition the model.
+DreamID-Omni's audio conditioning sequence grows with the complete input
+duration, so long recordings increase denoising memory and step time.
+
+Model-specific request translation (for example, DreamID-Omni's two negative
+prompts and its 16 kHz input requirement) is declared in `model_extras`. Model
+outputs must provide `fps` and, when audio is present, `audio_sample_rate`.
+`--fps` and `--audio-sample-rate` are available as explicit metadata overrides.
+The shared encoder normalizes supported video tensor layouts, so this runner
+does not branch on a model name or assume a model-specific frame layout.
 
 ## Local CLI Usage
 ### Download the Model locally
@@ -147,7 +166,8 @@ in `vllm_omni/model_extras/magi_human.py`, routed via `extra_args`):
 Key arguments:
 - `--prompt`: text description (string).
 - `--model`: path to the model local directory.
-- `--height/--width`: output resolution (defaults 704 * 1024).
+- `--model-class-name`: optional diffusion pipeline class override.
+- `--height/--width`: output resolution (defaults 704 * 1280).
 - `--image-path`: path to the input image list.
 - `--audio-path`: path to the input audio list, indicate the timbre of the output video.
 - `--cfg-parallel-size`: number of parallel cfg parallel (defaults 1).
@@ -157,6 +177,9 @@ Key arguments:
 - `--num-inference-steps`: number of denoising steps (defaults 45).
 - `--video-negative-prompt`: negative prompt for video generation.
 - `--audio-negative-prompt`: negative prompt for audio generation.
+- `--extra-body`: JSON object containing parameters declared by the selected model.
+- `--fps`: explicit output FPS override when a model does not return that metadata.
+- `--audio-sample-rate`: explicit output audio sample-rate override.
 - `--enable-cpu-offload`: enable CPU offload (defaults False).
 - `--cache-backend`: enable `cache_dit` for acceleration.
 - `--quantization`: online (dynamic) quantization method — `fp8` or `int8`.

examples/offline_inference/x_to_video_audio/x_to_video_audio.py

@@ -3,31 +3,97 @@
 
 import argparse
 import json
+import math
 import re
 import time
+from collections.abc import Mapping
+from dataclasses import dataclass
+from pathlib import Path
 from typing import Any
 
 import numpy as np
+import torch
 from PIL import Image
 from vllm.multimodal.media.audio import load_audio
 
-from vllm_omni.diffusion.utils.media_utils import mux_video_audio_bytes
+from vllm_omni.diffusion.data import DiffusionParallelConfig
+from vllm_omni.diffusion.utils.param_utils import apply_declared_extra_args
 from vllm_omni.entrypoints.omni import Omni
+from vllm_omni.entrypoints.openai.video_api_utils import encode_video_bytes
 from vllm_omni.inputs.data import OmniDiffusionSamplingParams
+from vllm_omni.model_extras import (
+    build_x_to_video_audio_prompt as build_model_x_to_video_audio_prompt,
+)
+from vllm_omni.model_extras import (
+    get_extra_body_params,
+    get_input_audio_sample_rate,
+    get_model_class_name,
+    get_output_tensor_range,
+)
 
 IMAGE_EXTENSIONS = {".jpg", ".jpeg", ".png", ".bmp", ".webp", ".gif"}
 AUDIO_EXTENSIONS = {".wav", ".mp3", ".flac", ".m4a", ".aac", ".ogg"}
 
 
+@dataclass(frozen=True)
+class XToVideoAudioOutput:
+    """Canonical generated video+audio payload and its required metadata."""
+
+    video: Any
+    audio: Any | None
+    fps: float
+    audio_sample_rate: int | None
+    output_tensor_range: str = "negative_one_to_one"
+
+
+def parse_json_object(value: str) -> dict[str, Any]:
+    try:
+        parsed = json.loads(value)
+    except json.JSONDecodeError as exc:
+        raise argparse.ArgumentTypeError(f"--extra-body must be valid JSON: {exc}") from exc
+    if not isinstance(parsed, dict):
+        raise argparse.ArgumentTypeError("--extra-body must be a JSON object")
+    return parsed
+
+
+def build_x_to_video_audio_prompt(
+    prompt: str,
+    media_inputs: dict[str, Any] | None = None,
+) -> dict[str, Any]:
+    """Build the canonical request envelope for image/audio-to-video models."""
+    result: dict[str, Any] = {"prompt": prompt, "modalities": ["video"]}
+    if media_inputs:
+        result["multi_modal_data"] = media_inputs
+    return result
+
+
+def _clean_official_prompt(prompt: str) -> str:
+    """Remove metadata tags used by the official DreamID prompt fixtures."""
+    prompt = re.sub(
+        r"\[SPEAKER_TIMESTAMPS_START\].*?\[SPEAKER_TIMESTAMPS_END\]",
+        "",
+        prompt,
+        flags=re.DOTALL,
+    ).strip()
+    prompt = re.sub(
+        r"\[AUDIO_DESCRIPTION_START].*?\[AUDIO_DESCRIPTION_END]",
+        "",
+        prompt,
+        flags=re.DOTALL,
+    ).strip()
+    prompt = re.sub(r"\[[A-Z_]+\]", "", prompt)
+    return re.sub(r"\n\s*\n", "\n", prompt).strip()
+
+
 def parse_args() -> argparse.Namespace:
     parser = argparse.ArgumentParser(description="Offline inference for X -> video+audio models.")
     parser.add_argument("--model", required=True, help="Model ckpt root directory.")
     parser.add_argument(
         "--model-type",
-        default="dreamid-omni",
-        choices=["dreamid-omni", "magi-human"],
-        help="Model type.",
+        default=None,
+        help="Optional legacy model type override (for example, magi-human).",
     )
+    parser.add_argument("--model-class-name", default=None, help="Optional diffusion pipeline class override.")
     parser.add_argument("--prompt", default=None, help="Text prompt.")
 
     parser.add_argument("--image-path", type=str, nargs="+", help="list of image-path")
@@ -35,9 +101,9 @@ def parse_args() -> argparse.Namespace:
     parser.add_argument("--prompt-file", type=str, default=None, help="Text prompt in json format.")
     parser.add_argument(
         "--extra-body",
-        type=str,
+        type=parse_json_object,
         default=None,
-        help="[magi-human] JSON dict of model-specific extra params (declared in vllm_omni/model_extras/), "
+        help="JSON dict of model-specific extra params (declared in vllm_omni/model_extras/), "
         'merged into sampling extra_args. Example: \'{"image_path": "/path/to/img.jpg", "seconds": 5}\'.',
     )
 
@@ -70,13 +136,20 @@ def parse_args() -> argparse.Namespace:
         default="robotic, muffled, echo, distorted",
         help="Negative prompt for audio.",
     )
+    parser.add_argument("--fps", type=float, default=None, help="Override output FPS metadata.")
+    parser.add_argument(
+        "--audio-sample-rate",
+        type=int,
+        default=None,
+        help="Override output audio sample-rate metadata.",
+    )
     parser.add_argument("--output", default="dreamid_output.mp4", help="Output video path.")
     parser.add_argument(
         "--quantization",
         type=str,
         default=None,
         choices=["fp8", "int8"],
-        help=("Online (dynamic) quantization method for the DreamID-Omni transformer. "),
+        help="Online (dynamic) quantization method for the model transformer.",
     )
     parser.add_argument(
         "--enable-cpu-offload",
@@ -93,7 +166,7 @@ def parse_args() -> argparse.Namespace:
     parser.add_argument(
         "--use-hsdp",
         action="store_true",
-        help="Enable HSDP for DreamID-Omni fused transformer blocks.",
+        help="Enable HSDP for supported transformer blocks.",
     )
     parser.add_argument(
         "--hsdp-shard-size",
@@ -110,20 +183,105 @@ def parse_args() -> argparse.Namespace:
     return parser.parse_args()
 
 
-def load_image_and_audio(image_paths, audio_paths):
-    image = []
-    audio = []
+def load_image_and_audio(
+    image_paths: list[str] | None,
+    audio_paths: list[str] | None,
+    *,
+    audio_sample_rate: int | None = None,
+) -> tuple[list[Image.Image], list[tuple[Any, int]]]:
+    """Load complete media inputs into the canonical multimodal representation."""
+    images = []
+    audios = []
 
-    for path in image_paths:
+    for path in image_paths or []:
         with Image.open(path) as img:
-            img = img.convert("RGB")
-            image.append(img)
+            images.append(img.convert("RGB"))
 
-    for path in audio_paths:
-        audio_array, sr = load_audio(path, sr=16000)
-        audio_array = audio_array[int(sr * 1) : int(sr * 3)]
-        audio.append(audio_array)
-    return image, audio
+    for path in audio_paths or []:
+        waveform, sample_rate = load_audio(path, sr=audio_sample_rate)
+        audios.append((waveform, sample_rate))
+    return images, audios
+
+
+def extract_x_to_video_audio_output(
+    outputs: Any,
+    *,
+    fps: float | None = None,
+    audio_sample_rate: int | None = None,
+    output_tensor_range: str = "negative_one_to_one",
+) -> XToVideoAudioOutput:
+    """Normalize engine output without depending on a model's tensor layout."""
+    if isinstance(outputs, list):
+        if not outputs:
+            raise RuntimeError("No output returned from the model.")
+        result = outputs[0]
+    else:
+        result = outputs
+    if result is None:
+        raise RuntimeError("No output returned from the model.")
+
+    multimodal_output = getattr(result, "multimodal_output", None) or {}
+    if not isinstance(multimodal_output, Mapping):
+        raise RuntimeError("Model multimodal output must be a mapping.")
+    images = getattr(result, "images", None)
+    video = images[0] if images else multimodal_output.get("video")
+    if video is None:
+        raise RuntimeError("No video payload found in model output.")
+
+    resolved_fps = fps if fps is not None else multimodal_output.get("fps")
+    try:
+        resolved_fps = float(resolved_fps)
+    except (TypeError, ValueError) as exc:
+        raise RuntimeError("Model output must declare a valid 'fps', or pass --fps.") from exc
+    if not math.isfinite(resolved_fps) or resolved_fps <= 0:
+        raise RuntimeError("Output FPS must be finite and positive.")
+
+    audio = multimodal_output.get("audio")
+    resolved_sample_rate = audio_sample_rate
+    if resolved_sample_rate is None:
+        resolved_sample_rate = multimodal_output.get("audio_sample_rate")
+    if audio is not None:
+        try:
+            resolved_sample_rate = int(resolved_sample_rate)
+        except (TypeError, ValueError) as exc:
+            raise RuntimeError(
+                "Model output with audio must declare a valid 'audio_sample_rate', or pass --audio-sample-rate."
+            ) from exc
+        if resolved_sample_rate <= 0:
+            raise RuntimeError("Output audio sample rate must be positive.")
+    elif resolved_sample_rate is not None:
+        resolved_sample_rate = int(resolved_sample_rate)
+
+    return XToVideoAudioOutput(video, audio, resolved_fps, resolved_sample_rate, output_tensor_range)
+
+
+def _normalize_output_tensor_range(video: Any, source_range: str) -> Any:
+    """Normalize floating-point output according to the pipeline contract."""
+    if isinstance(video, torch.Tensor):
+        if not video.is_floating_point():
+            return video
+        video = video.detach().cpu().float().numpy()
+    if isinstance(video, np.ndarray):
+        if not np.issubdtype(video.dtype, np.floating):
+            return video
+        if source_range == "negative_one_to_one":
+            return np.clip(video, -1.0, 1.0) * 0.5 + 0.5
+        if source_range == "zero_to_one":
+            return np.clip(video, 0.0, 1.0)
+        raise ValueError(f"Unsupported floating-point tensor range: {source_range!r}")
+    if isinstance(video, list):
+        return [_normalize_output_tensor_range(frame, source_range) for frame in video]
+    return video
+
+
+def encode_x_to_video_audio_output(output: XToVideoAudioOutput) -> bytes:
+    """Encode any video layout supported by the shared video API utility."""
+    return encode_video_bytes(
+        _normalize_output_tensor_range(output.video, output.output_tensor_range),
+        output.fps,
+        output.audio,
+        output.audio_sample_rate,
+    )
 
 
 def _extract_peak_memory_mb(result: Any) -> float:
@@ -156,45 +314,18 @@ def main() -> None:
     if args.prompt_file:
         with open(args.prompt_file) as f:
             text_prompt = json.load(f)
-            text_prompt = re.sub(
-                r"\[SPEAKER_TIMESTAMPS_START\].*?\[SPEAKER_TIMESTAMPS_END\]", "", text_prompt, flags=re.DOTALL
-            ).strip()
-            text_prompt = re.sub(
-                r"\[AUDIO_DESCRIPTION_START].*?\[AUDIO_DESCRIPTION_END]", "", text_prompt, flags=re.DOTALL
-            ).strip()
-            text_prompt = re.sub(r"\[[A-Z_]+\]", "", text_prompt)
-            text_prompt = re.sub(r"\n\s*\n", "\n", text_prompt).strip()
+            if isinstance(text_prompt, str):
+                text_prompt = _clean_official_prompt(text_prompt)
+    if not isinstance(text_prompt, str):
+        raise ValueError("Prompt content must be a JSON string or --prompt text.")
 
-    if args.model_type == "magi-human":
-        prompt = text_prompt
-        extra_args = json.loads(args.extra_body) if args.extra_body else {}
-        sampling_params = OmniDiffusionSamplingParams(
-            height=args.height,
-            width=args.width,
-            num_inference_steps=args.num_inference_steps,
-            seed=args.seed,
-            extra_args=extra_args,
-        )
-    else:
-        image, audio = load_image_and_audio(args.image_path, args.audio_path)
-
-        prompt = {
-            "prompt": text_prompt,
-            "video_negative_prompt": args.video_negative_prompt,
-            "audio_negative_prompt": args.audio_negative_prompt,
-            "multi_modal_data": {"image": image, "audio": audio},
-        }
-
-        sampling_params = OmniDiffusionSamplingParams(
-            height=args.height,
-            width=args.width,
-            num_inference_steps=args.num_inference_steps,
-            seed=args.seed,
-            extra_args={
-                "solver_name": args.solver_name,
-                "shift": args.shift,
-            },
-        )
+    parallel_config = DiffusionParallelConfig(
+        cfg_parallel_size=args.cfg_parallel_size,
+        tensor_parallel_size=args.tensor_parallel_size,
+        use_hsdp=args.use_hsdp,
+        hsdp_shard_size=args.hsdp_shard_size,
+        hsdp_replicate_size=args.hsdp_replicate_size,
+    )
 
     cache_config = None
     if args.cache_backend == "cache_dit":
@@ -211,68 +342,83 @@ def main() -> None:
             "scm_steps_policy": "dynamic",
         }
 
-    omni_kwargs = dict(
+    omni_kwargs: dict[str, Any] = dict(
         model=args.model,
-        cfg_parallel_size=args.cfg_parallel_size,
-        tensor_parallel_size=args.tensor_parallel_size,
-        use_hsdp=args.use_hsdp,
-        hsdp_shard_size=args.hsdp_shard_size,
-        hsdp_replicate_size=args.hsdp_replicate_size,
-        model_type=args.model_type,
+        parallel_config=parallel_config,
         enable_cpu_offload=args.enable_cpu_offload,
         enable_layerwise_offload=args.enable_layerwise_offload,
         cache_backend=args.cache_backend,
         cache_config=cache_config,
     )
+    if args.model_type is not None:
+        omni_kwargs["model_type"] = args.model_type
+    if args.model_class_name is not None:
+        omni_kwargs["model_class_name"] = args.model_class_name
     if args.quantization is not None:
         omni_kwargs["quantization"] = args.quantization
     omni = Omni(**omni_kwargs)
-    start = time.perf_counter()
-    outputs = omni.generate(prompt, sampling_params)
-    elapsed = time.perf_counter() - start
+    try:
+        model_class_name = args.model_class_name or get_model_class_name(omni)
+        input_sample_rate = get_input_audio_sample_rate(model_class_name)
+        images, audios = load_image_and_audio(
+            args.image_path,
+            args.audio_path,
+            audio_sample_rate=input_sample_rate,
+        )
+        media_inputs: dict[str, Any] = {}
+        if images:
+            media_inputs["image"] = images
+        if audios:
+            media_inputs["audio"] = audios
 
-    if not outputs:
-        raise RuntimeError("No output returned from the model.")
-    result = outputs[0]
-    if not result.images:
-        raise RuntimeError("No video frames found in model output.")
-    generated_video = result.images[0]
-    mm = result.multimodal_output or {}
-    generated_audio = mm.get("audio")
+        canonical_prompt = build_x_to_video_audio_prompt(text_prompt, media_inputs)
+        prompt = build_model_x_to_video_audio_prompt(
+            model_class_name,
+            canonical_prompt,
+            {
+                "video_negative_prompt": args.video_negative_prompt,
+                "audio_negative_prompt": args.audio_negative_prompt,
+            },
+        )
 
-    if args.model_type == "magi-human":
-        # MagiHuman returns frames already as (F, H, W, C) uint8, ready to mux.
-        fps = float(mm.get("fps", 25))
-        sample_rate = int(mm.get("audio_sample_rate", 24000))
-        frames = generated_video
-    else:
-        fps = float(mm.get("fps", 24))
-        sample_rate = int(mm.get("audio_sample_rate", 16000))
-        # DreamID-Omni returns video as (C, F, H, W) float32 in [-1, 1].
-        # mux_video_audio_bytes expects (F, H, W, C) uint8.
-        if not isinstance(generated_video, np.ndarray) or generated_video.ndim != 4:
-            raise RuntimeError(f"Unexpected video shape: {getattr(generated_video, 'shape', None)}")
-        frames = generated_video.transpose(1, 2, 3, 0)
-        frames = (np.clip((frames + 1.0) / 2.0, 0.0, 1.0) * 255.0).round().astype(np.uint8)
+        sampling_params = OmniDiffusionSamplingParams(
+            height=args.height,
+            width=args.width,
+            num_inference_steps=args.num_inference_steps,
+            seed=args.seed,
+            extra_args={},
+        )
+        declared_extra_body_params = get_extra_body_params(model_class_name)
+        extra_body = dict(args.extra_body or {})
+        legacy_extra_body = {"solver_name": args.solver_name, "shift": args.shift}
+        for key, value in legacy_extra_body.items():
+            if key in declared_extra_body_params:
+                extra_body.setdefault(key, value)
+        if declared_extra_body_params:
+            apply_declared_extra_args(sampling_params, declared_extra_body_params, extra_body)
+        elif extra_body:
+            sampling_params.extra_args.update({key: value for key, value in extra_body.items() if value is not None})
 
-    audio_np = None
-    if generated_audio is not None:
-        audio_np = np.squeeze(np.asarray(generated_audio)).astype(np.float32)
+        start = time.perf_counter()
+        outputs = omni.generate(prompt, sampling_params)
+        elapsed = time.perf_counter() - start
+        output = extract_x_to_video_audio_output(
+            outputs,
+            fps=args.fps,
+            audio_sample_rate=args.audio_sample_rate,
+            output_tensor_range=get_output_tensor_range(model_class_name),
+        )
 
-    output_path = args.output
-    video_bytes = mux_video_audio_bytes(
-        frames,
-        audio_np,
-        fps=float(fps),
-        audio_sample_rate=sample_rate,
-    )
-    with open(output_path, "wb") as f:
-        f.write(video_bytes)
-    print(f"Saved generated video to {output_path}")
-    print(f"Total time: {elapsed:.2f}s")
-    peak_mb = _extract_peak_memory_mb(outputs)
-    if peak_mb:
-        print(f"Worker peak GPU memory (reserved): {peak_mb:.2f} MiB ({peak_mb / 1024:.2f} GiB)")
+        output_path = Path(args.output)
+        output_path.parent.mkdir(parents=True, exist_ok=True)
+        output_path.write_bytes(encode_x_to_video_audio_output(output))
+        print(f"Saved generated video to {output_path}")
+        print(f"Total time: {elapsed:.2f}s")
+        peak_mb = _extract_peak_memory_mb(outputs)
+        if peak_mb:
+            print(f"Worker peak GPU memory (reserved): {peak_mb:.2f} MiB ({peak_mb / 1024:.2f} GiB)")
+    finally:
+        omni.close()
 
 
 if __name__ == "__main__":

examples/online_serving/pi0/README.md

@@ -1,104 +0,0 @@
-# π0 (Pi-Zero) VLA — OpenPI realtime serving
-
-[π0](https://www.physicalintelligence.company/blog/pi0) is a Vision-Language-Action
-model from Physical Intelligence: multi-camera images + a language instruction +
-robot proprioceptive state → a continuous action chunk via flow-matching denoising
-(it does **not** emit text tokens). This example serves π0 over the OpenPI realtime
-websocket protocol at `/v1/realtime/robot/openpi`.
-
-## Install extras
-
-The core `pip install -e .` does not include the OpenPI client used here:
-
-- `openpi-client` (from the [openpi](https://github.com/Physical-Intelligence/openpi)
-  repo: `pip install -e packages/openpi-client`), `websockets`, `msgpack`, `msgpack-numpy`
-
-## Weights
-
-π0 uses the LeRobot `lerobot/pi0_base` checkpoint (HF, ~13 GB). Either let the server
-download it (`MODEL=lerobot/pi0_base`) or point at a local copy
-(`MODEL=/path/to/pi0_base`).
-
-## Run the server
-
-```bash
-vllm serve lerobot/pi0_base --omni --port 8000 \
-    --served-model-name pi0 \
-    --deploy-config vllm_omni/deploy/pi0.yaml \
-    --enforce-eager --disable-log-stats
-```
-
-The deploy config (`vllm_omni/deploy/pi0.yaml`) declares a single diffusion stage
-(`Pi0Pipeline`), float32, `max_num_seqs: 1`, and the `policy_server_config` handshake
-metadata (3 cameras, `joint_position`, action horizon 50, action dim 32).
-
-### Commonly adjusted `model_config` keys
-
-These live under `stages[0].model_config` in `vllm_omni/deploy/pi0.yaml`:
-
-| Key | Default | Meaning |
-|---|---|---|
-| `chunk_size` | `50` | Action-chunk length (timesteps) the model predicts per inference. |
-| `num_inference_steps` | `10` | Flow-matching Euler denoising steps. |
-| `max_action_dim` | `32` | Action dimensionality (state/action are padded to this). |
-| `max_state_dim` | `32` | Proprioceptive-state dimensionality (zero-padded to this). |
-| `image_resolution` | `[224, 224]` | Per-camera input size (square; SigLIP). |
-| `tokenizer_max_length` | `48` | Max PaliGemma prompt tokens. |
-| `max_cameras` | `3` | Camera slots the model attends to (real + `-1`-padded). |
-| `image_feature_keys` | 3 `observation.images.*` keys | Camera order the model attends to. |
-| `image_key_map` | `{}` | Map raw obs camera keys → `image_feature_keys` (empty = verbatim). |
-
-The `policy_server_config` block below them is the OpenPI handshake metadata
-advertised to the client; keep its `action_horizon` / `action_dim` /
-`image_resolution` in sync with the `model_config` values above. The e2e test
-`tests/e2e/online_serving/test_pi0_expansion.py::test_pi0_openpi_online` connects to a
-live server and asserts the advertised metadata matches these `pi0.yaml` values.
-
-## Run the client
-
-```bash
-python examples/online_serving/pi0/openpi_client.py --host 127.0.0.1 --port 8000 \
-    --prompt "pick up the red block and place it in the bin"
-```
-
-It connects, prints the server metadata, sends robot observations, and prints the
-returned `[action_horizon, action_dim] = [50, 32]` action chunks. Replace the blank
-cameras / zero state in `_make_dummy_obs` with real frames (HWC uint8) and
-proprioceptive state to drive a robot.
-
-### Observation format
-
-The client sends a flat dict per inference:
-
-```python
-{
-    "observation.images.base_0_rgb":       np.uint8[H, W, 3],
-    "observation.images.left_wrist_0_rgb": np.uint8[H, W, 3],
-    "observation.images.right_wrist_0_rgb":np.uint8[H, W, 3],
-    "state":   np.float32[state_dim],   # zero-padded to max_state_dim=32 server-side
-    "prompt":  "pick up the red block",
-    "session_id": "<uuid>",             # accepted but ignored (π0 is stateless)
-}
-```
-
-Camera keys must match the server's `image_feature_keys` (the checkpoint's
-`input_features` order). If your robot uses different camera names, set
-`model_config.image_key_map` in `pi0.yaml` to map raw obs keys → feature keys.
-
-## Correctness
-
-π0's flow-matching kernel is bit-for-bit matched to the LeRobot `PI0Policy` reference
-(`max|Δ| = 7.15e-07`, CPU/float32, fixed noise; see
-`tests/diffusion/models/pi0/test_pi0_parity.py::test_pi0_vllm_omni_vs_lerobot`) and is
-version-stable across transformers releases (the version-stability checks in
-`tests/diffusion/models/pi0/test_pi0_units.py`). An OpenPI websocket online-serving e2e
-lives in `tests/e2e/online_serving/test_pi0_expansion.py::test_pi0_openpi_online`.
-
-## Limitations
-
-- **Normalization stats**: `lerobot/pi0_base` uses identity normalization, which is
-  fully supported (state passes through, actions are returned in the model's space).
-  Per-dataset `norm_stats` declared in a checkpoint's `config.json` are honored, but
-  stats stored only in LeRobot's `policy_preprocessor.json` companion safetensors are
-  not yet loaded — a fine-tuned checkpoint that relies on those would need that bridge
-  added before its actions are in real-world units.

examples/online_serving/pi0/openpi_client.py

@@ -1,119 +0,0 @@
-#!/usr/bin/env python3
-# SPDX-License-Identifier: Apache-2.0
-# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
-"""Minimal OpenPI client for the π0 (Pi-Zero) VLA policy server.
-
-Connects to a running π0 server (see README.md), sends a single robot
-observation (3 cameras + state + language instruction), and prints the returned
-continuous action chunk ``[action_horizon, action_dim]``.
-
-The server speaks the OpenPI websocket protocol:
-    connect -> server sends msgpack(policy_server_config metadata)
-    infer   -> client sends msgpack(obs), server replies msgpack(actions ndarray)
-
-Usage:
-    python openpi_client.py --host 127.0.0.1 --port 8000 \
-        --prompt "pick up the red block"
-"""
-
-from __future__ import annotations
-
-import argparse
-import logging
-import sys
-import uuid
-from pathlib import Path
-from typing import Any
-
-import numpy as np
-
-try:
-    # Avoid shadowing the installed ``openpi_client`` package with this dir.
-    example_dir = str(Path(__file__).resolve().parent)
-    removed = sys.path and sys.path[0] == example_dir
-    if removed:
-        sys.path.pop(0)
-    try:
-        from openpi_client.websocket_client_policy import WebsocketClientPolicy
-    finally:
-        if removed:
-            sys.path.insert(0, example_dir)
-except ImportError as exc:  # pragma: no cover - runtime dependency guard
-    raise ImportError("π0 OpenPI example requires `openpi-client`.") from exc
-
-logger = logging.getLogger("pi0_openpi_client")
-
-DEFAULT_HOST = "127.0.0.1"
-DEFAULT_PORT = 8000
-DEFAULT_PATH = "/v1/realtime/robot/openpi"
-DEFAULT_PROMPT = "pick up the red block and place it in the bin"
-
-# π0 / pi0_base camera identities (must match the server's image_feature_keys).
-CAMERA_KEYS = (
-    "observation.images.base_0_rgb",
-    "observation.images.left_wrist_0_rgb",
-    "observation.images.right_wrist_0_rgb",
-)
-STATE_DIM = 32
-
-
-def _make_dummy_obs(prompt: str, session_id: str, image_size: int = 224) -> dict[str, Any]:
-    """Build a single π0 observation with blank cameras + zero state.
-
-    Replace the zeros with real camera frames (HWC uint8) and proprioceptive
-    state to drive an actual robot.
-    """
-    obs: dict[str, Any] = {cam: np.zeros((image_size, image_size, 3), dtype=np.uint8) for cam in CAMERA_KEYS}
-    obs["state"] = np.zeros(STATE_DIM, dtype=np.float32)
-    obs["prompt"] = prompt
-    obs["session_id"] = session_id
-    return obs
-
-
-def _ws_uri(host: str, port: int, path: str) -> str:
-    """Build the full websocket URI. WebsocketClientPolicy takes the whole URI as
-    ``host`` (it has no separate path arg)."""
-    if host.startswith(("ws://", "wss://")):
-        return host
-    return f"ws://{host}:{port}{path}"
-
-
-def _as_action_array(response: Any) -> np.ndarray:
-    if isinstance(response, dict) and response.get("type") == "error":
-        raise RuntimeError(f"Inference failed: {response.get('message', response)}")
-    return np.asarray(response, dtype=np.float32)
-
-
-def main() -> int:
-    parser = argparse.ArgumentParser(description="π0 OpenPI client")
-    parser.add_argument("--host", default=DEFAULT_HOST)
-    parser.add_argument("--port", type=int, default=DEFAULT_PORT)
-    parser.add_argument("--path", default=DEFAULT_PATH)
-    parser.add_argument("--prompt", default=DEFAULT_PROMPT)
-    parser.add_argument("--num-steps", type=int, default=2, help="Inference observations to send.")
-    args = parser.parse_args()
-    logging.basicConfig(level=logging.INFO, format="%(message)s")
-
-    policy = WebsocketClientPolicy(host=_ws_uri(args.host, args.port, args.path))
-    metadata = dict(policy.get_server_metadata())
-    logger.info("Server metadata: %s", metadata)
-
-    session_id = str(uuid.uuid4())
-    for step in range(args.num_steps):
-        obs = _make_dummy_obs(args.prompt, session_id)
-        actions = _as_action_array(policy.infer(obs))
-        logger.info(
-            "[step %d] actions shape=%s mean=%.4f std=%.4f",
-            step,
-            actions.shape,
-            float(actions.mean()),
-            float(actions.std()),
-        )
-        if not np.isfinite(actions).all():
-            raise RuntimeError("Server returned non-finite actions.")
-    logger.info("Done — π0 OpenPI round-trip OK.")
-    return 0
-
-
-if __name__ == "__main__":
-    raise SystemExit(main())

mkdocs.yml

@@ -79,7 +79,6 @@ plugins:
         user_guide/diffusion/step_execution.md: user_guide/diffusion/execution_modes.md
         user_guide/diffusion/cpu_offload_diffusion.md: user_guide/diffusion/cpu_offload.md
         user_guide/quantization/index.md: user_guide/quantization/overview.md
-        configuration/pd_disaggregation.md: features/pd_disaggregation.md
         design/feature/diffusion_request_level_batching.md: design/feature/diffusion_continuous_batching.md
         design/feature/diffusion_step_execution.md: design/feature/diffusion_continuous_batching.md
         design/feature/cpu_offloading.md: design/feature/offloader/README.md

recipes/MammothModa2/MammothModa2.md

@@ -81,7 +81,7 @@ MammothModa2 generation parameters as a JSON object through `--extra-body`:
 ```bash
 python examples/offline_inference/text_to_image/text_to_image.py \
   --model ./MammothModa2-Preview \
-  --deploy-config vllm_omni/deploy/mammoth_moda2.yaml \
+  --stage-configs-path vllm_omni/deploy/mammoth_moda2.yaml \
   --prompt "A stylish woman riding a motorcycle in NYC, movie poster style" \
   --height 1024 \
   --width 1024 \
@@ -132,44 +132,6 @@ ls -lh mammoth_t2i.png
 python -c "from PIL import Image; print(Image.open('mammoth_t2i.png').size)"
 ```
 
-### 1x AMD MI300X, MammothModa2 Preview
-
-#### Environment
-
-- OS: Linux 6.8.0-134-generic, x86_64
-- Container: official ROCm image built from `docker/Dockerfile.rocm`
-- Python: 3.12.13
-- PyTorch: 2.11.0+gitd0c8b1f
-- Driver / runtime: AMD 6.19.14.31400000 / ROCm 7.2.53211
-- GPU: one AMD Instinct MI300X, `gfx942:sramecc+:xnack-`, 191.69 GiB visible HBM
-- vLLM version: 0.27.0+rocm723
-- vLLM Omni version or commit: `73e1368c7bb940efe1a025859c9d6c8eeeb2e3f0`
-- Installed vLLM Omni package metadata: `0.27.0rc2.dev44+g55abdade9.rocm`
-
-#### Offline Commands
-
-The checked run used the committed stage split, with `gpu_memory_utilization` set to 0.5 for AR and 0.3 for DiT:
-
-```bash
-python3 examples/offline_inference/text_to_image/text_to_image.py \
-    --model bytedance-research/MammothModa2-Preview \
-    --deploy-config vllm_omni/deploy/mammoth_moda2.yaml \
-    --prompt "A stylish woman riding a motorcycle in NYC, movie poster style" \
-    --height 1024 \
-    --width 1024 \
-    --seed 42 \
-    --extra-body '{"text_guidance_scale": 4.0, "cfg_range": [0.0, 1.0], "num_inference_steps": 50}' \
-    --enable-diffusion-pipeline-profiler \
-    --log-stats \
-    --output mammoth_t2i.png
-```
-
-#### Verification
-
-The first request took 85.224 seconds. The AR stage generated 4,161 visual tokens in 72.996 seconds, and the DiT stage took 12.163 seconds. AR weight loading used 21.4 GiB and took 8.250 seconds. DiT weight loading used 5.49 GiB and took 1.824 seconds. The largest one second whole device memory sample was 106.57 GiB, including the AR KV cache reserved by the 0.5 memory setting.
-
-The output was a valid 1024 by 1024 RGB PNG.
-
 ## MammothModa2-Dev unified inference
 
 MammothModa2-Dev uses a Qwen3-VL AR backbone, while MammothModa2-Preview uses

recipes/MiniMaxAI/MiniMax-H3.md

@@ -416,10 +416,9 @@ For example, keep the first main block's attention projections in BF16 with:
 ```
 
 The structured option replaces `--quantization fp8`. Online FP8 can be used
-with H3 layerwise offload and with both DLO transfer paths. The default
-AllGather path uses the ordinary loader to finalize FP8 weights and scales
-before sharding them across ranks. `--dlo-no-use-allgather` instead retains
-complete rank-local tensors and avoids the synchronized request-wave contract.
+with H3 layerwise offload and with distributed layerwise offload's full-weight
+per-rank path (`--dlo-no-use-allgather`). The sharded DLO AllGather path is not
+supported for runtime-created FP8 weights.
 
 ## AMD ROCm (gfx942 / gfx950)
 
@@ -871,9 +870,8 @@ vllm serve "${MODEL_ROOT}/FL2VA" \
   H3 native `tile` mode only.
 - A U2 x Ring2 hybrid currently fails with an attention-mask length mismatch; use
   pure Ulysses.
-- Online FP8 with DLO AllGather temporarily materializes the complete FP8 model
-  in host memory on every rank during startup before retaining only each rank's
-  shard. Size startup host memory for that transient peak.
+- Runtime-created FP8 weights do not support the sharded DLO AllGather path;
+  use `--dlo-no-use-allgather` when combining online FP8 with DLO.
 - TeaCache and Cache-DiT cannot be enabled on the same server.
 - Image+audio Ref2VA accepts exactly one image and one audio reference.
 - Video Ref2VA accepts one or more video files, but not an additional standalone

recipes/MiniMaxAI/MiniMax-Music3.md

@@ -1,173 +0,0 @@
-# MiniMax Music 3
-
-> Text-to-music: lyrics plus a style caption in, a 32 kHz stereo song out
-
-## Summary
-
-- Vendor: MiniMax
-- Model: `MiniMaxAI/MiniMax-Music3`
-- Task: Text-to-music generation (vocals and instrumental together)
-- Mode: Online serving through the OpenAI-compatible `/v1/audio/speech` API
-- Maintainer: @linyueqian
-
-## When to use this recipe
-
-Use it to serve `MiniMaxAI/MiniMax-Music3` for full-song generation up to six
-minutes. This is not a TTS model with a musical voice: there is no reference
-speaker, no `voice`, and no `temperature`. What you control is the lyrics, the
-caption, the seed, and the length.
-
-## References
-
-- Model card: [MiniMaxAI/MiniMax-Music3](https://huggingface.co/MiniMaxAI/MiniMax-Music3)
-- Demo samples: [MiniMax Music 3 demo](https://minimax-ai.github.io/music3-demo/)
-- Deploy config: `vllm_omni/deploy/minimax_music3.yaml` (single GPU),
-  `vllm_omni/deploy/minimax_music3_2gpu.yaml` (split across two)
-
-## Architecture
-
-| Component | Spec |
-|---|---|
-| Backbone | Qwen3 decoder, 36 layers, hidden 4096, GQA 32/8, vocab 200k |
-| Depth decoder | 4 layers, hidden 4096, 16 heads; 7 residual codebooks of 1024 |
-| Frame | 8 codebooks; `c0` in the backbone vocabulary, `c1..c7` in the depth decoder |
-| AR guidance | Classifier-free, scale 1.5, `c0` masked to the conditioned branch's top 50 |
-| Acoustic stage | Flow-matching transformer (36 layers, dim 2048) plus a DAC-style decoder, 512x upsampling |
-| Solver | Euler, 30 steps, acoustic guidance scale 1.7 |
-| Window | 200 frames per acoustic chunk, 100-frame hop |
-| Frame rate | 25 frames per second |
-| Context length | 10,240 tokens |
-| Output | 32 kHz stereo WAV |
-
-Generation is two-stage. The backbone predicts one audio frame per decode step,
-each frame eight RVQ codebooks deep. Every 200 frames are handed to the
-flow-matching transformer, which solves a latent that the decoder turns into
-waveform.
-
-**Guidance doubles the rows.** Every request decodes twice: one row sees the
-real prompt, the other sees the same prompt with the caption-and-lyrics span
-replaced by `<|audio_cfg|>`. Both rows hold their own KV cache for the whole
-song, so a request costs twice the KV its length suggests. Size the pool for
-rows, not requests.
-
-## Hardware Support
-
-## GPU
-
-### 1x H200 141GB
-
-#### Environment
-
-- OS: Linux
-- Python: 3.12
-- CUDA: 12.9
-- vLLM version: 0.27.0
-- vLLM-Omni version or commit: branch `feat/minimax-music3`
-
-#### Command
-
-```bash
-vllm serve MiniMaxAI/MiniMax-Music3 \
-    --host 0.0.0.0 --port 8000 \
-    --trust-remote-code --omni
-```
-
-The default deploy config `vllm_omni/deploy/minimax_music3.yaml` is discovered
-automatically from the checkpoint's `model_type=minimax_music3`. It colocates
-both stages on device 0.
-
-#### Verification
-
-```bash
-curl -X POST http://localhost:8000/v1/audio/speech \
-  -H "Content-Type: application/json" \
-  -d '{
-    "model": "MiniMaxAI/MiniMax-Music3",
-    "input": "[Verse]\nWalking down the empty street at midnight\nStreetlights flicker like a broken dream\n[Chorus]\nAnd I keep on walking\nTill the morning finds me",
-    "instructions": "A melancholic lo-fi hip-hop track at 85 BPM in F minor: mellow Rhodes piano riff, soft vinyl crackle, dusty boom-bap drums with a laid-back swing, warm upright bass. Intimate bedroom production, gentle tape saturation, no bright cymbals.",
-    "seed": 1,
-    "max_new_tokens": 750
-  }' \
-  --output song.wav
-```
-
-Expected: a 30 second 32 kHz stereo WAV. `max_new_tokens` counts audio frames
-at 25 per second, so 750 frames is at most 30 seconds. It is a cap, not a
-target: the model ends the song itself when it emits the audio-end token, and a
-response shorter than the cap is the model finishing rather than a truncation.
-
-#### Notes
-
-- **Memory:** stage 0 takes `gpu_memory_utilization: 0.6` for the bf16 backbone
-  and its KV pool; stage 1 takes 0.2 for the transformer and decoder weights,
-  which sit outside any KV budget.
-- **Key flags:** none required. Guidance, the solver step count and the window
-  geometry are fixed by the checkpoint and are not request parameters.
-- **Known limitations:**
-  - The external API is non-streaming; `stream: true` is rejected.
-  - `max_model_len` is 10,240, so a maximum-length prompt (5,000 tokens) and a
-    maximum-length song (9,000 frames) cannot both fit. Long captions shorten
-    the maximum song. This is a property of the checkpoint.
-  - The acoustic stage runs in float32. bfloat16 is accepted but measurably
-    degrades the solver.
-
-### 2x H200 141GB
-
-Use `minimax_music3_2gpu.yaml` to put the acoustic stage on the second device:
-
-```bash
-vllm serve MiniMaxAI/MiniMax-Music3 \
-    --host 0.0.0.0 --port 8000 \
-    --trust-remote-code --omni \
-    --deploy-config minimax_music3_2gpu.yaml
-```
-
-Only the placement differs; both layouts run the acoustic stage in float32.
-
-## Writing the request
-
-Two fields carry the request, and both are required.
-
-- `input` is the **lyrics**. Structure tags (`[Verse]`, `[Chorus]`, `[Bridge]`,
-  `[Outro]`) steer the arrangement and are part of the prompt contract.
-  **Put a tag on its own line.** Normalization keeps only the tags on a line
-  that starts with one and drops the rest of that line, so lyrics written next
-  to a tag are silently lost:
-
-  ```text
-  "[Verse]\nWalking down the street"   ->   [start] [verse] Walking down the street
-  "[Verse] Walking down the street"    ->   [start] [verse]
-  ```
-
-- `instructions` is the **caption** describing genre, instrumentation, tempo,
-  mood and production. It is the strongest control available: vague captions
-  give generic arrangements. Long structured captions work well, since the
-  model was trained on descriptions covering global attributes, emotional
-  progression, vocal detail and arrangement.
-
-A request is deterministic in its seed: the same lyrics, caption, seed and
-length return the same audio. Changing only the seed gives another take of the
-same song.
-
-## Request parameters
-
-| Parameter | Type | Default | Description |
-|---|---|---|---|
-| `input` | string | required | Lyrics, non-empty. Tags on their own lines |
-| `instructions` | string | required | Caption, non-empty |
-| `seed` | int | `0` | Non-negative. Fixes the output for a given request |
-| `max_new_tokens` | int | `9000` | Cap on audio frames, 25 per second. Maximum 9,000 |
-| `response_format` | string | `"wav"` | Output container |
-
-## Parameters the model rejects
-
-The request is refused rather than silently ignored, so a mistake is visible
-immediately.
-
-| Parameter | Reason |
-|---|---|
-| `temperature`, `top_p`, `top_k`, `repetition_penalty` | Sampling is fixed: guidance at 1.5, then a seeded top-k 50 draw |
-| `voice` | There is no speaker to select; the vocal comes from the caption |
-| `ref_audio`, `ref_text`, `language`, `task_type` | No reference-audio conditioning and no language tag |
-| `speed` | Only `1.0`. Tempo belongs in the caption, e.g. "at 92 BPM" |
-| `stream: true` | The external API is non-streaming |

recipes/NVIDIA/SANA-WM.md

@@ -1,193 +0,0 @@
-# SANA-WM
-
-> Camera-controllable first-frame image-to-video world model.
-
-## Summary
-
-- Vendor: Efficient-Large-Model / NVlabs SANA
-- Model: `BBBBruce/SANA-WM_bidirectional-stage1-diffusers` (standard diffusers layout, converted offline from the NVlabs release; Stage-1 transformer + VAE only)
-- Task: First-frame image-to-video generation with camera control
-- Mode: Online serving with the OpenAI-compatible video API
-- Model weights: about 13 GB for the Stage-1 transformer (10 GB) and VAE (2.3 GB)
-- Local disk: reserve about 40 GB for the Hugging Face cache and runtime artifacts
-- Recommended GPU: 24 GB or larger CUDA GPU
-- Maintainer: Community
-
-## When to use this recipe
-
-Use this recipe when you want to serve SANA-WM through `/v1/videos` or
-`/v1/videos/sync`. The model takes a text prompt, a first-frame image, and
-either an action DSL string or explicit camera poses. vLLM-Omni serves the
-SANA-WM Stage-1 DiT, decoded through the SANA VAE. The optional LTX-2 refiner
-stage is not supported by this integration; it is a planned follow-up.
-
-## References
-
-- Upstream model card: <https://huggingface.co/Efficient-Large-Model/SANA-WM_bidirectional>
-- Video API: [`docs/serving/videos_api.md`](../../docs/serving/videos_api.md)
-
-## Hardware Support
-
-## GPU
-
-### 1x NVIDIA RTX PRO 6000 Blackwell 96GB
-
-#### Capacity
-
-- Model storage: the Stage-1 transformer is about 10 GB and the VAE about
-  2.3 GB. The Gemma text encoder is a separate 4.9 GB repo. The engine prefetches
-  the whole model repo at startup (`allow_patterns=["*"]`), which is why the
-  Stage-1 weights live in their own repo — the two-stage one carries an
-  additional 84 GB `refiner/` that this path never loads.
-- Text encoder: the pipeline tries `google/gemma-2-2b-it` first, then falls back
-  to the ungated mirror `Efficient-Large-Model/gemma-2-2b-it`. The first repo is
-  gated, so without an accepted licence and `hf auth login` you will see one
-  failed load before the fallback succeeds — that is expected, not an error to
-  chase. Set `VLLM_OMNI_SANA_WM_STAGE1_TEXT_ENCODER` to pin a specific repo or a
-  local path and skip both.
-- Disk sizing: provision about 40 GB of local disk or Hugging Face cache volume
-  so the model, temporary downloads, and generated artifacts fit without cache
-  eviction.
-- GPU sizing: the default 1280x704, 161-frame, 60-step serving profile peaks at
-  22.6 GB of device memory and takes about 133 s to generate on one RTX PRO
-  6000 Blackwell. The peak lands in the VAE decode, which is why the pipeline
-  forces VAE tiling on regardless of `vae_use_tiling` — without it the same
-  request costs about 9 GB more, and a 321-frame one OOMs outright. On smaller
-  GPUs, lower `width`, `height`, or `num_frames` before serving production
-  requests.
-
-#### Environment
-
-- OS: Linux
-- Python: 3.10+
-- Driver / runtime: NVIDIA driver with CUDA runtime supported by your PyTorch
-  build
-- Recommended operator library: Triton, installed through the vLLM/vLLM-Omni
-  Python environment
-- vLLM version: Match the repository requirements for your checkout
-- vLLM-Omni version or commit: Use the commit you are deploying from
-
-#### Command
-
-The repo ships the standard Diffusers layout (`model_index.json` +
-`transformer/`, `vae/`), and its `model_index.json` names `SanaWmPipeline`, so
-the pipeline class resolves on its own.
-
-```bash
-CUDA_VISIBLE_DEVICES=0 \
-vllm serve BBBBruce/SANA-WM_bidirectional-stage1-diffusers \
-  --omni \
-  --host 0.0.0.0 \
-  --port 8091
-```
-
-No deploy config: single-stage diffusion models are deliberately absent from
-`OMNI_PIPELINES` (`vllm_omni/config/pipeline_registry.py`), so stage resolution
-falls back to the default stage config and a YAML's stage settings — including
-`default_sampling_params` — would not be applied. The production generation
-settings therefore live in the model (`num_inference_steps=60`,
-`guidance_scale=5.0`); a request that omits a field gets them. The examples
-below still pass every field explicitly so the numbers are visible.
-
-If you point this at the older two-stage repo
-(`BBBBruce/SANA-WM_bidirectional-diffusers`), startup fails with `Model class
-SanaWmTwoStagesPipeline not found in diffusion model registry`, because that
-repo's `model_index.json` names a class this build does not register. Add
-`--model-class-name SanaWmPipeline` to override it.
-
-#### Verification
-
-Use a short smoke request first:
-
-```bash
-curl -sS -X POST http://localhost:8091/v1/videos/sync \
-  -H "Accept: video/mp4" \
-  -F "prompt=A slow forward camera move through a quiet city street." \
-  -F "negative_prompt=blurry, low quality, distorted, watermark" \
-  -F "input_reference=@/path/to/first_frame.png;type=image/png" \
-  -F "width=1280" \
-  -F "height=704" \
-  -F "num_frames=9" \
-  -F "fps=16" \
-  -F "num_inference_steps=2" \
-  -F "guidance_scale=5.0" \
-  -F "seed=42" \
-  --form-string 'extra_params={"sana_wm":{"action":"w-8","translation_speed":0.055,"rotation_speed_deg":1.2,"intrinsics":{"fx":640,"fy":640,"cx":640,"cy":352}}}' \
-  -o sana_wm_smoke.mp4
-```
-
-For a production-length request, note that the action durations must sum to
-`num_frames - 1` — the rollout includes the identity start pose — and a
-mismatch is rejected rather than padded or truncated:
-
-```bash
-curl -sS -X POST http://localhost:8091/v1/videos/sync \
-  -H "Accept: video/mp4" \
-  -F "prompt=A slow forward camera move through a quiet city street." \
-  -F "negative_prompt=blurry, low quality, distorted, watermark" \
-  -F "input_reference=@/path/to/first_frame.png;type=image/png" \
-  -F "width=1280" \
-  -F "height=704" \
-  -F "num_frames=161" \
-  -F "fps=16" \
-  -F "num_inference_steps=60" \
-  -F "guidance_scale=5.0" \
-  -F "seed=42" \
-  --form-string 'extra_params={"sana_wm":{"action":"w-160","translation_speed":0.055,"rotation_speed_deg":1.2,"intrinsics":{"fx":640,"fy":640,"cx":640,"cy":352}}}' \
-  -o sana_wm_output.mp4
-```
-
-Use `POST /v1/videos` instead when you want job storage and polling rather than
-inline MP4 bytes. It accepts the same form fields as `/v1/videos/sync`.
-
-```bash
-create_response=$(curl -sS -X POST http://localhost:8091/v1/videos \
-  -H "Accept: application/json" \
-  -F "prompt=A slow forward camera move through a quiet city street." \
-  -F "negative_prompt=blurry, low quality, distorted, watermark" \
-  -F "input_reference=@/path/to/first_frame.png;type=image/png" \
-  -F "width=1280" \
-  -F "height=704" \
-  -F "num_frames=161" \
-  -F "fps=16" \
-  -F "num_inference_steps=60" \
-  -F "guidance_scale=5.0" \
-  -F "seed=42" \
-  --form-string 'extra_params={"sana_wm":{"action":"w-160","translation_speed":0.055,"rotation_speed_deg":1.2,"intrinsics":{"fx":640,"fy":640,"cx":640,"cy":352}}}')
-
-video_id=$(echo "$create_response" | jq -r '.id')
-curl -sS "http://localhost:8091/v1/videos/${video_id}" | jq .
-curl -L "http://localhost:8091/v1/videos/${video_id}/content" -o sana_wm_output.mp4
-```
-
-#### Notes
-
-- Sequence parallelism is not supported. The bidirectional gated delta
-  recurrence carries state across frames, so a rank cannot denoise a slice
-  of the token sequence in isolation; supporting it needs a distributed scan
-  or an all-gather before the GDN blocks.
-
-- `input_reference` is required for the first frame. Use `image_reference` only
-  when you need a JSON-safe image URL or data URL instead of a multipart upload.
-- `sana_wm` must provide exactly one of `action` or `camera`.
-- Action strings use comma-separated `<keys>-<duration>` segments. Supported
-  keys are `w`, `a`, `s`, `d` for translation and `i`, `j`, `k`, `l` for
-  pitch/yaw rotation. The durations must sum to `num_frames - 1`.
-- Explicit camera control (alternative to `action`): pass
-  `"camera": {"poses": [...]}` where `poses` is a list of `num_frames`
-  camera-to-world 4x4 matrices (row-major, OpenCV `+X right, +Y down, +Z forward`
-  convention), e.g.
-  `extra_params={"sana_wm":{"camera":{"poses":[[[1,0,0,0],[0,1,0,0],[0,0,1,0],[0,0,0,1]], ...]},"intrinsics":{...}}}`.
-  Most callers should prefer `action`; explicit poses exist for callers that
-  already have a per-frame trajectory.
-- Explicit `intrinsics` are recommended and take the mapping form
-  `{"fx":640,"fy":640,"cx":640,"cy":352}` (for 1280x704). This `{fx,fy,cx,cy}`
-  mapping is the only accepted intrinsics form; omit `intrinsics` to derive them
-  from the output resolution. All four values must be finite, and `fx`/`fy`
-  must be positive — the ray map divides by them.
-- The video API returns decoded MP4 bytes and has no `output_type` field, so
-  raw Stage-1 latents are reachable only from the offline API
-  (`OmniDiffusionSamplingParams(output_type="latent")`); see
-  [`tests/e2e/offline_inference/test_sana_wm.py`](../../tests/e2e/offline_inference/test_sana_wm.py).
-  Putting `output_type` in `extra_params` does not work: it lands in
-  `sampling_params.extra_args`, while the pipeline reads the top-level field.

recipes/OpenBMB/MiniCPM-o-4_5.md

@@ -71,35 +71,6 @@ single-GPU entry point; `minicpmo_4_5_2gpu.yaml` is the recommended
 two-GPU profile. The removed fused two-stage implementation is not retained as
 a fallback because it would duplicate state machines and correctness paths.
 
-### Graph execution
-
-Graph boundaries follow each stage's state model:
-
-| Stage | Graph mode on Ascend | Eager boundary |
-| --- | --- | --- |
-| 0 Thinker | vLLM `PIECEWISE` | multimodal preprocessing and output routing |
-| 1 Talker | vLLM `PIECEWISE` | conditioning preprocessing, codec sampling, request-state updates |
-| 2 Code2Wav | inner exact-shape NPUGraph for the CFM DiT estimator | encoder, timestep embedding, HiFT/RNG, request parsing, stream-state commit |
-
-Stage 2 keeps `enforce_eager: true` for the generation runner because its
-Python request metadata and per-request cache dictionaries are not valid
-outer-graph inputs. The backend captures only the deterministic CFM DiT
-estimator, with an exact graph key for each tensor shape. Timestep embedding
-stays eager because the upstream implementation creates a CPU frequency tensor;
-HiFT stays eager because it generates random phase. The graph cache stores up
-to 32 entries by default, after which unseen shapes run eagerly.
-
-An ACL graph capture failure is fatal to that Stage-2 process because older
-torch-npu releases can leave allocator and RNG capture state invalid. Restart
-the service after a capture failure. To run without capture, set
-`code2wav_enable_npu_graph: false` under the Stage-2
-`platforms.npu.stages[].additional_config` block before startup. Tune the cache
-limit there with `code2wav_max_npu_graphs`. Graph mode also requires
-`ASCEND_LAUNCH_BLOCKING` to be unset or set to `0`.
-
-The inner NPUGraph is independent of the outer runner setting, so do not use a
-global `--enforce-eager` override when Stage 0/1 `PIECEWISE` replay is desired.
-
 ## GPU
 
 ### 1 x GPU (default — single command)
@@ -240,9 +211,9 @@ speech output (TTS)"** checkbox on / off.
   model processes still share one CUDA device.
 - `--trust-remote-code` is required — the HF repo ships a custom
   `MiniCPMO` config / model class.
-- Stage 0 Thinker and Stage 1 Talker enable vLLM CUDA Graphs. Stage 2 keeps its
-  request orchestration eager while using inner CFM and HiFT CUDA Graphs. On
-  Ascend, the exact-shape CFM DiT estimator instead uses inner NPUGraph replay.
+- Stage 0 Thinker and Stage 1 Talker enable vLLM CUDA Graphs. Stage 2 remains
+  eager because its request-owned Flow/HiFT caches and variable chunk/cache
+  shapes are not yet exposed through a static exact-shape graph wrapper.
 - All default stages use `max_num_seqs: 4` to reduce cross-process GPU
   contention. Talker AR
   state and Code2Wav caches are request-owned; Code2Wav batches only
@@ -253,8 +224,8 @@ speech output (TTS)"** checkbox on / off.
 - `StageRequestStats.batch_size` is a request-scoped placeholder, not the
   scheduler's execution batch.
 - Single-GPU co-location trades throughput for hardware density: Stage 0/1
-  CUDA Graph replay and Stage 2 vocoder kernels compete across three CUDA
-  contexts. Use the 8x4090 config or a custom multi-GPU mapping for
+  CUDA Graph replay and eager Stage 2 vocoder kernels compete across three
+  CUDA contexts. Use the 8x4090 config or a custom multi-GPU mapping for
   throughput-sensitive serving.
 
 ### 8 x RTX 4090 24GB (consumer-GPU layout)

recipes/Qwen/Qwen3-TTS.md

@@ -252,53 +252,3 @@ curl -X POST http://localhost:8091/v1/audio/speech \
     - stage_id: 1
       gpu_memory_utilization: 0.15
   ```
-
-### 1x AMD MI300X, 1.7B checkpoints
-
-#### Environment
-
-- OS: Linux 6.8.0-134-generic, x86_64
-- Container: official ROCm image built from `docker/Dockerfile.rocm`
-- Python: 3.12.13
-- PyTorch: 2.11.0+gitd0c8b1f
-- Driver / runtime: AMD 6.19.14.31400000 / ROCm 7.2.53211
-- GPU: one AMD Instinct MI300X, `gfx942:sramecc+:xnack-`, 191.69 GiB visible HBM
-- vLLM version: 0.27.0+rocm723
-- vLLM Omni version or commit: `3fecb6953ca8dc51210cc0421ef24552267a41ef`
-- Installed vLLM Omni package metadata: `0.27.0rc2.dev44+g55abdade9.rocm`
-- ONNX Runtime: onnxruntime-rocm 1.22.2.post3 with `ROCMExecutionProvider`
-- transformers: 5.15.0
-
-#### Command
-
-The MI300X tests also worked without `enforce_eager`. To use that setup, remove `enforce_eager: true` from stage 1 under `platforms.rocm` in `vllm_omni/deploy/qwen3_tts.yaml`.
-
-Set this environment variable before you start Qwen3-TTS:
-
-```bash
-export MIOPEN_FIND_MODE=FAST
-```
-
-#### Verification
-
-Each row records one offline request with one prompt and batch size one. Request time is the logged `e2e_wall_time_ms`, and stage 1 time is the logged `e2e_stage_1_wall_time_ms`.
-
-| Checkpoint | Code2Wav configuration | `MIOPEN_FIND_MODE` | Request time | Stage 1 time | Output duration | Real time factor | Maximum sampled device memory |
-| --- | --- | --- | ---: | ---: | ---: | ---: | ---: |
-| CustomVoice | Eager | Unset | 1.683 s | 1.613 s | 5.68 s | 0.30 | 60.91 GiB |
-| CustomVoice | Graphs allowed | `FAST` | 1.326 s | 1.259 s | 5.68 s | 0.23 | 62.19 GiB |
-| VoiceDesign | Eager | Unset | 1.195 s | 1.180 s | 4.40 s | 0.27 | 60.91 GiB |
-| VoiceDesign | Graphs allowed | `FAST` | 1.127 s | 1.054 s | 4.40 s | 0.26 | 62.21 GiB |
-| Base | Eager | Unset | 13.330 s | 13.263 s | 4.32 s | 3.09 | 61.74 GiB |
-| Base | Graphs allowed | `FAST` | 14.314 s | 14.245 s | 4.80 s | 2.98 | 62.84 GiB |
-
-All six processes finished successfully, and each output passed the 24 kHz mono WAV checks.
-
-#### Notes
-
-- The current ROCm config keeps Code2Wav in eager mode.
-- The tested alternative allows graphs and sets [`MIOPEN_FIND_MODE=FAST`](https://rocm.docs.amd.com/projects/MIOpen/en/develop/reference/env_variables.html). MIOpen uses a saved kernel choice when available and uses its immediate fallback otherwise.
-- CustomVoice used a graph for one of four logged Code2Wav batches. VoiceDesign used a graph for two of four batches. Base used no graphs for its four logged batches.
-- The recorded request time with graphs allowed was lower for CustomVoice and VoiceDesign and higher for Base. The Base runs generated different amounts of audio, so their request times are not a direct comparison.
-- Each setup was tested with one request, so the timing is not a performance benchmark.
-- GPU memory was sampled once per second with `rocm-smi`.

recipes/README.md

@@ -49,7 +49,6 @@ recipes/
 | [`Helios/Helios.md`](./Helios/Helios.md) | Text-to-video, image-to-video, and video-to-video generation | 1x NVIDIA H20 |
 | [`inclusionAI/Ming-flash-omni-2.0.md`](./inclusionAI/Ming-flash-omni-2.0.md) | Online serving for multimodal chat + standalone TTS | 4x H100 / 1x H100 80GB |
 | [`inclusionAI/Ming-omni-tts.md`](./inclusionAI/Ming-omni-tts.md) | Offline + online dense Ming TTS/audio generation | 1x H100 80GB / 1x AMD MI300X (ROCm 7.2) |
-| [`k2-fsa/OmniVoice.md`](./k2-fsa/OmniVoice.md) | Offline and online multilingual text-to-speech | 1x AMD MI300X (ROCm 7.2) |
 | [`IndexTeam/IndexTTS-2.md`](./IndexTeam/IndexTTS-2.md) | Online serving for voice-cloned TTS with optional emotion control | 1x L4 24GB or larger CUDA GPU |
 | [`IndexTeam/IndexTTS-2_5.md`](./IndexTeam/IndexTTS-2_5.md) | Online + offline multilingual voice-cloned TTS with native speed and emotion control | 1x NVIDIA H20 96GB |
 | [`MiniMaxAI/MiniMax-H3.md`](./MiniMaxAI/MiniMax-H3.md) | T2VA, FL2VA, image+audio Ref2VA, and multi-video Ref2VA serving | 1x GPU with CPU offload / 4x B300 |
@@ -70,14 +69,13 @@ recipes/
 | [`OpenMOSS/MOSS-TTS.md`](./OpenMOSS/MOSS-TTS.md) | Online + offline multilingual TTS (MOSS-TTS family, 8B) | 1x H100 80GB |
 | [`NVIDIA/Nemotron-Labs-Audex.md`](./NVIDIA/Nemotron-Labs-Audex.md) | TTS / text-to-audio / audio understanding / cascaded S2S (Audex 2B + 30B-A3B) | 1x H100 80GB |
 | [`NVIDIA/NemotronLabs-VoiceChat.md`](./NVIDIA/NemotronLabs-VoiceChat.md) | Offline speech-to-speech voice chat (11B, frame-locked 12.5 Hz, 3-stage thinker/talker/code2wav) | 1x H100 80GB |
-| [`NVIDIA/SANA-WM.md`](./NVIDIA/SANA-WM.md) | First-frame image-to-video serving with camera control | 1x 24GB+ CUDA GPU; ~40GB disk |
 | [`Qwen/Qwen-Image.md`](./Qwen/Qwen-Image.md) | Text-to-image serving with step-wise continuous batching replay and ModelOpt mixed FP8/NVFP4 | 1x A100 80GB / 2x B200 |
 | [`Qwen/Qwen-Image-2512.md`](./Qwen/Qwen-Image-2512.md) | Text-to-image serving with step-wise continuous batching replay and ModelOpt FP8 / mixed FP8/NVFP4 | 1x A800 80GB / 2x B200 |
 | [`Qwen/Qwen-Image-Edit.md`](./Qwen/Qwen-Image-Edit.md) | Text-guided single-image editing | 1x or 2x H200 141GB |
 | [`Qwen/Qwen3-Omni.md`](./Qwen/Qwen3-Omni.md) | Online serving for multimodal chat | 1x A100 80GB |
-| [`Qwen/Qwen3-TTS.md`](./Qwen/Qwen3-TTS.md) | Text-to-speech serving (CustomVoice / VoiceDesign / Base) | 1x H100/A100 80GB / 1x AMD MI300X |
-| [`SenseNova/SenseNova-U1.md`](./SenseNova/SenseNova-U1.md) | Unified image generation and understanding | 1x or 2x H200 / 1x AMD MI300X |
-| [`StabilityAI/Stable-Audio-Open.md`](./StabilityAI/Stable-Audio-Open.md) | Offline + online text-to-audio generation (Stable Audio Open) | 1x RTX 4090 24GB / 1x AMD MI300X |
+| [`Qwen/Qwen3-TTS.md`](./Qwen/Qwen3-TTS.md) | Text-to-speech serving (CustomVoice / VoiceDesign / Base) | 1x H100/A100 80GB |
+| [`SenseNova/SenseNova-U1.md`](./SenseNova/SenseNova-U1.md) | Unified image generation and understanding | 1x H200 (144GB) |
+| [`StabilityAI/Stable-Audio-Open.md`](./StabilityAI/Stable-Audio-Open.md) | Offline + online text-to-audio generation (Stable Audio Open) | 1x RTX 4090 24GB |
 | [`Tencent/Covo-Audio-Chat.md`](./Tencent/Covo-Audio-Chat.md) | Online serving for audio chat | 1x A100 80GB |
 | [`Tencent/HunyuanImage-3.0-Instruct.md`](./Tencent/HunyuanImage-3.0-Instruct.md) | DiT-only text-to-image serving and benchmark, including ModelOpt mixed FP8/NVFP4 | 4x H100/H800 80GB / 2x B200 |
 | [`Wan-AI/Wan2.2-T2V.md`](./Wan-AI/Wan2.2-T2V.md) | Text-to-video serving (Wan2.2 14B) with Skip-Softmax sparse attention | 1x datacenter Blackwell (B200/B300) |

recipes/SenseNova/SenseNova-U1.md

@@ -178,48 +178,6 @@ with the expected 1536×2720 resolution.
 - The LLM transformer uses `QKVParallelLinear` and `MergedColumnParallelLinear`
   for fused QKV and gate/up projections with TP support.
 
-### 1x AMD MI300X 192GB
-
-#### Environment
-
-- OS: Linux 6.8.0-134-generic, x86_64
-- Container: official ROCm image built from `docker/Dockerfile.rocm`
-- Python: 3.12.13
-- PyTorch: 2.11.0+gitd0c8b1f
-- Driver / runtime: AMD 6.19.14.31400000 / ROCm 7.2.53211
-- GPU: one AMD Instinct MI300X, `gfx942:sramecc+:xnack-`, 191.69 GiB visible HBM
-- vLLM version: 0.27.0+rocm723
-- vLLM Omni version or commit: `a704c8759c96e123c0d7c89b11f120b1c0f120cf`
-- Installed vLLM Omni package metadata: `0.27.0rc2.dev44+g55abdade9.rocm`
-
-#### Command
-
-```bash
-python3 examples/offline_inference/text_to_image/text_to_image.py \
-    --model SenseNova/SenseNova-U1-8B-MoT \
-    --prompt "Close portrait of an elderly woman by a farmhouse window, textured skin, gentle smile, warm natural light, emotional documentary look." \
-    --width 1536 \
-    --height 2720 \
-    --seed 42 \
-    --num-inference-steps 50 \
-    --cfg-scale 4.0 \
-    --extra-body '{"think": true, "cfg_norm": "none", "timestep_shift": 3.0, "t_eps": 0.02}' \
-    --enable-diffusion-pipeline-profiler \
-    --log-stats \
-    --output sensenova_u1_mi300x.png
-```
-
-#### Verification
-
-The command completed and wrote a valid 1536 by 2720 RGB PNG.
-
-#### Notes
-
-- Generation with 50 inference steps took 34.011 seconds.
-- Model loading used 32.774 GiB and took 13.987 seconds.
-- The internal profiler recorded 35.67 GB reserved and 35.10 GB allocated for the request.
-- The highest one second whole device memory sample was 37.93 GiB.
-
 ## Online Serving
 
 SenseNova-U1 supports all four modalities via the OpenAI-compatible

recipes/StabilityAI/Stable-Audio-Open.md

@@ -164,50 +164,7 @@ PY
   `torchaudio==2.11.0+cu130`.
 - The `NIXL is not available`, `GLOO_SOCKET_IFNAME`, and `torchsde` boundary
   warnings observed during validation did not prevent successful generation.
-- The RTX 4090 entry was validated on one 24 GB GPU. The MI300X entry below covers one 192 GB GPU.
+- This recipe was validated on one RTX 4090 24GB GPU only. Other GPU counts,
+  ROCm, XPU, and NPU setups are not covered here.
 - Long generations, higher inference-step counts, and non-WAV response formats
   were not benchmarked in this recipe.
-
-#### 1x AMD MI300X 192GB
-
-##### Environment
-
-- OS: Linux 6.8.0-134-generic, x86_64
-- Container: official ROCm image built from `docker/Dockerfile.rocm`
-- Python: 3.12.13
-- PyTorch: 2.11.0+gitd0c8b1f
-- Driver / runtime: AMD 6.19.14.31400000 / ROCm 7.2.53211
-- GPU: one AMD Instinct MI300X, `gfx942:sramecc+:xnack-`, 191.69 GiB visible HBM
-- vLLM version: 0.27.0+rocm723
-- vLLM Omni version or commit: `73e1368c7bb940efe1a025859c9d6c8eeeb2e3f0`
-- Installed vLLM Omni package metadata: `0.27.0rc2.dev44+g55abdade9.rocm`
-
-##### Commands
-
-```bash
-python3 examples/offline_inference/text_to_audio/text_to_audio.py \
-    --model stabilityai/stable-audio-open-1.0 \
-    --prompt "A gentle piano melody with soft room ambience" \
-    --negative-prompt "Low quality, distorted, noisy" \
-    --seed 42 \
-    --guidance-scale 7.0 \
-    --audio-length 10.0 \
-    --num-inference-steps 50 \
-    --cache-backend tea_cache \
-    --enable-diffusion-pipeline-profiler \
-    --output stable_audio_10s.wav
-```
-
-##### Verification
-
-The command completed and wrote a valid 44.1 kHz stereo WAV with 10.00 seconds of audio.
-
-##### Notes
-
-- TeaCache ran with `rel_l1_thresh=0.2`.
-- Model loading used 2.7891 GiB and took 3.706 seconds.
-- Generation took 4.750 seconds, which gives a real time factor of 0.475 for the 10.00 second output.
-- The internal profiler recorded 15.65 GB reserved and 9.68 GB allocated for the request.
-- The highest one second whole device memory sample was 19.61 GiB.
-- The output RMS was 0.0887, and its peak absolute amplitude was 0.5761.
-- The full process took 384 seconds, including startup and compilation.

recipes/cosmos3/Cosmos3-DistOffload.md

@@ -189,10 +189,8 @@ wait
 
 - `--dlo-use-allgather` (default) requires all concurrent requests to have the same
   `num_inference_steps` (AllGather is a collective that needs all ranks synchronized)
-- Online per-tensor FP8 linears with DLO AllGather use the ordinary loader
-  instead of direct checkpoint mmap, then shard the finalized FP8 weights and
-  scales. Size host memory for a transient complete FP8 model on every rank at
-  startup
+- Online quantization (FP8) is incompatible with the DLO+AllGather mmap path;
+  use `--dlo-no-use-allgather` or disable online quantization
 - Tensor Parallel > 1 is rejected in the DLO+AllGather mmap path. TP with
   `--dlo-no-use-allgather` retains the standard TP loader and should be treated
   as an experimental compatibility path

recipes/inclusionAI/Ming-omni-tts.md

@@ -5,21 +5,20 @@
 ## Summary
 
 - Vendor: inclusionAI
-- Model: `inclusionAI/Ming-omni-tts-16.8B-A3B`, `inclusionAI/Ming-omni-tts-0.5B`
+- Model: `inclusionAI/Ming-omni-tts-0.5B`
 - Task: Text-to-speech with style, dialect, cloning, and multi-speaker controls
 - Mode: Online serving via the OpenAI-compatible `/v1/audio/speech` API
 - Maintainer: Community
 
 ## References
 
-- [Huggingface: inclusionAI/Ming-omni-tts-16.8B-A3B](https://huggingface.co/inclusionAI/Ming-omni-tts-16.8B-A3B)
-- [Huggingface: inclusionAI/Ming-omni-tts-0.5B](https://huggingface.co/inclusionAI/Ming-omni-tts-0.5B)
+- [Huggingface Model card](https://huggingface.co/inclusionAI/Ming-omni-tts-0.5B)
 - Upstream repository [inclusionAI/Ming-omni-tts](https://github.com/inclusionAI/Ming-omni-tts)
 
 
 ## Hardware Support
 
-This recipe documents a validated ROCm configuration and a CUDA configuration for the dense 0.5B and the MoE 16.8B two-stage TTS pipeline deployment.
+This recipe documents a validated ROCm configuration and a CUDA configuration for the dense 0.5B two-stage TTS pipeline deployment.
 Other hardware is welcome as community validation lands.
 
 ## CUDA
@@ -40,7 +39,11 @@ Other hardware is welcome as community validation lands.
 Launch the two-stage talker:
 
 ```bash
-vllm serve inclusionAI/Ming-omni-tts-16.8B-A3B --omni --port 8091
+vllm serve inclusionAI/Ming-omni-tts-0.5B \
+    --deploy-config vllm_omni/deploy/ming_tts.yaml \
+    --omni \
+    --port 8091 \
+    --enforce-eager
 ```
 
 #### Verification
@@ -51,7 +54,7 @@ Basic synthesis (save the WAV bytes):
 curl -X POST http://localhost:8091/v1/audio/speech \
     -H "Content-Type: application/json" \
     -d '{
-      "model": "inclusionAI/Ming-omni-tts-16.8B-A3B",
+      "model": "inclusionAI/Ming-omni-tts-0.5B",
       "input": "你好，这是 Ming 在线语音合成测试。",
       "response_format": "wav"
     }' --output ming_tts_basic.wav
@@ -71,7 +74,7 @@ Reference-audio zero-shot cloning:
 REF_AUDIO="$BASE/10002287-00000094.wav"
 jq -n --arg ref_audio "$REF_AUDIO" \
     '{
-       model: "inclusionAI/Ming-omni-tts-16.8B-A3B",
+       model: "inclusionAI/Ming-omni-tts-0.5B",
        input: "我们的愿景是构建未来服务业的数字化基础设施，为世界带来更多微小而美好的改变。",
        ref_audio: $ref_audio,
        ref_text: "在此奉劝大家别乱打美白针。",
@@ -104,7 +107,7 @@ jq -n \
     --arg a "$REF_A" \
     --arg b "$REF_B" \
     '{
-       model: "inclusionAI/Ming-omni-tts-16.8B-A3B",
+       model: "inclusionAI/Ming-omni-tts-0.5B",
        input: $input,
        ref_audio: [$a, $b],
        ref_text: $ref_text,
@@ -167,35 +170,6 @@ python examples/online_serving/text_to_speech/ming_tts/openai_speech_client.py \
 `--ref-audio` matches upstream `use_spk_emb=True`; do not add `--ref-text`
 for the dialect case.
 
-**Offline smoke result on ROCm 7.2:**
-
-A later offline smoke run used the same MI300X model with the following software:
-
-- GPU: one AMD Instinct MI300X with 191.69 GiB visible HBM and gfx942
-- Kernel: Linux 6.8.0-134-generic, x86_64
-- Python: 3.12.13
-- PyTorch: 2.11.0+gitd0c8b1f
-- ROCm or HIP: 7.2.53211
-- vLLM: 0.27.0+rocm723
-- vLLM Omni checkout: `73e1368c7bb940efe1a025859c9d6c8eeeb2e3f0`
-
-```bash
-python3 examples/offline_inference/text_to_speech/ming_tts/end2end.py \
-    --model inclusionAI/Ming-omni-tts-0.5B \
-    --deploy-config vllm_omni/deploy/ming_tts.yaml \
-    --case basic \
-    --text "你好，这是 AMD MI300X ROCm 配方验证。" \
-    --output-dir ./ming_tts_output \
-    --output-name ming_omni_tts_mi300x.wav \
-    --enforce-eager \
-    --log-stats \
-    --metadata-json ./ming_tts_manifest.json
-```
-
-The checked run produced a valid 44.1 kHz mono WAV with 4.80 seconds of audio, RMS 0.1449, and peak absolute amplitude 0.8621. The first request took 20.406 seconds, which gives a real time factor of 4.25. The stage weight loads used 1.31 GiB and 1.47 GiB and took 1.756 seconds and 1.264 seconds. The largest one second whole device memory sample was 91.05 GiB, including the reserved KV cache. Both stages used eager mode, and the AR stage selected `TRITON_ATTN`.
-
-The run used the official ROCm image built from `docker/Dockerfile.rocm`.
-
 ## Notes
 
 - The official ROCm image includes the platform dependencies.

recipes/k2-fsa/OmniVoice.md

@@ -1,98 +0,0 @@
-# OmniVoice
-
-> Offline and online text to speech with automatic voice, voice design, and reference voice cloning
-
-## Summary
-
-- Vendor: k2-fsa
-- Model: `k2-fsa/OmniVoice`
-- Task: Multilingual text to speech
-- Mode: Offline inference for all modes and online serving for automatic voice
-- Maintainer: Community
-
-## References
-
-- [Hugging Face model](https://huggingface.co/k2-fsa/OmniVoice)
-- [Offline text to speech examples](../../docs/user_guide/examples/offline_inference/text_to_speech.md#omnivoice)
-- [Online text to speech examples](../../docs/user_guide/examples/online_serving/text_to_speech.md#omnivoice)
-- [Deploy config](../../vllm_omni/deploy/omnivoice.yaml)
-
-## Requirements
-
-Download the checkpoint before the first run:
-
-```bash
-huggingface-cli download k2-fsa/OmniVoice
-```
-
-Automatic voice and voice design require `transformers>=4.57.0`. Reference voice cloning requires `transformers>=5.3.0`.
-
-The vLLM Omni deploy config runs the OmniVoice generator and decoder inside one float32 diffusion stage.
-
-## ROCm
-
-### 1x AMD MI300X
-
-#### Environment
-
-- GPU: one AMD Instinct MI300X with 191.69 GiB visible HBM and gfx942
-- Kernel: Linux 6.8.0-134-generic, x86_64
-- Python: 3.12.13
-- PyTorch: 2.11.0+gitd0c8b1f
-- ROCm or HIP: 7.2.53211
-- vLLM: 0.27.0+rocm723
-- vLLM Omni checkout: `73e1368c7bb940efe1a025859c9d6c8eeeb2e3f0`
-- transformers: 5.15.0
-
-The run used the official ROCm image built from `docker/Dockerfile.rocm`.
-
-#### Automatic voice
-
-Run the offline example from the repository root:
-
-```bash
-python3 examples/offline_inference/text_to_speech/omnivoice/end2end.py \
-    --model k2-fsa/OmniVoice \
-    --deploy-config vllm_omni/deploy/omnivoice.yaml \
-    --text "Hello, this is OmniVoice running on one AMD MI300X." \
-    --seed 42 \
-    --output omnivoice_mi300x.wav
-```
-
-The checked run produced a valid 24 kHz mono WAV with 3.60 seconds of audio, RMS 0.1194, and peak absolute amplitude 0.9129. The first request took 0.685 seconds, which gives a real time factor of 0.19. Model loading used 3.5739 GiB and took 9.564 seconds. The request used 4.56 GB reserved and 3.64 GB allocated according to the internal profiler, while the largest one second whole device sample was 8.33 GiB. The log states that the OmniVoice Triton kernels loaded.
-
-#### Reference voice cloning
-
-The offline interface also accepts a reference WAV and its transcript:
-
-```bash
-python3 examples/offline_inference/text_to_speech/omnivoice/end2end.py \
-    --model k2-fsa/OmniVoice \
-    --text "Hello, this is a cloned voice." \
-    --ref-audio ref.wav \
-    --ref-text "This is the reference transcription." \
-    --output cloned.wav
-```
-
-#### Voice design
-
-```bash
-python3 examples/offline_inference/text_to_speech/omnivoice/end2end.py \
-    --model k2-fsa/OmniVoice \
-    --text "Hello, this is a designed voice." \
-    --instruct "female, low pitch, British accent" \
-    --output designed.wav
-```
-
-## Online serving
-
-Online serving currently supports automatic voice:
-
-```bash
-vllm serve k2-fsa/OmniVoice --omni --port 8091 --trust-remote-code
-```
-
-```bash
-cd examples/online_serving/text_to_speech/omnivoice
-python speech_client.py --text "Hello, how are you?"
-```

recipes/lerobot/Pi0.md

@@ -1,108 +0,0 @@
-# Pi0 (π0, Pi-Zero) VLA
-
-> Vision-Language-Action policy serving over the OpenPI realtime websocket
-
-## Summary
-
-- Vendor: Physical Intelligence (checkpoint redistributed by LeRobot)
-- Model: `lerobot/pi0_base`
-- Task: Vision-Language-Action — multi-camera images + a language instruction +
-  robot proprioceptive state → a continuous action chunk
-  `[action_horizon, action_dim] = [50, 32]` via flow-matching denoising. π0 does
-  **not** emit text tokens.
-- Mode: Online serving via the OpenPI realtime robot API
-  (`/v1/realtime/robot/openpi` websocket)
-- Maintainer: Community
-
-## When to use this recipe
-
-Use this recipe to deploy `lerobot/pi0_base` as a robot policy server: a robot
-client connects over the OpenPI websocket and, per control step, sends an
-observation (camera frames + state + instruction) and receives an action chunk.
-
-A single pipeline class (`Pi0Pipeline`) serves the policy. π0 is stateless across
-calls (no KV reuse), so the protocol's `session_id` / `reset` are accepted but
-ignored.
-
-## Weights
-
-π0 uses the LeRobot `lerobot/pi0_base` checkpoint (HF, ~13 GB). Either let the
-server download it (`MODEL=lerobot/pi0_base`) or point at a local copy.
-
-## Hardware Support
-
-### 1x H200 141GB (Online serving)
-
-#### Environment
-
-- OS: Ubuntu 22.04+
-- Python: 3.12+
-- Driver / runtime: NVIDIA CUDA environment
-- vLLM / vLLM-Omni: use the commit you are deploying from
-- OpenPI client extras (for the example client): `openpi-client`
-  (`pip install -e packages/openpi-client` from the
-  [openpi](https://github.com/Physical-Intelligence/openpi) repo), `websockets`,
-  `msgpack`, `msgpack-numpy`
-
-#### Command
-
-```bash
-vllm serve lerobot/pi0_base \
-  --omni \
-  --host 0.0.0.0 --port 8000 \
-  --served-model-name pi0 \
-  --deploy-config vllm_omni/deploy/pi0.yaml \
-  --enforce-eager --disable-log-stats
-```
-
-The deploy config (`vllm_omni/deploy/pi0.yaml`) declares a single diffusion stage
-(`Pi0Pipeline`), `float32`, `max_num_seqs: 1`, and the `policy_server_config`
-handshake metadata (3 cameras, `joint_position`, action horizon 50, action dim
-32) that enables the OpenPI realtime serving layer.
-
-#### Verification
-
-Run the example client (sends a synthetic observation and prints the returned
-action chunk):
-
-```bash
-python examples/online_serving/pi0/openpi_client.py \
-  --host 127.0.0.1 --port 8000 \
-  --prompt "pick up the red block and place it in the bin"
-```
-
-It connects, prints the server metadata, sends robot observations, and prints the
-returned `[action_horizon, action_dim] = [50, 32]` action chunks. The observation
-is a flat dict per inference:
-
-```python
-{
-    "observation.images.base_0_rgb":        np.uint8[H, W, 3],
-    "observation.images.left_wrist_0_rgb":  np.uint8[H, W, 3],
-    "observation.images.right_wrist_0_rgb": np.uint8[H, W, 3],
-    "state":   np.float32[state_dim],   # zero-padded to max_state_dim=32 server-side
-    "prompt":  "pick up the red block",
-    "session_id": "<uuid>",             # accepted but ignored (π0 is stateless)
-}
-```
-
-Camera keys must match the server's `image_feature_keys` (the checkpoint's
-`input_features` order). If your robot uses different camera names, set
-`model_config.image_key_map` in `pi0.yaml` to map raw obs keys → feature keys.
-
-## Correctness
-
-π0's flow-matching kernel is bit-for-bit matched to the LeRobot `PI0Policy`
-reference (`max|Δ| = 7.15e-07`, CPU/float32, fixed noise; see
-`tests/diffusion/models/pi0/test_pi0_parity.py::test_pi0_vllm_omni_vs_lerobot`).
-
-## References
-
-- Model blog: <https://www.physicalintelligence.company/blog/pi0>
-- Checkpoint: <https://huggingface.co/lerobot/pi0_base>
-- Pipeline: [`vllm_omni/diffusion/models/pi0/pipeline_pi0.py`](../../vllm_omni/diffusion/models/pi0/pipeline_pi0.py)
-- Deploy config: [`vllm_omni/deploy/pi0.yaml`](../../vllm_omni/deploy/pi0.yaml)
-- Example client: [`examples/online_serving/pi0/`](../../examples/online_serving/pi0/)
-- Tests: CPU units + LeRobot parity in [`tests/diffusion/models/pi0/`](../../tests/diffusion/models/pi0/),
-  OpenPI websocket e2e in
-  [`tests/e2e/online_serving/test_pi0_expansion.py`](../../tests/e2e/online_serving/test_pi0_expansion.py)

tests/benchmarks/test_daily_omni_local_source.py

@@ -9,17 +9,13 @@ from pathlib import Path
 import pytest
 
 from vllm_omni.benchmarks.data_modules.daily_omni_dataset import (
-    DailyOmniDataset,
     daily_omni_local_qa_json,
     daily_omni_local_videos_dir,
-    ensure_daily_omni_hub_root,
     resolve_daily_omni_local_root,
 )
 
 pytestmark = [pytest.mark.core_model, pytest.mark.cpu, pytest.mark.benchmark]
 
-_HUB_ID = "liarliar/Daily-Omni"
-
 
 def _write_qa_json(root: Path) -> Path:
     qa = root / "qa.json"
@@ -27,28 +23,6 @@ def _write_qa_json(root: Path) -> Path:
     return qa
 
 
-def _stage_offline_hub_cache(tmp_path: Path, monkeypatch: pytest.MonkeyPatch, *, with_qa: bool = True) -> Path:
-    """Build a hub cache holding ``liarliar/Daily-Omni`` and cut off all network access.
-
-    Reproduces the air-gapped NPU runner: the dataset is pre-staged with ``hf download``, so it
-    only exists in the *hub* cache -- never in the ``datasets`` cache that ``load_dataset`` reads.
-    """
-    constants = pytest.importorskip("huggingface_hub").constants
-
-    cache = tmp_path / "hub"
-    repo_dir = cache / "datasets--liarliar--Daily-Omni"
-    snapshot = repo_dir / "snapshots" / "deadbeef"
-    snapshot.mkdir(parents=True)
-    if with_qa:
-        _write_qa_json(snapshot)
-    (repo_dir / "refs").mkdir()
-    (repo_dir / "refs" / "main").write_text("deadbeef", encoding="utf-8")
-
-    monkeypatch.setattr(constants, "HF_HUB_CACHE", str(cache))
-    monkeypatch.setattr(constants, "HF_HUB_OFFLINE", True)
-    return snapshot
-
-
 def test_resolve_local_root_returns_none_for_hub_id() -> None:
     assert resolve_daily_omni_local_root("liarliar/Daily-Omni") is None
     assert resolve_daily_omni_local_root("") is None
@@ -116,37 +90,3 @@ def test_missing_qa_json_reports_none(tmp_path: Path) -> None:
     root = resolve_daily_omni_local_root(str(tmp_path))
     assert daily_omni_local_qa_json(root) is None
     assert daily_omni_local_videos_dir(root) is None
-
-
-def test_hub_id_resolves_to_offline_cache(tmp_path: Path, monkeypatch: pytest.MonkeyPatch) -> None:
-    snapshot = _stage_offline_hub_cache(tmp_path, monkeypatch)
-
-    root = ensure_daily_omni_hub_root(_HUB_ID)
-
-    assert root == snapshot.resolve()
-    assert daily_omni_local_qa_json(root) == (snapshot / "qa.json").resolve()
-
-
-def test_dataset_loads_hub_id_without_network(tmp_path: Path, monkeypatch: pytest.MonkeyPatch) -> None:
-    """The regression this file exists for: a Hub id must not require ``load_dataset``."""
-    snapshot = _stage_offline_hub_cache(tmp_path, monkeypatch)
-    # Any use of `datasets` here means we went back to the path that cannot read the hub cache.
-    monkeypatch.setattr("vllm_omni.benchmarks.data_modules.daily_omni_dataset.load_dataset", None)
-
-    dataset = DailyOmniDataset(dataset_path=_HUB_ID, random_seed=0)
-
-    assert dataset.qa_json_path == (snapshot / "qa.json").resolve()
-    assert [row["video_id"] for row in dataset.data] == ["vid0"]
-
-
-def test_uncached_hub_id_reports_both_attempts(tmp_path: Path, monkeypatch: pytest.MonkeyPatch) -> None:
-    _stage_offline_hub_cache(tmp_path, monkeypatch, with_qa=False)
-    monkeypatch.setattr("vllm_omni.benchmarks.data_modules.daily_omni_dataset.load_dataset", None)
-
-    with pytest.raises(RuntimeError) as excinfo:
-        DailyOmniDataset(dataset_path="someone-else/Daily-Omni", random_seed=0)
-
-    message = str(excinfo.value)
-    assert "someone-else/Daily-Omni" in message
-    assert "snapshot_download failed" in message
-    assert "--daily-omni-qa-json" in message

tests/config/test_config_import_cycle.py

@@ -1,119 +0,0 @@
-# SPDX-License-Identifier: Apache-2.0
-# SPDX-FileCopyrightText: Copyright contributors to the vLLM-Omni project
-"""Guard the NPU pytest circular import through DiffusionOutput."""
-
-from __future__ import annotations
-
-import ast
-import os
-import subprocess
-import sys
-from pathlib import Path
-
-import pytest
-
-pytestmark = [pytest.mark.core_model, pytest.mark.cpu]
-REPO_ROOT = Path(__file__).resolve().parents[2]
-
-
-def _top_level_import_modules(path: Path) -> set[str]:
-    tree = ast.parse(path.read_text(encoding="utf-8"))
-    modules: set[str] = set()
-    for node in tree.body:
-        if isinstance(node, ast.ImportFrom) and node.module:
-            modules.add(node.module)
-        elif isinstance(node, ast.Import):
-            modules.update(alias.name for alias in node.names)
-    return modules
-
-
-def _assert_isolated_import_succeeds(script: str) -> None:
-    env = os.environ.copy()
-    env.pop("CUDA_VISIBLE_DEVICES", None)
-    env.pop("HIP_VISIBLE_DEVICES", None)
-    result = subprocess.run(
-        [sys.executable, "-c", script],
-        cwd=REPO_ROOT,
-        env=env,
-        capture_output=True,
-        text=True,
-        timeout=120,
-        check=False,
-    )
-    assert result.returncode == 0, result.stderr
-
-
-def test_config_package_does_not_eagerly_import_pipeline_registry() -> None:
-    imported = _top_level_import_modules(REPO_ROOT / "vllm_omni" / "config" / "__init__.py")
-    assert "vllm_omni.config.config_factory" not in imported
-    assert "vllm_omni.config.pipeline_registry" not in imported
-
-
-def test_config_package_lazy_exports_still_resolve() -> None:
-    from vllm_omni.config import StageConfigFactory, register_pipeline
-    from vllm_omni.config.config_factory import StageConfigFactory as DirectStageConfigFactory
-    from vllm_omni.config.pipeline_registry import register_pipeline as direct_register_pipeline
-
-    assert StageConfigFactory is DirectStageConfigFactory
-    assert register_pipeline is direct_register_pipeline
-
-
-def test_npu_platform_defers_minimax_h3_encoder_patch() -> None:
-    src = (REPO_ROOT / "vllm_omni" / "platforms" / "npu" / "platform.py").read_text(encoding="utf-8")
-    init_start = src.index("def __init__(self) -> None:")
-    runtime_start = src.index("def init_diffusion_model_runner_runtime")
-    init_src = src[init_start:runtime_start]
-    runtime_src = src[runtime_start:]
-    assert "apply_minimax_h3_qwen3vl_patch" not in init_src
-    assert "apply_minimax_h3_qwen3vl_patch" in runtime_src
-    assert "apply_minimax_h3_qwen3vl_swiglu_patch" in runtime_src
-
-
-def test_lora_config_import_does_not_load_pipeline_registry() -> None:
-    """Import LoRAConfig and diffusion.data in a fresh interpreter.
-
-    The NPU failure was ``data.py`` importing LoRAConfig while still loading,
-    then ``vllm_omni.config`` re-entering ``pipeline_registry`` → PI0 →
-    DiffusionOutput. This child process must not preload those exports.
-    """
-    _assert_isolated_import_succeeds(
-        """
-import sys
-
-from vllm_omni.config.lora import LoRAConfig
-
-if "vllm_omni.config.pipeline_registry" in sys.modules:
-    raise SystemExit("LoRAConfig import loaded pipeline_registry")
-if "vllm_omni.diffusion.models.pi0.pipeline_pi0" in sys.modules:
-    raise SystemExit("LoRAConfig import loaded PI0_PIPELINE")
-
-from vllm_omni.diffusion.data import DiffusionOutput
-
-if DiffusionOutput is None or LoRAConfig is None:
-    raise SystemExit("expected LoRAConfig and DiffusionOutput")
-"""
-    )
-
-
-def test_pipeline_registry_does_not_import_pi0_runtime() -> None:
-    """Resolving the pi0 topology must not import its runtime pipeline.
-
-    ``pipeline_pi0`` imports ``diffusion.data``. Importing it while
-    ``diffusion.data`` is only partially initialized closes the Ascend plugin's
-    circular-import loop, so the registry must depend only on a lightweight
-    pipeline-config module.
-    """
-    _assert_isolated_import_succeeds(
-        """
-import sys
-
-from vllm_omni.config.pipeline_registry import resolve_pipeline_config
-
-if "vllm_omni.diffusion.models.pi0.pipeline_pi0" in sys.modules:
-    raise SystemExit("pipeline_registry imported the pi0 runtime pipeline")
-
-pipeline = resolve_pipeline_config("pi0")
-if pipeline is None or pipeline.model_type != "pi0":
-    raise SystemExit("expected the registered pi0 pipeline config")
-"""
-    )

tests/config/test_omni_config.py

@@ -5,20 +5,11 @@
 from __future__ import annotations
 
 from dataclasses import fields
-from inspect import Parameter, signature
 from pathlib import Path
 
-import msgspec
 import pytest
 from pydantic import ValidationError
-from pydantic.fields import FieldInfo
 from transformers import Qwen3OmniMoeConfig
-from vllm.config import CacheConfig as VllmCacheConfig
-from vllm.config import CompilationConfig as VllmCompilationConfig
-from vllm.config import LoadConfig as VllmLoadConfig
-from vllm.config import ParallelConfig as VllmParallelConfig
-from vllm.config import ProfilerConfig as VllmProfilerConfig
-from vllm.config import SchedulerConfig as VllmSchedulerConfig
 
 from tests.helpers.stage_config import get_deploy_config_path
 from vllm_omni.config import omni_config as omni_config_module
@@ -49,7 +40,6 @@ from vllm_omni.config.stage_config import (
     merge_pipeline_deploy,
 )
 from vllm_omni.diffusion.diffusion_kv.config import DiffusionKVCacheMode
-from vllm_omni.engine.stage_engine_startup import _serialize_stage_config
 from vllm_omni.engine.stage_init_utils import build_legacy_engine_args_dict
 
 pytestmark = [pytest.mark.core_model, pytest.mark.cpu]
@@ -253,29 +243,9 @@ def test_from_pipeline_config_rejects_unowned_deploy_engine_extras(engine_extras
     ],
     ids=["global", "ar-stage", "generation-stage"],
 )
-def test_from_pipeline_config_accepts_upstream_cache_dtype_for_llm_stages(cli_overrides, stage_id):
-    stage = _from_pipeline_key("qwen3_tts", cli_overrides=cli_overrides).stage_by_id(stage_id)
-
-    assert stage.cache_config.cache_dtype == "fp8"
-    assert "cache_dtype" in stage.cache_config._omni_explicit_fields
-
-
-@pytest.mark.parametrize("field_name", ["data_parallel_master_ip", "data_parallel_address"])
-def test_from_pipeline_config_normalizes_nested_parallel_config_aliases(field_name):
-    pipeline = _resolve_pipeline_or_skip("qwen3_tts")
-    deploy = DeployConfig(
-        stages=[
-            StageDeployConfig(
-                stage_id=0,
-                engine_extras={"parallel_config": {field_name: "10.0.0.1"}},
-            )
-        ]
-    )
-
-    stage = VllmOmniConfig.from_pipeline_config(pipeline, user_deploy_config=deploy).stage_by_id(0)
-
-    assert stage.parallel_config.data_parallel_master_ip == "10.0.0.1"
-    assert "data_parallel_master_ip" in stage.parallel_config._omni_explicit_fields
+def test_from_pipeline_config_rejects_diffusion_only_cli_fields_for_llm_stages(cli_overrides, stage_id):
+    with pytest.raises(ValueError, match=rf"Stage {stage_id} .*no structured config owner: kv_cache_dtype"):
+        _from_pipeline_key("qwen3_tts", cli_overrides=cli_overrides)
 
 
 def test_from_pipeline_config_accepts_diffusion_only_cli_fields_for_diffusion_stage():
@@ -467,8 +437,6 @@ def test_vllm_omni_stage_config_public_fields_use_typed_stage_realizations():
         "connector_config",
         "runtime_config",
         "parallel_config",
-        "compilation_config",
-        "profiler_config",
         "quantization_config",
     }
     assert "diffusion_config" not in public_fields
@@ -487,47 +455,10 @@ def test_runtime_config_fields_match_structured_runtime_scope():
         "num_gpus",
         "log_level",
         "log_stats",
+        "profiler_config",
     }
 
 
-def test_upstream_config_compatible_fields_materialize_mapping_inputs():
-    compilation_config = {"backend": "eager"}
-    profiler_config = {"profiler": "cuda"}
-
-    stage_config = VllmOmniARStageConfig(
-        stage_pipeline_config=_resolve_pipeline_or_skip("qwen3_tts").stages[0],
-        compilation_config=compilation_config,
-        profiler_config=profiler_config,
-    )
-
-    assert isinstance(stage_config.compilation_config, VllmCompilationConfig)
-    assert stage_config.compilation_config.backend == "eager"
-    assert isinstance(stage_config.profiler_config, VllmProfilerConfig)
-    assert stage_config.profiler_config.profiler == "cuda"
-
-
-def test_upstream_config_compatible_fields_validate_mapping_inputs():
-    with pytest.raises(ValidationError):
-        VllmOmniARStageConfig(
-            stage_pipeline_config=_resolve_pipeline_or_skip("qwen3_tts").stages[0],
-            profiler_config={"delay_iterations": -1},
-        )
-
-
-def test_upstream_config_compatible_fields_keep_prebuilt_config_types():
-    compilation_config = VllmCompilationConfig(backend="eager")
-    profiler_config = VllmProfilerConfig(profiler="cuda")
-
-    stage_config = VllmOmniARStageConfig(
-        stage_pipeline_config=_resolve_pipeline_or_skip("qwen3_tts").stages[0],
-        compilation_config=compilation_config,
-        profiler_config=profiler_config,
-    )
-
-    assert stage_config.compilation_config is compilation_config
-    assert stage_config.profiler_config is profiler_config
-
-
 def test_sub_config_fields_match_structured_scopes():
     assert {f.name for f in fields(OmniStageModelConfig)} == {
         "model",
@@ -558,6 +489,7 @@ def test_sub_config_fields_match_structured_scopes():
         "enforce_eager",
         "max_cudagraph_capture_size",
         "enable_flashinfer_autotune",
+        "compilation_config",
         "enable_multithread_weight_load",
         "num_weight_load_threads",
         "disable_autocast",
@@ -567,25 +499,29 @@ def test_sub_config_fields_match_structured_scopes():
         "model_subdir",
         "tokenizer_subdir",
     }
-    vllm_load_fields = {f.name for f in fields(VllmLoadConfig)}
-    assert issubclass(OmniStageLoadConfig, VllmLoadConfig)
-    assert {f.name for f in fields(OmniStageLoadConfig)} == vllm_load_fields | {
+    assert {f.name for f in fields(OmniStageLoadConfig)} == {
         "tokenizer",
+        "download_dir",
         "skip_tokenizer_init",
+        "load_format",
         "tokenizer_mode",
         "config_format",
         "skip_mm_profiling",
     }
-    assert OmniStageLoadConfig(load_format="PT").load_format == "pt"
-    assert issubclass(OmniStageCacheConfig, VllmCacheConfig)
-    assert {f.name for f in fields(OmniStageCacheConfig)} == {f.name for f in fields(VllmCacheConfig)} | {
+    assert {f.name for f in fields(OmniStageCacheConfig)} == {
+        "kv_cache_memory_bytes",
+        "gpu_memory_utilization",
+        "enable_prefix_caching",
         "disable_hybrid_kv_cache_manager",
         "mm_processor_cache_gb",
         "mamba_ssm_cache_dtype",
     }
-    assert issubclass(OmniStageSchedulerConfig, VllmSchedulerConfig)
-    assert {f.name for f in fields(OmniStageSchedulerConfig)} == {f.name for f in fields(VllmSchedulerConfig)} | {
+    assert {f.name for f in fields(OmniStageSchedulerConfig)} == {
+        "max_num_seqs",
+        "max_num_batched_tokens",
         "max_model_len",
+        "enable_chunked_prefill",
+        "async_scheduling",
     }
     assert {f.name for f in fields(OmniStageConnectorConfig)} == {
         "async_chunk",
@@ -594,10 +530,17 @@ def test_sub_config_fields_match_structured_scopes():
         "output_connectors",
         "input_connectors",
     }
-    vllm_parallel_fields = {f.name for f in fields(VllmParallelConfig)}
-    assert issubclass(OmniStageParallelConfig, VllmParallelConfig)
-    assert {f.name for f in fields(OmniStageParallelConfig)} == vllm_parallel_fields
-    assert {f.name for f in fields(OmniStageDiffusionParallelConfig)} == vllm_parallel_fields | {
+    assert {f.name for f in fields(OmniStageParallelConfig)} == {
+        "pipeline_parallel_size",
+        "data_parallel_size",
+        "tensor_parallel_size",
+        "enable_expert_parallel",
+        "world_size",
+    }
+    assert {f.name for f in fields(OmniStageDiffusionParallelConfig)} == {
+        "pipeline_parallel_size",
+        "data_parallel_size",
+        "tensor_parallel_size",
         "sequence_parallel_size",
         "ulysses_degree",
         "ring_degree",
@@ -611,86 +554,20 @@ def test_sub_config_fields_match_structured_scopes():
         "mask_sp_padding",
         "hsdp_shard_size",
         "hsdp_replicate_size",
+        "enable_expert_parallel",
+        "world_size",
     }
 
 
-def test_inherited_sub_configs_initialize_transport_safe_derived_fields():
-    scheduler_config = OmniStageSchedulerConfig(max_num_batched_tokens=4096)
-    assert scheduler_config.max_num_encoder_input_tokens == 4096
-    assert scheduler_config.encoder_cache_size == 4096
-
-    unresolved_scheduler_config = OmniStageSchedulerConfig()
-    assert unresolved_scheduler_config.max_num_encoder_input_tokens is None
-    assert unresolved_scheduler_config.encoder_cache_size is None
-
-    parallel_config = OmniStageParallelConfig(data_parallel_size=3, data_parallel_rank=2)
-    assert parallel_config.data_parallel_index == 2
-
-    diffusion_parallel_config = OmniStageDiffusionParallelConfig(data_parallel_size=3, data_parallel_rank=2)
-    assert diffusion_parallel_config.data_parallel_index == 2
-
-
-@pytest.mark.parametrize(
-    "config_cls",
-    [
-        OmniStageLoadConfig,
-        OmniStageCacheConfig,
-        OmniStageSchedulerConfig,
-        OmniStageParallelConfig,
-        OmniStageDiffusionParallelConfig,
-    ],
-)
-def test_inherited_sub_configs_are_keyword_only(config_cls):
-    assert all(parameter.kind is Parameter.KEYWORD_ONLY for parameter in signature(config_cls).parameters.values())
-
-    with pytest.raises(TypeError):
-        config_cls(1)
-
-
-@pytest.mark.parametrize(
-    "config_cls",
-    [
-        OmniStageLoadConfig,
-        OmniStageCacheConfig,
-        OmniStageSchedulerConfig,
-        OmniStageParallelConfig,
-        OmniStageDiffusionParallelConfig,
-    ],
-)
-def test_inherited_sub_configs_remain_msgpack_transport_safe(config_cls):
-    config = config_cls()
-
-    uninitialized_fields = [
-        config_field.name
-        for config_field in fields(config)
-        if isinstance(getattr(config, config_field.name), FieldInfo)
-    ]
-    assert uninitialized_fields == []
-
-    msgspec.msgpack.encode(_serialize_stage_config(config))
-
-
-def test_structured_llm_stage_registration_payloads_remain_msgpack_transport_safe():
-    omni_config = _from_pipeline_key("qwen3_tts")
-
-    for stage_config in omni_config.stage_configs:
-        payload = {
-            "stage_id": stage_config.stage_id,
-            "stage_config": _serialize_stage_config(stage_config),
-        }
-        msgspec.msgpack.encode(payload)
-
-
 def test_diffusion_parallel_config_fields_cover_legacy_surface():
     from vllm_omni.diffusion.data import DiffusionParallelConfig
 
     legacy_fields = {f.name for f in fields(DiffusionParallelConfig)}
     structured_fields = {f.name for f in fields(OmniStageDiffusionParallelConfig)}
     expected_upstream_fields = {"mask_sp_padding"}
-    vllm_parallel_fields = {f.name for f in fields(VllmParallelConfig)}
 
     assert legacy_fields | expected_upstream_fields <= structured_fields
-    assert structured_fields - legacy_fields - expected_upstream_fields - vllm_parallel_fields == set()
+    assert structured_fields - legacy_fields - expected_upstream_fields == {"world_size"}
 
 
 def test_diffusion_parallel_config_keeps_current_diffusion_parallel_surface():
@@ -733,7 +610,8 @@ def test_diffusion_parallel_config_matches_diffusion_parallel_world_size_for_vae
 
 def test_diffusion_parallel_config_supports_diffusion_hsdp_auto_sharding():
     cfg = OmniStageDiffusionParallelConfig(
-        ulysses_degree=4,
+        pipeline_parallel_size=2,
+        ulysses_degree=2,
         use_hsdp=True,
         hsdp_shard_size=-1,
         hsdp_replicate_size=2,
@@ -744,10 +622,10 @@ def test_diffusion_parallel_config_supports_diffusion_hsdp_auto_sharding():
 
 
 def test_diffusion_parallel_config_rejects_hsdp_with_tp_or_dp():
-    with pytest.raises(ValueError, match="not compatible with TP"):
+    with pytest.raises(ValueError, match="cannot be used with TP or DP"):
         OmniStageDiffusionParallelConfig(tensor_parallel_size=2, use_hsdp=True, hsdp_shard_size=2)
 
-    with pytest.raises(ValueError, match="not compatible with DP"):
+    with pytest.raises(ValueError, match="cannot be used with TP or DP"):
         OmniStageDiffusionParallelConfig(data_parallel_size=2, use_hsdp=True, hsdp_shard_size=2)
 
 
@@ -1200,21 +1078,3 @@ def test_diffusion_config_projection_keeps_mapping_quantization_config_serializa
     cfg = omni_config_module._DiffusionConfigProjection.from_kwargs(quantization_config=quantization_config)
 
     assert cfg.quantization_config == quantization_config
-
-
-def test_diffusion_quantization_mapping_reaches_terminal_config(monkeypatch):
-    from vllm_omni.diffusion.data import OmniDiffusionConfig
-
-    quantization_config = {"method": "int8", "activation_scheme": "dynamic"}
-    cfg = omni_config_module._DiffusionConfigProjection.from_kwargs(
-        quantization_config=quantization_config,
-    )
-
-    # Exercise the terminal construction performed by the future typed startup
-    # path without probing ports or loading remote model metadata.
-    monkeypatch.setattr(OmniDiffusionConfig, "_resolve_master_port", lambda _self: 29500)
-    monkeypatch.setattr(OmniDiffusionConfig, "enrich_config", lambda _self: None)
-    cfg.enrich_config()
-
-    assert cfg.quantization_config is not None
-    assert cfg.quantization_config.get_name() == "int8"

tests/dfx/reliability/README.md

@@ -7,10 +7,12 @@ The detailed RFC document is maintained outside this repository (local/internal
 - Qwen3 tests: `test_reliability_qwen3_omni.py`
 - Wan2.2 tests: `test_reliability_wan22.py`
 - HunyuanImage tests: `test_reliability_hunyuan_image.py`
+- VoxCPM2 tests: `test_reliability_voxcpm2.py`
 
 ```bash
 pytest --collect-only tests/dfx/reliability
 pytest -s -v tests/dfx/reliability/test_reliability_qwen3_omni.py -m slow
 pytest -s -v tests/dfx/reliability/test_reliability_wan22.py -m slow
 pytest -s -v tests/dfx/reliability/test_reliability_hunyuan_image.py -m slow
+pytest -s -v tests/dfx/reliability/test_reliability_voxcpm2.py -m slow
 ```

tests/dfx/reliability/invalid_param_test/test_invalid_audio_speech.py

@@ -976,68 +976,3 @@ def test_voices_delete_invalid_requests(
             "err_message": err_message,
         }
     )
-
-
-# ─── POST /v1/audio/speech · MiniMax Music 3 ───
-
-# Text-to-music on the speech endpoint. Most of the speech contract does not
-# apply: there is no speaker, no reference audio and no temperature, and the
-# lyrics alone are not a request. Each case below pins a rejection so a caller
-# who assumes the usual TTS shape finds out immediately.
-_MINIMAX_MUSIC3_SPEECH = [
-    pytest.param(
-        OmniServerParams(
-            model="MiniMaxAI/MiniMax-Music3",
-            stage_config_path=get_deploy_config_path("minimax_music3.yaml"),
-            server_args=_SPEECH_SERVER_ARGS,
-        ),
-        id="minimax_music3",
-        marks=hardware_marks(res={"cuda": "H100"}),
-    ),
-]
-
-_MINIMAX_MUSIC3_LYRICS = "[Verse]\nWalking down the empty street at midnight"
-_MINIMAX_MUSIC3_CAPTION = "A melancholic lo-fi hip-hop track at 85 BPM in F minor."
-# 25 frames per second, capped at six minutes of song.
-_MINIMAX_MUSIC3_MAX_FRAMES = 9000
-
-
-@pytest.mark.parametrize(
-    "overrides, err_message",
-    [
-        pytest.param({"instructions": None}, ("instructions",), id="caption_missing"),
-        pytest.param({"instructions": "   "}, ("instructions",), id="caption_blank"),
-        pytest.param({"input": "   "}, ("input",), id="lyrics_blank"),
-        pytest.param({"voice": "alloy"}, ("voice",), id="voice_rejected"),
-        pytest.param({"speed": 1.5}, ("speed",), id="speed_rejected"),
-        pytest.param(
-            {"max_new_tokens": _MINIMAX_MUSIC3_MAX_FRAMES + 1},
-            ("max_new_tokens", str(_MINIMAX_MUSIC3_MAX_FRAMES)),
-            id="max_new_tokens_over_cap",
-        ),
-        pytest.param({"max_new_tokens": 0}, ("max_new_tokens",), id="max_new_tokens_zero"),
-    ],
-)
-@pytest.mark.parametrize("omni_server", _MINIMAX_MUSIC3_SPEECH, indirect=True)
-def test_minimax_music3_speech_invalid_field_values(
-    omni_server: OmniServer,
-    openai_client: OpenAIClientHandler,
-    overrides: dict[str, object],
-    err_message: str | tuple[str, ...],
-) -> None:
-    body: dict[str, Any] = {
-        "model": omni_server.model,
-        "input": _MINIMAX_MUSIC3_LYRICS,
-        "instructions": _MINIMAX_MUSIC3_CAPTION,
-        "seed": 1,
-        "max_new_tokens": 250,
-        "response_format": "wav",
-    }
-    for key, value in overrides.items():
-        if value is None:
-            body.pop(key, None)
-        else:
-            body[key] = value
-    openai_client.send_audio_speech_http_request(
-        {"json": body, "timeout": 120, "err_code": 400, "err_message": err_message}
-    )

tests/dfx/reliability/test_reliability_qwen3_omni.py

@@ -71,7 +71,7 @@ def _default_oom_device_spec() -> str:
 
 OOM_INJECTION_CONFIG = {
     "device": _default_oom_device_spec(),
-    "target_mem_ratio": 0.95,
+    "target_mem_ratio": 0.92,
     "hold_seconds": 0,
     "startup_timeout_sec": 20,
     "strict": False,

tests/dfx/reliability/test_reliability_voxcpm2.py

@@ -0,0 +1,508 @@
+"""
+VoxCPM2 reliability integration tests.
+
+Scenarios follow ``docs/user_guide/fault_injection_reliability_matrix.md``.
+"""
+
+from __future__ import annotations
+
+import concurrent.futures
+import os
+import time
+from pathlib import Path
+from typing import Any
+
+import pytest
+
+from tests.dfx.conftest import (
+    assert_fault_exception,
+    create_reliability_omni_server_params,
+    resolve_oom_device_spec,
+)
+from tests.dfx.reliability.helpers import (
+    FaultInjector,
+    assert_no_server_tree_process_residual_and_gpu_release,
+    extract_openai_error_contract_from_bytes,
+    get_health_raw,
+    inject_gpu_oom,
+    make_process_kill_fault_injector,
+    make_server_root_kill_fault_injector,
+    make_server_tree_kill_fault_injector,
+    post_json_raw,
+    run_fault_injection_with_rate_load,
+    stop_gpu_oom_hogs,
+    worker_residual_timeout_after_kill_signal,
+)
+from tests.helpers.mark import hardware_test
+from tests.helpers.media import load_test_audio_data_url
+from tests.helpers.runtime import OpenAIClientHandler
+
+RELIABILITY_SCENARIOS: list[dict[str, Any]] = [
+    {
+        "test_name": "voxcpm2_reliability_default",
+        "server_params": {
+            "model": "openbmb/VoxCPM2",
+            "stage_config_name": "voxcpm2.yaml",
+            # VoxCPM2 uses a custom HF tokenizer; CLI default trust_remote_code=False
+            # would otherwise override deploy YAML settings during server startup.
+            "server_args": ["--trust-remote-code", "--disable-log-stats"],
+        },
+    },
+]
+
+DEPLOY_CONFIGS_DIR = Path(__file__).resolve().parent.parent.parent.parent / "vllm_omni" / "deploy"
+OOM_INJECTION_CONFIG = {
+    "target_mem_ratio": 0.96,
+    "hold_seconds": 0,
+    "startup_timeout_sec": 20,
+    "strict": False,
+}
+FAULT_ERROR_KEYWORDS = (
+    "the request failed",
+    "oom",
+    "out of memory",
+    "cuda",
+    "orchestrator",
+    "timeout",
+    "connection",
+    "500",
+    "503",
+)
+PROCESS_KILL_ERROR_KEYWORDS = (
+    "timeout",
+    "did not complete within",
+    "connection",
+    "engine",
+    "orchestrator",
+    "dead",
+    "internal",
+    "500",
+    "503",
+)
+RUNTIME_WORKER_PATTERN = "VLLM::"
+SERVE_SIGNAL_PARAMS = [
+    pytest.param("SIGTERM", id="sigterm"),
+    pytest.param("SIGINT", id="sigint"),
+    pytest.param("SIGKILL", id="sigkill"),
+]
+SERVE_WITH_LOAD_SIGNAL_PARAMS = [
+    pytest.param("SIGTERM", id="sigterm"),
+    pytest.param("SIGINT", id="sigint"),
+    pytest.param("SIGKILL", id="sigkill"),
+]
+TREE_SIGNAL_PARAMS = [
+    pytest.param("SIGTERM", id="sigterm"),
+    pytest.param("SIGKILL", id="sigkill"),
+]
+TREE_WITH_LOAD_SIGNAL_PARAMS = [
+    pytest.param("SIGTERM", id="sigterm"),
+    pytest.param("SIGKILL", id="sigkill"),
+]
+WORKER_SIGNAL_FAULT_PARAMS = [
+    pytest.param(
+        make_process_kill_fault_injector(
+            grep_patterns=RUNTIME_WORKER_PATTERN,
+            signal_name="SIGTERM",
+            limit=1,
+            post_kill_wait_seconds=2.0,
+        ),
+        id="runtime_process_chain_sigterm",
+    ),
+    pytest.param(
+        make_process_kill_fault_injector(
+            grep_patterns=RUNTIME_WORKER_PATTERN,
+            signal_name="SIGKILL",
+            limit=1,
+            post_kill_wait_seconds=2.0,
+        ),
+        id="runtime_process_chain_sigkill",
+    ),
+]
+
+VOXCPM2_PARAMS = create_reliability_omni_server_params(RELIABILITY_SCENARIOS, DEPLOY_CONFIGS_DIR)
+INFLIGHT_INJECTION_REQUEST_RATE = 0.3
+INFLIGHT_INJECTION_REQUEST_COUNT = 10
+_SPEECH_FF_HTTP_TIMEOUT_SEC = 30
+_SPEECH_FF_MAX_ELAPSED_SEC = 30
+
+
+def _speech_request_config(omni_server: Any) -> dict[str, Any]:
+    return {
+        "model": omni_server.model,
+        "input": "The weather is nice today, perfect for a walk in the park.",
+        "stream": False,
+        "timeout": 120.0,
+        "response_format": "wav",
+        "voice": "default",
+    }
+
+
+def _submit_speech_request(openai_client: OpenAIClientHandler, omni_server: Any) -> None:
+    openai_client.send_audio_speech_request(_speech_request_config(omni_server), request_num=1)
+
+
+def _oom_speech_request_config(omni_server: Any) -> dict[str, Any]:
+    base = "The weather is nice today, perfect for a walk in the park. "
+    return {
+        "model": omni_server.model,
+        "input": base * 80,
+        "max_new_tokens": 4096,
+        "ref_audio": load_test_audio_data_url("qwen3_tts/clone_2.wav"),
+        "stream": False,
+        "timeout": 300.0,
+        "response_format": "wav",
+        "voice": "default",
+    }
+
+
+def _submit_oom_speech_request(openai_client: OpenAIClientHandler, omni_server: Any) -> None:
+    openai_client.send_audio_speech_request(_oom_speech_request_config(omni_server), request_num=1)
+
+
+def _assert_post_fault_speech_fast_fail(host: str, port: int, *, model: str, scenario: str) -> None:
+    payload = {
+        "model": model,
+        "input": "post-fault fast-fail check",
+        "voice": "default",
+    }
+    start = time.monotonic()
+    try:
+        status, body = post_json_raw(host, port, "/v1/audio/speech", payload, timeout_sec=20)
+        elapsed = time.monotonic() - start
+        assert elapsed < 15, f"[{scenario} fast_fail] /v1/audio/speech did not fail fast: {elapsed:.2f}s"
+        assert status >= 500, (
+            f"[{scenario} fast_fail] expected server-side error after fault, got status={status}, body={body[:200]!r}"
+        )
+    except Exception:  # noqa: BLE001
+        elapsed = time.monotonic() - start
+        assert elapsed < 15, f"[{scenario} fast_fail] exception was too slow after fault: {elapsed:.2f}s"
+
+
+def _assert_post_fault_health_terminal(host: str, port: int, *, scenario: str) -> None:
+    deadline = time.monotonic() + 20.0
+    last_observation = ""
+    while time.monotonic() < deadline:
+        try:
+            status, body = get_health_raw(host, port, timeout_sec=5)
+            last_observation = f"http={status}, body={body[:200]!r}"
+            if status == 503:
+                return
+        except Exception as exc:  # noqa: BLE001
+            last_observation = f"exception={exc!r}"
+            return
+        time.sleep(0.5)
+    pytest.fail(f"[{scenario} health] no terminal post-fault health observed: {last_observation}")
+
+
+@pytest.mark.slow
+@pytest.mark.tts
+@pytest.mark.skip(reason="issue#4285")
+@hardware_test(res={"cuda": "L4"}, num_cards=1)
+@pytest.mark.parametrize("omni_server_function", VOXCPM2_PARAMS, indirect=True)
+def test_reliability_fault_gpu_oom_speech_request_failure(
+    omni_server_function,
+    openai_client_function,
+) -> None:
+    stage_config_path = getattr(omni_server_function, "stage_config_path", None)
+    device_spec = resolve_oom_device_spec(OOM_INJECTION_CONFIG, stage_config_path)
+    handle = inject_gpu_oom(
+        device=device_spec,
+        target_mem_ratio=OOM_INJECTION_CONFIG["target_mem_ratio"],
+        hold_seconds=OOM_INJECTION_CONFIG["hold_seconds"],
+        startup_timeout_sec=OOM_INJECTION_CONFIG["startup_timeout_sec"],
+        strict=OOM_INJECTION_CONFIG["strict"],
+    )
+    try:
+        try:
+            _submit_oom_speech_request(openai_client_function, omni_server_function)
+        except Exception as exc:
+            assert_fault_exception(exc, FAULT_ERROR_KEYWORDS)
+        else:
+            pytest.fail("expected /v1/audio/speech request failure during GPU OOM injection")
+    finally:
+        stop_gpu_oom_hogs(handle)
+
+
+@pytest.mark.slow
+@pytest.mark.tts
+@pytest.mark.skip(reason="issue#4285")
+@hardware_test(res={"cuda": "L4"}, num_cards=1)
+@pytest.mark.parametrize("omni_server_function", VOXCPM2_PARAMS, indirect=True)
+def test_reliability_fault_gpu_oom_speech_error_contract(
+    omni_server_function,
+) -> None:
+    stage_config_path = getattr(omni_server_function, "stage_config_path", None)
+    device_spec = resolve_oom_device_spec(OOM_INJECTION_CONFIG, stage_config_path)
+    handle = inject_gpu_oom(
+        device=device_spec,
+        target_mem_ratio=OOM_INJECTION_CONFIG["target_mem_ratio"],
+        hold_seconds=OOM_INJECTION_CONFIG["hold_seconds"],
+        startup_timeout_sec=OOM_INJECTION_CONFIG["startup_timeout_sec"],
+        strict=OOM_INJECTION_CONFIG["strict"],
+    )
+    host = omni_server_function.host
+    port = omni_server_function.port
+    oom_config = _oom_speech_request_config(omni_server_function)
+    payload = {
+        "model": oom_config["model"],
+        "input": oom_config["input"],
+        "voice": oom_config["voice"],
+        "max_new_tokens": oom_config["max_new_tokens"],
+        "ref_audio": oom_config["ref_audio"],
+    }
+    try:
+        status, body = post_json_raw(host, port, "/v1/audio/speech", payload, timeout_sec=300)
+    finally:
+        stop_gpu_oom_hogs(handle)
+
+    assert status >= 500, f"expected speech runtime error under OOM, got status={status}"
+    speech_error = extract_openai_error_contract_from_bytes(body)
+    assert speech_error is not None, f"speech error payload not OpenAI-compatible: {body[:300]!r}"
+    assert "code" in speech_error, f"speech error lacks code field: {speech_error!r}"
+
+
+@pytest.mark.slow
+@pytest.mark.tts
+@pytest.mark.skipif(os.name == "nt", reason="process-kill injection helper is POSIX-only")
+@pytest.mark.parametrize(
+    "fault_injector",
+    WORKER_SIGNAL_FAULT_PARAMS,
+    indirect=True,
+)
+@pytest.mark.parametrize("omni_server_function", VOXCPM2_PARAMS, indirect=True)
+def test_reliability_fault_process_kill_request_failure(
+    omni_server_after_fault_function,
+    openai_client_function,
+) -> None:
+    try:
+        _submit_speech_request(openai_client_function, omni_server_after_fault_function)
+    except Exception as exc:
+        assert_fault_exception(exc, PROCESS_KILL_ERROR_KEYWORDS)
+    else:
+        pytest.fail("expected /v1/audio/speech request failure after process-kill injection")
+
+
+@pytest.mark.slow
+@pytest.mark.tts
+@pytest.mark.skipif(os.name == "nt", reason="process-kill injection helper is POSIX-only")
+@pytest.mark.parametrize(
+    "fault_injector",
+    WORKER_SIGNAL_FAULT_PARAMS,
+    indirect=True,
+)
+@pytest.mark.parametrize("omni_server_function", VOXCPM2_PARAMS, indirect=True)
+def test_reliability_fault_process_kill_fast_fail_and_concurrent(
+    omni_server_after_fault_function,
+    openai_client_function,
+) -> None:
+    host = omni_server_after_fault_function.host
+    port = omni_server_after_fault_function.port
+    model = omni_server_after_fault_function.model
+    payload = {
+        "model": model,
+        "input": "Say hello in one short sentence.",
+        "voice": "default",
+    }
+
+    start = time.monotonic()
+    try:
+        status, body = post_json_raw(
+            host,
+            port,
+            "/v1/audio/speech",
+            payload,
+            timeout_sec=_SPEECH_FF_HTTP_TIMEOUT_SEC,
+        )
+        elapsed = time.monotonic() - start
+        assert elapsed <= _SPEECH_FF_MAX_ELAPSED_SEC + 1.0, (
+            f"[process_kill fast_fail] /v1/audio/speech did not fail fast after fault: {elapsed:.2f}s"
+        )
+        assert status >= 500, (
+            "[process_kill fast_fail] expected server-side failure after fault, "
+            f"got status={status}, body={body[:200]!r}"
+        )
+    except Exception:
+        elapsed = time.monotonic() - start
+        assert elapsed <= _SPEECH_FF_MAX_ELAPSED_SEC + 1.0, (
+            f"[process_kill fast_fail] /v1/audio/speech exception was too slow after fault: {elapsed:.2f}s"
+        )
+
+    start = time.monotonic()
+    with concurrent.futures.ThreadPoolExecutor(max_workers=3) as executor:
+        futures = [
+            executor.submit(_submit_speech_request, openai_client_function, omni_server_after_fault_function)
+            for _ in range(3)
+        ]
+        done, pending = concurrent.futures.wait(
+            futures,
+            timeout=30,
+            return_when=concurrent.futures.ALL_COMPLETED,
+        )
+
+    elapsed = time.monotonic() - start
+    assert not pending, f"[process_kill concurrent] some fault-time speech requests hung: pending={len(pending)}"
+    assert elapsed < 30, f"[process_kill concurrent] fault-time speech request convergence is too slow: {elapsed:.2f}s"
+
+    fault_observed = False
+    for future in done:
+        try:
+            future.result()
+        except Exception as exc:
+            fault_observed = True
+            assert_fault_exception(exc, PROCESS_KILL_ERROR_KEYWORDS)
+    assert fault_observed, (
+        "[process_kill concurrent] expected at least one /v1/audio/speech request to fail after fault"
+    )
+
+
+@pytest.mark.slow
+@pytest.mark.tts
+@pytest.mark.skipif(os.name == "nt", reason="process-kill injection helper is POSIX-only")
+@pytest.mark.parametrize(
+    "fault_injector",
+    WORKER_SIGNAL_FAULT_PARAMS,
+    indirect=True,
+)
+@pytest.mark.parametrize("omni_server_function", VOXCPM2_PARAMS, indirect=True)
+def test_reliability_fault_process_kill_worker_with_load_request_failure(
+    omni_server_function,
+    openai_client_function,
+    fault_injector: FaultInjector,
+) -> None:
+    scenario = "kill_worker_with_load"
+    load_result = run_fault_injection_with_rate_load(
+        submit_request=lambda: _submit_speech_request(openai_client_function, omni_server_function),
+        inject_fault=lambda: fault_injector(omni_server_function),
+        num_requests=INFLIGHT_INJECTION_REQUEST_COUNT,
+        request_rate=INFLIGHT_INJECTION_REQUEST_RATE,
+        completion_timeout_sec=120.0,
+    )
+    assert load_result["failure_observed"], (
+        f"[{scenario}] expected at least one load request failure after fault; load_result={load_result}"
+    )
+    host = omni_server_function.host
+    port = omni_server_function.port
+    _assert_post_fault_speech_fast_fail(host, port, model=omni_server_function.model, scenario=scenario)
+    _assert_post_fault_health_terminal(host, port, scenario=scenario)
+
+
+@pytest.mark.slow
+@pytest.mark.tts
+@hardware_test(res={"cuda": "L4"}, num_cards=1)
+@pytest.mark.skipif(os.name == "nt", reason="process-kill injection helper is POSIX-only")
+@pytest.mark.parametrize("signal_name", SERVE_SIGNAL_PARAMS)
+@pytest.mark.parametrize("omni_server_function", VOXCPM2_PARAMS, indirect=True)
+def test_reliability_fault_process_kill_serve_root_no_load_fast_fail_and_cleanup(
+    omni_server_function,
+    signal_name: str,
+) -> None:
+    scenario = f"kill_serve_root_no_load_{signal_name.lower()}"
+    injector = make_server_root_kill_fault_injector(signal_name=signal_name, post_kill_wait_seconds=2.0)
+    injector(omni_server_function)
+    host = omni_server_function.host
+    port = omni_server_function.port
+    _assert_post_fault_speech_fast_fail(host, port, model=omni_server_function.model, scenario=scenario)
+    _assert_post_fault_health_terminal(host, port, scenario=scenario)
+    assert_no_server_tree_process_residual_and_gpu_release(
+        omni_server_function,
+        scenario=scenario,
+        timeout_sec=worker_residual_timeout_after_kill_signal(signal_name),
+    )
+
+
+@pytest.mark.slow
+@pytest.mark.tts
+@hardware_test(res={"cuda": "L4"}, num_cards=1)
+@pytest.mark.skipif(os.name == "nt", reason="process-kill injection helper is POSIX-only")
+@pytest.mark.parametrize("signal_name", SERVE_WITH_LOAD_SIGNAL_PARAMS)
+@pytest.mark.parametrize("omni_server_function", VOXCPM2_PARAMS, indirect=True)
+def test_reliability_fault_process_kill_serve_root_with_load_fast_fail_and_cleanup(
+    omni_server_function,
+    openai_client_function,
+    signal_name: str,
+) -> None:
+    scenario = f"kill_serve_root_with_load_{signal_name.lower()}"
+    injector = make_server_root_kill_fault_injector(signal_name=signal_name, post_kill_wait_seconds=2.0)
+    load_result = run_fault_injection_with_rate_load(
+        submit_request=lambda: _submit_speech_request(openai_client_function, omni_server_function),
+        inject_fault=lambda: injector(omni_server_function),
+        num_requests=INFLIGHT_INJECTION_REQUEST_COUNT,
+        request_rate=INFLIGHT_INJECTION_REQUEST_RATE,
+        completion_timeout_sec=300.0,
+    )
+    assert load_result["failure_observed"], (
+        f"[{scenario}] expected at least one load request failure after fault; load_result={load_result}"
+    )
+    host = omni_server_function.host
+    port = omni_server_function.port
+    _assert_post_fault_speech_fast_fail(host, port, model=omni_server_function.model, scenario=scenario)
+    _assert_post_fault_health_terminal(host, port, scenario=scenario)
+    assert_no_server_tree_process_residual_and_gpu_release(
+        omni_server_function,
+        scenario=scenario,
+        timeout_sec=worker_residual_timeout_after_kill_signal(signal_name),
+    )
+
+
+@pytest.mark.slow
+@pytest.mark.tts
+@pytest.mark.skipif(os.name == "nt", reason="process-kill injection helper is POSIX-only")
+@pytest.mark.parametrize("signal_name", TREE_SIGNAL_PARAMS)
+@pytest.mark.parametrize("omni_server_function", VOXCPM2_PARAMS, indirect=True)
+def test_reliability_fault_process_kill_tree_no_load_fast_fail_and_cleanup(
+    omni_server_function,
+    signal_name: str,
+) -> None:
+    scenario = f"kill_serve_tree_no_load_{signal_name.lower()}"
+    injector = make_server_tree_kill_fault_injector(
+        signal_name=signal_name,
+        post_kill_wait_seconds=2.0,
+        inter_kill_wait_seconds=0.1,
+    )
+    injector(omni_server_function)
+    host = omni_server_function.host
+    port = omni_server_function.port
+    _assert_post_fault_speech_fast_fail(host, port, model=omni_server_function.model, scenario=scenario)
+    _assert_post_fault_health_terminal(host, port, scenario=scenario)
+    assert_no_server_tree_process_residual_and_gpu_release(
+        omni_server_function,
+        scenario=scenario,
+        timeout_sec=worker_residual_timeout_after_kill_signal(signal_name),
+    )
+
+
+@pytest.mark.slow
+@pytest.mark.tts
+@pytest.mark.skipif(os.name == "nt", reason="process-kill injection helper is POSIX-only")
+@pytest.mark.parametrize("signal_name", TREE_WITH_LOAD_SIGNAL_PARAMS)
+@pytest.mark.parametrize("omni_server_function", VOXCPM2_PARAMS, indirect=True)
+def test_reliability_fault_process_kill_tree_with_load_fast_fail_and_cleanup(
+    omni_server_function,
+    openai_client_function,
+    signal_name: str,
+) -> None:
+    scenario = f"kill_serve_tree_with_load_{signal_name.lower()}"
+    injector = make_server_tree_kill_fault_injector(
+        signal_name=signal_name,
+        post_kill_wait_seconds=2.0,
+        inter_kill_wait_seconds=0.1,
+    )
+    load_result = run_fault_injection_with_rate_load(
+        submit_request=lambda: _submit_speech_request(openai_client_function, omni_server_function),
+        inject_fault=lambda: injector(omni_server_function),
+        num_requests=INFLIGHT_INJECTION_REQUEST_COUNT,
+        request_rate=INFLIGHT_INJECTION_REQUEST_RATE,
+        completion_timeout_sec=120.0,
+    )
+    assert load_result["failure_observed"], (
+        f"[{scenario}] expected at least one load request failure after fault; load_result={load_result}"
+    )
+    host = omni_server_function.host
+    port = omni_server_function.port
+    _assert_post_fault_speech_fast_fail(host, port, model=omni_server_function.model, scenario=scenario)
+    _assert_post_fault_health_terminal(host, port, scenario=scenario)
+    assert_no_server_tree_process_residual_and_gpu_release(
+        omni_server_function,
+        scenario=scenario,
+        timeout_sec=worker_residual_timeout_after_kill_signal(signal_name),
+    )

tests/dfx/stability/scripts/test_stability_voxcpm2.py

@@ -0,0 +1,69 @@
+"""
+VoxCPM2 stability: OmniServer + ``vllm bench serve --omni`` for a fixed duration.
+
+Configuration: ``tests/dfx/stability/tests/test_voxcpm2.json``.
+"""
+
+from __future__ import annotations
+
+from pathlib import Path
+
+import pytest
+
+from tests.dfx.conftest import (
+    create_benchmark_indices,
+    create_test_parameter_mapping,
+    create_unique_server_params,
+    load_configs,
+)
+from tests.dfx.stability.helpers import _run_one_vllm_bench_batch, run_stability_benchmark_loop
+
+STABILITY_DIR = Path(__file__).resolve().parent.parent
+DEPLOY_CONFIGS_DIR = STABILITY_DIR / "deploy"
+CONFIG_FILE_PATH = str(STABILITY_DIR / "tests" / "test_voxcpm2.json")
+DEFAULT_NUM_PROMPTS_PER_BATCH = 20
+STABILITY_SERVER_TIMEOUT_ARGS = ["--stage-init-timeout", "600"]
+
+try:
+    BENCHMARK_CONFIGS = load_configs(CONFIG_FILE_PATH)
+except FileNotFoundError:
+    BENCHMARK_CONFIGS = []
+
+test_params = create_unique_server_params(BENCHMARK_CONFIGS, DEPLOY_CONFIGS_DIR) if BENCHMARK_CONFIGS else []
+server_to_benchmark_mapping = create_test_parameter_mapping(BENCHMARK_CONFIGS) if BENCHMARK_CONFIGS else {}
+benchmark_indices = create_benchmark_indices(BENCHMARK_CONFIGS, server_to_benchmark_mapping)
+
+
+@pytest.mark.slow
+@pytest.mark.tts
+@pytest.mark.parametrize("omni_server", test_params, indirect=True)
+@pytest.mark.parametrize("stability_benchmark_params", benchmark_indices, indirect=True)
+def test_stability_voxcpm2(omni_server, stability_benchmark_params):
+    test_name = stability_benchmark_params["test_name"]
+    params = stability_benchmark_params["params"]
+    duration_sec = params.get("duration_sec", 300)
+    num_prompts_per_batch = params.get("num_prompts_per_batch", DEFAULT_NUM_PROMPTS_PER_BATCH)
+    request_rate = params.get("request_rate")
+    max_concurrency = params.get("max_concurrency")
+
+    bench_params = {
+        k: v
+        for k, v in params.items()
+        if k not in ("duration_sec", "request_rate", "max_concurrency", "num_prompts_per_batch")
+    }
+
+    result = run_stability_benchmark_loop(
+        host=omni_server.host,
+        port=omni_server.port,
+        model=omni_server.model,
+        duration_sec=duration_sec,
+        params=bench_params,
+        request_rate=request_rate,
+        max_concurrency=max_concurrency,
+        result_dir=str(STABILITY_DIR),
+        num_prompts_per_batch=num_prompts_per_batch,
+        run_one_batch=_run_one_vllm_bench_batch,
+    )
+
+    assert result.get("failed", 0) == 0, f"[{test_name}] Failed requests detected: {result.get('errors', [])}"
+    assert result.get("completed", 0) > 0, f"[{test_name}] No requests completed"

tests/diffusion/diffusion_kv/test_initialization.py

@@ -1,297 +0,0 @@
-# SPDX-License-Identifier: Apache-2.0
-# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
-
-from types import SimpleNamespace
-from unittest.mock import Mock
-
-import pytest
-import torch
-from vllm.v1.kv_cache_interface import FullAttentionSpec
-
-import vllm_omni.diffusion.diffusion_kv.initialization as initialization
-import vllm_omni.diffusion.vllm_config as diffusion_vllm_config
-from vllm_omni.diffusion.diffusion_kv.config import DiffusionKVCacheMode
-from vllm_omni.diffusion.sched.request_scheduler import RequestScheduler
-from vllm_omni.quantization import build_quant_config
-
-pytestmark = [pytest.mark.core_model, pytest.mark.cpu, pytest.mark.diffusion]
-
-
-class _Executor:
-    def __init__(self) -> None:
-        self.specs = [{"layer0": object()}, {"layer0": object()}]
-        self.memory = [1024, 1024]
-        self.configs = None
-        self.profile_requests = None
-
-    def get_kv_cache_specs(self):
-        return self.specs
-
-    def determine_available_kv_memory(self, profile_requests):
-        self.profile_requests = profile_requests
-        return self.memory
-
-    def set_kv_cache_configs(self, configs) -> None:
-        self.configs = configs
-
-
-def _od_config(**overrides):
-    values = dict(
-        model="test-diffusion",
-        diffusion_kv_mode=DiffusionKVCacheMode.PAGED_SCHEDULER,
-        dtype=torch.bfloat16,
-        max_model_len=64,
-        quantization_config=None,
-        tf_model_config=None,
-        enforce_eager=True,
-        is_moe=False,
-        additional_config={},
-        profiler_config=None,
-        parallel_config=SimpleNamespace(
-            tensor_parallel_size=1,
-            pipeline_parallel_size=1,
-            data_parallel_size=1,
-            enable_expert_parallel=False,
-        ),
-        gpu_memory_utilization=0.9,
-        kv_cache_memory_bytes=None,
-        max_num_batched_tokens=64,
-        num_gpus=1,
-    )
-    values.update(overrides)
-    return SimpleNamespace(**values)
-
-
-def test_control_plane_builds_worker_and_scheduler_configs(monkeypatch) -> None:
-    executor = _Executor()
-    od_config = _od_config(num_gpus=2)
-    vllm_config = SimpleNamespace(
-        model_config=SimpleNamespace(max_model_len=4096),
-        max_in_flight_tokens=128,
-    )
-    worker_configs = [object(), object()]
-    scheduler_kv_cache_config = object()
-    profile_requests = [object()]
-    monkeypatch.setattr(initialization, "create_diffusion_vllm_config", lambda *args, **kwargs: vllm_config)
-    build = Mock(return_value=(worker_configs, scheduler_kv_cache_config))
-    monkeypatch.setattr(
-        initialization,
-        "build_native_kv_cache_configs",
-        build,
-    )
-    monkeypatch.setattr(initialization, "resolve_kv_cache_block_sizes", lambda *args: (16, 16))
-
-    result = initialization.initialize_diffusion_kv_control_plane(
-        executor,
-        od_config,
-        profile_requests=profile_requests,
-        device=torch.device("cpu"),
-    )
-
-    build.assert_called_once_with(vllm_config, executor.specs, executor.memory)
-    assert result == (scheduler_kv_cache_config, 16, 16, vllm_config)
-    assert executor.configs == worker_configs
-    assert executor.profile_requests is profile_requests
-
-
-def test_dense_mode_skips_worker_kv_initialization(monkeypatch) -> None:
-    executor = _Executor()
-    od_config = _od_config(diffusion_kv_mode=DiffusionKVCacheMode.DENSE_LEGACY)
-    get_device = Mock(side_effect=AssertionError("dense mode must not resolve a KV device"))
-    monkeypatch.setattr(initialization.current_omni_platform, "get_torch_device", get_device)
-
-    result = initialization.initialize_diffusion_kv_control_plane(executor, od_config)
-
-    assert result is None
-    assert executor.configs is None
-    get_device.assert_not_called()
-
-
-def test_control_plane_rejects_rank_count_mismatch(monkeypatch) -> None:
-    executor = _Executor()
-    executor.memory = [1024]
-    od_config = _od_config(num_gpus=2)
-    monkeypatch.setattr(initialization, "create_diffusion_vllm_config", lambda *args, **kwargs: object())
-
-    with pytest.raises(ValueError, match="rank count mismatch"):
-        initialization.initialize_diffusion_kv_control_plane(
-            executor,
-            od_config,
-            profile_requests=[object()],
-            device=torch.device("cpu"),
-        )
-
-
-def test_paged_control_plane_requires_prepared_profile_request() -> None:
-    with pytest.raises(ValueError, match="prepared memory profile requests"):
-        initialization.initialize_diffusion_kv_control_plane(
-            _Executor(),
-            _od_config(),
-            device=torch.device("cpu"),
-        )
-
-
-def test_native_config_builder_produces_matching_worker_and_scheduler_pools() -> None:
-    od_config = _od_config()
-    vllm_config = diffusion_vllm_config.create_diffusion_vllm_config(torch.device("cpu"), od_config)
-    spec = FullAttentionSpec(
-        block_size=vllm_config.cache_config.block_size,
-        num_kv_heads=2,
-        head_size=8,
-        dtype=torch.bfloat16,
-        non_causal=True,
-    )
-    num_blocks = 32
-
-    worker_configs, scheduler_kv_cache_config = initialization.build_native_kv_cache_configs(
-        vllm_config,
-        [{"layer0": spec}],
-        [spec.page_size_bytes * num_blocks],
-    )
-
-    assert len(worker_configs) == 1
-    assert worker_configs[0].num_blocks == num_blocks
-    assert scheduler_kv_cache_config.num_blocks == num_blocks
-    assert worker_configs[0].kv_cache_groups == scheduler_kv_cache_config.kv_cache_groups
-
-
-def test_native_control_plane_config_initializes_scheduler() -> None:
-    od_config = _od_config()
-    vllm_config = diffusion_vllm_config.create_diffusion_vllm_config(torch.device("cpu"), od_config)
-    spec = FullAttentionSpec(
-        block_size=vllm_config.cache_config.block_size,
-        num_kv_heads=2,
-        head_size=8,
-        dtype=torch.bfloat16,
-        non_causal=True,
-    )
-    _, scheduler_kv_cache_config = initialization.build_native_kv_cache_configs(
-        vllm_config,
-        [{"layer0": spec}],
-        [spec.page_size_bytes * 32],
-    )
-
-    scheduler = RequestScheduler()
-    scheduler.initialize(
-        od_config,
-        kv_cache_config=scheduler_kv_cache_config,
-        scheduler_block_size=spec.block_size,
-        hash_block_size=spec.block_size,
-        kv_vllm_config=vllm_config,
-    )
-
-    assert scheduler._diffusion_kv_manager is not None
-    scheduler.close()
-
-
-def test_dense_config_keeps_new_paged_sizing_inputs_inactive() -> None:
-    od_config = _od_config(
-        diffusion_kv_mode=DiffusionKVCacheMode.DENSE_LEGACY,
-        parallel_config=SimpleNamespace(
-            tensor_parallel_size=1,
-            pipeline_parallel_size=4,
-            data_parallel_size=1,
-            enable_expert_parallel=False,
-        ),
-        gpu_memory_utilization=0.5,
-        kv_cache_memory_bytes=4096,
-        max_num_batched_tokens=64,
-    )
-    vllm_config = diffusion_vllm_config.create_base_diffusion_vllm_config(torch.device("cpu"), od_config)
-    original_pp_size = vllm_config.parallel_config.pipeline_parallel_size
-    original_memory_utilization = vllm_config.cache_config.gpu_memory_utilization
-    original_memory_bytes = vllm_config.cache_config.kv_cache_memory_bytes
-    original_token_budget = vllm_config.scheduler_config.max_num_batched_tokens
-
-    diffusion_vllm_config.configure_diffusion_vllm_config(vllm_config, od_config)
-
-    assert vllm_config.parallel_config.pipeline_parallel_size == original_pp_size
-    assert vllm_config.cache_config.gpu_memory_utilization == original_memory_utilization
-    assert vllm_config.cache_config.kv_cache_memory_bytes == original_memory_bytes
-    assert vllm_config.scheduler_config.max_num_batched_tokens == original_token_budget
-
-
-def test_paged_config_forwards_gpu_memory_utilization_to_native_cache_config() -> None:
-    vllm_config = diffusion_vllm_config.create_diffusion_vllm_config(
-        torch.device("cpu"),
-        _od_config(gpu_memory_utilization=0.42),
-    )
-
-    assert vllm_config.cache_config.gpu_memory_utilization == 0.42
-
-
-def test_diffusion_vllm_model_config_supplies_dtype_for_quant_methods() -> None:
-    quantization_config = build_quant_config(
-        {
-            "quant_method": "modelopt",
-            "quant_algo": "FP8",
-            "ignore": [],
-        }
-    )
-    model_config = diffusion_vllm_config._make_diffusion_vllm_model_config(
-        _od_config(
-            diffusion_kv_mode=DiffusionKVCacheMode.DENSE_LEGACY,
-            quantization_config=quantization_config,
-        )
-    )
-
-    assert model_config.dtype is torch.bfloat16
-    assert model_config.quantization == "modelopt"
-    assert model_config.quantization_config is quantization_config
-    assert model_config.is_quantized
-    assert model_config.is_moe is False
-    assert model_config.is_nvfp4_quantized() is False
-    assert model_config.original_max_model_len == model_config.max_model_len
-
-    unquantized_model_config = diffusion_vllm_config._make_diffusion_vllm_model_config(
-        _od_config(
-            diffusion_kv_mode=DiffusionKVCacheMode.DENSE_LEGACY,
-            quantization_config=None,
-        )
-    )
-    assert not unquantized_model_config.is_quantized
-
-
-def test_minus_one_max_model_len_uses_model_limit() -> None:
-    od_config = SimpleNamespace(
-        max_model_len=-1,
-        tf_model_config=SimpleNamespace(max_position_embeddings=4096),
-        diffusion_kv_mode=DiffusionKVCacheMode.PAGED_SCHEDULER,
-    )
-
-    assert diffusion_vllm_config.resolve_diffusion_max_model_len(od_config) == 4096
-
-    model_config = diffusion_vllm_config._make_diffusion_vllm_model_config(
-        _od_config(
-            max_model_len=-1,
-            tf_model_config=SimpleNamespace(max_position_embeddings=4096),
-        )
-    )
-    assert model_config.max_model_len == 4096
-    assert model_config.original_max_model_len == -1
-
-
-def test_minus_one_max_model_len_is_auto_fitted_by_native_cache_sizing() -> None:
-    vllm_config = diffusion_vllm_config.create_diffusion_vllm_config(
-        torch.device("cpu"),
-        _od_config(
-            max_model_len=-1,
-            tf_model_config=SimpleNamespace(max_position_embeddings=4096),
-        ),
-    )
-    spec = FullAttentionSpec(
-        block_size=vllm_config.cache_config.block_size,
-        num_kv_heads=2,
-        head_size=8,
-        dtype=torch.bfloat16,
-        non_causal=True,
-    )
-
-    initialization.build_native_kv_cache_configs(
-        vllm_config,
-        [{"layer0": spec}],
-        [spec.page_size_bytes * 16],
-    )
-
-    assert vllm_config.model_config.original_max_model_len == -1
-    assert vllm_config.model_config.max_model_len == 256

tests/diffusion/diffusion_kv/test_manager.py

@@ -1,211 +0,0 @@
-# SPDX-License-Identifier: Apache-2.0
-# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
-
-import pytest
-import torch
-import vllm.v1.core.single_type_kv_cache_manager as native_kv_managers
-from vllm.v1.kv_cache_interface import FullAttentionSpec, KVCacheConfig, KVCacheGroupSpec, KVCacheTensor
-
-from vllm_omni.diffusion.diffusion_kv.manager import DiffusionKVAdmissionError, DiffusionKVCacheManager
-from vllm_omni.diffusion.diffusion_kv.request import DiffusionKVContext, DiffusionKVRequest
-
-pytestmark = [pytest.mark.core_model, pytest.mark.cpu, pytest.mark.diffusion]
-
-BLOCK_SIZE = 4
-
-
-def _config(num_blocks: int) -> KVCacheConfig:
-    native_kv_managers.register_all_kvcache_specs(None)
-    spec = FullAttentionSpec(
-        block_size=BLOCK_SIZE,
-        num_kv_heads=2,
-        head_size=8,
-        dtype=torch.bfloat16,
-    )
-    return KVCacheConfig(
-        num_blocks=num_blocks,
-        kv_cache_tensors=[KVCacheTensor(size=spec.page_size_bytes * num_blocks, shared_by=["layer0"])],
-        kv_cache_groups=[KVCacheGroupSpec(layer_names=["layer0"], kv_cache_spec=spec)],
-    )
-
-
-def _request(public_id: str, sequence_id: int, *, seq_len: int = 8, kv_contexts=()) -> DiffusionKVRequest:
-    return DiffusionKVRequest(
-        f"{public_id}/diffusion-kv/{sequence_id}",
-        sequence_id=sequence_id,
-        prefix_len=4,
-        target_len=4,
-        seq_len=seq_len,
-        kv_contexts=kv_contexts,
-    )
-
-
-def _manager(num_blocks: int, *, max_model_len: int = 64) -> DiffusionKVCacheManager:
-    return DiffusionKVCacheManager(
-        _config(num_blocks),
-        max_model_len=max_model_len,
-        scheduler_block_size=BLOCK_SIZE,
-        hash_block_size=BLOCK_SIZE,
-    )
-
-
-def test_reserve_and_free_multi_cfg_request() -> None:
-    manager = _manager(8)
-    free_before = manager.native_manager.block_pool.get_num_free_blocks()
-    requests = (_request("public", 0), _request("public", 1))
-
-    metadata = manager.reserve_request("public", requests)
-
-    assert metadata is not None
-    assert metadata.request_id == "public"
-    assert metadata.allocation_generation == 1
-    assert [sequence.sequence_id for sequence in metadata.sequences] == [0, 1]
-    assert [sequence.prefix_len for sequence in metadata.sequences] == [4, 4]
-    assert [sequence.target_len for sequence in metadata.sequences] == [4, 4]
-    assert [sequence.seq_len for sequence in metadata.sequences] == [8, 8]
-    assert [len(sequence.block_ids[0]) for sequence in metadata.sequences] == [2, 2]
-    assert all(sequence.context_ids == () for sequence in metadata.sequences)
-    assert metadata.contexts == ()
-    assert metadata == manager.get_metadata("public")
-    assert manager.native_manager.block_pool.get_num_free_blocks() == free_before - 4
-
-    manager.free_request("public")
-    assert manager.native_manager.block_pool.get_num_free_blocks() == free_before
-
-
-def test_impossible_cfg_allocation_rolls_back_and_fails_fast() -> None:
-    manager = _manager(3)
-    free_before = manager.native_manager.block_pool.get_num_free_blocks()
-
-    with pytest.raises(DiffusionKVAdmissionError, match="cannot fit even when the block pool is empty"):
-        manager.reserve_request(
-            "public",
-            (_request("public", 0), _request("public", 1)),
-        )
-
-    assert manager.has_request("public") is False
-    assert manager.native_manager.block_pool.get_num_free_blocks() == free_before
-
-
-def test_impossible_cfg_allocation_fails_fast_while_pool_is_busy() -> None:
-    manager = _manager(5)
-    assert manager.reserve_request("running", (_request("running", 0),)) is not None
-    free_before = manager.native_manager.block_pool.get_num_free_blocks()
-
-    with pytest.raises(DiffusionKVAdmissionError, match="required_blocks=6, available_blocks=4"):
-        manager.reserve_request(
-            "impossible",
-            tuple(_request("impossible", sequence_id) for sequence_id in range(3)),
-        )
-
-    assert manager.has_request("impossible") is False
-    assert manager.native_manager.block_pool.get_num_free_blocks() == free_before
-
-
-def test_temporary_capacity_pressure_returns_none() -> None:
-    manager = _manager(3)
-    assert manager.reserve_request("running", (_request("running", 0),)) is not None
-
-    allocation = manager.reserve_request("waiting", (_request("waiting", 0),))
-
-    assert allocation is None
-    assert manager.has_request("waiting") is False
-
-
-def test_partial_cfg_rollback_can_retry_after_capacity_is_released() -> None:
-    manager = _manager(5)
-    assert manager.reserve_request("running", (_request("running", 0),)) is not None
-    waiting = (_request("waiting", 0), _request("waiting", 1))
-
-    assert manager.reserve_request("waiting", waiting) is None
-    assert manager.has_request("waiting") is False
-
-    manager.free_request("running")
-    metadata = manager.reserve_request("waiting", waiting)
-
-    assert metadata is not None
-    assert [len(sequence.block_ids[0]) for sequence in metadata.sequences] == [2, 2]
-
-
-def test_reallocated_request_gets_a_new_allocation_generation() -> None:
-    manager = _manager(3)
-    request = (_request("public", 0),)
-
-    first = manager.reserve_request("public", request)
-    assert first is not None
-    manager.free_request("public")
-    second = manager.reserve_request("public", request)
-
-    assert second is not None
-    assert second.allocation_generation == first.allocation_generation + 1
-
-
-def test_cfg_allocation_exception_rolls_back(monkeypatch) -> None:
-    manager = _manager(8)
-    free_before = manager.native_manager.block_pool.get_num_free_blocks()
-    native_allocate = manager.native_manager.allocate_slots
-    num_calls = 0
-
-    def fail_second_allocation(*args, **kwargs):
-        nonlocal num_calls
-        num_calls += 1
-        if num_calls == 2:
-            raise RuntimeError("injected allocation failure")
-        return native_allocate(*args, **kwargs)
-
-    monkeypatch.setattr(manager.native_manager, "allocate_slots", fail_second_allocation)
-
-    with pytest.raises(RuntimeError, match="injected allocation failure"):
-        manager.reserve_request(
-            "public",
-            (_request("public", 0), _request("public", 1)),
-        )
-
-    assert manager.has_request("public") is False
-    assert manager.native_manager.block_pool.get_num_free_blocks() == free_before
-
-
-def test_rejects_independent_context_until_role_routing_is_implemented() -> None:
-    manager = _manager(8)
-    context = DiffusionKVContext(context_id="text", cache_role="cross.text", num_tokens=8)
-
-    with pytest.raises(DiffusionKVAdmissionError, match="DiffusionKVContext"):
-        manager.reserve_request(
-            "public",
-            (_request("public", 0, kv_contexts=(context,)),),
-        )
-
-
-def test_rejects_sequence_longer_than_admission_bound() -> None:
-    manager = _manager(8, max_model_len=8)
-
-    with pytest.raises(DiffusionKVAdmissionError, match="exceeds max_model_len"):
-        manager.reserve_request("public", (_request("public", 0, seq_len=9),))
-
-
-@pytest.mark.parametrize(
-    "requests",
-    [
-        (_request("public", 1),),
-        (_request("public", 1), _request("public", 0)),
-        (_request("public", 0), _request("public", 2)),
-    ],
-)
-def test_rejects_ambiguous_sequence_order(requests) -> None:
-    manager = _manager(8)
-
-    with pytest.raises(ValueError, match="ordered by contiguous sequence_id"):
-        manager.reserve_request("public", requests)
-
-
-def test_close_releases_every_public_request() -> None:
-    manager = _manager(8)
-    free_before = manager.native_manager.block_pool.get_num_free_blocks()
-    assert manager.reserve_request("first", (_request("first", 0),)) is not None
-    assert manager.reserve_request("second", (_request("second", 0),)) is not None
-
-    manager.close()
-
-    assert manager.native_manager.block_pool.get_num_free_blocks() == free_before
-    assert manager.has_request("first") is False
-    assert manager.has_request("second") is False

tests/diffusion/diffusion_kv/test_request.py

@@ -107,8 +107,11 @@ def test_request_rejects_non_context_values() -> None:
 def _manager(*, enable_caching: bool) -> KVCacheManager:
     # vLLM 0.26 lazily registers its built-in KV specs. Another test may have
     # already registered an out-of-tree spec, making the registry non-empty
-    # before FullAttentionSpec is registered. Initialize the native mappings.
-    native_kv_managers.register_all_kvcache_specs(None)
+    # before FullAttentionSpec is registered. Initialize the native mappings
+    # explicitly; vLLM 0.25 has no registration hook and needs no setup.
+    register_all = getattr(native_kv_managers, "register_all_kvcache_specs", None)
+    if register_all is not None:
+        register_all(None)
 
     spec = FullAttentionSpec(
         block_size=BLOCK_SIZE,

tests/diffusion/diffusion_kv/test_spec_discovery.py

@@ -1,259 +0,0 @@
-# SPDX-License-Identifier: Apache-2.0
-# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
-
-from contextlib import nullcontext
-from types import SimpleNamespace
-from unittest.mock import Mock
-
-import pytest
-import torch
-from torch import nn
-from vllm.v1.kv_cache_interface import FullAttentionSpec, KVCacheConfig, KVCacheGroupSpec, KVCacheTensor
-
-import vllm_omni.diffusion.worker.diffusion_worker as diffusion_worker_module
-from vllm_omni.diffusion.attention.layer import Attention
-from vllm_omni.diffusion.diffusion_kv.config import DiffusionKVCacheMode
-from vllm_omni.diffusion.worker.diffusion_model_runner import DiffusionModelRunner
-from vllm_omni.diffusion.worker.diffusion_worker import DiffusionWorker
-
-pytestmark = [pytest.mark.core_model, pytest.mark.cpu, pytest.mark.diffusion]
-
-
-class _Backend:
-    @classmethod
-    def indexes_kv_by_block_stride(cls) -> bool:
-        return True
-
-
-def _attention(*, enabled: bool) -> Attention:
-    attention = Attention.__new__(Attention)
-    nn.Module.__init__(attention)
-    attention.paged_kv_cache_role = "primary" if enabled else None
-    attention.paged_kv_cache_dtype = torch.bfloat16
-    attention.num_kv_heads = 2
-    attention.head_size = 8
-    attention.causal = False
-    attention.attn_backend = _Backend
-    return attention
-
-
-def _runner(attention: Attention) -> DiffusionModelRunner:
-    runner = object.__new__(DiffusionModelRunner)
-    runner.od_config = SimpleNamespace(diffusion_kv_mode=DiffusionKVCacheMode.PAGED_SCHEDULER)
-    runner.vllm_config = SimpleNamespace(
-        cache_config=SimpleNamespace(block_size=4),
-        model_config=SimpleNamespace(dtype=torch.float16),
-    )
-    pipeline = nn.Module()
-    pipeline.image_attention = attention
-    runner.pipeline = pipeline
-    return runner
-
-
-def test_runner_discovers_native_spec_from_loaded_attention() -> None:
-    specs = _runner(_attention(enabled=True)).get_kv_cache_spec()
-
-    assert set(specs) == {"image_attention"}
-    spec = specs["image_attention"]
-    assert isinstance(spec, FullAttentionSpec)
-    assert spec.block_size == 4
-    assert spec.num_kv_heads == 2
-    assert spec.head_size == 8
-    assert spec.dtype is torch.bfloat16
-    assert spec.indexes_kv_by_block_stride is True
-    assert spec.non_causal is True
-
-
-def test_runner_rejects_paged_mode_without_cache_enabled_attention() -> None:
-    runner = _runner(_attention(enabled=False))
-
-    with pytest.raises(RuntimeError, match="no cache-enabled Attention"):
-        runner.get_kv_cache_spec()
-
-
-def test_runner_retains_matching_rank_local_config() -> None:
-    runner = _runner(_attention(enabled=True))
-    spec = runner.get_kv_cache_spec()["image_attention"]
-    config = KVCacheConfig(
-        num_blocks=8,
-        kv_cache_tensors=[KVCacheTensor(size=spec.page_size_bytes * 8, shared_by=["image_attention"])],
-        kv_cache_groups=[KVCacheGroupSpec(layer_names=["image_attention"], kv_cache_spec=spec)],
-    )
-
-    runner.set_kv_cache_config(config)
-
-    assert runner.kv_cache_config is config
-
-
-def test_runner_rejects_rank_local_config_for_different_layers() -> None:
-    runner = _runner(_attention(enabled=True))
-    spec = runner.get_kv_cache_spec()["image_attention"]
-    config = KVCacheConfig(
-        num_blocks=8,
-        kv_cache_tensors=[KVCacheTensor(size=spec.page_size_bytes * 8, shared_by=["other_attention"])],
-        kv_cache_groups=[KVCacheGroupSpec(layer_names=["other_attention"], kv_cache_spec=spec)],
-    )
-
-    with pytest.raises(ValueError, match="layer mismatch"):
-        runner.set_kv_cache_config(config)
-
-
-def test_worker_selects_its_rank_local_config() -> None:
-    worker = object.__new__(DiffusionWorker)
-    worker.rank = 1
-    worker.od_config = SimpleNamespace(num_gpus=2)
-    worker.model_runner = SimpleNamespace(set_kv_cache_config=lambda config: setattr(worker, "installed", config))
-    configs = [object(), object()]
-
-    worker.set_kv_cache_configs(configs)
-
-    assert worker.installed is configs[1]
-    with pytest.raises(ValueError, match="rank count mismatch"):
-        worker.set_kv_cache_configs(configs[:1])
-
-
-def test_worker_honors_explicit_kv_memory_budget(monkeypatch) -> None:
-    worker = object.__new__(DiffusionWorker)
-    worker.rank = 0
-    worker.vllm_config = SimpleNamespace(cache_config=SimpleNamespace(kv_cache_memory_bytes=4096))
-    worker.init_snapshot = SimpleNamespace(free_memory=16384)
-    worker.requested_memory = 8192
-    worker.model_runner = SimpleNamespace(profile_run=Mock(), model_memory_usage=1024)
-    monkeypatch.setattr(diffusion_worker_module, "_all_gather_rank_values", lambda value: [value])
-    profile_request = object()
-
-    assert worker.determine_available_kv_memory(profile_request) == [4096]
-    worker.model_runner.profile_run.assert_called_once_with(profile_request)
-
-
-def test_worker_treats_zero_kv_memory_budget_as_profiled_auto_sizing(monkeypatch) -> None:
-    worker = object.__new__(DiffusionWorker)
-    worker.rank = 0
-    worker.init_snapshot = SimpleNamespace(free_memory=900)
-    worker.requested_memory = 750
-    worker.vllm_config = SimpleNamespace(
-        cache_config=SimpleNamespace(
-            kv_cache_memory_bytes=0,
-            gpu_memory_utilization=0.75,
-        )
-    )
-    worker.model_runner = SimpleNamespace(profile_run=Mock(), model_memory_usage=100)
-    monkeypatch.setattr(diffusion_worker_module, "_all_gather_rank_values", lambda value: [value])
-    profile_result = SimpleNamespace(
-        non_kv_cache_memory=400,
-        after_profile=SimpleNamespace(free_memory=500),
-    )
-    monkeypatch.setattr(
-        diffusion_worker_module,
-        "memory_profiling",
-        lambda snapshot, weights_memory: nullcontext(profile_result),
-    )
-    profile_request = object()
-
-    assert worker.determine_available_kv_memory(profile_request) == [350]
-    worker.model_runner.profile_run.assert_called_once_with(profile_request)
-
-
-@pytest.mark.parametrize("non_kv_cache_memory", [750, 800])
-def test_worker_rejects_non_positive_profiled_kv_memory(monkeypatch, non_kv_cache_memory: int) -> None:
-    worker = object.__new__(DiffusionWorker)
-    worker.rank = 0
-    worker.init_snapshot = SimpleNamespace(free_memory=900)
-    worker.requested_memory = 750
-    worker.vllm_config = SimpleNamespace(
-        cache_config=SimpleNamespace(
-            kv_cache_memory_bytes=None,
-            gpu_memory_utilization=0.75,
-        )
-    )
-    worker.model_runner = SimpleNamespace(profile_run=Mock(), model_memory_usage=100)
-    monkeypatch.setattr(diffusion_worker_module, "_all_gather_rank_values", lambda value: [value])
-    profile_result = SimpleNamespace(
-        non_kv_cache_memory=non_kv_cache_memory,
-        after_profile=SimpleNamespace(free_memory=100),
-    )
-    monkeypatch.setattr(
-        diffusion_worker_module,
-        "memory_profiling",
-        lambda snapshot, weights_memory: nullcontext(profile_result),
-    )
-    profile_request = object()
-
-    with pytest.raises(
-        RuntimeError,
-        match=(
-            "rank 0: RuntimeError: No memory remains for Diffusion KV cache after profiling: "
-            f"requested_memory=750 bytes, non_kv_cache_memory={non_kv_cache_memory} bytes"
-        ),
-    ):
-        worker.determine_available_kv_memory(profile_request)
-
-    worker.model_runner.profile_run.assert_called_once_with(profile_request)
-
-
-def test_worker_profiles_activation_headroom_instead_of_current_residency(monkeypatch) -> None:
-    worker = object.__new__(DiffusionWorker)
-    worker.rank = 0
-    worker.init_snapshot = SimpleNamespace(free_memory=1200)
-    worker.requested_memory = 1000
-    worker.vllm_config = SimpleNamespace(
-        cache_config=SimpleNamespace(
-            kv_cache_memory_bytes=None,
-            gpu_memory_utilization=0.8,
-        )
-    )
-    worker.model_runner = SimpleNamespace(profile_run=Mock(), model_memory_usage=250)
-    monkeypatch.setattr(diffusion_worker_module, "_all_gather_rank_values", lambda value: [value])
-    profile_result = SimpleNamespace(
-        non_kv_cache_memory=650,
-        after_profile=SimpleNamespace(free_memory=550),
-    )
-    profiling_calls = []
-
-    def profile(snapshot, *, weights_memory):
-        profiling_calls.append((snapshot, weights_memory))
-        return nullcontext(profile_result)
-
-    monkeypatch.setattr(
-        diffusion_worker_module,
-        "memory_profiling",
-        profile,
-    )
-    profile_request = object()
-
-    assert worker.determine_available_kv_memory(profile_request) == [350]
-    assert profiling_calls == [(worker.init_snapshot, 250)]
-    worker.model_runner.profile_run.assert_called_once_with(profile_request)
-
-
-def test_worker_rejects_profile_without_preload_snapshot(monkeypatch) -> None:
-    worker = object.__new__(DiffusionWorker)
-    worker.init_snapshot = None
-    worker.requested_memory = None
-    worker.vllm_config = SimpleNamespace(cache_config=SimpleNamespace(kv_cache_memory_bytes=None))
-    worker.model_runner = SimpleNamespace()
-    monkeypatch.setattr(diffusion_worker_module, "_all_gather_rank_values", lambda value: [value])
-
-    with pytest.raises(RuntimeError, match="snapshot was not captured before model loading"):
-        worker.determine_available_kv_memory(object())
-
-
-def test_rank_probe_gathers_local_failure_before_raising(monkeypatch) -> None:
-    gathered = []
-
-    def gather(local_result):
-        gathered.append(local_result)
-        return [local_result, (True, "peer-result")]
-
-    def fail_local_probe():
-        raise ValueError("local failure")
-
-    monkeypatch.setattr(diffusion_worker_module, "_all_gather_rank_values", gather)
-
-    with pytest.raises(RuntimeError, match="rank 0: ValueError: local failure"):
-        diffusion_worker_module._run_and_gather_rank_values(
-            "test probe",
-            fail_local_probe,
-        )
-
-    assert gathered == [(False, "ValueError: local failure")]

tests/diffusion/diffusion_kv/test_worker_contract.py

@@ -85,15 +85,19 @@ def test_new_request_data_carries_scheduler_allocation_atomically() -> None:
     assert new_req.diffusion_kv_metadata is metadata
 
 
-def test_paged_request_without_metadata_fails_before_request_forward() -> None:
+def test_paged_request_without_metadata_reaches_request_forward() -> None:
     runner = make_runner(DiffusionKVCacheMode.PAGED_SCHEDULER)
-    runner._execute_request_list = Mock()
+    expected_output = DiffusionOutput(output=None)
+    runner._execute_request_list = Mock(
+        return_value=SimpleNamespace(
+            runner_outputs=[SimpleNamespace(result=expected_output)],
+        )
+    )
     req = SimpleNamespace(request_id="req-0")
 
-    with pytest.raises(ValueError, match="requires Diffusion KV metadata"):
-        runner.execute_model(req)
+    assert runner.execute_model(req) is expected_output
 
-    runner._execute_request_list.assert_not_called()
+    runner._execute_request_list.assert_called_once()
 
 
 def test_dense_request_rejects_metadata_before_request_forward() -> None:
@@ -137,18 +141,18 @@ def test_request_rpc_rejects_envelope_request_identity_mismatch() -> None:
     assert calls == []
 
 
-def test_batch_path_rejects_missing_metadata_before_forward() -> None:
+def test_batch_path_allows_missing_metadata_before_forward() -> None:
     runner = make_runner(DiffusionKVCacheMode.PAGED_SCHEDULER)
-    runner._execute_request_list = Mock()
+    expected_output = object()
+    runner._execute_request_list = Mock(return_value=expected_output)
     new_req = NewRequestData(
         request_id="req-0",
         req=SimpleNamespace(request_id="req-0"),
     )
 
-    with pytest.raises(ValueError, match="requires Diffusion KV metadata"):
-        runner.execute_model_batch(make_scheduler_output(new_req), runner.od_config)
+    assert runner.execute_model_batch(make_scheduler_output(new_req), runner.od_config) is expected_output
 
-    runner._execute_request_list.assert_not_called()
+    runner._execute_request_list.assert_called_once()
 
 
 def test_batch_path_rejects_envelope_request_identity_mismatch() -> None:
@@ -166,7 +170,7 @@ def test_batch_path_rejects_envelope_request_identity_mismatch() -> None:
     runner._execute_request_list.assert_not_called()
 
 
-def test_step_path_rejects_missing_metadata_before_step_support_check() -> None:
+def test_step_path_allows_missing_metadata_before_step_support_check() -> None:
     runner = make_runner(DiffusionKVCacheMode.PAGED_SCHEDULER)
     runner.pipeline = object()
     runner._supports_step_mode = Mock(return_value=False)
@@ -175,10 +179,10 @@ def test_step_path_rejects_missing_metadata_before_step_support_check() -> None:
         req=SimpleNamespace(request_id="req-0"),
     )
 
-    with pytest.raises(ValueError, match="requires Diffusion KV metadata"):
+    with pytest.raises(ValueError, match="does not support step execution"):
         runner.execute_stepwise(make_scheduler_output(new_req))
 
-    runner._supports_step_mode.assert_not_called()
+    runner._supports_step_mode.assert_called_once()
 
 
 def test_step_path_rejects_envelope_request_identity_mismatch() -> None:
@@ -284,42 +288,3 @@ def test_dense_worker_call_does_not_extend_model_runner_signature() -> None:
     worker.execute_model(req, SimpleNamespace())
 
     assert calls == [(req, None)]
-
-
-def test_dlo_worker_selects_request_and_metadata_from_same_envelope(monkeypatch) -> None:
-    calls: list[tuple] = []
-
-    class ModelRunner:
-        def execute_model(self, req, **kwargs):
-            calls.append((req, kwargs))
-            return DiffusionOutput(output=None)
-
-    worker = object.__new__(DiffusionWorker)
-    worker.model_runner = ModelRunner()
-    worker.lora_manager = None
-    worker._get_profiler = lambda: None
-    monkeypatch.setattr(
-        "vllm_omni.diffusion.distributed.parallel_state.get_data_parallel_rank",
-        lambda: 1,
-    )
-    req_0 = SimpleNamespace(request_id="req-0")
-    req_1 = SimpleNamespace(request_id="req-1")
-    metadata_0 = make_metadata("req-0")
-    metadata_1 = make_metadata("req-1")
-    envelopes = [
-        NewRequestData(request_id="req-0", req=req_0, diffusion_kv_metadata=metadata_0),
-        NewRequestData(request_id="req-1", req=req_1, diffusion_kv_metadata=metadata_1),
-    ]
-
-    result = worker.execute_model(envelopes, SimpleNamespace())
-
-    assert calls == [
-        (
-            req_1,
-            {
-                "kv_prefetch_job": None,
-                "diffusion_kv_metadata": metadata_1,
-            },
-        )
-    ]
-    assert result["dp_rank"] == 1

tests/diffusion/distributed/test_distributed_vae_executor.py

@@ -74,6 +74,14 @@ def mock_dist(monkeypatch: pytest.MonkeyPatch):
     monkeypatch.setattr(dist, "broadcast", lambda *args, **kwargs: None)
 
 
+@pytest.fixture(autouse=True)
+def mock_dit_group(monkeypatch: pytest.MonkeyPatch):
+    monkeypatch.setattr(
+        "vllm_omni.diffusion.distributed.autoencoders.distributed_vae_executor.get_dit_group",
+        lambda: None,
+    )
+
+
 @pytest.fixture(autouse=True)
 def mock_dist_vae_executor(monkeypatch: pytest.MonkeyPatch):
     monkeypatch.setattr(DistributedVaeExecutor, "gather_tensors", lambda self, x: [x])

tests/diffusion/distributed/test_expert_parallel_layout.py

@@ -92,8 +92,9 @@ def test_moe_ep_maps_diffusion_sp_cfg_dp_to_vllm_groups(monkeypatch):
     monkeypatch.setattr(parallel_state, "get_forward_context", lambda: fake_forward_context)
     monkeypatch.setattr(parallel_state, "init_model_parallel_group", fake_init_model_parallel_group)
     monkeypatch.setattr(parallel_state, "init_vllm_model_parallel_group", fake_init_vllm_model_parallel_group)
+    monkeypatch.setattr(parallel_state, "init_dit_group", lambda *_args, **_kwargs: None)
 
-    for name in ("_DP", "_CFG", "_SP", "_PP", "_EXPERT_PARALLEL_GROUP_RANKS"):
+    for name in ("_DP", "_CFG", "_SP", "_PP", "_FS", "_EXPERT_PARALLEL_GROUP_RANKS"):
         monkeypatch.setattr(parallel_state, name, None)
     for name in ("_TP", "_PCP", "_DP", "_EP", "_PP"):
         monkeypatch.setattr(parallel_state.vllm_parallel_state, name, None, raising=False)
@@ -228,8 +229,9 @@ def test_cfg_parallel_keeps_diffusion_dp_without_ep(monkeypatch):
     monkeypatch.setattr(parallel_state.torch.distributed, "new_group", lambda ranks: tuple(ranks))
     monkeypatch.setattr(parallel_state, "get_world_group", lambda: fake_world_group)
     monkeypatch.setattr(parallel_state, "init_model_parallel_group", fake_init_model_parallel_group)
+    monkeypatch.setattr(parallel_state, "init_dit_group", lambda *_args, **_kwargs: None)
 
-    for name in ("_DP", "_CFG", "_SP", "_PP", "_EXPERT_PARALLEL_GROUP_RANKS"):
+    for name in ("_DP", "_CFG", "_SP", "_PP", "_FS", "_EXPERT_PARALLEL_GROUP_RANKS"):
         monkeypatch.setattr(parallel_state, name, None)
     for name in ("_TP", "_PCP", "_DP", "_EP", "_PP"):
         monkeypatch.setattr(parallel_state.vllm_parallel_state, name, None, raising=False)
@@ -257,33 +259,6 @@ def test_cfg_parallel_keeps_diffusion_dp_without_ep(monkeypatch):
     ]
 
 
-@pytest.mark.cpu
-@pytest.mark.core_model
-def test_destroy_model_parallel_clears_vllm_pipeline_group(monkeypatch):
-    """A destroyed diffusion PP group must not block the next initialization."""
-
-    class _DestroyableGroup:
-        def __init__(self) -> None:
-            self.destroy_calls = 0
-
-        def destroy(self) -> None:
-            self.destroy_calls += 1
-
-    pipeline_group = _DestroyableGroup()
-    for name in ("_DP", "_CFG", "_SP"):
-        monkeypatch.setattr(parallel_state, name, None)
-    monkeypatch.setattr(parallel_state, "_PP", pipeline_group)
-    monkeypatch.setattr(parallel_state, "_EXPERT_PARALLEL_GROUP_RANKS", None)
-    for name in ("_DP", "_PCP", "_TP", "_EP"):
-        monkeypatch.setattr(parallel_state.vllm_parallel_state, name, None, raising=False)
-    monkeypatch.setattr(parallel_state.vllm_parallel_state, "_PP", pipeline_group, raising=False)
-
-    parallel_state.destroy_model_parallel()
-
-    assert pipeline_group.destroy_calls == 1
-    assert parallel_state.vllm_parallel_state._PP is None
-
-
 @pytest.mark.cpu
 @pytest.mark.core_model
 def test_non_moe_ep_fails_before_vllm_ep_remap(monkeypatch):
@@ -320,8 +295,9 @@ def test_non_moe_ep_fails_before_vllm_ep_remap(monkeypatch):
     monkeypatch.setattr(parallel_state, "get_world_group", lambda: fake_world_group)
     monkeypatch.setattr(parallel_state, "get_forward_context", lambda: fake_forward_context)
     monkeypatch.setattr(parallel_state, "init_model_parallel_group", fake_init_model_parallel_group)
+    monkeypatch.setattr(parallel_state, "init_dit_group", lambda *_args, **_kwargs: None)
 
-    for name in ("_DP", "_CFG", "_SP", "_PP", "_EXPERT_PARALLEL_GROUP_RANKS"):
+    for name in ("_DP", "_CFG", "_SP", "_PP", "_FS", "_EXPERT_PARALLEL_GROUP_RANKS"):
         monkeypatch.setattr(parallel_state, name, None)
     for name in ("_TP", "_PCP", "_DP", "_EP", "_PP"):
         monkeypatch.setattr(parallel_state.vllm_parallel_state, name, None, raising=False)

tests/diffusion/distributed/test_hsdp.py

@@ -234,7 +234,7 @@ class TestDiffusionParallelConfigHSDP:
 
     def test_hsdp_cannot_use_with_tp(self):
         """Test that HSDP and Tensor Parallelism cannot be used together."""
-        with pytest.raises(ValueError, match="not compatible with TP"):
+        with pytest.raises(ValueError, match="cannot be used with TP or DP"):
             DiffusionParallelConfig(
                 tensor_parallel_size=2,
                 use_hsdp=True,
@@ -243,7 +243,7 @@ class TestDiffusionParallelConfigHSDP:
 
     def test_hsdp_cannot_use_with_dp(self):
         """Test that HSDP and Data Parallelism cannot be used together."""
-        with pytest.raises(ValueError, match="not compatible with DP"):
+        with pytest.raises(ValueError, match="cannot be used with TP or DP"):
             DiffusionParallelConfig(
                 data_parallel_size=2,
                 use_hsdp=True,
@@ -264,6 +264,158 @@ class TestDiffusionParallelConfigHSDP:
         assert config.hsdp_shard_size == 2  # auto: 4 // 2
 
 
+class TestStandaloneHSDPDetection:
+    """Tests for standalone HSDP detection and dit_parallel_size calculation.
+
+    These tests verify the logic used in initialize_model_parallel() to detect
+    standalone HSDP mode and calculate effective parallel sizes.
+
+    Standalone HSDP is when all non-HSDP parallelism dimensions are 1.
+    """
+
+    @staticmethod
+    def compute_standalone_hsdp_params(
+        data_parallel_size: int = 1,
+        cfg_parallel_size: int = 1,
+        sequence_parallel_size: int = 1,
+        pipeline_parallel_size: int = 1,
+        tensor_parallel_size: int = 1,
+        fully_shard_degree: int = 1,
+        hsdp_replicate_size: int = 1,
+    ) -> dict:
+        """Compute standalone HSDP detection parameters.
+
+        This mirrors the logic in initialize_model_parallel().
+        """
+        dit_parallel_size = (
+            data_parallel_size
+            * cfg_parallel_size
+            * sequence_parallel_size
+            * pipeline_parallel_size
+            * tensor_parallel_size
+        )
+
+        # Check for standalone HSDP: all non-HSDP parallelism dimensions are 1
+        is_standalone_hsdp = dit_parallel_size == 1 and fully_shard_degree > 1
+
+        # For standalone HSDP: use (fully_shard_degree * hsdp_replicate_size)
+        if is_standalone_hsdp:
+            effective_dit_parallel_size = fully_shard_degree * hsdp_replicate_size
+        else:
+            effective_dit_parallel_size = dit_parallel_size
+
+        effective_dp_size = (fully_shard_degree * hsdp_replicate_size) if is_standalone_hsdp else data_parallel_size
+
+        return {
+            "original_dit_parallel_size": dit_parallel_size,
+            "is_standalone_hsdp": is_standalone_hsdp,
+            "effective_dit_parallel_size": effective_dit_parallel_size,
+            "effective_dp_size": effective_dp_size,
+        }
+
+    def test_standalone_hsdp_basic(self):
+        """Test basic standalone HSDP detection (shard_size=4, replicate=1)."""
+        result = self.compute_standalone_hsdp_params(
+            fully_shard_degree=4,
+            hsdp_replicate_size=1,
+        )
+        assert result["original_dit_parallel_size"] == 1
+        assert result["is_standalone_hsdp"] is True
+        assert result["effective_dit_parallel_size"] == 4
+        assert result["effective_dp_size"] == 4
+
+    def test_standalone_hsdp_with_replicate(self):
+        """Test standalone HSDP with replication (shard_size=4, replicate=2)."""
+        result = self.compute_standalone_hsdp_params(
+            fully_shard_degree=4,
+            hsdp_replicate_size=2,
+        )
+        assert result["original_dit_parallel_size"] == 1
+        assert result["is_standalone_hsdp"] is True
+        assert result["effective_dit_parallel_size"] == 8  # 4 * 2
+        assert result["effective_dp_size"] == 8
+
+    def test_combined_hsdp_sp_not_standalone(self):
+        """Test HSDP combined with SP is NOT detected as standalone.
+
+        This is a regression test for the bug where the condition
+        `dit_parallel_size == fully_shard_degree` incorrectly matched
+        combined modes like SP=4 + HSDP=4.
+        """
+        result = self.compute_standalone_hsdp_params(
+            sequence_parallel_size=4,
+            fully_shard_degree=4,
+            hsdp_replicate_size=1,
+        )
+        assert result["original_dit_parallel_size"] == 4
+        assert result["is_standalone_hsdp"] is False
+        # Should NOT override dp_size for combined mode
+        assert result["effective_dp_size"] == 1  # original data_parallel_size
+
+    def test_combined_hsdp_cfg_not_standalone(self):
+        """Test HSDP combined with CFG is NOT detected as standalone."""
+        result = self.compute_standalone_hsdp_params(
+            cfg_parallel_size=2,
+            fully_shard_degree=4,
+            hsdp_replicate_size=1,
+        )
+        assert result["original_dit_parallel_size"] == 2
+        assert result["is_standalone_hsdp"] is False
+        assert result["effective_dp_size"] == 1
+
+    def test_combined_hsdp_pp_not_standalone(self):
+        """Test HSDP combined with PP is NOT detected as standalone."""
+        result = self.compute_standalone_hsdp_params(
+            pipeline_parallel_size=2,
+            fully_shard_degree=4,
+            hsdp_replicate_size=1,
+        )
+        assert result["original_dit_parallel_size"] == 2
+        assert result["is_standalone_hsdp"] is False
+        assert result["effective_dp_size"] == 1
+
+    def test_no_hsdp_not_standalone(self):
+        """Test that no HSDP (fully_shard_degree=1) is NOT standalone."""
+        result = self.compute_standalone_hsdp_params(
+            fully_shard_degree=1,
+        )
+        assert result["original_dit_parallel_size"] == 1
+        assert result["is_standalone_hsdp"] is False
+        assert result["effective_dp_size"] == 1
+
+    def test_combined_multiple_parallelism_not_standalone(self):
+        """Test HSDP combined with multiple parallelism is NOT standalone."""
+        result = self.compute_standalone_hsdp_params(
+            sequence_parallel_size=2,
+            cfg_parallel_size=2,
+            fully_shard_degree=4,
+            hsdp_replicate_size=1,
+        )
+        assert result["original_dit_parallel_size"] == 4  # 2 * 2
+        assert result["is_standalone_hsdp"] is False
+        assert result["effective_dp_size"] == 1
+
+    def test_standalone_hsdp_large_shard(self):
+        """Test standalone HSDP with large shard size."""
+        result = self.compute_standalone_hsdp_params(
+            fully_shard_degree=8,
+            hsdp_replicate_size=1,
+        )
+        assert result["is_standalone_hsdp"] is True
+        assert result["effective_dit_parallel_size"] == 8
+        assert result["effective_dp_size"] == 8
+
+    def test_standalone_hsdp_large_replicate(self):
+        """Test standalone HSDP with large replicate size."""
+        result = self.compute_standalone_hsdp_params(
+            fully_shard_degree=4,
+            hsdp_replicate_size=4,
+        )
+        assert result["is_standalone_hsdp"] is True
+        assert result["effective_dit_parallel_size"] == 16  # 4 * 4
+        assert result["effective_dp_size"] == 16
+
+
 class TestHSDPShardConditions:
     """Tests for _hsdp_shard_conditions matching logic."""
 

tests/diffusion/model_loader/test_diffusers_loader.py

@@ -14,7 +14,7 @@ from huggingface_hub import snapshot_download
 from vllm.config.load import LoadConfig
 
 from vllm_omni.diffusion.config import get_current_diffusion_config, get_current_diffusion_config_or_none
-from vllm_omni.diffusion.data import DiffusionParallelConfig, OmniDiffusionConfig
+from vllm_omni.diffusion.data import OmniDiffusionConfig
 from vllm_omni.diffusion.model_loader.diffusers_loader import DiffusersPipelineLoader
 from vllm_omni.diffusion.model_loader.host_weight_plan import (
     HostWeightPlan,
@@ -91,17 +91,6 @@ def _make_loader_with_weights(weight_names: list[str]) -> DiffusersPipelineLoade
     return loader
 
 
-def _make_dlo_online_quant_config() -> OmniDiffusionConfig:
-    return OmniDiffusionConfig(
-        model="",
-        dtype=torch.float32,
-        quantization_config="fp8",
-        parallel_config=DiffusionParallelConfig(data_parallel_size=2),
-        enable_distributed_layerwise_offload=True,
-        dlo_use_allgather=True,
-    )
-
-
 def test_strict_check_only_validates_source_prefix_parameters():
     model = _DummyPipelineModel(source_prefix="transformer.")
     loader = _make_loader_with_weights(["transformer.weight"])
@@ -405,78 +394,6 @@ def test_dlo_plan_fallback_runs_ordinary_loader(monkeypatch):
     assert loader.take_host_weight_plan() is None
 
 
-def test_dlo_allgather_online_fp8_uses_ordinary_loader(monkeypatch):
-    from vllm.model_executor.layers.quantization.online.fp8 import (
-        Fp8PerTensorOnlineLinearMethod,
-    )
-
-    import vllm_omni.diffusion.model_loader.diffusers_loader as loader_mod
-
-    od_config = _make_dlo_online_quant_config()
-    loader = DiffusersPipelineLoader(LoadConfig(), od_config)
-    model = nn.Module()
-    model.transformer = nn.Linear(2, 2, bias=False)
-    model.transformer.quant_method = object.__new__(Fp8PerTensorOnlineLinearMethod)
-    model.transformer.quant_method.uses_meta_device = True
-    calls: list[object] = []
-    allowlist_models: list[nn.Module] = []
-
-    original_allowlist_check = loader._unsupported_dlo_allgather_online_quant_methods
-
-    def check_allowlist(candidate: nn.Module) -> tuple[str, ...]:
-        allowlist_models.append(candidate)
-        return original_allowlist_check(candidate)
-
-    loader._init_from_load_format = lambda *_args, **_kwargs: model  # type: ignore[method-assign]
-    monkeypatch.setattr(loader, "_unsupported_dlo_allgather_online_quant_methods", check_allowlist)
-    loader._request_offload_after_quant = lambda _model: 1  # type: ignore[method-assign]
-    loader.load_weights = (  # type: ignore[method-assign]
-        lambda _model, *, stream_online_quant_to_cpu=False: calls.append(("load", stream_online_quant_to_cpu))
-    )
-    loader._process_weights_after_loading = lambda *_args: calls.append("process")  # type: ignore[method-assign]
-    loader._apply_skip_softmax_calibration = lambda _model: None  # type: ignore[method-assign]
-    monkeypatch.setattr(
-        loader_mod,
-        "build_checkpoint_mmap_plan",
-        lambda *_args, **_kwargs: HostWeightPlanResult(
-            None,
-            "online quantization requires the ordinary loader",
-        ),
-    )
-
-    assert loader.load_model(load_device="cpu", device=torch.device("cpu")) is model
-    assert allowlist_models == [model]
-    assert calls == [("load", True), "process"]
-    assert loader.take_host_weight_plan() is None
-
-
-def test_dlo_allgather_rejects_unvalidated_online_quant_method(monkeypatch):
-    import vllm_omni.diffusion.model_loader.diffusers_loader as loader_mod
-
-    class UnsupportedOnlineMethod:
-        uses_meta_device = True
-
-    od_config = _make_dlo_online_quant_config()
-    loader = DiffusersPipelineLoader(LoadConfig(), od_config)
-    model = nn.Module()
-    model.transformer = nn.Linear(2, 2, bias=False)
-    model.transformer.quant_method = UnsupportedOnlineMethod()
-
-    loader._init_from_load_format = lambda *_args, **_kwargs: model  # type: ignore[method-assign]
-    loader._apply_skip_softmax_calibration = lambda _model: None  # type: ignore[method-assign]
-    monkeypatch.setattr(
-        loader_mod,
-        "build_checkpoint_mmap_plan",
-        lambda *_args, **_kwargs: HostWeightPlanResult(
-            None,
-            "online quantization requires the ordinary loader",
-        ),
-    )
-
-    with pytest.raises(ValueError, match="per-tensor FP8 linears"):
-        loader.load_model(load_device="cpu")
-
-
 def test_hsdp_processes_quantized_weights_before_sharding(mocker):
     import vllm_omni.diffusion.model_loader.diffusers_loader as loader_mod
     from vllm_omni.diffusion.offloader.module_collector import PipelineModules

tests/diffusion/models/hunyuan_image3/test_image_kv_cache_manager.py

@@ -118,14 +118,6 @@ def _call_mgr(
     )
 
 
-def test_cache_manager_registers_attention_without_adding_dense_state() -> None:
-    mgr = _make_cache_mgr()
-
-    assert isinstance(mgr, nn.Module)
-    assert dict(mgr.named_modules())["attn"] is mgr.attn
-    assert mgr.state_dict() == {}
-
-
 # ============================================================
 # Test 1: No AR KV — basic cache → reuse
 # ============================================================

tests/diffusion/models/pi0/test_pi0_parity.py

@@ -1,366 +0,0 @@
-#!/usr/bin/env python
-# SPDX-License-Identifier: Apache-2.0
-# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
-"""π0 LeRobot parity (in-process): the bit-for-bit correctness oracle.
-
-Verifies that vllm-omni's ``Pi0ForActionPrediction`` produces bit-for-bit
-matching action chunks with LeRobot's ``PI0Policy`` when fed:
-  * the same weights (``lerobot/pi0_base``)
-  * the same pre-processed inputs (images, masks, tokens, state)
-  * the same initial noise tensor
-
-π0 is a flow-matching model (Euler-integrated ODE from t=1 → t=0 with a fixed
-``num_steps``), so the output is deterministic once the noise is fixed and
-``torch.allclose`` on the final action chunk is a valid oracle. This is the
-authoritative correctness oracle for the vllm-omni π0 port (max|Δ| < 1e-4).
-
-The OpenPI websocket online-serving e2e lives separately in
-``tests/e2e/online_serving/test_pi0_expansion.py``.
-
-Run in a SEPARATE ``lerobot[pi]`` venv (avoids dep conflict with the vllm-omni
-env), with the vllm-omni ``pi0`` package importable::
-
-    python -m pytest tests/diffusion/models/pi0/test_pi0_parity.py -v -s
-
-Skipped automatically when LeRobot is not installed (e.g. the vllm-omni env).
-The run is CPU/float32 with fixed defaults; the only override is
-``PI0_PARITY_MODEL_PATH`` (a local pi0_base dir in LeRobot format, to skip the
-HF download; defaults to ``lerobot/pi0_base``).
-"""
-
-from __future__ import annotations
-
-import copy
-import importlib.util
-import os
-
-import pytest
-import torch
-
-# local_model: needs real weights + a lerobot venv (transformers 5.3.0), so it
-# runs locally rather than in the ready-CI (see docs/contributing/ci/CI_5levels.md).
-# Additionally gated on lerobot being importable.
-_HAS_LEROBOT = importlib.util.find_spec("lerobot") is not None
-
-pytestmark = [pytest.mark.local_model, pytest.mark.diffusion]
-
-
-# ─── Config (fixed; matches LeRobot defaults for ``lerobot/pi0_base``) ──
-DEVICE = "cpu"
-DTYPE_STR = "float32"
-ATOL = 1e-4
-NUM_STEPS = 10
-BATCH_SIZE = 2
-ACTION_DIM = 32
-STATE_DIM = 32
-ACTION_HORIZON = 50
-MAX_TOKEN_LEN = 48
-
-# The only knob: point at a local pi0_base dir (LeRobot format) to skip the HF
-# download; defaults to the HF repo id.
-MODEL_PATH = os.environ.get("PI0_PARITY_MODEL_PATH", "lerobot/pi0_base")
-
-
-def _resolve_checkpoint_dir() -> str:
-    """Return a local dir containing the pi0_base checkpoint (download if needed)."""
-    if os.path.isdir(MODEL_PATH):
-        return MODEL_PATH
-    from huggingface_hub import snapshot_download
-
-    return snapshot_download(repo_id=MODEL_PATH, repo_type="model")
-
-
-# ─── Dummy dataset stats (identity) ──────────────────────────────────
-def _dummy_dataset_stats() -> dict:
-    return {
-        "observation.state": {
-            "mean": torch.zeros(STATE_DIM),
-            "std": torch.ones(STATE_DIM),
-            "q01": torch.zeros(STATE_DIM),
-            "q99": torch.ones(STATE_DIM),
-        },
-        "action": {
-            "mean": torch.zeros(ACTION_DIM),
-            "std": torch.ones(ACTION_DIM),
-            "q01": torch.zeros(ACTION_DIM),
-            "q99": torch.ones(ACTION_DIM),
-        },
-        "images": {
-            cam: {
-                "mean": torch.zeros(3, 224, 224),
-                "std": torch.ones(3, 224, 224),
-                "q01": torch.zeros(3, 224, 224),
-                "q99": torch.ones(3, 224, 224),
-            }
-            for cam in ("base_0_rgb", "left_wrist_0_rgb", "right_wrist_0_rgb")
-        },
-    }
-
-
-def _create_dummy_batch(batch_size: int = BATCH_SIZE, device: str = DEVICE) -> dict:
-    """Reproducible dummy inputs — identical across both implementations."""
-    g = torch.Generator(device="cpu").manual_seed(0)
-    prompt = "Pick up the red block and place it in the bin"
-    return {
-        "observation.state": torch.randn(batch_size, STATE_DIM, generator=g, dtype=torch.float32).to(device),
-        "action": torch.randn(batch_size, ACTION_HORIZON, ACTION_DIM, generator=g, dtype=torch.float32).to(device),
-        "observation.images.base_0_rgb": torch.rand(batch_size, 3, 224, 224, generator=g, dtype=torch.float32).to(
-            device
-        ),
-        "observation.images.left_wrist_0_rgb": torch.rand(batch_size, 3, 224, 224, generator=g, dtype=torch.float32).to(
-            device
-        ),
-        "observation.images.right_wrist_0_rgb": torch.rand(
-            batch_size, 3, 224, 224, generator=g, dtype=torch.float32
-        ).to(device),
-        "task": [prompt for _ in range(batch_size)],
-    }
-
-
-# ─── LeRobot instantiation ────────────────────────────────────────────
-def _instantiate_lerobot():
-    from lerobot.policies.pi0 import PI0Policy
-    from lerobot.policies.pi0.processor_pi0 import make_pi0_pre_post_processors
-
-    policy = PI0Policy.from_pretrained(MODEL_PATH, strict=True)
-    policy.to(DEVICE)
-    policy.config.device = DEVICE
-    policy.eval()
-
-    pre, post = make_pi0_pre_post_processors(config=policy.config, dataset_stats=_dummy_dataset_stats())
-    return policy, pre, post
-
-
-# ─── vllm-omni instantiation ──────────────────────────────────────────
-def _instantiate_vllm_omni():
-    """Build the vllm-omni π0 model in isolation (no pipeline, no engine)."""
-    from vllm_omni.diffusion.models.pi0 import Pi0Config, Pi0ForActionPrediction
-
-    cfg = Pi0Config(
-        max_action_dim=ACTION_DIM,
-        max_state_dim=STATE_DIM,
-        chunk_size=ACTION_HORIZON,
-        num_inference_steps=NUM_STEPS,
-        dtype=DTYPE_STR,
-    )
-    model = Pi0ForActionPrediction(cfg)
-    model.to(DEVICE).eval()
-    _load_lerobot_weights(model)
-    return model
-
-
-def _load_lerobot_weights(model):
-    """Feed the ``lerobot/pi0_base`` safetensors into the vllm-omni model.
-
-    ``Pi0ForActionPrediction.load_weights`` handles the leading ``model.``
-    prefix and the flat→nested / lm_head→embed_tokens remaps, so we can pass
-    the raw checkpoint dict.
-    """
-    import safetensors.torch
-
-    cache_dir = _resolve_checkpoint_dir()
-    path = os.path.join(cache_dir, "model.safetensors")
-    state = safetensors.torch.load_file(path)
-    model.load_weights(list(state.items()))
-
-
-# ─── Helpers to extract LeRobot's pre-processed inputs ────────────────
-def _extract_lerobot_model_inputs(lerobot_policy, processed_batch):
-    """Mimic what ``PI0Policy.predict_action_chunk`` feeds into
-    ``self.model.sample_actions``. We use these *exact* tensors for vllm-omni
-    so any divergence must come from the core model, not preprocessing.
-    """
-    images, img_masks = lerobot_policy._preprocess_images(processed_batch)
-    from lerobot.utils.constants import (
-        OBS_LANGUAGE_ATTENTION_MASK,
-        OBS_LANGUAGE_TOKENS,
-    )
-
-    lang_tokens = processed_batch[OBS_LANGUAGE_TOKENS]
-    lang_masks = processed_batch[OBS_LANGUAGE_ATTENTION_MASK]
-    state = lerobot_policy.prepare_state(processed_batch)
-    return images, img_masks, lang_tokens, lang_masks, state
-
-
-# ─── Shared fixed-noise sampler ───────────────────────────────────────
-def _make_fixed_noise(batch_size: int, device: str) -> torch.Tensor:
-    g = torch.Generator(device="cpu").manual_seed(42)
-    return torch.randn(batch_size, ACTION_HORIZON, ACTION_DIM, generator=g, dtype=torch.float32).to(device)
-
-
-# ─── Main test ────────────────────────────────────────────────────────
-@pytest.mark.skipif(not _HAS_LEROBOT, reason="lerobot not installed (run in a lerobot venv).")
-def test_pi0_vllm_omni_vs_lerobot():
-    print("\n[parity] Instantiating LeRobot…")
-    lerobot_policy, lerobot_pre, _ = _instantiate_lerobot()
-
-    print("[parity] Instantiating vllm-omni…")
-    omni_model = _instantiate_vllm_omni()
-
-    print("[parity] Preparing shared inputs…")
-    raw_batch = _create_dummy_batch()
-    processed_batch = lerobot_pre(copy.deepcopy(raw_batch))
-    images, img_masks, lang_tokens, lang_masks, state = _extract_lerobot_model_inputs(lerobot_policy, processed_batch)
-    noise = _make_fixed_noise(raw_batch["observation.state"].shape[0], DEVICE)
-
-    print(f"[parity] state.shape={state.shape}  lang_tokens.shape={lang_tokens.shape}")
-    print(f"[parity] images[0].shape={images[0].shape} (num_cams={len(images)})")
-    print(f"[parity] noise.shape={noise.shape}  dtype={noise.dtype}")
-
-    # ── LeRobot forward ──
-    print("[parity] Running LeRobot sample_actions…")
-    with torch.no_grad():
-        lerobot_actions = lerobot_policy.model.sample_actions(
-            images,
-            img_masks,
-            lang_tokens,
-            lang_masks,
-            state,
-            noise=noise,
-            num_steps=NUM_STEPS,
-        )
-    print(
-        f"[parity] LeRobot actions: shape={lerobot_actions.shape} "
-        f"mean={lerobot_actions.mean().item():.6f} std={lerobot_actions.std().item():.6f}"
-    )
-
-    # ── vllm-omni forward ──
-    print("[parity] Running vllm-omni sample_actions…")
-    with torch.no_grad():
-        omni_actions = omni_model.sample_actions(
-            images=images,
-            image_masks=img_masks,
-            lang_tokens=lang_tokens,
-            lang_masks=lang_masks,
-            state=state,
-            noise=noise,
-            num_steps=NUM_STEPS,
-        )
-    print(
-        f"[parity] vllm-omni actions: shape={omni_actions.shape} "
-        f"mean={omni_actions.mean().item():.6f} std={omni_actions.std().item():.6f}"
-    )
-
-    # ── Compare ──
-    diff = (lerobot_actions.float() - omni_actions.float()).abs()
-    max_diff = diff.max().item()
-    mean_diff = diff.mean().item()
-    print(f"[parity] |Δ| max={max_diff:.2e}  mean={mean_diff:.2e}  atol={ATOL:.1e}")
-    close = torch.allclose(lerobot_actions.float(), omni_actions.float(), atol=ATOL)
-    print(f"[parity] torch.allclose(atol={ATOL}): {close}")
-
-    if not close:
-        print("\n[parity] ⚠️  Outputs diverge — running per-stage diagnostics…")
-        _diagnose_divergence(
-            lerobot_policy.model,
-            omni_model,
-            images,
-            img_masks,
-            lang_tokens,
-            lang_masks,
-            state,
-            noise,
-        )
-
-    assert close, (
-        f"vllm-omni vs LeRobot actions differ beyond atol={ATOL}. max_diff={max_diff:.2e}  mean_diff={mean_diff:.2e}"
-    )
-
-
-# ─── Per-stage divergence diagnostics ─────────────────────────────────
-@torch.no_grad()
-def _diagnose_divergence(
-    lerobot_flow_model,
-    omni_model,
-    images,
-    img_masks,
-    lang_tokens,
-    lang_masks,
-    state,
-    noise,
-):
-    """Localize a numerical mismatch to a specific pipeline stage:
-    1. Prefix embeddings (SigLIP / embed_tokens / projector) — image vs lang.
-    2. Prefix KV cache layer 0 — PaliGemma LM attention.
-    3. A single denoise_step velocity at t=1.0 — action expert forward.
-    """
-    from vllm_omni.diffusion.models.pi0.modeling_pi0 import (
-        make_att_2d_masks,
-        prepare_attention_masks_4d,
-    )
-
-    # ── Stage 1: prefix embeddings ──
-    lr_prefix_embs, lr_prefix_pad, lr_prefix_att = lerobot_flow_model.embed_prefix(
-        images, img_masks, lang_tokens, lang_masks
-    )
-    sg_prefix_embs, sg_prefix_pad, sg_prefix_att = omni_model.embed_prefix(images, img_masks, lang_tokens, lang_masks)
-    total_diff = (lr_prefix_embs.float() - sg_prefix_embs.float()).abs().max().item()
-    print(f"[diag] prefix_embs max |Δ| = {total_diff:.2e}   (shape={tuple(sg_prefix_embs.shape)})")
-    print(f"[diag] prefix_pad_masks equal: {torch.equal(lr_prefix_pad, sg_prefix_pad)}")
-    print(f"[diag] prefix_att_masks equal: {torch.equal(lr_prefix_att.bool(), sg_prefix_att.bool())}")
-
-    num_cams = len(images)
-    img_len = 256 * num_cams
-    lang_len = lr_prefix_embs.shape[1] - img_len
-    img_diff = (lr_prefix_embs[:, :img_len].float() - sg_prefix_embs[:, :img_len].float()).abs().max().item()
-    lang_diff = (lr_prefix_embs[:, img_len:].float() - sg_prefix_embs[:, img_len:].float()).abs().max().item()
-    print(f"[diag]   image slice [:{img_len}] max|Δ| = {img_diff:.2e}   (num_cams={num_cams})")
-    print(f"[diag]   lang slice  [{img_len}:] max|Δ| = {lang_diff:.2e}   (len={lang_len})")
-
-    def _stats(name, t):
-        t = t.float()
-        print(
-            f"[diag] {name} prefix_embs: mean={t.mean().item():+.4f} "
-            f"std={t.std().item():.4f} min={t.min().item():+.2f} "
-            f"max={t.max().item():+.2f}"
-        )
-
-    _stats("LeRobot ", lr_prefix_embs)
-    _stats("vllm-omni", sg_prefix_embs)
-
-    # ── Stage 2: prefix KV cache ──
-    prefix_att_2d = make_att_2d_masks(sg_prefix_pad, sg_prefix_att)
-    prefix_pos = torch.cumsum(sg_prefix_pad, dim=1) - 1
-    prefix_att_4d = prepare_attention_masks_4d(prefix_att_2d)
-
-    lerobot_flow_model.paligemma_with_expert.paligemma.language_model.config._attn_implementation = "eager"
-    _, lr_kv = lerobot_flow_model.paligemma_with_expert.forward(
-        attention_mask=prefix_att_4d,
-        position_ids=prefix_pos,
-        past_key_values=None,
-        inputs_embeds=[lr_prefix_embs, None],
-        use_cache=True,
-    )
-    _, sg_kv = omni_model.paligemma_with_expert.forward(
-        attention_mask=prefix_att_4d,
-        position_ids=prefix_pos,
-        past_key_values=None,
-        inputs_embeds=[sg_prefix_embs, None],
-        use_cache=True,
-    )
-
-    def _layer0_kv(cache):
-        if isinstance(cache, list):
-            return cache[0]
-        layer0 = cache.layers[0]
-        return layer0.keys, layer0.values
-
-    try:
-        lr_k, lr_v = _layer0_kv(lr_kv)
-        sg_k, sg_v = _layer0_kv(sg_kv)
-        dk = (lr_k.float() - sg_k.float()).abs().max().item()
-        dv = (lr_v.float() - sg_v.float()).abs().max().item()
-        print(f"[diag] prefix KV layer0  K max|Δ|={dk:.2e}  V max|Δ|={dv:.2e}")
-    except Exception as e:  # noqa: BLE001
-        print(f"[diag] could not extract prefix KV for comparison: {e}")
-
-    # ── Stage 3: a single denoise_step at t=1.0 ──
-    bsize = state.shape[0]
-    t = torch.ones(bsize, dtype=torch.float32, device=state.device)
-    lr_vt = lerobot_flow_model.denoise_step(state, sg_prefix_pad, lr_kv, noise, t)
-    sg_vt = omni_model.denoise_step(state, sg_prefix_pad, sg_kv, noise, t)
-    print(f"[diag] denoise_step(t=1) v_t max|Δ| = {(lr_vt.float() - sg_vt.float()).abs().max().item():.2e}")
-
-
-if __name__ == "__main__":
-    test_pi0_vllm_omni_vs_lerobot()

tests/diffusion/models/pi0/test_pi0_units.py

@@ -1,585 +0,0 @@
-# SPDX-License-Identifier: Apache-2.0
-# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
-"""CPU unit tests for the π0 (Pi-Zero) VLA model.
-
-Deliberately lightweight: pure utility code (config plumbing, mask building,
-sinusoidal embeddings, normalization, image preprocessing, weight-load remap)
-that runs on CPU with no weights and no lerobot. The full LeRobot parity check
-lives in ``tests/diffusion/models/pi0/test_pi0_parity.py`` and requires a model download.
-
-    pytest tests/diffusion/models/pi0/test_pi0_units.py -v
-"""
-
-from __future__ import annotations
-
-import logging
-import math
-
-import pytest
-import torch
-
-from vllm_omni.diffusion.models.pi0.config import Pi0Config
-from vllm_omni.diffusion.models.pi0.modeling_pi0 import (
-    OPENPI_ATTENTION_MASK_VALUE,
-    Pi0ForActionPrediction,
-    _apply_norm,
-    _build_norm_buffers,
-    create_sinusoidal_pos_embedding,
-    get_gemma_config,
-    make_att_2d_masks,
-    prepare_attention_masks_4d,
-)
-from vllm_omni.diffusion.models.pi0.processor_pi0 import (
-    Pi0ImageProcessor,
-    build_model_inputs,
-    pil_image_to_tensor,
-    resize_with_pad,
-)
-
-pytestmark = [pytest.mark.core_model, pytest.mark.cpu]
-
-
-# ----------------------------------------------------------------------------
-# Pi0Config dataclass + checkpoint resolver
-# ----------------------------------------------------------------------------
-_LEROBOT_CFG = {
-    "type": "pi0",
-    "paligemma_variant": "gemma_2b",
-    "action_expert_variant": "gemma_300m",
-    "chunk_size": 50,
-    "max_action_dim": 32,
-    "max_state_dim": 32,
-    "num_inference_steps": 10,
-    "image_resolution": [224, 224],
-    "tokenizer_max_length": 48,
-    "dtype": "float32",
-    # Training-only keys that must be filtered out of the dataclass.
-    "optimizer_lr": 2.5e-5,
-    "scheduler_warmup_steps": 1000,
-    "input_features": {
-        "observation.images.base_0_rgb": {"type": "VISUAL", "shape": [3, 224, 224]},
-        "observation.images.left_wrist_0_rgb": {"type": "VISUAL", "shape": [3, 224, 224]},
-        "observation.images.right_wrist_0_rgb": {"type": "VISUAL", "shape": [3, 224, 224]},
-        "observation.state": {"type": "STATE", "shape": [32]},
-    },
-}
-
-_EXPECTED_CAMERA_ORDER = [
-    "observation.images.base_0_rgb",
-    "observation.images.left_wrist_0_rgb",
-    "observation.images.right_wrist_0_rgb",
-]
-
-
-def test_config_camera_order_from_input_features():
-    c = Pi0Config.from_model_config(_LEROBOT_CFG)
-    assert c.image_feature_keys == _EXPECTED_CAMERA_ORDER
-
-
-def test_config_filters_training_only_keys():
-    c = Pi0Config.from_model_config(_LEROBOT_CFG)
-    assert not hasattr(c, "optimizer_lr")
-    assert not hasattr(c, "scheduler_warmup_steps")
-
-
-def test_config_coerces_image_resolution_to_tuple():
-    c = Pi0Config.from_model_config(_LEROBOT_CFG)
-    assert c.image_resolution == (224, 224)
-    assert isinstance(c.image_resolution, tuple)
-
-
-def test_config_non_square_resolution_raises():
-    with pytest.raises(ValueError):
-        Pi0Config(image_resolution=(224, 256))
-
-
-@pytest.mark.parametrize("bad", [(224,), (224, 224, 3), 224])
-def test_config_malformed_resolution_raises(bad):
-    with pytest.raises(ValueError):
-        Pi0Config(image_resolution=bad)
-
-
-def test_config_defaults_and_empty():
-    c = Pi0Config.from_model_config(None)
-    assert c.chunk_size == 50
-    assert c.max_action_dim == 32
-    assert c.max_state_dim == 32
-    assert c.num_inference_steps == 10
-    assert c.image_resolution == (224, 224)
-    assert c.tokenizer_max_length == 48
-    assert c.max_cameras == 3
-    assert c.norm_stats is None
-    assert c.image_key_map == {}
-
-
-def test_config_deploy_yaml_override():
-    """Deploy-yaml model_config is authoritative (e.g. force dtype/variant)."""
-    c = Pi0Config.from_model_config({**_LEROBOT_CFG, "dtype": "bfloat16", "num_inference_steps": 4})
-    assert c.dtype == "bfloat16"
-    assert c.num_inference_steps == 4
-
-
-# ----------------------------------------------------------------------------
-# Kernel utilities: sinusoidal timestep embedding
-# ----------------------------------------------------------------------------
-def test_sinusoidal_embedding_shape():
-    emb = create_sinusoidal_pos_embedding(torch.tensor([0.0, 0.5, 1.0]), dimension=64)
-    assert emb.shape == (3, 64)
-
-
-def test_sinusoidal_embedding_values_at_zero():
-    emb = create_sinusoidal_pos_embedding(torch.tensor([0.0]), dimension=4)
-    assert torch.allclose(emb[0, :2].float(), torch.zeros(2), atol=1e-6)
-    assert torch.allclose(emb[0, 2:].float(), torch.ones(2), atol=1e-6)
-
-
-def test_sinusoidal_embedding_odd_dim_raises():
-    with pytest.raises(ValueError):
-        create_sinusoidal_pos_embedding(torch.tensor([1.0]), dimension=3)
-
-
-def test_sinusoidal_embedding_non_1d_raises():
-    with pytest.raises(ValueError):
-        create_sinusoidal_pos_embedding(torch.tensor([[0.0, 0.5]]), dimension=4)
-
-
-def test_sinusoidal_embedding_distinct_timesteps_differ():
-    emb = create_sinusoidal_pos_embedding(torch.tensor([0.1, 0.2, 0.5, 0.9]), dimension=16)
-    for i in range(emb.shape[0]):
-        for j in range(i + 1, emb.shape[0]):
-            assert not torch.allclose(emb[i], emb[j])
-
-
-# ----------------------------------------------------------------------------
-# Kernel utilities: attention mask construction
-# ----------------------------------------------------------------------------
-def test_make_att_2d_masks_bidirectional():
-    result = make_att_2d_masks(torch.ones(1, 4, dtype=torch.bool), torch.zeros(1, 4, dtype=torch.long))
-    assert result.all()
-
-
-def test_make_att_2d_masks_causal():
-    result = make_att_2d_masks(torch.ones(1, 4, dtype=torch.bool), torch.ones(1, 4, dtype=torch.long))
-    expected = torch.tril(torch.ones(4, 4, dtype=torch.bool))
-    assert torch.equal(result[0], expected)
-
-
-def test_make_att_2d_masks_prefix_lm():
-    result = make_att_2d_masks(torch.ones(1, 5, dtype=torch.bool), torch.tensor([[0, 0, 0, 1, 1]], dtype=torch.long))
-    assert result[0, 0, 0]
-    assert not result[0, 0, 3]
-    assert result[0, 3, 0]
-    assert not result[0, 3, 4]
-
-
-def test_make_att_2d_masks_respects_padding():
-    pad = torch.tensor([[1, 1, 1, 0]], dtype=torch.bool)  # last token is padding
-    att = torch.zeros(1, 4, dtype=torch.long)  # full bidirectional
-    result = make_att_2d_masks(pad, att)
-    assert not result[0, :, 3].any(), "Nothing should attend to padded key"
-    assert not result[0, 3, :].any(), "Padded query should attend to nothing"
-
-
-def test_prepare_attention_masks_4d():
-    mask_2d = torch.tensor([[[True, False], [True, True]]])
-    mask_4d = prepare_attention_masks_4d(mask_2d)
-    assert mask_4d.shape == (1, 1, 2, 2)
-    assert mask_4d[0, 0, 0, 0].item() == 0.0
-    assert mask_4d[0, 0, 1, 1].item() == 0.0
-    expected = torch.tensor(OPENPI_ATTENTION_MASK_VALUE, dtype=mask_4d.dtype)
-    assert torch.equal(mask_4d[0, 0, 0, 1], expected)
-
-
-def test_attention_mask_value_finite_and_negative():
-    assert math.isfinite(OPENPI_ATTENTION_MASK_VALUE)
-    assert OPENPI_ATTENTION_MASK_VALUE < -1e30
-
-
-# ----------------------------------------------------------------------------
-# Normalization round-trips
-# ----------------------------------------------------------------------------
-def test_build_norm_buffers_missing():
-    assert _build_norm_buffers(None, "state") is None
-    assert _build_norm_buffers({}, "state") is None
-    assert _build_norm_buffers({"state": None}, "state") is None
-
-
-def test_build_norm_buffers_mean_std():
-    stats = {"state": {"mode": "mean_std", "mean": [1.0, 2.0], "std": [0.5, 0.5]}}
-    out = _build_norm_buffers(stats, "state")
-    assert out["mode"] == "mean_std"
-    assert torch.equal(out["mean"], torch.tensor([1.0, 2.0]))
-    assert torch.equal(out["std"], torch.tensor([0.5, 0.5]))
-
-
-def test_build_norm_buffers_min_max():
-    stats = {"action": {"mode": "min_max", "min": [-1.0, 0.0], "max": [1.0, 2.0]}}
-    out = _build_norm_buffers(stats, "action")
-    assert out["mode"] == "min_max"
-    assert torch.equal(out["min"], torch.tensor([-1.0, 0.0]))
-    assert torch.equal(out["max"], torch.tensor([1.0, 2.0]))
-
-
-def test_apply_norm_mean_std_forward_inverse():
-    stats = _build_norm_buffers({"s": {"mode": "mean_std", "mean": [0.5, -0.2], "std": [2.0, 0.1]}}, "s")
-    x = torch.tensor([[1.0, 0.3]])
-    z = _apply_norm(x, stats, inverse=False)
-    assert torch.allclose(z, torch.tensor([[0.25, 5.0]]), atol=1e-5)
-    x_back = _apply_norm(z, stats, inverse=True)
-    assert torch.allclose(x_back, x, atol=1e-5)
-
-
-def test_apply_norm_min_max_forward_inverse():
-    stats = _build_norm_buffers({"s": {"mode": "min_max", "min": [0.0, -1.0], "max": [1.0, 1.0]}}, "s")
-    x = torch.tensor([[0.25, 0.5]])
-    z = _apply_norm(x, stats, inverse=False)
-    assert torch.allclose(z, torch.tensor([[-0.5, 0.5]]), atol=1e-5)
-    x_back = _apply_norm(z, stats, inverse=True)
-    assert torch.allclose(x_back, x, atol=1e-5)
-
-
-def test_apply_norm_preserves_padded_tail():
-    stats = _build_norm_buffers({"s": {"mode": "mean_std", "mean": [0.0, 0.0], "std": [1.0, 1.0]}}, "s")
-    x = torch.tensor([[1.0, 2.0, 99.0, -99.0]])
-    z = _apply_norm(x, stats, inverse=False)
-    assert torch.allclose(z[:, :2], x[:, :2])
-    assert torch.allclose(z[:, 2:], x[:, 2:])
-
-
-def test_apply_norm_noop_when_stats_none():
-    x = torch.randn(2, 5)
-    assert torch.equal(_apply_norm(x, None, inverse=False), x)
-    assert torch.equal(_apply_norm(x, None, inverse=True), x)
-
-
-# ----------------------------------------------------------------------------
-# Gemma variant configs
-# ----------------------------------------------------------------------------
-def test_gemma_2b_dims():
-    c = get_gemma_config("gemma_2b")
-    assert c.width == 2048
-    assert c.depth == 18
-
-
-def test_gemma_300m_dims():
-    c = get_gemma_config("gemma_300m")
-    assert c.width == 1024
-    assert c.depth == 18
-
-
-def test_gemma_unknown_variant_raises():
-    with pytest.raises(ValueError):
-        get_gemma_config("gemma_7b")
-
-
-# ----------------------------------------------------------------------------
-# load_weights remap rules (tiny model; needs torch + transformers)
-# ----------------------------------------------------------------------------
-def _tiny_pi0_model():
-    """Construct a tiny π0 model for remap tests (full Gemma depth but small
-    action/state dims — the PaliGemma backbone is always full-size)."""
-    cfg = Pi0Config(max_action_dim=8, max_state_dim=8, chunk_size=2)
-    return Pi0ForActionPrediction(cfg)
-
-
-@pytest.fixture(autouse=True)
-def _silence_pi0_loader(caplog):
-    # The remap tests feed a single synthetic weight; the loader's
-    # "missing params" warning is expected noise here.
-    logging.getLogger("vllm_omni.diffusion.models.pi0.modeling_pi0").setLevel(logging.ERROR)
-    yield
-
-
-@pytest.mark.slow
-def test_remap_paligemma_submodules_to_nested_layout():
-    model = _tiny_pi0_model()
-    params = dict(model.named_parameters())
-    target = next(k for k in params if k.startswith("paligemma_with_expert.paligemma.model.vision_tower."))
-    flat_key = "model." + target.replace(
-        "paligemma_with_expert.paligemma.model.vision_tower.",
-        "paligemma_with_expert.paligemma.vision_tower.",
-    )
-    payload = torch.full(params[target].shape, 0.12345)
-    before = params[target].detach().clone()
-    model.load_weights([(flat_key, payload)])
-    after = dict(model.named_parameters())[target]
-    assert not torch.equal(before, after), "flat→nested remap broken"
-    assert torch.allclose(after, payload)
-
-
-@pytest.mark.slow
-def test_lm_head_remap_to_embed_tokens():
-    model = _tiny_pi0_model()
-    params = dict(model.named_parameters())
-    target = "paligemma_with_expert.paligemma.model.language_model.embed_tokens.weight"
-    assert target in params
-    payload = torch.full(params[target].shape, 0.5)
-    before = params[target].detach().clone()
-    model.load_weights([("model.paligemma_with_expert.paligemma.lm_head.weight", payload)])
-    after = dict(model.named_parameters())[target]
-    assert not torch.equal(before, after), "lm_head→embed_tokens remap broken"
-    assert torch.allclose(after, payload)
-
-
-@pytest.mark.slow
-def test_vision_tower_version_robust_remap():
-    """transformers >=5.4 flattens SigLIP from vision_tower.vision_model.* to
-    vision_tower.*; the lerobot/pi0_base checkpoint uses the vision_model infix.
-    load_weights must adapt so vision weights aren't silently dropped (which
-    would leave the vision encoder at random init)."""
-    model = _tiny_pi0_model()
-    params = dict(model.named_parameters())
-    # A representative vision-tower param as it exists on the model.
-    target = next(
-        k
-        for k in params
-        if k.startswith("paligemma_with_expert.paligemma.model.vision_tower.") and "patch_embedding.weight" in k
-    )
-    vt = "paligemma_with_expert.paligemma.model.vision_tower."
-    rest = target[len(vt) :]
-    # Build the checkpoint key in whichever layout differs from the model's.
-    if rest.startswith("vision_model."):
-        ckpt_key = vt + rest[len("vision_model.") :]
-    else:
-        ckpt_key = vt + "vision_model." + rest
-    payload = torch.full(params[target].shape, 0.073)
-    before = params[target].detach().clone()
-    model.load_weights([("model." + ckpt_key, payload)])
-    after = dict(model.named_parameters())[target]
-    assert not torch.equal(before, after), "vision_tower version-robust remap broken"
-    assert torch.allclose(after, payload)
-
-
-# ----------------------------------------------------------------------------
-# Image processor
-# ----------------------------------------------------------------------------
-def test_resize_with_pad_square():
-    out = resize_with_pad(torch.ones(1, 3, 224, 224), 224, 224)
-    assert out.shape == (1, 3, 224, 224)
-
-
-def test_resize_with_pad_wide():
-    """Wide source (H < W): padding applied to top/bottom (rows)."""
-    out = resize_with_pad(torch.zeros(1, 3, 100, 200), 224, 224)
-    assert out.shape == (1, 3, 224, 224)
-    assert out[0, 0, 0, 112].item() == -1.0  # top row is padded
-
-
-def test_resize_with_pad_tall():
-    """Tall source (H > W): padding applied to left/right (cols)."""
-    out = resize_with_pad(torch.zeros(1, 3, 400, 100), 224, 224)
-    assert out.shape == (1, 3, 224, 224)
-    assert out[0, 0, 112, 0].item() == -1.0  # left-most column padded
-
-
-def test_resize_with_pad_rejects_non_4d():
-    with pytest.raises(ValueError):
-        resize_with_pad(torch.zeros(3, 100, 200), 224, 224)
-
-
-def test_pil_to_tensor():
-    import numpy as np
-    from PIL import Image
-
-    img = Image.fromarray(np.full((10, 10, 3), 128, dtype=np.uint8))
-    t = pil_image_to_tensor(img)
-    assert t.shape == (1, 3, 10, 10)
-    expected = 128.0 / 255.0 * 2.0 - 1.0
-    assert abs(t[0, 0, 5, 5].item() - expected) < 1e-4
-
-
-def test_pil_to_tensor_rgba_to_rgb():
-    import numpy as np
-    from PIL import Image
-
-    rgba = np.full((8, 8, 4), 200, dtype=np.uint8)
-    img = Image.fromarray(rgba, mode="RGBA")
-    t = pil_image_to_tensor(img)
-    assert t.shape == (1, 3, 8, 8)
-
-
-def test_make_empty_image():
-    proc = Pi0ImageProcessor(image_size=224)
-    empty = proc.make_empty_image()
-    assert empty.shape == (1, 3, 224, 224)
-    assert (empty == -1.0).all()
-
-
-def test_preprocess_single_resizes_ndarray():
-    import numpy as np
-
-    proc = Pi0ImageProcessor(image_size=224)
-    arr = np.full((120, 120, 3), 100, dtype=np.uint8)
-    t = proc.preprocess_single(arr)
-    assert t.shape == (1, 3, 224, 224)
-
-
-# ----------------------------------------------------------------------------
-# build_model_inputs: camera ordering, padding, masks, state
-# ----------------------------------------------------------------------------
-class _FakeTokenizer:
-    """Minimal stand-in for the PaliGemma tokenizer (no download)."""
-
-    def __call__(
-        self, text, padding=None, max_length=None, truncation=None, add_special_tokens=None, return_tensors=None
-    ):
-        ids = [2] + [ord(c) % 100 for c in text][: max_length - 1]
-        ids = ids[:max_length]
-        attn = [1] * len(ids)
-        while len(ids) < max_length:
-            ids.append(0)
-            attn.append(0)
-        return {"input_ids": ids, "attention_mask": attn}
-
-
-def test_build_model_inputs_camera_order_and_padding():
-    import numpy as np
-
-    config = Pi0Config.from_model_config(_LEROBOT_CFG)  # 3-camera config
-    device = torch.device("cpu")
-    # Provide only 2 of the 3 declared cameras.
-    robot_obs = {
-        "images": {
-            "observation.images.base_0_rgb": np.full((224, 224, 3), 50, dtype=np.uint8),
-            "observation.images.left_wrist_0_rgb": np.full((224, 224, 3), 80, dtype=np.uint8),
-        },
-        "state": [0.1] * 14,  # raw state shorter than max_state_dim=32
-        "prompt": "pick up the block",
-    }
-    images, image_masks, lang_tokens, lang_masks, state = build_model_inputs(
-        robot_obs, config, _FakeTokenizer(), device
-    )
-    assert len(images) == 3
-    assert all(img.shape == (1, 3, 224, 224) for img in images)
-    masks = [bool(m.item()) for m in image_masks]
-    assert masks == [True, True, False]
-    # The padded (3rd) camera must be all -1.
-    assert (images[2] == -1.0).all()
-    assert lang_tokens.shape == (1, 48)
-    assert lang_masks.shape == (1, 48)
-    assert state.shape == (1, 32)
-    # State padded with zeros beyond the 14 real entries.
-    assert torch.allclose(state[0, 14:], torch.zeros(18))
-
-
-def test_build_model_inputs_state_truncated():
-    config = Pi0Config.from_model_config(_LEROBOT_CFG)
-    robot_obs = {"images": {}, "state": [0.5] * 40, "prompt": "x"}
-    _, _, _, _, state = build_model_inputs(robot_obs, config, _FakeTokenizer(), torch.device("cpu"))
-    assert state.shape == (1, 32)
-
-
-def test_build_model_inputs_image_key_map():
-    import numpy as np
-
-    cfg = Pi0Config.from_model_config({**_LEROBOT_CFG, "image_key_map": {"cam_high": "observation.images.base_0_rgb"}})
-    robot_obs = {
-        "images": {"cam_high": np.full((224, 224, 3), 60, dtype=np.uint8)},
-        "prompt": "go",
-    }
-    _, image_masks, _, _, _ = build_model_inputs(robot_obs, cfg, _FakeTokenizer(), torch.device("cpu"))
-    masks = [bool(m.item()) for m in image_masks]
-    assert masks == [True, False, False]  # only base_0_rgb (mapped) is present
-
-
-# ----------------------------------------------------------------------------
-# Version-stability / self-consistency (gated on a real pi0_base checkpoint)
-#
-# The LeRobot parity gate (test_pi0_parity.py) only validates correctness at
-# transformers 5.3.0 (where LeRobot loads pi0_base cleanly). Production vllm-omni
-# ships transformers 5.6.2, and LeRobot is not a usable reference there (it
-# incorrectly loads the SigLIP vision tower at >=5.4). So we pin the kernel's deterministic
-# output to golden values established at 5.3.0 and re-check them here: running on
-# the production transformers proves the port is version-stable and guards the
-# vision_tower / embed_scale version-robust paths against regressions.
-#
-#   PI0_SELFCONSIST_CKPT=/path/to/pi0_base \
-#       pytest tests/diffusion/models/pi0/test_pi0_units.py -k version_stable -v -s
-# ----------------------------------------------------------------------------
-import os  # noqa: E402
-
-_SELFCONSIST_CKPT = os.environ.get("PI0_SELFCONSIST_CKPT")
-
-# Golden values established at transformers 5.3.0 (where parity = 7.15e-07).
-_GOLDEN_SUM = -67.808113
-_GOLDEN_STD = 0.252789
-_GOLDEN_MEAN = -0.021190
-_GOLDEN_EMBED_IMAGE_STD = 2.17603
-
-
-def _selfconsist_inputs(device):
-    # NB: draw order is load-bearing — the golden values were established with
-    # state drawn first, then the 3 images, then lang (shared RNG, seed 0).
-    g = torch.Generator(device="cpu").manual_seed(0)
-    b = 2
-    state = torch.randn(b, 32, generator=g)
-    images = [torch.rand(b, 3, 224, 224, generator=g) for _ in range(3)]
-    masks = [torch.ones(b, dtype=torch.bool) for _ in range(3)]
-    lang = torch.randint(0, 1000, (b, 48), generator=g)
-    lmask = torch.ones(b, 48, dtype=torch.bool)
-    ng = torch.Generator(device="cpu").manual_seed(42)
-    noise = torch.randn(b, 50, 32, generator=ng)
-    return (
-        [im.to(device) for im in images],
-        [m.to(device) for m in masks],
-        lang.to(device),
-        lmask.to(device),
-        state.to(device),
-        noise.to(device),
-    )
-
-
-def _selfconsist_model(device):
-    import safetensors.torch
-
-    model = Pi0ForActionPrediction(
-        Pi0Config(max_action_dim=32, max_state_dim=32, chunk_size=50, num_inference_steps=10, dtype="float32")
-    )
-    state = safetensors.torch.load_file(os.path.join(_SELFCONSIST_CKPT, "model.safetensors"))
-    filled = model.load_weights(list(state.items()))
-    expected = {n for n, _ in model.named_parameters() if "rotary_emb" not in n and not n.endswith(".inv_freq")}
-    missing = expected - (filled or set())
-    assert not missing, f"{len(missing)} params received no weight (e.g. {sorted(missing)[:3]})"
-    return model.to(device).eval()
-
-
-@pytest.mark.slow
-@pytest.mark.skipif(not _SELFCONSIST_CKPT, reason="Set PI0_SELFCONSIST_CKPT to a pi0_base dir.")
-@torch.no_grad()
-def test_version_stable_actions():
-    device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
-    model = _selfconsist_model(device)
-    images, masks, lang, lmask, state, noise = _selfconsist_inputs(device)
-    actions = (
-        model.sample_actions(
-            images=images,
-            image_masks=masks,
-            lang_tokens=lang,
-            lang_masks=lmask,
-            state=state,
-            noise=noise,
-            num_steps=10,
-        )
-        .float()
-        .cpu()
-    )
-    s, m, std = actions.sum().item(), actions.mean().item(), actions.std().item()
-    assert abs(s - _GOLDEN_SUM) < 0.5, f"action sum drifted: {s} vs {_GOLDEN_SUM}"
-    assert abs(std - _GOLDEN_STD) < 0.01, f"action std drifted: {std} vs {_GOLDEN_STD}"
-    assert abs(m - _GOLDEN_MEAN) < 0.01, f"action mean drifted: {m} vs {_GOLDEN_MEAN}"
-
-
-@pytest.mark.slow
-@pytest.mark.skipif(not _SELFCONSIST_CKPT, reason="Set PI0_SELFCONSIST_CKPT to a pi0_base dir.")
-@torch.no_grad()
-def test_version_stable_embed_image():
-    """Directly guards the SigLIP vision-tower load across the >=5.4 flattening."""
-    device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
-    model = _selfconsist_model(device)
-    g = torch.Generator(device="cpu").manual_seed(7)
-    img = torch.rand(1, 3, 224, 224, generator=g, dtype=torch.float32).to(device)
-    std = model.paligemma_with_expert.embed_image(img).float().std().item()
-    assert abs(std - _GOLDEN_EMBED_IMAGE_STD) < 0.05, (
-        f"embed_image std drifted: {std} vs {_GOLDEN_EMBED_IMAGE_STD} — "
-        "likely a vision_tower weight-load regression under this transformers version."
-    )

tests/diffusion/models/sana_wm/test_cam_reduce_scatter.py

@@ -1,129 +0,0 @@
-# SPDX-License-Identifier: Apache-2.0
-# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
-
-"""Equivalence tests for SANA-WM's camera-branch reduce-scatter.
-
-``out_proj_cam`` is row-parallel, so every rank produces a partial sum over the
-full hidden width, while the GDN stream it is added to is TP-local. Instead of
-all-reducing and then discarding all but one slice, the branch reduce-scatters
-straight to the local shard. These tests pin the shard each rank ends up with,
-including the bias slice, on both code paths:
-
-* the process-group path, which drives ``torch.distributed.reduce_scatter_tensor``
-  directly because vLLM-Omni's diffusion ``GroupCoordinator`` does not implement
-  ``reduce_scatter``;
-* the all-reduce-and-slice fallback taken when the coordinator exposes no
-  ``device_group``.
-
-CPU-only: the collectives and the rank/world-size lookups are mocked, so a
-single process can simulate an arbitrary TP degree.
-"""
-
-from types import SimpleNamespace
-from unittest.mock import patch
-
-import pytest
-import torch
-
-from vllm_omni.diffusion.models.sana_wm.sana_wm_transformer import SanaWmSelfAttention
-
-pytestmark = [pytest.mark.core_model, pytest.mark.cpu]
-
-_MODULE = "vllm_omni.diffusion.models.sana_wm.sana_wm_transformer"
-
-HIDDEN = 24
-
-
-def _host(bias: torch.Tensor | None) -> SimpleNamespace:
-    """A stand-in ``self`` exposing only what the method reads."""
-    return SimpleNamespace(out_proj_cam=SimpleNamespace(bias=bias))
-
-
-def _fake_reduce_scatter_tensor(rank: int, tp_size: int):
-    """Model ``reduce_scatter_tensor``: split dim 0 and keep chunk ``rank``.
-
-    The test drives a single process, so the other ranks' partials are already
-    folded into the input; scattering is all this has to reproduce.
-    """
-
-    def _impl(output: torch.Tensor, input_: torch.Tensor, group=None) -> None:
-        output.copy_(input_.chunk(tp_size, dim=0)[rank])
-
-    return _impl
-
-
-def _tp_group(with_device_group: bool):
-    return SimpleNamespace(device_group=object() if with_device_group else None)
-
-
-@pytest.mark.parametrize("tp_size", [2, 4])
-@pytest.mark.parametrize("rank", [0, 1])
-def test_process_group_path_matches_all_reduce_then_slice(tp_size: int, rank: int) -> None:
-    """The scattered shard equals the slice the all-reduce path would take."""
-    local_width = HIDDEN // tp_size
-    torch.manual_seed(0)
-    contrib = torch.randn(2, 3, HIDDEN)
-    bias = torch.randn(HIDDEN)
-    reference = torch.zeros(2, 3, local_width)
-
-    with (
-        patch(f"{_MODULE}.get_tensor_model_parallel_world_size", return_value=tp_size),
-        patch(f"{_MODULE}.get_tensor_model_parallel_rank", return_value=rank),
-        patch(f"{_MODULE}.get_tp_group", return_value=_tp_group(True)),
-        patch("torch.distributed.reduce_scatter_tensor", side_effect=_fake_reduce_scatter_tensor(rank, tp_size)),
-    ):
-        out = SanaWmSelfAttention._reduce_scatter_cam_contrib(_host(bias), contrib, reference)
-
-    start = rank * local_width
-    expected = (contrib + bias)[..., start : start + local_width]
-    assert out.shape == (2, 3, local_width)
-    torch.testing.assert_close(out, expected)
-
-
-@pytest.mark.parametrize("rank", [0, 1])
-def test_fallback_path_matches_the_process_group_path(rank: int) -> None:
-    """Without a device_group the method all-reduces and slices, same result."""
-    tp_size = 2
-    local_width = HIDDEN // tp_size
-    torch.manual_seed(1)
-    contrib = torch.randn(2, 3, HIDDEN)
-    bias = torch.randn(HIDDEN)
-    reference = torch.zeros(2, 3, local_width)
-
-    with (
-        patch(f"{_MODULE}.get_tensor_model_parallel_world_size", return_value=tp_size),
-        patch(f"{_MODULE}.get_tensor_model_parallel_rank", return_value=rank),
-        patch(f"{_MODULE}.get_tp_group", return_value=_tp_group(False)),
-        patch(f"{_MODULE}.tensor_model_parallel_all_reduce", side_effect=lambda t: t),
-    ):
-        out = SanaWmSelfAttention._reduce_scatter_cam_contrib(_host(bias), contrib, reference)
-
-    start = rank * local_width
-    torch.testing.assert_close(out, (contrib + bias)[..., start : start + local_width])
-
-
-def test_tp1_adds_bias_without_a_collective() -> None:
-    contrib = torch.randn(2, 3, HIDDEN)
-    bias = torch.randn(HIDDEN)
-
-    with (
-        patch(f"{_MODULE}.get_tensor_model_parallel_world_size", return_value=1),
-        patch("torch.distributed.reduce_scatter_tensor") as collective,
-    ):
-        out = SanaWmSelfAttention._reduce_scatter_cam_contrib(_host(bias), contrib, contrib)
-
-    assert collective.call_count == 0
-    torch.testing.assert_close(out, contrib + bias)
-
-
-def test_rejects_a_width_that_does_not_shard() -> None:
-    """A local width that is not 1/tp of the contribution is a wiring bug."""
-    contrib = torch.randn(1, 2, HIDDEN)
-    reference = torch.zeros(1, 2, HIDDEN // 3)  # tp=2 would need HIDDEN // 2
-
-    with (
-        patch(f"{_MODULE}.get_tensor_model_parallel_world_size", return_value=2),
-        patch("torch.distributed.reduce_scatter_tensor"),
-        pytest.raises(ValueError, match="camera contribution width mismatch"),
-    ):
-        SanaWmSelfAttention._reduce_scatter_cam_contrib(_host(None), contrib, reference)

tests/diffusion/models/sana_wm/test_fused_qk_rms_norm.py

@@ -1,111 +0,0 @@
-# SPDX-License-Identifier: Apache-2.0
-# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
-
-"""Equivalence tests for SANA-WM's TensorParallelRMSNorm q/k fusion.
-
-`fused_qk_rms_norm` packs the q and k per-token sum-of-squares into one tensor
-and issues a single tensor-parallel all-reduce instead of two. These tests lock
-in that it is numerically identical to applying `norm_q`/`norm_k` separately,
-and that it really issues exactly one (fused) all-reduce.
-
-CPU-only: the collective is mocked, so a single process simulates one rank
-(all-reduce is identity). This exercises the cat/split logic a real multi-rank
-run depends on.
-"""
-
-from unittest.mock import patch
-
-import pytest
-import torch
-from torch import nn
-
-from vllm_omni.diffusion.layers.norm import RMSNorm
-from vllm_omni.diffusion.models.sana_wm.sana_wm_transformer import (
-    TensorParallelRMSNorm,
-    fused_qk_rms_norm,
-)
-
-pytestmark = [pytest.mark.core_model, pytest.mark.cpu]
-
-_MODULE = "vllm_omni.diffusion.models.sana_wm.sana_wm_transformer"
-
-
-def _make_norms(dim: int, tp_size: int) -> tuple[TensorParallelRMSNorm, TensorParallelRMSNorm]:
-    """Two norms with distinct random weights so a swapped q/k slice is caught."""
-    torch.manual_seed(0)
-    norm_q = TensorParallelRMSNorm(dim, tp_size=tp_size)
-    norm_k = TensorParallelRMSNorm(dim, tp_size=tp_size)
-    norm_q.weight.data.normal_()
-    norm_k.weight.data.normal_()
-    return norm_q, norm_k
-
-
-def _identity_all_reduce(tensor: torch.Tensor) -> torch.Tensor:
-    """Single-rank all-reduce: the reduction over one rank is the identity."""
-    return tensor
-
-
-def test_fused_matches_separate() -> None:
-    dim = 64
-    norm_q, norm_k = _make_norms(dim, tp_size=2)
-    q = torch.randn(2, 5, dim)
-    k = torch.randn(2, 5, dim)
-
-    with patch(f"{_MODULE}.tensor_model_parallel_all_reduce", side_effect=_identity_all_reduce):
-        ref_q, ref_k = norm_q(q), norm_k(k)
-        fused_q, fused_k = fused_qk_rms_norm(norm_q, norm_k, q, k)
-
-    torch.testing.assert_close(fused_q, ref_q, rtol=0, atol=0)
-    torch.testing.assert_close(fused_k, ref_k, rtol=0, atol=0)
-
-
-def test_fused_issues_one_all_reduce() -> None:
-    dim = 32
-    norm_q, norm_k = _make_norms(dim, tp_size=4)
-    q = torch.randn(1, 3, dim)
-    k = torch.randn(1, 3, dim)
-
-    with patch(f"{_MODULE}.tensor_model_parallel_all_reduce", side_effect=_identity_all_reduce) as collective:
-        fused_qk_rms_norm(norm_q, norm_k, q, k)
-        assert collective.call_count == 1
-        # Both sums travel in one tensor: last dim is 2, not 1.
-        assert collective.call_args.args[0].shape[-1] == 2
-
-        collective.reset_mock()
-        norm_q(q), norm_k(k)
-        assert collective.call_count == 2
-
-
-@pytest.mark.parametrize(
-    "norms",
-    [
-        (RMSNorm(16), RMSNorm(16)),
-        (nn.Identity(), nn.Identity()),
-    ],
-    ids=["plain_rms_norm", "identity"],
-)
-def test_falls_back_for_non_tp_norms(norms: tuple[nn.Module, nn.Module]) -> None:
-    """qk_norm=False and TP=1 both hand in norms the fused path cannot use."""
-    norm_q, norm_k = norms
-    q = torch.randn(2, 4, 16)
-    k = torch.randn(2, 4, 16)
-
-    with patch(f"{_MODULE}.tensor_model_parallel_all_reduce", side_effect=_identity_all_reduce) as collective:
-        fused_q, fused_k = fused_qk_rms_norm(norm_q, norm_k, q, k)
-        assert collective.call_count == 0
-
-    torch.testing.assert_close(fused_q, norm_q(q), rtol=0, atol=0)
-    torch.testing.assert_close(fused_k, norm_k(k), rtol=0, atol=0)
-
-
-def test_tp1_skips_the_collective() -> None:
-    norm_q, norm_k = _make_norms(16, tp_size=1)
-    q = torch.randn(2, 4, 16)
-    k = torch.randn(2, 4, 16)
-
-    with patch(f"{_MODULE}.tensor_model_parallel_all_reduce", side_effect=_identity_all_reduce) as collective:
-        fused_q, fused_k = fused_qk_rms_norm(norm_q, norm_k, q, k)
-        assert collective.call_count == 0
-
-    torch.testing.assert_close(fused_q, norm_q(q), rtol=0, atol=0)
-    torch.testing.assert_close(fused_k, norm_k(k), rtol=0, atol=0)

tests/diffusion/models/sana_wm/test_request.py

@@ -1,142 +0,0 @@
-# SPDX-License-Identifier: Apache-2.0
-# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
-"""Unit tests for SANA-WM request normalization."""
-
-import pytest
-
-from vllm_omni.diffusion.models.sana_wm.request import (
-    SANA_WM_DEFAULT_HEIGHT,
-    SANA_WM_DEFAULT_NUM_FRAMES,
-    SANA_WM_DEFAULT_WIDTH,
-    normalize_sana_wm_payload,
-)
-
-pytestmark = [pytest.mark.core_model, pytest.mark.cpu]
-
-
-def _prompt(**sana_wm):
-    """A valid request, with the action rollout kept in step with num_frames."""
-    block = {"num_frames": SANA_WM_DEFAULT_NUM_FRAMES, **sana_wm}
-    block.setdefault("action", f"w-{block['num_frames'] - 1}")
-    return {
-        "prompt": "a quiet city street",
-        "multi_modal_data": {"image": object()},
-        "sana_wm": block,
-    }
-
-
-def _identity_poses(count):
-    return [[[1, 0, 0, 0], [0, 1, 0, 0], [0, 0, 1, 0], [0, 0, 0, 1]] for _ in range(count)]
-
-
-def _camera_prompt(**camera):
-    prompt = _prompt(num_frames=9)
-    prompt["sana_wm"].pop("action")
-    prompt["sana_wm"]["camera"] = {"poses": _identity_poses(9), **camera}
-    return prompt
-
-
-def test_defaults_are_latent_aligned():
-    payload = normalize_sana_wm_payload(_prompt())["additional_information"]["sana_wm"]
-    assert payload["num_frames"] == SANA_WM_DEFAULT_NUM_FRAMES
-    assert payload["height"] == SANA_WM_DEFAULT_HEIGHT
-    assert payload["width"] == SANA_WM_DEFAULT_WIDTH
-
-
-@pytest.mark.parametrize("num_frames", [9, 33, 161])
-def test_accepts_aligned_geometry(num_frames):
-    payload = normalize_sana_wm_payload(_prompt(num_frames=num_frames))["additional_information"]["sana_wm"]
-    assert payload["num_frames"] == num_frames
-
-
-@pytest.mark.parametrize("num_frames", [8, 10, 160])
-def test_rejects_unaligned_num_frames(num_frames):
-    # The VAE compresses 8x temporally; an unaligned count would be floored.
-    with pytest.raises(ValueError, match="num_frames"):
-        normalize_sana_wm_payload(_prompt(num_frames=num_frames))
-
-
-@pytest.mark.parametrize(("height", "width"), [(700, 1280), (704, 1270), (703, 1279)])
-def test_rejects_unaligned_resolution(height, width):
-    with pytest.raises(ValueError, match="divisible by 32"):
-        normalize_sana_wm_payload(_prompt(height=height, width=width))
-
-
-def test_normalization_is_idempotent():
-    once = normalize_sana_wm_payload(_prompt(num_frames=33))
-    twice = normalize_sana_wm_payload(once)
-    assert once["additional_information"]["sana_wm"] == twice["additional_information"]["sana_wm"]
-
-
-def test_rejects_action_shorter_than_num_frames():
-    # Used to shrink the camera tensor silently, then fail deep in the
-    # transformer on a camera/latent frame mismatch.
-    with pytest.raises(ValueError, match="rolls out 33 frames but num_frames is 161"):
-        normalize_sana_wm_payload(_prompt(action="w-32"))
-
-
-def test_rejects_action_longer_than_num_frames():
-    with pytest.raises(ValueError, match="rolls out 161 frames but num_frames is 9"):
-        normalize_sana_wm_payload(_prompt(num_frames=9, action="w-160"))
-
-
-def test_rejects_oversized_action_without_expanding_it():
-    # A 12-byte request body: expanding this rollout before checking the frame
-    # count allocated ~16 GB. Rejection has to be arithmetic, so this test also
-    # fails (by hanging/OOM) if the duration is ever expanded again.
-    with pytest.raises(ValueError, match="rolls out 1000000001 frames"):
-        normalize_sana_wm_payload(_prompt(action="w-1000000000"))
-
-
-@pytest.mark.parametrize("action", ["", "   ", "w-0", "w-abc", "w", "x-4"])
-def test_rejects_malformed_action(action):
-    with pytest.raises(ValueError):
-        normalize_sana_wm_payload(_prompt(action=action))
-
-
-@pytest.mark.parametrize("value", [0, -1, float("nan"), float("inf")])
-def test_rejects_unusable_focal_length(value):
-    # compute_raymap divides by fx/fy; a zero or non-finite focal length turns
-    # the whole Plucker map into NaN and the response into a NaN video.
-    intrinsics = {"fx": value, "fy": 640, "cx": 640, "cy": 352}
-    with pytest.raises(ValueError, match="fx"):
-        normalize_sana_wm_payload(_prompt(intrinsics=intrinsics))
-
-
-@pytest.mark.parametrize("value", [float("nan"), float("inf")])
-def test_rejects_non_finite_principal_point(value):
-    intrinsics = {"fx": 640, "fy": 640, "cx": value, "cy": 352}
-    with pytest.raises(ValueError, match="must be finite"):
-        normalize_sana_wm_payload(_prompt(intrinsics=intrinsics))
-
-
-@pytest.mark.parametrize("value", [float("nan"), float("inf")])
-def test_rejects_non_finite_speeds(value):
-    # `parsed <= 0` is False for NaN, so the positivity check alone let it pass.
-    with pytest.raises(ValueError, match="must be finite"):
-        normalize_sana_wm_payload(_prompt(translation_speed=value))
-
-
-def test_rejects_non_finite_pose_entry():
-    prompt = _camera_prompt()
-    prompt["sana_wm"]["camera"]["poses"][3][2][1] = float("nan")
-    with pytest.raises(ValueError, match="must be finite"):
-        normalize_sana_wm_payload(prompt)
-
-
-def test_accepts_default_camera_metadata():
-    payload = normalize_sana_wm_payload(_camera_prompt())["additional_information"]["sana_wm"]
-    assert payload["camera"]["poses"]
-    # The unread fields are not carried into the canonical payload.
-    assert set(payload["camera"]) == {"poses"}
-
-
-@pytest.mark.parametrize(
-    ("field", "value"),
-    [("format", "w2c_4x4"), ("coordinate_system", "opengl")],
-)
-def test_rejects_unsupported_camera_convention(field, value):
-    # Nothing downstream reads these, so accepting one would mean interpreting
-    # the poses under the wrong convention.
-    with pytest.raises(ValueError, match=field.split("_")[0]):
-        normalize_sana_wm_payload(_camera_prompt(**{field: value}))

tests/diffusion/models/sensenova_u1/test_sensenova_u1_logits_processor.py

@@ -1,72 +0,0 @@
-# SPDX-License-Identifier: Apache-2.0
-# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
-"""Regression test: SenseNovaU1ForCausalLM instantiation under the diffusion
-config shim must not crash on missing ``head_dtype`` and the resulting
-LogitsProcessor must have ``head_dtype is None`` (= use model dtype)."""
-
-import os
-
-import pytest
-import torch
-from vllm.config import DeviceConfig, VllmConfig, set_current_vllm_config
-from vllm.distributed.parallel_state import (
-    cleanup_dist_env_and_memory,
-    init_distributed_environment,
-    initialize_model_parallel,
-)
-
-from vllm_omni.diffusion.models.sensenova_u1.sensenova_u1_transformer import (
-    SenseNovaU1ForCausalLM,
-)
-from vllm_omni.diffusion.vllm_config import _DiffusionVllmModelConfig
-from vllm_omni.transformers_utils.configs.sensenova_u1 import (
-    SenseNovaU1Config,
-)
-
-pytestmark = [pytest.mark.core_model, pytest.mark.diffusion, pytest.mark.cpu]
-
-
-def _build_tiny_sensenova_u1(quant_config=None, prefix="model"):
-    """Build a minimal SenseNovaU1ForCausalLM for unit tests."""
-    config = SenseNovaU1Config(
-        llm_config={
-            "hidden_size": 64,
-            "num_attention_heads": 2,
-            "num_hidden_layers": 1,
-            "intermediate_size": 128,
-            "vocab_size": 32,
-            "max_position_embeddings": 128,
-            "max_position_embeddings_hw": 128,
-        },
-    )
-    return SenseNovaU1ForCausalLM(config.llm_config, quant_config=quant_config, prefix=prefix)
-
-
-def test_logits_processor_head_dtype_under_diffusion_shim():
-    # Initialize the distributed environment
-    os.environ.setdefault("MASTER_ADDR", "localhost")
-    os.environ.setdefault("MASTER_PORT", "29543")
-    init_distributed_environment(world_size=1, rank=0, local_rank=0, distributed_init_method="env://")
-    initialize_model_parallel()
-
-    # Create a vLLM config with a wrapped diffusion config
-    fake_diff_config = _DiffusionVllmModelConfig(
-        model="sensenova-u1-test",
-        dtype=torch.bfloat16,
-        max_model_len=8192,
-        original_max_model_len=8192,
-    )
-    vllm_config = VllmConfig(
-        device_config=DeviceConfig(device="cpu"),
-    )
-    vllm_config.modelconfig = fake_diff_config  # type: ignore[assignment]
-
-    # Build a tiny version of the Causal LM component
-    with set_current_vllm_config(vllm_config):
-        model = _build_tiny_sensenova_u1()
-
-    # Ensure that we have a set head_dtype attribute. Currently, the head_dtype
-    # is set to None
-    head_dtype = model.logits_processor.head_dtype
-    assert head_dtype is None
-    cleanup_dist_env_and_memory()

tests/diffusion/offloader/test_distributed_layerwise_backend.py

@@ -185,79 +185,6 @@ class TestDistributedLayerwiseOffloadHook:
         assert next_block.weight.stride() == expected_stride
         assert torch.equal(next_block.weight, expected)
 
-    def test_allgather_reconstructs_online_fp8_weight_and_scale(
-        self,
-        patched_offload_runtime,
-        monkeypatch,
-    ):
-        """DLO must reconstruct finalized online-FP8 tensors byte-for-byte."""
-
-        class OnlineFp8Block(nn.Module):
-            def __init__(self):
-                super().__init__()
-                logical_weight = torch.tensor(
-                    [
-                        [1.0, 2.0, 3.0, 4.0],
-                        [5.0, 6.0, 7.0, 8.0],
-                        [9.0, 10.0, 11.0, 12.0],
-                    ],
-                    dtype=torch.float32,
-                ).to(torch.float8_e4m3fn)
-                self.weight = nn.Parameter(logical_weight.t(), requires_grad=False)
-                self.weight_scale = nn.Parameter(torch.tensor([0.25]), requires_grad=False)
-
-        current_block = OnlineFp8Block()
-        rank0_block = OnlineFp8Block()
-        rank1_block = OnlineFp8Block()
-        expected_weight = rank0_block.weight.detach().clone()
-        expected_scale = rank0_block.weight_scale.detach().clone()
-        expected_stride = rank0_block.weight.stride()
-
-        rank0_hook = DistributedLayerwiseOffloadHook(
-            next_block=rank0_block,
-            device=torch.device("cpu"),
-            dp_group=object(),
-            dp_size=2,
-            rank=0,
-            copy_stream=DummyStream(),
-            comm_stream=DummyStream(),
-            pin_memory=False,
-        )
-        rank1_hook = DistributedLayerwiseOffloadHook(
-            next_block=rank1_block,
-            device=torch.device("cpu"),
-            dp_group=object(),
-            dp_size=2,
-            rank=1,
-            copy_stream=DummyStream(),
-            comm_stream=DummyStream(),
-            pin_memory=False,
-        )
-        rank0_hook.initialize_hook(current_block)
-        rank1_hook.initialize_hook(OnlineFp8Block())
-        rank0_hook.gpu_shard_buffers = [
-            {dtype: torch.empty_like(shard) for dtype, shard in rank0_hook.cpu_shards.items()} for _ in range(2)
-        ]
-
-        def fake_allgather(output, local_shard, *, group):
-            del group
-            remote_shard = rank1_hook.cpu_shards[local_shard.dtype]
-            shard_size = local_shard.numel()
-            output[:shard_size].copy_(local_shard)
-            output[shard_size : 2 * shard_size].copy_(remote_shard)
-
-        monkeypatch.setattr(torch.distributed, "all_gather_into_tensor", fake_allgather)
-        rank0_hook.prefetch_layer(slot=0, non_blocking=False)
-
-        assert rank0_block.weight.stride() == expected_stride
-        torch.testing.assert_close(
-            rank0_block.weight.float(),
-            expected_weight.float(),
-            rtol=0,
-            atol=0,
-        )
-        torch.testing.assert_close(rank0_block.weight_scale, expected_scale)
-
     def test_dtensor_wrapper_preserved_across_prefetch_and_offload(self, dist_group, patched_offload_runtime):
         """DTensor wrapper should be preserved through prefetch/offload cycle."""
         current_block = TinyBlock(_make_values(1.0))

tests/diffusion/test_diffusion_config_propagation.py

@@ -92,30 +92,6 @@ class TestParallelConfigPropagation:
         od = _roundtrip_diffusion_config(model="x", parallel_config=pc)
         assert od.num_gpus == 2
 
-    def test_dp_is_inferred_from_num_gpus(self):
-        pc = DiffusionParallelConfig(tensor_parallel_size=2, ulysses_degree=2)
-        od = OmniDiffusionConfig.from_kwargs(model="x", parallel_config=pc, num_gpus=8)
-        assert od.parallel_config.data_parallel_size == 2
-        assert od.parallel_config.world_size == 8
-
-    def test_explicit_dp_is_validated_against_num_gpus(self):
-        pc = DiffusionParallelConfig(tensor_parallel_size=2, data_parallel_size=2)
-        od = OmniDiffusionConfig.from_kwargs(model="x", parallel_config=pc, num_gpus=4)
-        assert od.parallel_config.data_parallel_size == 2
-
-        with pytest.raises(ValueError, match="does not match WORLD-derived value"):
-            OmniDiffusionConfig.from_kwargs(
-                model="x",
-                parallel_config=DiffusionParallelConfig(tensor_parallel_size=2, data_parallel_size=2),
-                num_gpus=8,
-            )
-
-    def test_hsdp_does_not_infer_ordinary_dp(self):
-        pc = DiffusionParallelConfig(use_hsdp=True, hsdp_shard_size=4)
-        od = OmniDiffusionConfig.from_kwargs(model="x", parallel_config=pc, num_gpus=4)
-        assert od.parallel_config.data_parallel_size == 1
-        assert od.parallel_config.world_size == 4
-
 
 class TestCreateDefaultDiffusion:
     """Verify engine_args structure from create_default_diffusion."""
@@ -154,21 +130,6 @@ class TestCreateDefaultDiffusion:
 
         assert od.diffusion_kv_mode is DiffusionKVCacheMode.PAGED_SCHEDULER
 
-    def test_diffusion_kv_sizing_fields_roundtrip(self, monkeypatch):
-        monkeypatch.setattr(OmniDiffusionConfig, "_resolve_master_port", lambda self: 29500)
-        od = _roundtrip_diffusion_config(
-            model="x",
-            kv_cache_memory_bytes=4096,
-            gpu_memory_utilization=0.75,
-            max_num_batched_tokens=2048,
-            max_model_len=4096,
-        )
-
-        assert od.kv_cache_memory_bytes == 4096
-        assert od.gpu_memory_utilization == 0.75
-        assert od.max_num_batched_tokens == 2048
-        assert od.max_model_len == 4096
-
 
 def test_qwen_image_edit_plus_sets_generic_multimodal_limit():
     od_config = OmniDiffusionConfig(model="Qwen/Qwen-Image-Edit-2511", model_class_name="QwenImageEditPlusPipeline")

tests/diffusion/test_diffusion_engine_cleanup.py

@@ -161,83 +161,9 @@ def test_scheduler_initialization_failure_closes_scheduler_and_executor(monkeypa
     assert exc_info.value is initialization_error
     custom_scheduler.initialize.assert_called_once_with(od_config)
     custom_scheduler.close.assert_called_once_with()
-
-
-def test_init_shuts_down_executor_when_kv_control_plane_initialization_fails(
-    monkeypatch: pytest.MonkeyPatch,
-) -> None:
-    fake_executor = SimpleNamespace(shutdown=Mock())
-    monkeypatch.setattr(DiffusionEngine, "_init_process_hooks", lambda self, config: None)
-    monkeypatch.setattr(
-        DiffusionEngine,
-        "_resolve_execution_mode",
-        lambda self, config: DiffusionExecutionMode.REQUEST_BATCH,
-    )
-    monkeypatch.setattr(
-        DiffusionEngine, "_init_executor", lambda self, config: setattr(self, "executor", fake_executor)
-    )
-    monkeypatch.setattr(
-        "vllm_omni.diffusion.diffusion_engine.initialize_diffusion_kv_control_plane",
-        Mock(side_effect=RuntimeError("KV initialization failed")),
-    )
-
-    with pytest.raises(RuntimeError, match="KV initialization failed"):
-        DiffusionEngine(SimpleNamespace())
-
     fake_executor.shutdown.assert_called_once_with()
 
 
-def test_paged_init_profiles_before_scheduler_initialization(monkeypatch: pytest.MonkeyPatch) -> None:
-    events = []
-    expected_profile_requests = [object()]
-    fake_executor = SimpleNamespace(shutdown=Mock())
-    kv_cache_config = object()
-    kv_vllm_config = object()
-    od_config = SimpleNamespace(diffusion_kv_mode="paged_scheduler")
-
-    monkeypatch.setattr(DiffusionEngine, "_init_process_hooks", lambda self, config: None)
-    monkeypatch.setattr(
-        DiffusionEngine,
-        "_resolve_execution_mode",
-        lambda self, config: DiffusionExecutionMode.REQUEST_BATCH,
-    )
-    monkeypatch.setattr(
-        DiffusionEngine,
-        "_init_executor",
-        lambda self, config: setattr(self, "executor", fake_executor),
-    )
-    monkeypatch.setattr(
-        DiffusionEngine,
-        "_prepare_diffusion_kv_profile_requests",
-        lambda self: events.append("prepare-profile") or expected_profile_requests,
-    )
-
-    def initialize(executor, config, *, profile_requests):
-        events.append("profile-workers")
-        assert executor is fake_executor
-        assert config is od_config
-        assert profile_requests is expected_profile_requests
-        return kv_cache_config, 16, 16, kv_vllm_config
-
-    monkeypatch.setattr(
-        "vllm_omni.diffusion.diffusion_engine.initialize_diffusion_kv_control_plane",
-        initialize,
-    )
-
-    def init_scheduler(self, config, scheduler, *args, **kwargs):
-        events.append("initialize-scheduler")
-        self.scheduler = SimpleNamespace(close=Mock())
-
-    monkeypatch.setattr(DiffusionEngine, "_init_scheduler", init_scheduler)
-    monkeypatch.setattr(DiffusionEngine, "_init_runtime_state", lambda self: None)
-    monkeypatch.setattr(DiffusionEngine, "_init_execute_fn", lambda self: None)
-    monkeypatch.setattr(DiffusionEngine, "_log_execution_mode", lambda self, config: None)
-
-    DiffusionEngine(od_config)
-
-    assert events == ["prepare-profile", "profile-workers", "initialize-scheduler"]
-
-
 @pytest.mark.asyncio
 async def test_step_compatibility_wrapper_returns_final_batch() -> None:
     engine = _make_engine()

tests/diffusion/test_diffusion_engine_dummy_run.py

@@ -1,17 +1,9 @@
 # SPDX-License-Identifier: Apache-2.0
 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project
 
-from types import SimpleNamespace
-from unittest.mock import Mock
-
 import pytest
 
 from vllm_omni.diffusion import io_support
-from vllm_omni.diffusion.diffusion_engine import DiffusionEngine, DiffusionExecutionMode
-from vllm_omni.diffusion.diffusion_kv.config import DiffusionKVCacheMode
-from vllm_omni.diffusion.diffusion_kv.request import DiffusionKVRequest
-from vllm_omni.diffusion.request import OmniDiffusionRequest
-from vllm_omni.inputs.data import OmniDiffusionSamplingParams
 
 pytestmark = [pytest.mark.core_model, pytest.mark.cpu, pytest.mark.diffusion]
 
@@ -63,151 +55,3 @@ def test_dummy_run_num_frames_uses_audio_input_fallback(monkeypatch: pytest.Monk
     )
 
     assert io_support.get_dummy_run_num_frames("unknown", supports_audio_input=True) == 2
-
-
-def test_dummy_run_image_count_resolves_hunyuan_architecture_alias() -> None:
-    assert io_support.get_dummy_run_num_image_inputs("HunyuanImage3ForCausalMM") == 3
-    assert io_support.get_dummy_run_num_image_inputs("unknown") == 1
-
-
-def test_dense_mode_does_not_build_kv_profile_request() -> None:
-    engine = object.__new__(DiffusionEngine)
-    engine.od_config = SimpleNamespace(diffusion_kv_mode=DiffusionKVCacheMode.DENSE_LEGACY)
-    engine._make_dummy_request = Mock(side_effect=AssertionError("dense mode must not profile"))
-
-    assert engine._prepare_diffusion_kv_profile_requests() is None
-    engine._make_dummy_request.assert_not_called()
-
-
-@pytest.mark.parametrize(
-    ("execution_mode", "uses_dlo_dp", "max_num_seqs", "expected_profile_requests"),
-    [
-        (DiffusionExecutionMode.STEP_BATCH, False, 2, 2),
-        (DiffusionExecutionMode.REQUEST_BATCH, True, 4, 1),
-        (DiffusionExecutionMode.STEP_BATCH, True, 3, 3),
-    ],
-)
-def test_paged_kv_profile_requests_match_per_rank_batch(
-    execution_mode: DiffusionExecutionMode,
-    uses_dlo_dp: bool,
-    max_num_seqs: int,
-    expected_profile_requests: int,
-) -> None:
-    engine = object.__new__(DiffusionEngine)
-    engine.execution_mode = execution_mode
-    engine.od_config = SimpleNamespace(
-        diffusion_kv_mode=DiffusionKVCacheMode.PAGED_SCHEDULER,
-        model_class_name="HunyuanImage3ForCausalMM",
-        max_num_seqs=max_num_seqs,
-        parallel_config=SimpleNamespace(data_parallel_size=max_num_seqs if uses_dlo_dp else 1),
-        enable_distributed_layerwise_offload=uses_dlo_dp,
-        dlo_use_allgather=True,
-    )
-    request = OmniDiffusionRequest(
-        prompt="profile",
-        request_id="profile-request",
-        sampling_params=OmniDiffusionSamplingParams(num_inference_steps=1),
-    )
-    engine._make_dummy_request = Mock(return_value=request)
-    prepared_layout = object()
-    scheduler_kv_state = (
-        DiffusionKVRequest(
-            "profile-request/diffusion-kv/0",
-            sequence_id=0,
-            prefix_len=3,
-            target_len=8,
-            seq_len=12,
-        ),
-        DiffusionKVRequest(
-            "profile-request/diffusion-kv/1",
-            sequence_id=1,
-            prefix_len=4,
-            target_len=8,
-            seq_len=13,
-        ),
-    )
-
-    def preprocess(req: OmniDiffusionRequest) -> OmniDiffusionRequest:
-        req.prepared_layout = prepared_layout
-        req.diffusion_kv_requests = scheduler_kv_state
-        return req
-
-    engine._prepare_request_for_admission = Mock(side_effect=preprocess)
-
-    result = engine._prepare_diffusion_kv_profile_requests()
-
-    assert result is not None
-    assert len(result) == expected_profile_requests
-    assert len({profile_request.request_id for profile_request in result}) == expected_profile_requests
-    assert all(profile_request.is_dummy_run() for profile_request in result)
-    assert all(profile_request.prepared_layout is prepared_layout for profile_request in result)
-    assert all(profile_request.diffusion_kv_requests is None for profile_request in result)
-    assert len({id(profile_request.sampling_params) for profile_request in result}) == expected_profile_requests
-    assert engine._diffusion_kv_profile_limits == (2, 13, 8)
-    assert len(result) * engine._diffusion_kv_profile_limits[0] == expected_profile_requests * 2
-    engine._make_dummy_request.assert_called_once_with(
-        height=1024,
-        width=1024,
-        guidance_scale=5.0,
-        num_image_inputs=3,
-    )
-    engine._prepare_request_for_admission.assert_called_once_with(request)
-
-
-@pytest.mark.parametrize(
-    ("num_sequences", "seq_len", "target_len"),
-    [
-        (3, 16, 8),
-        (1, 17, 8),
-        (1, 16, 9),
-    ],
-)
-def test_paged_kv_admission_rejects_shape_beyond_profile_envelope(
-    num_sequences: int,
-    seq_len: int,
-    target_len: int,
-) -> None:
-    engine = object.__new__(DiffusionEngine)
-    engine._diffusion_kv_profile_limits = (2, 16, 8)
-    engine.pre_process_func = lambda request: request
-    request = OmniDiffusionRequest(
-        prompt="too large",
-        request_id="too-large",
-        sampling_params=OmniDiffusionSamplingParams(num_inference_steps=1),
-        diffusion_kv_requests=tuple(
-            DiffusionKVRequest(
-                f"too-large/diffusion-kv/{sequence_id}",
-                sequence_id=sequence_id,
-                prefix_len=0,
-                target_len=target_len,
-                seq_len=seq_len,
-            )
-            for sequence_id in range(num_sequences)
-        ),
-    )
-
-    with pytest.raises(ValueError, match="exceeds the startup memory-profile envelope"):
-        engine._prepare_request_for_admission(request)
-
-
-def test_paged_kv_admission_accepts_shape_at_profile_envelope() -> None:
-    engine = object.__new__(DiffusionEngine)
-    engine._diffusion_kv_profile_limits = (2, 16, 8)
-    engine.pre_process_func = lambda request: request
-    request = OmniDiffusionRequest(
-        prompt="fits",
-        request_id="fits",
-        sampling_params=OmniDiffusionSamplingParams(num_inference_steps=1),
-        diffusion_kv_requests=tuple(
-            DiffusionKVRequest(
-                f"fits/diffusion-kv/{sequence_id}",
-                sequence_id=sequence_id,
-                prefix_len=8,
-                target_len=8,
-                seq_len=16,
-            )
-            for sequence_id in range(2)
-        ),
-    )
-
-    assert engine._prepare_request_for_admission(request) is request

tests/diffusion/test_diffusion_engine_rpc_routing.py

@@ -473,38 +473,6 @@ async def test_cancelled_future_does_not_kill_busy_loop():
         _stop_engine(engine)
 
 
-@pytest.mark.asyncio
-async def test_native_kv_reservation_error_wakes_stream_without_killing_busy_loop():
-    loop = asyncio.get_running_loop()
-    engine = _make_engine_with_loop(loop)
-
-    def fail_reservation(*args, **kwargs):
-        raise RuntimeError("native allocation bug")
-
-    engine.scheduler._diffusion_kv_manager = SimpleNamespace(
-        has_request=lambda request_id: False,
-        reserve_request=fail_reservation,
-        free_request=lambda request_id: None,
-    )
-    request = _make_request("kv-error")
-    request.diffusion_kv_requests = (object(),)
-    try:
-        response_stream = engine.async_add_req_and_stream_response(request)
-        output_queue = engine._out_streams[request.request_id]
-        for _ in range(300):
-            if not output_queue.empty():
-                break
-            await asyncio.sleep(0.01)
-        assert not output_queue.empty(), "terminal KV allocation error was not delivered to the request stream"
-
-        output = await anext(response_stream)
-        assert output.error == "native allocation bug"
-        assert output.finished
-        assert engine.worker_thread.is_alive()
-    finally:
-        _stop_engine(engine)
-
-
 # ─────────────────────────── _RpcTask basics ───────────────────────────────
 
 

tests/diffusion/test_diffusion_model_runner.py

@@ -3,7 +3,6 @@
 
 from contextlib import contextmanager
 from types import SimpleNamespace
-from unittest.mock import Mock
 
 import pytest
 import torch
@@ -513,84 +512,6 @@ def test_execute_model_accepts_bare_diffusion_output_from_single_request_pipelin
     assert runner.pipeline.last_req.num_reqs == 1
 
 
-@pytest.mark.core_model
-@pytest.mark.cpu
-def test_profile_run_executes_forward_without_scheduler_kv_validation(monkeypatch):
-    runner = _make_runner(cache_backend=None, cache_backend_name="none")
-    request = _make_request()
-    runner._validate_diffusion_kv_metadata = Mock(side_effect=AssertionError("profile must bypass admission"))
-    record_names = []
-    original_execute = runner._execute_request_list
-
-    def execute(*args, **kwargs):
-        record_names.append(kwargs["record_name"])
-        return original_execute(*args, **kwargs)
-
-    runner._execute_request_list = execute
-    monkeypatch.setattr(model_runner_module, "set_forward_context", _noop_forward_context)
-    reset_peak_memory_stats = Mock()
-    monkeypatch.setattr(
-        model_runner_module.current_omni_platform,
-        "reset_peak_memory_stats",
-        reset_peak_memory_stats,
-    )
-    monkeypatch.setattr(model_runner_module.current_omni_platform, "max_memory_reserved", lambda: 0)
-    monkeypatch.setattr(model_runner_module.current_omni_platform, "max_memory_allocated", lambda: 0)
-    synchronize = Mock()
-    monkeypatch.setattr(model_runner_module.current_omni_platform, "synchronize", synchronize)
-
-    runner.profile_run([request])
-
-    assert runner.pipeline.forward_calls == 1
-    assert record_names == ["pipeline_memory_profile"]
-    runner._validate_diffusion_kv_metadata.assert_not_called()
-    reset_peak_memory_stats.assert_not_called()
-    synchronize.assert_called_once_with()
-
-
-@pytest.mark.core_model
-@pytest.mark.cpu
-def test_profile_run_executes_maximum_step_batch_without_resetting_peak(monkeypatch):
-    runner = _make_runner(cache_backend=None, cache_backend_name=None)
-    runner.pipeline = _FinalOnlyStepPipeline()
-    runner.od_config.step_execution = True
-    requests = [_make_request(), _make_request()]
-    requests[0].request_id = "profile-0"
-    requests[1].request_id = "profile-1"
-    for request in requests:
-        request.sampling_params.num_inference_steps = 1
-
-    observed_batch_rows = []
-    original_denoise_step = runner.pipeline.denoise_step
-
-    def denoise_step(input_batch, states):
-        observed_batch_rows.append((len(states), int(input_batch.latents.shape[0])))
-        return original_denoise_step(input_batch, states)
-
-    runner.pipeline.denoise_step = denoise_step
-    runner._validate_diffusion_kv_metadata = Mock(side_effect=AssertionError("profile must bypass admission"))
-    monkeypatch.setattr(model_runner_module, "set_forward_context", _noop_forward_context)
-    reset_peak_memory_stats = Mock()
-    monkeypatch.setattr(
-        model_runner_module.current_omni_platform,
-        "reset_peak_memory_stats",
-        reset_peak_memory_stats,
-    )
-    monkeypatch.setattr(model_runner_module.current_omni_platform, "max_memory_reserved", lambda: 0)
-    monkeypatch.setattr(model_runner_module.current_omni_platform, "max_memory_allocated", lambda: 0)
-    synchronize = Mock()
-    monkeypatch.setattr(model_runner_module.current_omni_platform, "synchronize", synchronize)
-
-    runner.profile_run(requests)
-
-    assert observed_batch_rows == [(2, 2)]
-    assert runner.state_cache == {}
-    assert runner.input_batch is None
-    runner._validate_diffusion_kv_metadata.assert_not_called()
-    reset_peak_memory_stats.assert_not_called()
-    synchronize.assert_called_once_with()
-
-
 class _BatchPipeline:
     """Pipeline returning a configurable list of outputs from forward()."""
 

tests/diffusion/test_diffusion_request.py

@@ -52,7 +52,6 @@ def test_dummy_run_request_is_identified_by_reserved_request_id():
 
     assert req.is_dummy_run()
     assert OmniDiffusionRequest.is_dummy_run_request_id(DUMMY_DIFFUSION_REQUEST_ID)
-    assert OmniDiffusionRequest.is_dummy_run_request_id(f"{DUMMY_DIFFUSION_REQUEST_ID}/kv-profile-1")
 
 
 def test_non_dummy_request_is_not_identified_as_dummy_run():

tests/diffusion/test_diffusion_scheduler.py

@@ -8,14 +8,12 @@ from types import SimpleNamespace
 
 import pytest
 import torch
-import vllm.v1.core.single_type_kv_cache_manager as native_kv_managers
 from pytest_mock import MockerFixture
-from vllm.v1.kv_cache_interface import FullAttentionSpec, KVCacheConfig, KVCacheGroupSpec, KVCacheTensor
 
 from vllm_omni.diffusion.data import DiffusionOutput, DiffusionRequestAbortedError
 from vllm_omni.diffusion.diffusion_engine import DiffusionEngine, DiffusionExecutionMode
 from vllm_omni.diffusion.diffusion_kv.config import DiffusionKVCacheMode
-from vllm_omni.diffusion.diffusion_kv.request import DiffusionKVRequest
+from vllm_omni.diffusion.diffusion_kv.request import DiffusionKVContext, DiffusionKVRequest
 from vllm_omni.diffusion.request import OmniDiffusionRequest
 from vllm_omni.diffusion.sched import (
     BaseScheduler,
@@ -89,52 +87,6 @@ def _cached_ids(sched_output) -> list[str]:
     return list(sched_output.scheduled_cached_reqs.request_ids)
 
 
-def _initialize_paged_scheduler(
-    scheduler: BaseScheduler,
-    *,
-    num_blocks: int = 64,
-    max_num_seqs: int = 1,
-) -> None:
-    native_kv_managers.register_all_kvcache_specs(None)
-    spec = FullAttentionSpec(
-        block_size=4,
-        num_kv_heads=2,
-        head_size=8,
-        dtype=torch.bfloat16,
-    )
-    config = KVCacheConfig(
-        num_blocks=num_blocks,
-        kv_cache_tensors=[KVCacheTensor(size=spec.page_size_bytes * num_blocks, shared_by=["layer0"])],
-        kv_cache_groups=[KVCacheGroupSpec(layer_names=["layer0"], kv_cache_spec=spec)],
-    )
-    scheduler.initialize(
-        SimpleNamespace(
-            diffusion_kv_mode=DiffusionKVCacheMode.PAGED_SCHEDULER,
-            max_model_len=64,
-            max_num_seqs=max_num_seqs,
-        ),
-        kv_cache_config=config,
-        scheduler_block_size=4,
-        hash_block_size=4,
-        kv_vllm_config=SimpleNamespace(
-            model_config=SimpleNamespace(max_model_len=64),
-            max_in_flight_tokens=64,
-        ),
-    )
-
-
-def _attach_diffusion_kv(request: OmniDiffusionRequest, *, seq_len: int = 8) -> None:
-    request.diffusion_kv_requests = (
-        DiffusionKVRequest(
-            f"{request.request_id}/diffusion-kv/0",
-            sequence_id=0,
-            prefix_len=4,
-            target_len=4,
-            seq_len=seq_len,
-        ),
-    )
-
-
 class _StubScheduler:
     def __init__(self, request: OmniDiffusionRequest, output) -> None:
         self._request = request
@@ -448,16 +400,18 @@ class TestRequestScheduler:
         assert self.scheduler.has_requests() is False
 
     def test_diffusion_kv_request_moves_to_scheduler_state(self) -> None:
-        _initialize_paged_scheduler(self.scheduler)
+        self.scheduler.initialize(SimpleNamespace(diffusion_kv_mode=DiffusionKVCacheMode.PAGED_SCHEDULER))
         request = _make_request("diffusion-kv")
         prepared_layout = object()
         request.prepared_layout = prepared_layout
+        text_context = DiffusionKVContext(context_id="text", cache_role="cross.text", num_tokens=8)
         kv_request = DiffusionKVRequest(
             "diffusion-kv/diffusion-kv/0",
             sequence_id=0,
             prefix_len=4,
             target_len=8,
             seq_len=16,
+            kv_contexts=(text_context,),
         )
         request.diffusion_kv_requests = (kv_request,)
 
@@ -467,179 +421,12 @@ class TestRequestScheduler:
 
         assert state is not None
         assert state.diffusion_kv_requests == (kv_request,)
+        assert state.diffusion_kv_requests[0].kv_contexts == (text_context,)
         assert request.diffusion_kv_requests is None
         assert scheduler_output.scheduled_new_reqs[0].req.prepared_layout is prepared_layout
-        metadata = scheduler_output.scheduled_new_reqs[0].diffusion_kv_metadata
-        assert metadata is not None
-        assert metadata.request_id == request.request_id
-        assert len(metadata.sequences[0].block_ids[0]) == 4
-
-    def test_diffusion_kv_capacity_backpressures_fifo_until_blocks_are_freed(self) -> None:
-        _initialize_paged_scheduler(self.scheduler, num_blocks=3, max_num_seqs=2)
-        first_request = _make_request("first")
-        second_request = _make_request("second")
-        _attach_diffusion_kv(first_request)
-        _attach_diffusion_kv(second_request)
-        self.scheduler.add_request(first_request)
-        self.scheduler.add_request(second_request)
-
-        first_output = self.scheduler.schedule()
-
-        assert _new_ids(first_output) == ["first"]
-        assert first_output.num_waiting_reqs == 1
-
-        self.scheduler.update_from_output(first_output, _make_request_output("first"))
-        second_output = self.scheduler.schedule()
-
-        assert _new_ids(second_output) == ["second"]
-        assert second_output.num_waiting_reqs == 0
-        # The same output carries both the release and the newly admitted block
-        # table; the future Worker data plane must process finished ids first.
-        assert second_output.finished_req_ids == {"first"}
-        assert second_output.scheduled_new_reqs[0].diffusion_kv_metadata is not None
-
-    def test_impossible_diffusion_kv_capacity_finishes_only_that_request(self) -> None:
-        _initialize_paged_scheduler(self.scheduler, num_blocks=2, max_num_seqs=2)
-        impossible = _make_request("impossible")
-        schedulable = _make_request("schedulable")
-        _attach_diffusion_kv(impossible, seq_len=8)
-        schedulable.diffusion_kv_requests = (
-            DiffusionKVRequest(
-                "schedulable/diffusion-kv/0",
-                sequence_id=0,
-                prefix_len=0,
-                target_len=4,
-                seq_len=4,
-            ),
-        )
-        self.scheduler.add_request(impossible)
-        self.scheduler.add_request(schedulable)
-
-        sched_output = self.scheduler.schedule()
-
-        failed_state = self.scheduler.get_request_state("impossible")
-        assert failed_state is not None
-        assert failed_state.status == DiffusionRequestStatus.FINISHED_ERROR
-        assert failed_state.error is not None
-        assert "cannot fit even when the block pool is empty" in failed_state.error
-        assert sched_output.finished_req_ids == {"impossible"}
-        assert _new_ids(sched_output) == ["schedulable"]
-
-        finished = self.scheduler.update_from_output(
-            sched_output,
-            _make_request_output("schedulable"),
-        )
-
-        assert finished == {"impossible", "schedulable"}
-
-    def test_impossible_diffusion_kv_capacity_does_not_block_waiters_under_load(self) -> None:
-        _initialize_paged_scheduler(self.scheduler, num_blocks=4, max_num_seqs=2)
-        running = _make_request("running")
-        _attach_diffusion_kv(running)
-        self.scheduler.add_request(running)
-        assert _new_ids(self.scheduler.schedule()) == ["running"]
-
-        impossible = _make_request("impossible")
-        impossible.diffusion_kv_requests = tuple(
-            DiffusionKVRequest(
-                f"impossible/diffusion-kv/{sequence_id}",
-                sequence_id=sequence_id,
-                prefix_len=0,
-                target_len=4,
-                seq_len=4,
-            )
-            for sequence_id in range(4)
-        )
-        schedulable = _make_request("schedulable")
-        schedulable.diffusion_kv_requests = (
-            DiffusionKVRequest(
-                "schedulable/diffusion-kv/0",
-                sequence_id=0,
-                prefix_len=0,
-                target_len=4,
-                seq_len=4,
-            ),
-        )
-        self.scheduler.add_request(impossible)
-        self.scheduler.add_request(schedulable)
-
-        sched_output = self.scheduler.schedule()
-
-        failed_state = self.scheduler.get_request_state("impossible")
-        assert failed_state is not None
-        assert failed_state.status == DiffusionRequestStatus.FINISHED_ERROR
-        assert failed_state.error is not None
-        assert "required_blocks=4, available_blocks=3" in failed_state.error
-        assert sched_output.finished_req_ids == {"impossible"}
-        assert _new_ids(sched_output) == ["schedulable"]
-
-    def test_diffusion_kv_internal_allocation_error_finishes_request(self, monkeypatch) -> None:
-        _initialize_paged_scheduler(self.scheduler)
-        request = _make_request("native-error")
-        _attach_diffusion_kv(request)
-        self.scheduler.add_request(request)
-        manager = self.scheduler._diffusion_kv_manager
-        assert manager is not None
-
-        def raise_native_error(*args, **kwargs):
-            raise ValueError("native allocation bug")
-
-        monkeypatch.setattr(
-            manager,
-            "reserve_request",
-            raise_native_error,
-        )
-
-        sched_output = self.scheduler.schedule()
-
-        state = self.scheduler.get_request_state("native-error")
-        assert state is not None
-        assert state.status == DiffusionRequestStatus.FINISHED_ERROR
-        assert state.error == "native allocation bug"
-        assert sched_output.finished_req_ids == {"native-error"}
-        assert sched_output.scheduled_request_ids == []
-
-    def test_diffusion_kv_preemption_retains_allocation(self) -> None:
-        _initialize_paged_scheduler(self.scheduler, num_blocks=3)
-        request = _make_request("preempted")
-        _attach_diffusion_kv(request)
-        self.scheduler.add_request(request)
-        self.scheduler.schedule()
-        manager = self.scheduler._diffusion_kv_manager
-        assert manager is not None
-        metadata = manager.get_metadata("preempted")
-
-        assert self.scheduler.preempt_request("preempted") is True
-        resumed_output = self.scheduler.schedule()
-
-        assert resumed_output.scheduled_new_reqs == []
-        assert _cached_ids(resumed_output) == ["preempted"]
-        assert manager.get_metadata("preempted") == metadata
-
-    @pytest.mark.parametrize(
-        "status",
-        [
-            DiffusionRequestStatus.FINISHED_COMPLETED,
-            DiffusionRequestStatus.FINISHED_ABORTED,
-            DiffusionRequestStatus.FINISHED_ERROR,
-        ],
-    )
-    def test_diffusion_kv_terminal_status_releases_allocation(self, status: DiffusionRequestStatus) -> None:
-        _initialize_paged_scheduler(self.scheduler, num_blocks=3)
-        request = _make_request("terminal")
-        _attach_diffusion_kv(request)
-        self.scheduler.add_request(request)
-        self.scheduler.schedule()
-        manager = self.scheduler._diffusion_kv_manager
-        assert manager is not None
-        assert manager.has_request("terminal") is True
-
-        self.scheduler.finish_requests("terminal", status)
-
-        assert manager.has_request("terminal") is False
 
     def test_paged_scheduler_rejects_missing_kv_request(self) -> None:
-        _initialize_paged_scheduler(self.scheduler)
+        self.scheduler.initialize(SimpleNamespace(diffusion_kv_mode=DiffusionKVCacheMode.PAGED_SCHEDULER))
         request = _make_request("diffusion-kv")
 
         with pytest.raises(ValueError, match="did not produce DiffusionKVRequest"):
@@ -656,7 +443,7 @@ class TestRequestScheduler:
         ],
     )
     def test_diffusion_kv_request_rejects_legacy_dense_kv(self, owner_name: str, field_name: str) -> None:
-        _initialize_paged_scheduler(self.scheduler)
+        self.scheduler.initialize(SimpleNamespace(diffusion_kv_mode=DiffusionKVCacheMode.PAGED_SCHEDULER))
         request = _make_request("diffusion-kv")
         owner = request if owner_name == "request" else request.sampling_params
         setattr(owner, field_name, object())
@@ -1134,45 +921,6 @@ class TestDiffusionEngine:
         assert engine.scheduler.max_num_running_reqs == 1
         fake_executor_cls.assert_called_once_with(od_config)
 
-    def test_initializes_paged_scheduler_from_native_control_plane(
-        self,
-        monkeypatch: pytest.MonkeyPatch,
-        mocker: MockerFixture,
-    ) -> None:
-        od_config = SimpleNamespace(model_class_name="mock_model", streaming_output=False)
-        fake_executor_cls = mocker.Mock(return_value=mocker.Mock())
-        scheduler = mocker.Mock()
-        kv_cache_config = object()
-        kv_vllm_config = object()
-
-        monkeypatch.setattr(
-            "vllm_omni.diffusion.diffusion_engine.get_diffusion_post_process_func",
-            lambda *args, **kwargs: None,
-        )
-        monkeypatch.setattr(
-            "vllm_omni.diffusion.diffusion_engine.get_diffusion_pre_process_func",
-            lambda *args, **kwargs: None,
-        )
-        monkeypatch.setattr(
-            "vllm_omni.diffusion.diffusion_engine.DiffusionExecutor.get_class",
-            lambda *args, **kwargs: fake_executor_cls,
-        )
-        monkeypatch.setattr(
-            "vllm_omni.diffusion.diffusion_engine.initialize_diffusion_kv_control_plane",
-            lambda *args, **kwargs: (kv_cache_config, 16, 16, kv_vllm_config),
-        )
-
-        engine = DiffusionEngine(od_config, scheduler=scheduler)
-
-        assert engine.scheduler is scheduler
-        scheduler.initialize.assert_called_once_with(
-            od_config,
-            kv_cache_config=kv_cache_config,
-            scheduler_block_size=16,
-            hash_block_size=16,
-            kv_vllm_config=kv_vllm_config,
-        )
-
     def test_scheduler_alias_keeps_default_request_scheduler(self) -> None:
         scheduler = Scheduler()
         scheduler.initialize(SimpleNamespace())
@@ -1239,21 +987,6 @@ class TestDiffusionEngine:
         assert output.aborted is True
         assert output.abort_message == "Request req-finalize aborted."
 
-    def test_finalize_finished_request_returns_scheduler_admission_error(self) -> None:
-        engine = DiffusionEngine.__new__(DiffusionEngine)
-        engine.scheduler = StepScheduler()
-        engine.scheduler.initialize(SimpleNamespace())
-
-        req_id = engine.scheduler.add_request(_make_request("req-error"))
-        engine.scheduler._finish_requests(
-            {req_id: DiffusionRequestStatus.FINISHED_ERROR},
-            {req_id: "KV request cannot fit"},
-        )
-
-        output = engine._finalize_finished_request(req_id)
-
-        assert output.error == "KV request cannot fit"
-
     @pytest.mark.asyncio
     async def test_streaming_runner_output_notifies_each_chunk(self) -> None:
         engine = DiffusionEngine.__new__(DiffusionEngine)

tests/diffusion/test_diffusion_worker.py

@@ -13,7 +13,11 @@ import pytest
 import torch
 from pytest_mock import MockerFixture
 
-from vllm_omni.diffusion.worker.diffusion_worker import DiffusionWorker
+from vllm_omni.diffusion.worker.diffusion_worker import (
+    DiffusionWorker,
+    _create_diffusion_worker_vllm_config,
+    _make_diffusion_vllm_model_config,
+)
 
 pytestmark = [pytest.mark.core_model, pytest.mark.diffusion, pytest.mark.gpu]
 
@@ -54,6 +58,34 @@ def mock_gpu_worker(mocker: MockerFixture, mock_od_config):
     return worker
 
 
+def test_diffusion_vllm_model_config_supplies_dtype_for_quant_methods():
+    from types import SimpleNamespace
+
+    from vllm_omni.quantization import build_quant_config
+
+    od_config = SimpleNamespace(
+        model="dummy",
+        dtype=torch.bfloat16,
+        quantization_config=build_quant_config(
+            {
+                "quant_method": "modelopt",
+                "quant_algo": "FP8",
+                "ignore": [],
+            }
+        ),
+        tf_model_config=SimpleNamespace(),
+        enforce_eager=True,
+        is_moe=False,
+    )
+
+    model_config = _make_diffusion_vllm_model_config(od_config)
+
+    assert model_config.dtype is torch.bfloat16
+    assert model_config.quantization == "modelopt"
+    assert model_config.quantization_config is od_config.quantization_config
+    assert model_config.is_quantized()
+
+
 class TestDiffusionWorkerSleep:
     """Test DiffusionWorker.sleep method."""
 
@@ -297,3 +329,49 @@ class TestDiffusionWorkerWakeUp:
         mock_buffer2.data.copy_.assert_not_called()
 
         assert result is True
+
+
+class TestWorkerVllmConfigAdditionalConfig:
+    """Test worker-side VllmConfig construction with additional_config."""
+
+    def test_create_diffusion_worker_vllm_config_passes_additional_config(self, mocker: MockerFixture):
+        mock_vllm_config = mocker.Mock()
+        mock_vllm_cls = mocker.patch(
+            "vllm_omni.diffusion.worker.diffusion_worker.VllmConfig",
+            return_value=mock_vllm_config,
+        )
+        od_config = mocker.Mock(additional_config={"torchair_graph_config": {"enabled": True}})
+
+        result = _create_diffusion_worker_vllm_config(torch.device("cpu"), od_config)
+
+        assert result is mock_vllm_config
+        assert mock_vllm_cls.call_args.kwargs["additional_config"] == od_config.additional_config
+
+    def test_create_diffusion_worker_vllm_config_falls_back_when_constructor_rejects_additional_config(
+        self, mocker: MockerFixture
+    ):
+        mock_vllm_config = mocker.Mock()
+        mock_vllm_cls = mocker.patch(
+            "vllm_omni.diffusion.worker.diffusion_worker.VllmConfig",
+            side_effect=[
+                TypeError("VllmConfig.__init__() got an unexpected keyword argument 'additional_config'"),
+                mock_vllm_config,
+            ],
+        )
+        od_config = mocker.Mock(additional_config={"ascend_scheduler_config": {"foo": "bar"}})
+
+        result = _create_diffusion_worker_vllm_config(torch.device("cpu"), od_config)
+
+        assert result is mock_vllm_config
+        assert mock_vllm_cls.call_count == 2
+        assert getattr(mock_vllm_config, "additional_config") == od_config.additional_config
+
+    def test_create_diffusion_worker_vllm_config_reraises_other_type_errors(self, mocker: MockerFixture):
+        mocker.patch(
+            "vllm_omni.diffusion.worker.diffusion_worker.VllmConfig",
+            side_effect=TypeError("additional_config values must be hashable"),
+        )
+        od_config = mocker.Mock(additional_config={"ascend_scheduler_config": {"foo": "bar"}})
+
+        with pytest.raises(TypeError, match="hashable"):
+            _create_diffusion_worker_vllm_config(torch.device("cpu"), od_config)

tests/diffusion/test_multiproc_engine_concurrency.py

@@ -413,26 +413,6 @@ class TestRequestModeDispatch:
         assert [output.result.error for output in result.runner_outputs] == ["A", "B"]
         executor.collective_rpc.assert_called_once()
 
-    def test_dlo_dp_forwards_request_metadata_envelopes_as_one_wave(self):
-        executor, _, _ = _make_executor(num_gpus=2)
-        executor.od_config = SimpleNamespace(
-            step_execution=False,
-            parallel_config=SimpleNamespace(data_parallel_size=2),
-            enable_distributed_layerwise_offload=True,
-            dlo_use_allgather=True,
-        )
-        executor.collective_rpc = Mock(return_value=[_tagged_output("A"), _tagged_output("B")])
-        scheduler_output = _make_sched_output("A", "B")
-        for new_req in scheduler_output.scheduled_new_reqs:
-            new_req.diffusion_kv_metadata = SimpleNamespace(request_id=new_req.request_id)
-
-        result = executor.execute_request(scheduler_output)
-
-        assert [output.result.error for output in result.runner_outputs] == ["A", "B"]
-        forwarded_envelopes = executor.collective_rpc.call_args.kwargs["args"][0]
-        assert forwarded_envelopes is scheduler_output.scheduled_new_reqs
-        assert [envelope.diffusion_kv_metadata.request_id for envelope in forwarded_envelopes] == ["A", "B"]
-
     @pytest.mark.parametrize(
         ("field", "value"),
         [

tests/e2e/offline_inference/test_sana_wm.py

@@ -1,159 +0,0 @@
-# SPDX-License-Identifier: Apache-2.0
-# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
-
-from __future__ import annotations
-
-import os
-from typing import Any
-
-import numpy as np
-import pytest
-
-pytestmark = [
-    pytest.mark.advanced_model,
-    pytest.mark.diffusion,
-    pytest.mark.gpu,
-]
-
-
-def _make_sana_wm_e2e_image() -> Any:
-    from PIL import Image
-
-    from vllm_omni.diffusion.models.sana_wm import SANA_WM_OUTPUT_HEIGHT, SANA_WM_OUTPUT_WIDTH
-
-    return Image.new("RGB", (SANA_WM_OUTPUT_WIDTH, SANA_WM_OUTPUT_HEIGHT), (96, 128, 160))
-
-
-def _sana_wm_e2e_tensor_parallel_size() -> int:
-    return int(
-        os.environ.get(
-            "SANA_WM_E2E_TENSOR_PARALLEL_SIZE",
-            os.environ.get("SANA_WM_E2E_TP", "1"),
-        )
-    )
-
-
-def _coerce_video_array(video: Any) -> np.ndarray:
-    import torch
-
-    if isinstance(video, list):
-        assert video, "SANA-WM e2e produced an empty video list."
-        video = video[0]
-    if isinstance(video, torch.Tensor):
-        video = video.detach().cpu().float().numpy()
-    video_array = np.asarray(video)
-    if video_array.ndim == 5:
-        assert video_array.shape[0] == 1
-        video_array = video_array[0]
-    assert video_array.ndim == 4
-    return video_array
-
-
-def _assert_sana_wm_e2e_shape(
-    video: np.ndarray,
-    *,
-    output_type: str,
-    num_frames: int,
-) -> None:
-    if output_type == "latent":
-        # Latent output is channel-first: (C, T, H, W) after removing the
-        # batch dimension. Stage-1 emits LTX-2 latents with 128 channels.
-        assert video.shape[0] == 128
-        assert 0 < video.shape[1] <= num_frames
-        return
-    assert 0 < video.shape[0] <= num_frames
-
-
-def _run_sana_wm_e2e(*, output_type: str) -> np.ndarray:
-    import torch
-
-    from vllm_omni.diffusion.models.sana_wm import (
-        SANA_WM_MODEL_ID,
-        SANA_WM_OUTPUT_HEIGHT,
-        SANA_WM_OUTPUT_WIDTH,
-    )
-    from vllm_omni.entrypoints.omni import Omni
-    from vllm_omni.inputs.data import OmniDiffusionSamplingParams
-    from vllm_omni.outputs import OmniRequestOutput
-
-    if not torch.cuda.is_available():
-        pytest.skip("Sana-WM e2e requires CUDA.")
-
-    model = os.environ.get("SANA_WM_E2E_MODEL", SANA_WM_MODEL_ID)
-    model_class_name = os.environ.get("SANA_WM_E2E_MODEL_CLASS", "SanaWmPipeline")
-    num_frames = int(os.environ.get("SANA_WM_E2E_NUM_FRAMES", "9"))
-    stage1_steps = int(
-        os.environ.get(
-            "SANA_WM_E2E_STAGE1_STEPS",
-            os.environ.get("SANA_WM_E2E_NUM_INFERENCE_STEPS", "1"),
-        )
-    )
-    action = os.environ.get("SANA_WM_E2E_ACTION", f"w-{max(num_frames - 1, 1)}")
-    tensor_parallel_size = _sana_wm_e2e_tensor_parallel_size()
-    image = _make_sana_wm_e2e_image()
-    omni = Omni(
-        model=model,
-        model_class_name=model_class_name,
-        enforce_eager=True,
-        tensor_parallel_size=tensor_parallel_size,
-    )
-    extra_args: dict[str, Any] = {}
-    # The native Stage-1 path is latent-token capped; raise it for full-size
-    # e2e runs (704x1280 exceeds the default 4096-token cap).
-    native_max_tokens = os.environ.get("SANA_WM_E2E_NATIVE_MAX_TOKENS", "").strip()
-    if native_max_tokens:
-        extra_args["sana_wm_native_max_tokens"] = int(native_max_tokens)
-    # Return the requested output type so the shape assertion matches the
-    # pipeline output instead of the raw Stage-1 latent default.
-
-    output = omni.generate(
-        {
-            "prompt": "A slow forward camera move through a quiet city street.",
-            "multi_modal_data": {"image": image},
-            "sana_wm": {
-                "action": action,
-                "num_frames": num_frames,
-                "translation_speed": 0.055,
-                "rotation_speed_deg": 1.2,
-                # Avoid the optional Pi3X dependency by passing deterministic
-                # camera intrinsics directly.
-                "intrinsics": {
-                    "fx": SANA_WM_OUTPUT_WIDTH / 2,
-                    "fy": SANA_WM_OUTPUT_WIDTH / 2,
-                    "cx": SANA_WM_OUTPUT_WIDTH / 2,
-                    "cy": SANA_WM_OUTPUT_HEIGHT / 2,
-                },
-            },
-        },
-        OmniDiffusionSamplingParams(
-            height=SANA_WM_OUTPUT_HEIGHT,
-            width=SANA_WM_OUTPUT_WIDTH,
-            num_frames=num_frames,
-            seed=0,
-            fps=16,
-            num_inference_steps=stage1_steps,
-            guidance_scale=1.0,
-            guidance_scale_provided=True,
-            output_type=output_type,
-            extra_args=extra_args,
-        ),
-    )
-
-    request_output = output[0] if isinstance(output, list) else output
-    assert isinstance(request_output, OmniRequestOutput)
-    assert request_output.error is None
-    assert request_output.images
-
-    frames = request_output.images[0]
-    video = _coerce_video_array(frames)
-    _assert_sana_wm_e2e_shape(video, output_type=output_type, num_frames=num_frames)
-    return video
-
-
-def test_sana_wm_native_generates_video() -> None:
-    output_type = os.environ.get("SANA_WM_E2E_OUTPUT_TYPE", "np")
-    video = _run_sana_wm_e2e(output_type=output_type)
-    if output_type == "latent":
-        assert video.ndim == 4
-    else:
-        assert video.shape[-1] in (3, 4)

tests/e2e/online_serving/test_minimax_music3.py

@@ -1,112 +0,0 @@
-# SPDX-License-Identifier: Apache-2.0
-# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
-"""E2E online tests for MiniMax Music 3 text-to-music.
-
-This is a music model on the speech endpoint, so the request shape differs
-from every other TTS model here: ``input`` carries the lyrics, ``instructions``
-carries the caption that decides genre and arrangement, and both are required.
-
-Weekly rather than per-PR: the checkpoint is 27 GB across five components and a
-single request occupies two decode rows, because classifier-free guidance is
-mandatory for this model. Rejection cases live in
-``tests/dfx/reliability/invalid_param_test/test_invalid_audio_speech.py``.
-"""
-
-import os
-
-os.environ["VLLM_WORKER_MULTIPROC_METHOD"] = "spawn"
-
-import pytest
-
-from tests.helpers.mark import hardware_test
-from tests.helpers.runtime import OmniServerParams
-from tests.helpers.stage_config import get_deploy_config_path
-
-MODEL = "MiniMaxAI/MiniMax-Music3"
-DEFAULT_AUDIO_SPEECH_TIMEOUT_S = 600.0
-
-# 25 frames per second, so 250 frames is ten seconds and 750 is thirty. The
-# byte floors sit below the true payload (32 kHz stereo, 16-bit) to leave room
-# for the model ending a song before the cap.
-_FRAMES_10S = 250
-_FRAMES_30S = 750
-_MIN_BYTES_10S = 500_000
-_MIN_BYTES_30S = 1_500_000
-
-LYRICS = (
-    "[Verse]\nWalking down the empty street at midnight\n"
-    "Streetlights flicker like a broken dream\n"
-    "[Chorus]\nAnd I keep on walking\nTill the morning finds me"
-)
-
-CAPTION = (
-    "A melancholic lo-fi hip-hop track at 85 BPM in F minor: mellow Rhodes piano riff, "
-    "soft vinyl crackle, dusty boom-bap drums with a laid-back swing, warm upright bass. "
-    "Intimate bedroom production, gentle tape saturation, no bright cymbals."
-)
-
-
-tts_server_params = [
-    pytest.param(
-        OmniServerParams(
-            model=MODEL,
-            stage_config_path=get_deploy_config_path("minimax_music3.yaml"),
-            server_args=["--trust-remote-code", "--disable-log-stats"],
-        ),
-        id="minimax_music3",
-    )
-]
-
-
-@pytest.mark.slow
-@pytest.mark.tts
-@hardware_test(res={"cuda": "H100"}, num_cards={"cuda": 1})
-@pytest.mark.parametrize("omni_server", tts_server_params, indirect=True)
-def test_text_to_music_001(omni_server, openai_client) -> None:
-    """Baseline: lyrics plus caption in, one ten-second song out.
-
-    Deploy Setting: minimax_music3.yaml
-    Input Modal: text (lyrics) + text (caption)
-    Output Modal: audio, 32 kHz stereo
-    Input Setting: stream=False, seed pinned, 250 frames
-    """
-    request_config = {
-        "model": omni_server.model,
-        "input": LYRICS,
-        "instructions": CAPTION,
-        "seed": 1,
-        "max_new_tokens": _FRAMES_10S,
-        "stream": False,
-        "response_format": "wav",
-        "timeout": DEFAULT_AUDIO_SPEECH_TIMEOUT_S,
-        "min_audio_bytes": _MIN_BYTES_10S,
-    }
-    openai_client.send_audio_speech_request(request_config)
-
-
-@pytest.mark.slow
-@pytest.mark.tts
-@hardware_test(res={"cuda": "H100"}, num_cards={"cuda": 1})
-@pytest.mark.parametrize("omni_server", tts_server_params, indirect=True)
-def test_text_to_music_multi_window_002(omni_server, openai_client) -> None:
-    """A song long enough to span several acoustic windows.
-
-    The acoustic stage decodes overlapping 200-frame windows on a 100-frame
-    hop and crops the overlap back off, so a request that fits in one window
-    never exercises the stitching. 750 frames covers seven of them.
-
-    Deploy Setting: minimax_music3.yaml
-    Input Setting: stream=False, seed pinned, 750 frames
-    """
-    request_config = {
-        "model": omni_server.model,
-        "input": LYRICS,
-        "instructions": CAPTION,
-        "seed": 1,
-        "max_new_tokens": _FRAMES_30S,
-        "stream": False,
-        "response_format": "wav",
-        "timeout": DEFAULT_AUDIO_SPEECH_TIMEOUT_S,
-        "min_audio_bytes": _MIN_BYTES_30S,
-    }
-    openai_client.send_audio_speech_request(request_config)

tests/e2e/online_serving/test_pi0_expansion.py

@@ -1,85 +0,0 @@
-# SPDX-License-Identifier: Apache-2.0
-# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
-
-"""E2E online serving test for π0 (Pi-Zero) OpenPI websocket serving.
-
-Boots ``vllm serve --omni --deploy-config pi0.yaml`` and drives the real OpenPI
-websocket (``/v1/realtime/robot/openpi``) — the same wire path a robot uses
-(handshake metadata → send observation → receive action chunk). Mirrors
-``tests/e2e/online_serving/test_dreamzero_expansion.py``. Needs a GPU + a
-pi0_base checkpoint; skipped in CI unless explicitly run.
-
-The in-process LeRobot parity oracle lives separately in
-``tests/diffusion/models/pi0/test_pi0_parity.py``.
-"""
-
-from __future__ import annotations
-
-import os
-
-import pytest
-
-from tests.helpers.mark import hardware_test
-from tests.helpers.runtime import (
-    OmniServerParams,
-    get_open_port,
-    pi0_openpi_require_dependencies,
-    pi0_openpi_run_policy_session,
-    pi0_openpi_validate_session_result,
-)
-
-os.environ["VLLM_WORKER_MULTIPROC_METHOD"] = "spawn"
-os.environ["VLLM_TEST_CLEAN_GPU_MEMORY"] = "0"
-
-MODEL = "lerobot/pi0_base"
-
-
-test_params = [
-    OmniServerParams(
-        model=MODEL,
-        port=8092,
-        server_args=[
-            "--deploy-config",
-            "vllm_omni/deploy/pi0.yaml",
-            "--served-model-name",
-            "pi0",
-            "--enforce-eager",
-            "--disable-log-stats",
-        ],
-        env_dict={
-            "ATTENTION_BACKEND": "torch",
-            "DIFFUSION_ATTENTION_BACKEND": "TORCH_SDPA",
-            "VLLM_DISABLE_COMPILE_CACHE": "1",
-            "MASTER_PORT": str(get_open_port()),
-        },
-    )
-]
-
-
-@pytest.mark.full_model
-@pytest.mark.diffusion
-@hardware_test(res={"cuda": "H100"})
-@pytest.mark.parametrize("omni_server", test_params, indirect=True)
-def test_pi0_openpi_online(omni_server):
-    try:
-        pi0_openpi_require_dependencies()
-    except ModuleNotFoundError as exc:
-        pytest.skip(str(exc))
-
-    result = pi0_openpi_run_policy_session(
-        host=omni_server.host,
-        port=omni_server.port,
-        prompt="pick up the red block and place it in the bin",
-        session_id="pi0-online-e2e",
-        num_steps=2,
-    )
-
-    # Asserts every returned chunk is [50, 32] + finite, and the handshake
-    # metadata matches pi0.yaml's policy_server_config.
-    pi0_openpi_validate_session_result(result)
-
-    metadata = result["metadata"]
-    assert tuple(metadata["image_resolution"]) == (224, 224)
-    assert metadata["needs_wrist_camera"] is True
-    assert metadata["needs_session_id"] is False
-    assert metadata["action_space"] == "joint_position"

tests/e2e/online_serving/test_qwen3_omni_expansion.py

@@ -452,31 +452,11 @@ def test_audio_in_video_default_loader_sampling_regression(omni_server, openai_c
 @hardware_test(res={"cuda": "H100", "rocm": "MI325"}, num_cards=2)
 @pytest.mark.parametrize("omni_server", test_params, indirect=True)
 def test_one_word_prompt_001(omni_server, openai_client) -> None:
-    """Catastrophic-regression gate on one-word pronunciation.
-
-    Whether a one-word answer is pronounced intelligibly is sampled, not guaranteed: the
-    talker runs at temperature 0.9 (the upstream default, see qwen3_omni_moe.yaml stage 1),
-    so asserting it per request gates on a random variable, which is what the old 10-retry
-    loop was compensating for.
-
-    Sends a fixed 40 requests and requires at least 29 to transcribe as the expected
-    word. 85.6% is a predeclared minimum acceptable success probability, taken from a
-    measured HF reference run (77/90); vLLM-Omni itself measured 92.2%. Assuming 40
-    independent Bernoulli trials, >=29/40 has a nominal rejection probability of 0.9%
-    at p=.856, with 92.9% power at p=.60 and 55.9% at p=.70.
-
-    Those are conditional on p, not a measured false-failure rate for this CI: 77/90 is
-    one point estimate (Wilson 76.8%-91.4%), and at the low end of that interval the gate
-    would reject about 20% of runs. Pinning down a real rate would take repeated
-    report-only runs on the target hardware.
-
-    A drop to 70% is caught only about half the time, so this is a smoke gate for gross
-    pipeline breakage, not a quality tracker -- resolving moderate drift needs ~80+
-    samples and belongs in an accuracy benchmark. See #5395.
-
+    """
     Input Modal: text only (one-word answer constraint).
     Output Modal: text, audio (default ``modalities``); ``key_words`` only assert on text.
     Input Setting: stream=True
+    Datasets: single request
     """
     messages = dummy_messages_from_mix_data(
         system_prompt=get_system_prompt(),
@@ -494,13 +474,17 @@ def test_one_word_prompt_001(omni_server, openai_client) -> None:
         "transcript_language": "en",
     }
 
-    # The judge runs per request so mismatches can be counted, which also transcribes
-    # concurrently where the previous request_num=5 path judged serially -- worth watching
-    # on tighter cards, since each judge loads whisper on the same device as the talker and
-    # code2wav stages.
-    openai_client.send_omni_request(
-        request_config, request_num=40, min_successes=29, max_concurrency=get_max_batch_size()
-    )
+    # Retry only when assert_omni_response fails on text/audio cosine similarity (see tests/helpers/assertions.py).
+    _similarity_assert_msg = "The audio content is not same as the text"
+    _max_retries = 10
+    for attempt in range(_max_retries):
+        try:
+            openai_client.send_omni_request(request_config, request_num=get_max_batch_size())
+            break
+        except AssertionError as e:
+            if _similarity_assert_msg not in str(e) or attempt == _max_retries - 1:
+                raise
+            print(f"Similarity assertion failed, retrying {attempt + 2}/{_max_retries}: {e!r}")
 
 
 @hardware_test(res={"cuda": "H100", "rocm": "MI325"}, num_cards=2)

tests/e2e/online_serving/test_sana_wm.py

@@ -1,106 +0,0 @@
-# SPDX-License-Identifier: Apache-2.0
-# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
-
-"""
-Online serving smoke for SANA-WM image-to-video (first-frame I2V via ``/v1/videos``).
-
-Boots the server straight from the model id — the Stage-1 repo's
-``model_index.json`` names ``SanaWmPipeline``, so the class resolves on its own
-— submits one async ``/v1/videos`` job with a first-frame reference image, and
-asserts the job completes and returns video bytes.
-
-From ``tests/``::
-
-    pytest -s -v e2e/online_serving/test_sana_wm.py -m "advanced_model and diffusion" --run-level=advanced_model
-"""
-
-import base64
-import json
-import os
-from io import BytesIO
-
-import pytest
-
-from tests.helpers.mark import hardware_marks
-from tests.helpers.runtime import OmniServer, OmniServerParams, OpenAIClientHandler
-from vllm_omni.diffusion.models.sana_wm import (
-    SANA_WM_MODEL_ID,
-    SANA_WM_OUTPUT_HEIGHT,
-    SANA_WM_OUTPUT_WIDTH,
-)
-
-os.environ["VLLM_WORKER_MULTIPROC_METHOD"] = "spawn"
-
-MODEL = SANA_WM_MODEL_ID
-PROMPT = "A slow forward camera move through a quiet city street."
-NEGATIVE_PROMPT = "blurry, low quality, distorted, watermark"
-
-# Smoke knobs: a handful of frames / steps so the job finishes quickly.
-SMOKE_NUM_FRAMES = 9
-SMOKE_NUM_INFERENCE_STEPS = 2
-# First-frame I2V needs a camera trajectory; "w-<n>" is the forward-move action DSL.
-# Passed via the ``sana_wm`` form field (a JSON string parsed server-side); explicit
-# intrinsics avoid the optional Pi3X camera-calibration dependency.
-SANA_WM_PARAMS = {
-    "action": f"w-{SMOKE_NUM_FRAMES - 1}",
-    "translation_speed": 0.055,
-    "rotation_speed_deg": 1.2,
-    "intrinsics": {
-        "fx": SANA_WM_OUTPUT_WIDTH / 2,
-        "fy": SANA_WM_OUTPUT_WIDTH / 2,
-        "cx": SANA_WM_OUTPUT_WIDTH / 2,
-        "cy": SANA_WM_OUTPUT_HEIGHT / 2,
-    },
-}
-
-SINGLE_CARD_FEATURE_MARKS = hardware_marks(res={"cuda": "H100"})
-
-
-def _first_frame_data_url() -> str:
-    """A solid-color first frame encoded as a PNG ``data:`` URL for I2V."""
-    from PIL import Image
-
-    image = Image.new("RGB", (SANA_WM_OUTPUT_WIDTH, SANA_WM_OUTPUT_HEIGHT), (96, 128, 160))
-    buffer = BytesIO()
-    image.save(buffer, format="PNG")
-    encoded = base64.b64encode(buffer.getvalue()).decode("ascii")
-    return f"data:image/png;base64,{encoded}"
-
-
-def _get_diffusion_feature_cases(model: str):
-    """Single default server row."""
-    return [
-        pytest.param(
-            OmniServerParams(model=model),
-            id="default",
-            marks=SINGLE_CARD_FEATURE_MARKS,
-        ),
-    ]
-
-
-@pytest.mark.advanced_model
-@pytest.mark.diffusion
-@pytest.mark.parametrize("omni_server", _get_diffusion_feature_cases(MODEL), indirect=True)
-def test_image_to_video_001(omni_server: OmniServer, openai_client: OpenAIClientHandler) -> None:
-    """Default SANA-WM I2V smoke: async ``/v1/videos`` job completes and returns video bytes."""
-    request_config = {
-        "model": omni_server.model,
-        "image_reference": _first_frame_data_url(),
-        "form_data": {
-            "prompt": PROMPT,
-            "negative_prompt": NEGATIVE_PROMPT,
-            # The camera payload rides inside extra_params; there is no
-            # top-level sana_wm form field on the video endpoints, so sending
-            # one is silently dropped and the request fails in the
-            # preprocessor for missing camera control.
-            "extra_params": json.dumps({"sana_wm": SANA_WM_PARAMS}),
-            "height": SANA_WM_OUTPUT_HEIGHT,
-            "width": SANA_WM_OUTPUT_WIDTH,
-            "num_frames": SMOKE_NUM_FRAMES,
-            "fps": 16,
-            "num_inference_steps": SMOKE_NUM_INFERENCE_STEPS,
-            "guidance_scale": 5.0,
-            "seed": 42,
-        },
-    }
-    openai_client.send_video_diffusion_request(request_config)

tests/engine/test_async_omni_engine_input.py

@@ -371,106 +371,3 @@ def test_build_add_request_message_releases_preselected_replica_on_preprocess_er
         )
 
     assert stage_pool.get_bound_replica_id("req-error") is None
-
-
-def test_cfg_companion_build_failure_admits_nothing(mocker: MockerFixture):
-    """A guided request is all-or-nothing.
-
-    Admitting the parent before its companion is built leaves an orphan: a
-    model whose guidance is mandatory never completes the pair, so the request
-    holds scheduler and KV capacity for the scheduler's whole hold budget and
-    then produces no audio. The build has to raise before anything is enqueued.
-    """
-    engine = object.__new__(AsyncOmniEngine)
-    params = SamplingParams(max_tokens=8)
-    engine.prompt_expand_func = mocker.Mock(side_effect=ValueError("cannot build the null twin"))
-    engine.supported_tasks = ("generate",)
-    engine.input_processor = mocker.Mock()
-    engine.request_queue = mocker.Mock()
-
-    with pytest.raises(ValueError, match="null twin"):
-        engine._build_cfg_companions(
-            parent_id="req",
-            original_prompt={"prompt": "positive"},
-            stage0_params=params,
-            sampling_params_list=[params],
-        )
-
-    engine.request_queue.sync_q.put.assert_not_called()
-
-
-def test_cfg_companion_processing_failure_admits_nothing(mocker: MockerFixture):
-    """The same holds when expansion succeeds but input processing does not."""
-    engine = object.__new__(AsyncOmniEngine)
-    params = SamplingParams(max_tokens=8)
-    engine.prompt_expand_func = lambda *_args: [
-        ExpandedPrompt(
-            prompt={"prompt": "negative"},
-            role="cfg_text",
-            request_id_suffix="__cfg_text",
-        )
-    ]
-    engine.supported_tasks = ("generate",)
-    engine.input_processor = mocker.Mock()
-    engine.input_processor.process_inputs.side_effect = RuntimeError("tokenizer rejected the prompt")
-    engine.request_queue = mocker.Mock()
-
-    with pytest.raises(RuntimeError, match="tokenizer rejected"):
-        engine._build_cfg_companions(
-            parent_id="req",
-            original_prompt={"prompt": "positive"},
-            stage0_params=params,
-            sampling_params_list=[params],
-        )
-
-    engine.request_queue.sync_q.put.assert_not_called()
-
-
-def test_cfg_companion_build_returns_messages_without_enqueueing(mocker: MockerFixture):
-    """Construction and admission are separate steps."""
-    engine = object.__new__(AsyncOmniEngine)
-    params = SamplingParams(max_tokens=8)
-    engine.prompt_expand_func = lambda *_args: [
-        ExpandedPrompt(
-            prompt={"prompt": "negative"},
-            role="cfg_text",
-            request_id_suffix="__cfg_text",
-        )
-    ]
-    engine.supported_tasks = ("generate",)
-    engine.input_processor = mocker.Mock()
-    engine.input_processor.process_inputs.return_value = _make_engine_core_request("req__cfg_text")
-    engine.request_queue = mocker.Mock()
-
-    companions = engine._build_cfg_companions(
-        parent_id="req",
-        original_prompt={"prompt": "positive"},
-        stage0_params=params,
-        sampling_params_list=[params],
-    )
-
-    assert len(companions) == 1
-    assert companions[0].companion_id == "req__cfg_text"
-    assert companions[0].parent_id == "req"
-    engine.request_queue.sync_q.put.assert_not_called()
-
-
-def test_cfg_expansion_returning_nothing_is_not_an_error(mocker: MockerFixture):
-    """An unguided request expands to no companions and admits normally."""
-    engine = object.__new__(AsyncOmniEngine)
-    params = SamplingParams(max_tokens=8)
-    engine.prompt_expand_func = lambda *_args: []
-    engine.supported_tasks = ("generate",)
-    engine.input_processor = mocker.Mock()
-    engine.request_queue = mocker.Mock()
-
-    assert (
-        engine._build_cfg_companions(
-            parent_id="req",
-            original_prompt={"prompt": "positive"},
-            stage0_params=params,
-            sampling_params_list=[params],
-        )
-        == []
-    )
-    engine.request_queue.sync_q.put.assert_not_called()

tests/engine/test_stage_engine_args.py

@@ -8,23 +8,8 @@ from pathlib import Path
 
 import pytest
 from pydantic.fields import FieldInfo
-from vllm.config import CacheConfig as VllmCacheConfig
-from vllm.config import CompilationConfig as VllmCompilationConfig
-from vllm.config import LoadConfig as VllmLoadConfig
-from vllm.config import ParallelConfig as VllmParallelConfig
-from vllm.config import ProfilerConfig as VllmProfilerConfig
-from vllm.config import SchedulerConfig as VllmSchedulerConfig
-from vllm.engine.arg_utils import EngineArgs
 
-from vllm_omni.config.omni_config import (
-    _LLM_STAGE_ENGINE_FIELDS,
-    OmniStageCacheConfig,
-    OmniStageLoadConfig,
-    OmniStageParallelConfig,
-    OmniStageSchedulerConfig,
-    VllmOmniARStageConfig,
-    VllmOmniConfig,
-)
+from vllm_omni.config.omni_config import VllmOmniConfig
 from vllm_omni.config.pipeline_registry import OMNI_PIPELINES, resolve_pipeline_config
 from vllm_omni.config.stage_config import (
     DeployConfig,
@@ -68,30 +53,6 @@ def _effective_backend_values(config_cls: type, engine_args: dict) -> dict[str,
 
 _LLM_BACKEND_FIELDS = frozenset(field.name for field in fields(OmniEngineArgs))
 _DIFFUSION_BACKEND_FIELDS = frozenset(field.name for field in fields(OmniDiffusionConfig))
-_OMNI_ONLY_LLM_STAGE_ENGINE_FIELDS = frozenset(
-    {
-        "active_stream_window",
-        "codec_frame_rate_hz",
-        "custom_voice_dir",
-        "devices",
-        "disable_autocast",
-        "enable_multithread_weight_load",
-        "env",
-        "has_sampling_extra_args",
-        "log_level",
-        "log_stats",
-        "model_arch",
-        "model_subdir",
-        "num_gpus",
-        "num_replicas",
-        "num_weight_load_threads",
-        "omni_kv_config",
-        "parallel_config",
-        "subtalker_sampling_params",
-        "task_type",
-        "tokenizer_subdir",
-    }
-)
 
 
 @pytest.fixture(autouse=True)
@@ -236,100 +197,6 @@ def _legacy_and_typed_stages(
     return legacy_stages, omni_config
 
 
-def test_llm_stage_engine_field_schema_tracks_upstream_engine_args():
-    upstream_engine_fields = frozenset(field.name for field in fields(EngineArgs))
-
-    assert _LLM_STAGE_ENGINE_FIELDS - upstream_engine_fields == _OMNI_ONLY_LLM_STAGE_ENGINE_FIELDS
-
-
-def test_typed_llm_projection_discovers_explicit_upstream_config_fields():
-    stage_config = VllmOmniARStageConfig(
-        stage_pipeline_config=StagePipelineConfig(stage_id=0, model_stage="test"),
-        load_config=OmniStageLoadConfig(ignore_patterns=["*.bin"]),
-        cache_config=OmniStageCacheConfig(block_size=32, cache_dtype="fp8"),
-        scheduler_config=OmniStageSchedulerConfig(
-            prefill_schedule_interval=2,
-            policy="priority",
-        ),
-        parallel_config=OmniStageParallelConfig(
-            data_parallel_master_ip="10.0.0.1",
-            enable_dbo=True,
-        ),
-    )
-
-    engine_args = stage_init_utils._project_omni_stage_engine_args(stage_config)
-
-    assert engine_args["ignore_patterns"] == ["*.bin"]
-    assert engine_args["block_size"] == 32
-    assert engine_args["kv_cache_dtype"] == "fp8"
-    assert engine_args["prefill_schedule_interval"] == 2
-    assert engine_args["scheduling_policy"] == "priority"
-    assert engine_args["data_parallel_address"] == "10.0.0.1"
-    assert engine_args["enable_dbo"] is True
-
-
-def test_typed_llm_projection_does_not_emit_inherited_upstream_defaults():
-    stage_config = VllmOmniARStageConfig(
-        stage_pipeline_config=StagePipelineConfig(stage_id=0, model_stage="test"),
-    )
-
-    engine_args = stage_init_utils._project_omni_stage_engine_args(stage_config)
-
-    inherited_defaults = {
-        "ignore_patterns",
-        "block_size",
-        "kv_cache_dtype",
-        "prefill_schedule_interval",
-        "scheduling_policy",
-        "data_parallel_address",
-        "data_parallel_rank",
-        "data_parallel_size_local",
-        "data_parallel_rpc_port",
-        "enable_dbo",
-    }
-    assert inherited_defaults.isdisjoint(engine_args)
-
-
-def test_typed_llm_projection_rejects_explicit_fields_owned_by_another_boundary():
-    stage_config = VllmOmniARStageConfig(
-        stage_pipeline_config=StagePipelineConfig(stage_id=0, model_stage="test"),
-        scheduler_config=OmniStageSchedulerConfig(scheduler_cls="test.CustomScheduler"),
-    )
-
-    with pytest.raises(ValueError, match=r"no EngineArgs projection: scheduler_cls"):
-        stage_init_utils._project_omni_stage_engine_args(stage_config)
-
-
-@pytest.mark.parametrize("stage_id", [0, 1], ids=["ar", "generation"])
-def test_typed_llm_engine_args_preserve_upstream_config_objects(tmp_path, stage_id):
-    pipeline, deploy, model = _engine_arg_inputs(tmp_path)
-    compilation_config = VllmCompilationConfig(backend="eager")
-    profiler_config = VllmProfilerConfig(profiler="cuda")
-    stage_prefix = f"stage_{stage_id}_"
-    omni_config = VllmOmniConfig.from_pipeline_config(
-        pipeline,
-        user_deploy_config=copy.deepcopy(deploy),
-        cli_overrides={
-            "model": model,
-            f"{stage_prefix}compilation_config": compilation_config,
-            f"{stage_prefix}profiler_config": profiler_config,
-        },
-    )
-
-    stage_config = omni_config.stage_by_id(stage_id)
-    typed_args = build_engine_args_dict_from_omni_stage_config(stage_config, model)
-
-    assert isinstance(stage_config.compilation_config, VllmCompilationConfig)
-    assert isinstance(typed_args["compilation_config"], VllmCompilationConfig)
-    assert typed_args["compilation_config"] is not stage_config.compilation_config
-    assert typed_args["compilation_config"].backend == "eager"
-
-    assert isinstance(stage_config.profiler_config, VllmProfilerConfig)
-    assert isinstance(typed_args["profiler_config"], VllmProfilerConfig)
-    assert typed_args["profiler_config"] is not stage_config.profiler_config
-    assert typed_args["profiler_config"].profiler == "cuda"
-
-
 def test_typed_llm_engine_args_preserve_legacy_adapter_behavior(tmp_path):
     pipeline, deploy, model = _engine_arg_inputs(tmp_path)
     legacy_stages, omni_config = _legacy_and_typed_stages(pipeline, deploy, model)
@@ -355,13 +222,6 @@ def test_typed_llm_engine_args_preserve_legacy_adapter_behavior(tmp_path):
         assert {name: typed_args[name] for name in legacy_args} == legacy_args
 
     thinker_args = typed_args_by_stage[0]
-    inherited_vllm_fields = {
-        field.name
-        for config_cls in (VllmLoadConfig, VllmCacheConfig, VllmSchedulerConfig, VllmParallelConfig)
-        for field in fields(config_cls)
-    }
-    topology_projected_fields = {"scheduler_cls"}
-    assert (inherited_vllm_fields - _LLM_STAGE_ENGINE_FIELDS - topology_projected_fields).isdisjoint(thinker_args)
     assert thinker_args["model"] == str(tmp_path / "stage-model" / "ar-model")
     assert thinker_args["tokenizer"] == str(tmp_path / "stage-model" / "ar-tokenizer")
     assert thinker_args["stage_id"] == 0

tests/entrypoints/openai_api/test_api_server_guards.py

@@ -1,810 +0,0 @@
-# SPDX-License-Identifier: Apache-2.0
-# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
-"""API-server surface guards for the OpenAI Omni entrypoint.
-
-Purpose
--------
-Freeze the *observable* server contract so helper moves (P0.2), route
-extraction (P0.3), and later endpoint-family PRs cannot silently change:
-
-* which routes exist,
-* how upstream routes are overridden at assembly time,
-* unavailable / missing-handler envelopes,
-* shared engine-error JSON fields,
-* app-state keys wired by ``omni_init_app_state``.
-
-CPU-only (fakes / TestClient). No checked-in OpenAPI golden JSON.
-
-Common wrong changes these tests are meant to catch
-----------------------------------------------------
-* Route accidentally removed, renamed, method-changed, or double-registered
-  after a move (census / assembled-app checks).
-* OpenAPI drops an Omni HTTP path while the router still has it (or the reverse).
-* Assembly forgets to strip upstream ``/v1/chat/completions`` (batch) /
-  ``/v1/models``, or also deletes unrelated upstream routes.
-* Profiler / Omni router not mounted; storage not started; Omni engine
-  exception handlers not registered.
-* Timestamp middleware stops stamping HTTP, or starts mutating WebSocket scopes.
-* Missing-handler WS/HTTP responses change status, JSON shape, or wording
-  (clients treat these as capability probes).
-* Engine error payloads lose ``request_id`` / ``error_stage_id``.
-* Pure-diffusion vs multi-stage init drops an ``app.state`` key, or leaves a
-  handler as ``None`` while the route is still registered.
-
-Not covered here (add elsewhere if needed)
-------------------------------------------
-* Deep per-endpoint success behavior (family UTs).
-* Byte-identical helper bodies after a move (P0.2 AST integrity).
-* Live GPU / speech byte-identity checks.
-"""
-
-from __future__ import annotations
-
-import asyncio
-import json
-from contextlib import asynccontextmanager
-from types import SimpleNamespace
-
-import pytest
-from fastapi import FastAPI, HTTPException
-from fastapi.routing import APIRoute, APIWebSocketRoute
-from fastapi.testclient import TestClient
-from starlette.datastructures import State
-from starlette.requests import Request
-from starlette.websockets import WebSocketDisconnect
-from vllm.v1.engine.exceptions import EngineDeadError, EngineGenerateError
-
-from vllm_omni.entrypoints.openai import api_server
-
-pytestmark = [pytest.mark.core_model, pytest.mark.cpu]
-
-# FastAPI/Starlette auto-adds HEAD on every GET route. Including HEAD here
-# would invent ("HEAD", path) entries for every GET and fail the current census.
-# TODO: extend this set and the expected census if @router.head / @router.options appear.
-_HTTP_METHODS = {"GET", "POST", "DELETE", "PUT", "PATCH"}
-
-# Exact Omni router census on main. Update intentionally when routes change.
-_EXPECTED_ROUTER_ROUTES = {
-    ("POST", "/v1/chat/completions"),
-    ("POST", "/v1/chat/completions/batch"),
-    ("POST", "/v1/audio/speech"),
-    ("POST", "/v1/audio/speech/batch"),
-    ("POST", "/v1/audio/generate"),
-    ("GET", "/v1/audio/voices"),
-    ("POST", "/v1/audio/voices"),
-    ("DELETE", "/v1/audio/voices/{name}"),
-    ("WEBSOCKET", "/v1/audio/speech/stream"),
-    ("WEBSOCKET", "/v1/video/chat/stream"),
-    ("WEBSOCKET", "/v1/realtime/video"),
-    ("WEBSOCKET", "/v1/realtime"),
-    ("WEBSOCKET", "/v1/realtime/robot/openpi"),
-    ("WEBSOCKET", "/v1/duplex"),
-    ("GET", "/health"),
-    ("GET", "/v1/models"),
-    ("POST", "/v1/images/generations"),
-    ("POST", "/v1/images/edits"),
-    ("POST", "/v1/videos"),
-    ("POST", "/v1/videos/sync"),
-    ("GET", "/v1/videos"),
-    ("GET", "/v1/videos/{video_id}"),
-    ("DELETE", "/v1/videos/{video_id}"),
-    ("GET", "/v1/videos/{video_id}/content"),
-    ("POST", "/v1/omni/sleep"),
-    ("POST", "/v1/omni/wakeup"),
-}
-_EXPECTED_PROFILER_ROUTES = {
-    ("POST", "/start_profile"),
-    ("POST", "/stop_profile"),
-}
-
-# App-state keys that must remain available across refactor phases.
-# Presence is not enough: "not wired" is written as ``state.X = None``.
-_DIFFUSION_APP_STATE_KEYS = {
-    "engine_client",
-    "log_stats",
-    "args",
-    "sleeping_stages",
-    "stage_configs",
-    "vllm_config",
-    "diffusion_engine",
-    "openai_serving_models",
-    "serving_tokenization",
-    "openai_serving_chat",
-    "openai_serving_chat_batch",
-    "openai_serving_speech",
-    "openai_serving_audio_generate",
-    "openai_serving_video",
-    "openai_streaming_video_output",
-    "openai_serving_duplex",
-    "openai_streaming_speech",
-    "openai_streaming_video",
-    "openai_serving_realtime_robot",
-    "enable_server_load_tracking",
-    "server_load_metrics",
-}
-_DIFFUSION_MUST_BE_NONE = {
-    "vllm_config",
-    "serving_tokenization",
-    "openai_serving_duplex",
-    "openai_streaming_speech",
-    "openai_streaming_video",
-}
-_DIFFUSION_MUST_BE_WIRED = _DIFFUSION_APP_STATE_KEYS - _DIFFUSION_MUST_BE_NONE
-_MULTISTAGE_APP_STATE_KEYS = {
-    "engine_client",
-    "log_stats",
-    "args",
-    "sleeping_stages",
-    "stage_configs",
-    "vllm_config",
-    "openai_serving_models",
-    "serving_tokenization",
-    "online_renderer",
-    "openai_serving_chat",
-    "openai_serving_chat_batch",
-    "openai_serving_speech",
-    "openai_serving_audio_generate",
-    "openai_streaming_speech",
-    "openai_streaming_video",
-    "openai_serving_duplex",
-    "openai_serving_realtime",
-    "openai_serving_video",
-    "openai_serving_realtime_robot",
-    "enable_server_load_tracking",
-    "server_load_metrics",
-}
-_MULTISTAGE_MUST_BE_NONE = {
-    "openai_serving_duplex",
-    "openai_serving_realtime_robot",
-}
-_MULTISTAGE_MUST_BE_WIRED = _MULTISTAGE_APP_STATE_KEYS - _MULTISTAGE_MUST_BE_NONE
-
-
-def _route_entries(routes) -> list[tuple[str, str]]:
-    entries: list[tuple[str, str]] = []
-    for route in routes:
-        if isinstance(route, APIRoute):
-            for method in sorted((route.methods or set()) & _HTTP_METHODS):
-                entries.append((method, route.path))
-        elif isinstance(route, APIWebSocketRoute):
-            entries.append(("WEBSOCKET", route.path))
-    return entries
-
-
-def _route_manifest(routes) -> set[tuple[str, str]]:
-    return set(_route_entries(routes))
-
-
-def _assert_unique_route_entries(routes) -> None:
-    entries = _route_entries(routes)
-    duplicates = sorted({item for item in entries if entries.count(item) > 1})
-    assert not duplicates, f"duplicate route registrations: {duplicates}"
-
-
-def _matching_http_routes(routes, method: str, path: str) -> list[APIRoute]:
-    return [
-        route
-        for route in routes
-        if isinstance(route, APIRoute) and route.path == path and method in (route.methods or set())
-    ]
-
-
-def _single_http_route(routes, method: str, path: str) -> APIRoute:
-    matches = _matching_http_routes(routes, method, path)
-    assert len(matches) == 1, f"expected exactly one {method} {path}, got {len(matches)}"
-    return matches[0]
-
-
-def _assert_app_state_snapshot(
-    state,
-    *,
-    expected_keys: set[str],
-    must_be_wired: set[str],
-    must_be_none: set[str],
-) -> None:
-    assert must_be_wired.isdisjoint(must_be_none)
-    assert must_be_wired | must_be_none <= expected_keys
-    present = {key for key in expected_keys if hasattr(state, key)}
-    assert present == expected_keys
-    not_wired = sorted(key for key in must_be_wired if getattr(state, key) is None)
-    assert not not_wired, f"app.state keys registered but not wired: {not_wired}"
-    unexpectedly_set = sorted(key for key in must_be_none if getattr(state, key) is not None)
-    assert not unexpectedly_set, f"app.state keys expected to stay None: {unexpectedly_set}"
-
-
-def _minimal_args(**overrides) -> SimpleNamespace:
-    args = SimpleNamespace(
-        tool_parser_plugin=None,
-        reasoning_parser_plugin=None,
-        reasoning_parser=None,
-        structured_outputs_config=SimpleNamespace(reasoning_parser=None),
-        enable_ssl_refresh=False,
-        host="127.0.0.1",
-        port=8000,
-        uvicorn_log_level="info",
-        disable_uvicorn_access_log=True,
-        ssl_keyfile=None,
-        ssl_certfile=None,
-        ssl_ca_certs=None,
-        ssl_cert_reqs=0,
-        ssl_ciphers=None,
-        h11_max_incomplete_event_size=None,
-        h11_max_header_count=None,
-        log_config_file=None,
-        disable_access_log_for_endpoints=None,
-        disable_log_stats=True,
-        enable_log_requests=False,
-        max_log_len=None,
-        served_model_name=None,
-        model="demo-model",
-        enable_server_load_tracking=False,
-        log_error_stack=False,
-        chat_template=None,
-        chat_template_content_format="auto",
-        trust_request_chat_template=False,
-        tool_server=None,
-        lora_modules=None,
-        response_role="assistant",
-        default_chat_template_kwargs=None,
-        return_tokens_as_token_ids=False,
-        enable_auto_tool_choice=False,
-        exclude_tools_when_tool_choice_none=False,
-        tool_call_parser=None,
-        enable_prompt_tokens_details=False,
-        enable_force_include_usage=False,
-        enable_log_outputs=False,
-        enable_log_deltas=False,
-        tokens_only=False,
-        shutdown_timeout=0,
-        stage_configs_path=None,
-        deploy_config=None,
-    )
-    for key, value in overrides.items():
-        setattr(args, key, value)
-    return args
-
-
-def _request_for(app: FastAPI, *, method: str = "GET", path: str = "/") -> Request:
-    return Request(
-        {
-            "type": "http",
-            "method": method,
-            "path": path,
-            "headers": [],
-            "app": app,
-            "state": {},
-        }
-    )
-
-
-class _FakeSocket:
-    def __init__(self) -> None:
-        self.closed = False
-
-    def close(self) -> None:
-        self.closed = True
-
-
-class _FakeEngineClient:
-    def __init__(self, *, stage_configs=None, endpoint_restrictions=None, vllm_config=None) -> None:
-        self.stage_configs = stage_configs if stage_configs is not None else []
-        self.endpoint_restrictions = endpoint_restrictions if endpoint_restrictions is not None else {}
-        self.model_config = SimpleNamespace()
-        self.input_processor = object()
-        self.renderer = object()
-        self.errored = False
-        self.vllm_config = vllm_config
-
-    async def get_vllm_config(self):
-        return self.vllm_config
-
-    async def get_supported_tasks(self) -> tuple[str, ...]:
-        return ("generate",)
-
-    async def get_tokenizer(self):
-        return SimpleNamespace(chat_template="dummy")
-
-    async def check_health(self) -> None:
-        return None
-
-
-def _marker(name: str):
-    return SimpleNamespace(name=name)
-
-
-def test_router_route_manifest_is_explicit_and_complete() -> None:
-    """Lock the Omni + profiler router census.
-
-    Fails if a route is added/removed/renamed, a method changes, or a
-    WebSocket path disappears during helper/router moves. Also fails if the
-    same public route is registered more than once, which a set manifest alone
-    would collapse.
-    """
-    # TODO(P0.3): retarget this object-level census when routes move onto family
-    # routers. The assembled-app census is the client-facing lock that must survive.
-    _assert_unique_route_entries(api_server.router.routes)
-    _assert_unique_route_entries(api_server.profiler_router.routes)
-    assert _route_manifest(api_server.router.routes) == _EXPECTED_ROUTER_ROUTES
-    assert _route_manifest(api_server.profiler_router.routes) == _EXPECTED_PROFILER_ROUTES
-
-
-def test_router_openapi_paths_cover_http_manifest() -> None:
-    """Lock that OpenAPI advertises every HTTP census route/method.
-
-    Fails if a decorator still registers a route but OpenAPI loses the path
-    (or method), e.g. after moving route bodies / response models.
-    """
-    app = FastAPI()
-    app.include_router(api_server.router)
-    app.include_router(api_server.profiler_router)
-    paths = app.openapi().get("paths", {})
-
-    http_routes = {
-        (method, path) for method, path in _EXPECTED_ROUTER_ROUTES | _EXPECTED_PROFILER_ROUTES if method != "WEBSOCKET"
-    }
-    for method, path in sorted(http_routes):
-        assert path in paths, f"missing OpenAPI path {path}"
-        assert method.lower() in paths[path], f"missing OpenAPI method {method} {path}"
-
-
-@pytest.mark.asyncio
-async def test_api_server_assembly_replaces_upstream_routes_and_mounts_omni_router(monkeypatch) -> None:
-    """Lock worker assembly: override, mount, storage, handlers, full census.
-
-    Fails if assembly stops replacing upstream chat/batch/models, deletes
-    unrelated upstream routes, forgets Omni/profiler mounts, skips storage
-    start, or drops Omni ``EngineDeadError`` / ``EngineGenerateError`` handlers.
-    """
-    captured: dict[str, object] = {}
-
-    def fake_build_openai_app(args, supported_tasks):
-        app = FastAPI()
-
-        @app.post("/v1/chat/completions")
-        async def upstream_chat():
-            return {"owner": "upstream"}
-
-        @app.post("/v1/chat/completions/batch")
-        async def upstream_batch():
-            return {"owner": "upstream"}
-
-        @app.get("/v1/models")
-        async def upstream_models():
-            return {"owner": "upstream"}
-
-        @app.get("/v1/upstream-only")
-        async def upstream_only():
-            return {"owner": "upstream"}
-
-        captured["supported_tasks"] = supported_tasks
-        return app
-
-    @asynccontextmanager
-    async def fake_build_async_omni(*_args, **_kwargs):
-        yield _FakeEngineClient(
-            stage_configs=[{"engine_args": {"profiler_config": {"profiler": "torch"}}}],
-            vllm_config=SimpleNamespace(parallel_config=SimpleNamespace(_api_process_rank=0)),
-        )
-
-    async def fake_omni_init_app_state(engine_client, state, args):
-        state.engine_client = engine_client
-        state.initialized_by_omni = True
-
-    async def fake_storage_start():
-        captured["storage_started"] = True
-
-    async def fake_serve_http(app, **_kwargs):
-        captured["served_app"] = app
-        task = asyncio.create_task(asyncio.sleep(0))
-        await asyncio.sleep(0)
-        return task
-
-    monkeypatch.setattr(api_server, "build_openai_app", fake_build_openai_app)
-    monkeypatch.setattr(api_server, "build_async_omni", fake_build_async_omni)
-    monkeypatch.setattr(api_server, "omni_init_app_state", fake_omni_init_app_state)
-    monkeypatch.setattr(
-        api_server,
-        "shutdown_unsupported_routes",
-        lambda app, restrictions: captured.setdefault("restrictions", restrictions),
-    )
-    monkeypatch.setattr(api_server, "STORAGE_MANAGER", SimpleNamespace(start=fake_storage_start))
-    monkeypatch.setattr(api_server, "serve_http", fake_serve_http)
-
-    sock = _FakeSocket()
-    await api_server.omni_run_server_worker("127.0.0.1:8000", sock, _minimal_args())
-
-    served_wrapper = captured["served_app"]
-    # Timestamp middleware wraps the FastAPI app.
-    assert hasattr(served_wrapper, "_inner")
-    served_app = served_wrapper._inner
-    routes = served_app.routes
-    _assert_unique_route_entries(routes)
-    manifest = _route_manifest(routes)
-
-    assert captured["supported_tasks"] == ("generate",)
-    assert captured["storage_started"] is True
-    assert captured["restrictions"] == {}
-    assert sock.closed is True
-    assert served_app.state.initialized_by_omni is True
-
-    # Upstream override: unrelated route kept; Omni owns overridden pairs once.
-    # Endpoint identity ensures the surviving route is the Omni handler, not
-    # the fake upstream handler with the same method/path.
-    assert ("GET", "/v1/upstream-only") in manifest
-    assert _single_http_route(routes, "POST", "/v1/chat/completions").endpoint is api_server.create_chat_completion
-    assert (
-        _single_http_route(routes, "POST", "/v1/chat/completions/batch").endpoint
-        is api_server.create_batch_chat_completion
-    )
-    assert _single_http_route(routes, "GET", "/v1/models").endpoint is api_server.show_available_models
-
-    # Assembled app includes Omni router HTTP/WS surface + profiler.
-    for item in _EXPECTED_ROUTER_ROUTES:
-        assert item in manifest, f"assembled app missing {item}"
-    assert ("POST", "/start_profile") in manifest
-    assert ("POST", "/stop_profile") in manifest
-
-    assert served_app.exception_handlers[EngineGenerateError]
-    assert served_app.exception_handlers[EngineDeadError]
-
-    # Light OpenAPI check on the assembled FastAPI app (not the ASGI wrapper).
-    openapi_paths = served_app.openapi().get("paths", {})
-    assert "/v1/chat/completions" in openapi_paths
-    assert "/v1/models" in openapi_paths
-    assert "/health" in openapi_paths
-
-
-@pytest.mark.asyncio
-async def test_timestamp_middleware_stamps_http_and_passes_websocket(monkeypatch) -> None:
-    """Lock outermost timestamp middleware behavior.
-
-    Fails if HTTP requests lose ``request_timestamp``, or WebSocket scopes
-    start getting stamped / mutated incorrectly.
-    """
-    captured: dict[str, object] = {}
-
-    def fake_build_openai_app(args, supported_tasks):
-        return FastAPI()
-
-    @asynccontextmanager
-    async def fake_build_async_omni(*_args, **_kwargs):
-        yield _FakeEngineClient()
-
-    async def fake_omni_init_app_state(engine_client, state, args):
-        state.engine_client = engine_client
-
-    async def fake_serve_http(app, **_kwargs):
-        captured["served_app"] = app
-        task = asyncio.create_task(asyncio.sleep(0))
-        await asyncio.sleep(0)
-        return task
-
-    monkeypatch.setattr(api_server, "build_openai_app", fake_build_openai_app)
-    monkeypatch.setattr(api_server, "build_async_omni", fake_build_async_omni)
-    monkeypatch.setattr(api_server, "omni_init_app_state", fake_omni_init_app_state)
-    monkeypatch.setattr(api_server, "shutdown_unsupported_routes", lambda *_a, **_k: None)
-    monkeypatch.setattr(api_server, "STORAGE_MANAGER", SimpleNamespace(start=lambda: asyncio.sleep(0)))
-    monkeypatch.setattr(api_server, "serve_http", fake_serve_http)
-
-    await api_server.omni_run_server_worker("127.0.0.1:8000", _FakeSocket(), _minimal_args())
-    middleware = captured["served_app"]
-
-    http_scope = {"type": "http", "state": {}}
-    ws_scope = {"type": "websocket"}
-
-    async def _receive():
-        return {"type": "http.disconnect"}
-
-    async def _send(_message):
-        return None
-
-    called = {"http": False, "websocket": False}
-
-    async def inner(scope, receive, send):
-        called[scope["type"]] = True
-
-    middleware._inner = inner
-    await middleware(http_scope, _receive, _send)
-    await middleware(ws_scope, _receive, _send)
-
-    assert called["http"] is True
-    assert called["websocket"] is True
-    assert "request_timestamp" in http_scope["state"]
-    assert isinstance(http_scope["state"]["request_timestamp"], float)
-    assert "request_timestamp" not in ws_scope.get("state", {})
-
-
-@pytest.mark.parametrize(
-    ("path", "payload"),
-    [
-        ("/v1/audio/speech/stream", {"type": "error", "message": "Streaming speech is not available"}),
-        ("/v1/video/chat/stream", {"type": "error", "message": "Streaming video chat is not available"}),
-        ("/v1/realtime/video", {"type": "error", "message": "Streaming video generation is not available"}),
-        ("/v1/realtime", {"type": "error", "error": "Realtime API is not available", "code": "unsupported"}),
-        (
-            "/v1/realtime/robot/openpi",
-            {"type": "error", "error": "Robot policy not available", "code": "unsupported"},
-        ),
-        ("/v1/duplex", {"type": "error", "error": "Duplex API is not available", "code": "unsupported"}),
-    ],
-)
-def test_websocket_routes_emit_stable_unavailable_frames_and_close(path: str, payload: dict[str, str]) -> None:
-    """Lock missing-handler WebSocket capability probes.
-
-    Fails if unavailable JSON changes (keys/wording/code) or the socket stays
-    open after the error frame — common when WS wrappers move between modules.
-    """
-    app = FastAPI()
-    app.include_router(api_server.router)
-
-    with TestClient(app) as client:
-        with client.websocket_connect(path) as websocket:
-            assert websocket.receive_json() == payload
-            with pytest.raises(WebSocketDisconnect):
-                websocket.receive_text()
-
-
-def test_health_without_engine_returns_stable_unhealthy_response() -> None:
-    """Lock ``/health`` with no engine initialized.
-
-    Fails if status drifts from 503 or the unhealthy payload shape/reason
-    string changes during health-route extraction.
-    """
-    app = FastAPI()
-    response = asyncio.run(api_server.health(_request_for(app, path="/health")))
-
-    assert response.status_code == 503
-    assert json.loads(response.body) == {
-        "status": "unhealthy",
-        "reason": "No engine initialized",
-    }
-
-
-def test_models_without_handler_returns_empty_openai_list() -> None:
-    """Lock ``/v1/models`` when ``openai_serving_models`` is unset.
-
-    Fails if the empty listing becomes an error, or the OpenAI list envelope
-    changes (``object`` / ``data``).
-    """
-    app = FastAPI()
-    response = asyncio.run(api_server.show_available_models(_request_for(app, path="/v1/models")))
-
-    assert response.status_code == 200
-    assert json.loads(response.body) == {"object": "list", "data": []}
-
-
-def test_images_generation_without_engine_preserves_service_unavailable_error() -> None:
-    """Lock images generation when the multi-stage engine is missing.
-
-    Fails if the 503 detail string changes — clients/tests use it as a
-    capability signal when bootstrap failed.
-    """
-    app = FastAPI()
-    raw_request = _request_for(app, method="POST", path="/v1/images/generations")
-    request = api_server.ImageGenerationRequest(prompt="a cat", model="demo-model")
-
-    with pytest.raises(HTTPException) as exc_info:
-        asyncio.run(api_server.generate_images(request, raw_request))
-
-    assert exc_info.value.status_code == 503
-    assert exc_info.value.detail == "Multi-stage engine not initialized. Start server with a multi-stage omni model."
-
-
-def test_images_generation_without_multistage_chat_handler_preserves_unavailable_error(monkeypatch) -> None:
-    """Lock images generation when chat serving was not wired.
-
-    Fails if multi-stage images stop requiring ``openai_serving_chat``, or the
-    503 detail changes after images/chat ownership splits.
-    """
-    app = FastAPI()
-    app.state.engine_client = SimpleNamespace(stage_configs=[object(), object()])
-    app.state.stage_configs = app.state.engine_client.stage_configs
-    app.state.openai_serving_models = SimpleNamespace(base_model_paths=[SimpleNamespace(name="demo-model")])
-    app.state.openai_serving_chat = None
-    app.state.args = SimpleNamespace(max_generated_image_size=None)
-
-    raw_request = _request_for(app, method="POST", path="/v1/images/generations")
-    request = api_server.ImageGenerationRequest(prompt="a cat", model="demo-model")
-    monkeypatch.setattr(api_server, "get_stage_type", lambda _stage_cfg: "diffusion")
-
-    with pytest.raises(HTTPException) as exc_info:
-        asyncio.run(api_server.generate_images(request, raw_request))
-
-    assert exc_info.value.status_code == 503
-    assert exc_info.value.detail == "openai_serving_chat is not initialized for multi-stage image generation."
-
-
-def test_speech_without_handler_preserves_not_found_http_error() -> None:
-    """Lock speech HTTP when ``openai_serving_speech`` is unset.
-
-    Fails if missing speech becomes a different status/detail after TTS route
-    wrappers move out of ``api_server``.
-    """
-    app = FastAPI()
-    app.state.openai_serving_speech = None
-    raw_request = _request_for(app, method="POST", path="/v1/audio/speech")
-    request = api_server.OpenAICreateSpeechRequest(input="hello", model="demo-model", voice="alloy")
-
-    with pytest.raises(HTTPException) as exc_info:
-        asyncio.run(api_server.create_speech(request, raw_request))
-
-    assert exc_info.value.status_code == 404
-    assert exc_info.value.detail == "The model does not support Speech API"
-
-
-def test_engine_error_json_response_includes_request_and_stage_fields(monkeypatch) -> None:
-    """Lock shared ``EngineGenerateError`` JSON fields used by Omni handlers.
-
-    Fails if ``request_id`` / ``error_stage_id`` are dropped when error helpers
-    move (P0.2) or exception-handler wiring changes.
-
-    Exercises the envelope through the registered handler, not the private
-    ``_create_engine_error_json_response`` helper, so relocating that helper
-    in P0.2 is not a false red.
-    """
-    app = FastAPI()
-    # TODO(P0.2): retarget this call if ``_register_omni_exception_handlers``
-    # leaves api_server (red here is a rename, not a surface regression).
-    api_server._register_omni_exception_handlers(app)
-    app.state.engine_client = SimpleNamespace(errored=True, engine=SimpleNamespace(is_alive=lambda: False))
-    app.state.server = object()
-    app.state.args = SimpleNamespace(log_error_stack=False)
-
-    req = _request_for(app, method="POST", path="/v1/chat/completions")
-    req.state.request_metadata = SimpleNamespace(request_id="req-123")
-
-    monkeypatch.setattr(api_server, "terminate_if_errored", lambda **_kwargs: None)
-
-    exc = EngineGenerateError("boom")
-    exc.error_stage_id = 2  # type: ignore[attr-defined]
-    handler = app.exception_handlers[EngineGenerateError]
-    response = asyncio.run(handler(req, exc))
-    payload = json.loads(response.body)
-
-    assert "error" in payload
-    assert payload["error"]["request_id"] == "req-123"
-    assert payload["error"]["error_stage_id"] == 2
-    assert response.status_code >= 400
-
-
-def test_engine_dead_error_handler_registered_returns_json(monkeypatch) -> None:
-    """Lock Omni ``EngineDeadError`` handler registration + JSON envelope.
-
-    Fails if registration is skipped or the handler stops returning OpenAI-style
-    JSON with ``request_id``.
-    """
-    app = FastAPI()
-    # TODO(P0.2): retarget this call if ``_register_omni_exception_handlers``
-    # leaves api_server (red here is a rename, not a surface regression).
-    api_server._register_omni_exception_handlers(app)
-    app.state.engine_client = SimpleNamespace(errored=True, engine=SimpleNamespace(is_alive=lambda: False))
-    app.state.server = object()
-    app.state.args = SimpleNamespace(log_error_stack=False)
-
-    monkeypatch.setattr(api_server, "terminate_if_errored", lambda **_kwargs: None)
-
-    handler = app.exception_handlers[EngineDeadError]
-    req = _request_for(app)
-    req.state.request_metadata = SimpleNamespace(request_id="dead-1")
-    response = asyncio.run(handler(req, EngineDeadError()))
-    payload = json.loads(response.body)
-    assert payload["error"]["request_id"] == "dead-1"
-    assert response.status_code >= 400
-
-
-@pytest.mark.asyncio
-async def test_pure_diffusion_app_state_key_snapshot(monkeypatch) -> None:
-    """Lock pure-diffusion ``app.state`` keys after init, including live vs None.
-
-    Fails if bootstrap drops keys route owners still read (e.g. video/speech/
-    models), wires a handler that must stay None, or leaves a required
-    handler as None.
-    """
-    stage = SimpleNamespace(engine_args={})
-    engine = _FakeEngineClient(stage_configs=[stage])
-
-    def _for_diffusion_factory(label: str):
-        @classmethod
-        def _factory(cls, *args, **kwargs):
-            return _marker(label)
-
-        return _factory
-
-    monkeypatch.setattr(api_server, "get_stage_type", lambda _cfg: "diffusion")
-    monkeypatch.setattr(api_server.OmniOpenAIServingChat, "for_diffusion", _for_diffusion_factory("chat"))
-    monkeypatch.setattr(api_server.OmniOpenAIServingChatBatch, "for_diffusion", _for_diffusion_factory("chat_batch"))
-    monkeypatch.setattr(
-        api_server.OmniOpenAIServingAudioGenerate,
-        "for_diffusion",
-        _for_diffusion_factory("audio_generate"),
-    )
-    monkeypatch.setattr(api_server.OmniOpenAIServingVideo, "for_diffusion", _for_diffusion_factory("video"))
-    monkeypatch.setattr(api_server.OmniStreamingVideoOutputHandler, "__init__", lambda self, *a, **k: None)
-    monkeypatch.setattr(api_server.OmniOpenAIServingSpeech, "for_diffusion", _for_diffusion_factory("speech"))
-    monkeypatch.setattr(
-        api_server.ServingRealtimeRobotOpenPI,
-        "create_policy_server",
-        classmethod(lambda cls, *a, **k: _marker("openpi")),
-    )
-
-    state = State()
-    await api_server.omni_init_app_state(engine, state, _minimal_args())
-
-    _assert_app_state_snapshot(
-        state,
-        expected_keys=_DIFFUSION_APP_STATE_KEYS,
-        must_be_wired=_DIFFUSION_MUST_BE_WIRED,
-        must_be_none=_DIFFUSION_MUST_BE_NONE,
-    )
-    assert state.diffusion_engine is engine
-
-
-@pytest.mark.asyncio
-async def test_multistage_app_state_key_snapshot(monkeypatch) -> None:
-    """Lock multi-stage ``app.state`` keys after init, including live vs None.
-
-    Fails if chat/speech/video/realtime/tokenization keys disappear or are
-    left as None after init refactors — classic “route still registered,
-    handler never wired” failure mode.
-    """
-    engine = _FakeEngineClient(
-        stage_configs=[object(), object()],
-        vllm_config=SimpleNamespace(
-            lora_config=None,
-            model_config=SimpleNamespace(),
-            parallel_config=SimpleNamespace(_api_process_rank=0),
-        ),
-    )
-
-    class _FakeModels:
-        def __init__(self, *args, **kwargs):
-            self.base_model_paths = kwargs.get("base_model_paths") or []
-
-        async def init_static_loras(self):
-            return None
-
-    class _FakeCtor:
-        def __init__(self, *args, **kwargs):
-            pass
-
-        def warmup(self):
-            return None
-
-    class _FakeSpeech(_FakeCtor):
-        async def warmup(self):
-            return None
-
-    monkeypatch.setattr(api_server, "load_chat_template", lambda *_a, **_k: None)
-    monkeypatch.setattr(api_server, "process_lora_modules", lambda modules, _defaults: modules or [])
-    monkeypatch.setattr(api_server, "OpenAIServingModels", _FakeModels)
-    monkeypatch.setattr(api_server, "OnlineRenderer", _FakeCtor)
-    monkeypatch.setattr(api_server, "OpenAIServingResponses", _FakeCtor)
-    monkeypatch.setattr(api_server, "OmniOpenAIServingChat", _FakeCtor)
-    monkeypatch.setattr(api_server, "OmniOpenAIServingChatBatch", _FakeCtor)
-    monkeypatch.setattr(api_server, "OpenAIServingCompletion", _FakeCtor)
-    monkeypatch.setattr(api_server, "ServingPooling", _FakeCtor)
-    monkeypatch.setattr(api_server, "OpenAIServingEmbedding", _FakeCtor)
-    monkeypatch.setattr(api_server, "ServingClassification", _FakeCtor)
-    monkeypatch.setattr(api_server, "ServingScores", _FakeCtor)
-    monkeypatch.setattr(api_server, "ServingTokenization", _FakeCtor)
-    monkeypatch.setattr(api_server, "OpenAIServingTranscription", _FakeCtor)
-    monkeypatch.setattr(api_server, "OpenAIServingTranslation", _FakeCtor)
-    monkeypatch.setattr(api_server, "AnthropicServingMessages", _FakeCtor)
-    monkeypatch.setattr(api_server, "ServingTokens", _FakeCtor)
-    monkeypatch.setattr(api_server, "OmniOpenAIServingSpeech", _FakeSpeech)
-    monkeypatch.setattr(api_server, "OmniOpenAIServingAudioGenerate", _FakeCtor)
-    monkeypatch.setattr(api_server, "OmniStreamingSpeechHandler", _FakeCtor)
-    monkeypatch.setattr(api_server, "create_streaming_video_handler", lambda **_k: _marker("streaming_video"))
-    monkeypatch.setattr(api_server, "OpenAIServingRealtime", _FakeCtor)
-    monkeypatch.setattr(api_server, "OmniOpenAIServingVideo", _FakeCtor)
-    monkeypatch.setattr(api_server, "should_enable_duplex_endpoint", lambda *_a, **_k: False)
-
-    state = State()
-    await api_server.omni_init_app_state(engine, state, _minimal_args())
-
-    _assert_app_state_snapshot(
-        state,
-        expected_keys=_MULTISTAGE_APP_STATE_KEYS,
-        must_be_wired=_MULTISTAGE_MUST_BE_WIRED,
-        must_be_none=_MULTISTAGE_MUST_BE_NONE,
-    )

tests/entrypoints/openai_api/test_serving_speech.py

@@ -33,8 +33,6 @@ from vllm_omni.entrypoints.openai.protocol.audio import (
     OpenAICreateAudioGenerateRequest,
     OpenAICreateSpeechRequest,
     SpeechBatchItem,
-    SpeechInputTokenDetails,
-    SpeechTokenUsage,
     StreamingSpeechSessionConfig,
 )
 from vllm_omni.entrypoints.openai.serving_speech import (
@@ -157,7 +155,6 @@ class TestAudioMixin:
 # Helper to create mock model output for endpoint tests
 def create_mock_audio_output_for_test(
     request_id: str = "speech-mock-123",
-    metrics: dict | None = None,
 ) -> OmniRequestOutput:
     class MockCompletionOutput:
         def __init__(self, index: int = 0):
@@ -184,16 +181,11 @@ def create_mock_audio_output_for_test(
     audio_tensor = torch.sin(torch.linspace(0, 440 * 2 * torch.pi, num_samples))
     mock_request_output = MockRequestOutput(request_id=request_id, audio_tensor=audio_tensor)
 
-    if metrics is None:
-        metrics = {"stage_metrics": {"0": {"num_tokens_in": 0, "num_tokens_out": 5}}}
-
-    output = OmniRequestOutput.from_stage_output(
+    return OmniRequestOutput.from_stage_output(
         mock_request_output,
         stage_id=0,
         final_output_type="audio",
     )
-    output.metrics = metrics
-    return output
 
 
 def _write_custom_voice_manifest(root: Path, *, model_type: str, voices: dict) -> None:
@@ -241,7 +233,6 @@ def test_app(mocker: MockerFixture):
         models=mock_models,
         request_logger=mock_request_logger,
     )
-    speech_server._tts_tokenizer = lambda text, padding=False: {"input_ids": text.split()}
 
     # Skip TTS validation in tests (mock doesn't set up supported_speakers)
     speech_server._validate_tts_request = mocker.MagicMock(return_value=None)
@@ -379,37 +370,6 @@ class TestSpeechAPI:
         assert response.headers["content-type"] == "audio/wav"
         assert len(response.content) > 0
 
-    def test_create_speech_includes_token_usage_and_detail_headers(self, client):
-        payload = {
-            "input": "Hello world",
-            "model": "tts-model",
-            "voice": "alloy",
-            "response_format": "wav",
-        }
-        response = client.post("/v1/audio/speech", json=payload)
-        assert response.status_code == 200
-        assert response.headers["x-vllm-omni-input-tokens"] == "2"
-        assert response.headers["x-vllm-omni-output-tokens"] == "5"
-        assert response.headers["x-vllm-omni-total-tokens"] == "7"
-        assert response.headers["x-vllm-omni-input-text-tokens"] == "2"
-        assert response.headers["x-vllm-omni-input-audio-tokens"] == "0"
-
-    def test_build_speech_usage_headers_uses_usage_and_detail_field_names(self):
-        usage = SpeechTokenUsage(
-            input_tokens=7,
-            output_tokens=11,
-            total_tokens=18,
-            input_token_details=SpeechInputTokenDetails(text_tokens=3, audio_tokens=4),
-        )
-        headers = OmniOpenAIServingSpeech._build_speech_usage_headers(usage)
-        assert headers == {
-            "X-VLLM-OMNI-INPUT-TOKENS": "7",
-            "X-VLLM-OMNI-OUTPUT-TOKENS": "11",
-            "X-VLLM-OMNI-TOTAL-TOKENS": "18",
-            "X-VLLM-OMNI-INPUT-TEXT-TOKENS": "3",
-            "X-VLLM-OMNI-INPUT-AUDIO-TOKENS": "4",
-        }
-
     def test_create_speech_mp3_format(self, client):
         payload = {
             "input": "Hello world",
@@ -4059,7 +4019,9 @@ class TestTTSAsyncOffloading:
 
         assert adapter_model_type != legacy_tts_model_type
         assert any(
-            call.args and call.args[0] == "TTS speech request %s: model=%s" and call.args[2] == adapter_model_type
+            call.args
+            and call.args[0] == "TTS speech request %s: text=%r, model=%s"
+            and call.args[3] == adapter_model_type
             for call in log_info.call_args_list
         )
 

tests/entrypoints/openai_api/test_tts_adapter.py

@@ -22,10 +22,6 @@ from vllm_omni.entrypoints.openai.tts_adapters.indextts2 import (
     IndexTTS25Adapter,
     indextts2_conditioning_cache_salt,
 )
-from vllm_omni.entrypoints.openai.tts_adapters.moss_tts import (
-    MossTTSAdapter,
-    MossTTSNanoAdapter,
-)
 from vllm_omni.entrypoints.openai.tts_adapters.qwen3_tts import Qwen3TTSAdapter
 from vllm_omni.model_executor.models.indextts2 import prompt_utils
 from vllm_omni.model_executor.models.indextts2.tokenizer_v2_5 import (
@@ -98,20 +94,6 @@ def test_all_adapters_are_ar_or_diffusion():
         assert cls.backend in ("ar", "diffusion")
 
 
-@pytest.mark.parametrize("adapter_cls", [MossTTSAdapter, MossTTSNanoAdapter])
-def test_moss_tts_applies_request_max_new_tokens(adapter_cls):
-    adapter = adapter_cls(SimpleNamespace(server=object()))
-    stage_defaults = [SimpleNamespace(max_tokens=4096)]
-
-    overridden = adapter.apply_sampling_overrides(
-        stage_defaults,
-        SimpleNamespace(max_new_tokens=512),
-    )
-
-    assert overridden[0].max_tokens == 512
-    assert stage_defaults[0].max_tokens == 4096
-
-
 def test_qwen3_tts_metadata():
     assert Qwen3TTSAdapter.backend == "ar"
     assert issubclass(Qwen3TTSAdapter, ARTTSAdapter)

tests/entrypoints/openai_api/test_tts_detection.py

@@ -153,7 +153,6 @@ _PIPELINE_STAGES = [
     "indextts2_talker",
     "latent_generator",
     "llm",
-    "minimax_music3_ar",
     "ming_tts",
     "moss_tts",
     "moss_tts_codec",

tests/entrypoints/openai_api/test_video_api_utils.py

@@ -2,122 +2,16 @@
 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project
 """Unit tests for OpenAI-compatible video API encoding helpers."""
 
-import base64
-from io import BytesIO
-
-import httpx
 import numpy as np
 import pytest
 import torch
-from PIL import Image
-from vllm import envs
 
 from vllm_omni.diffusion.postprocess import rife_interpolator
 from vllm_omni.entrypoints.openai import video_api_utils
-from vllm_omni.entrypoints.openai.errors import InvalidInputReferenceError
 
 pytestmark = [pytest.mark.core_model, pytest.mark.cpu]
 
 
-def _png_bytes() -> bytes:
-    buffer = BytesIO()
-    Image.new("RGB", (2, 1), color=(12, 34, 56)).save(buffer, format="PNG")
-    return buffer.getvalue()
-
-
-def _install_http_transport(monkeypatch, handler, client_kwargs):
-    async_client = httpx.AsyncClient
-
-    def _client_factory(*args, **kwargs):
-        client_kwargs.append(kwargs.copy())
-        return async_client(*args, transport=httpx.MockTransport(handler), **kwargs)
-
-    monkeypatch.setattr(video_api_utils.httpx, "AsyncClient", _client_factory)
-
-
-@pytest.mark.asyncio
-async def test_decode_image_url_follows_redirects_when_allowed(monkeypatch):
-    requested_paths = []
-    client_kwargs = []
-
-    def _handler(request):
-        requested_paths.append(request.url.path)
-        if request.url.path == "/redirect.png":
-            return httpx.Response(302, headers={"location": "/image.png"})
-        return httpx.Response(200, content=_png_bytes(), headers={"content-type": "image/png"})
-
-    monkeypatch.setattr(envs, "VLLM_MEDIA_URL_ALLOW_REDIRECTS", True)
-    _install_http_transport(monkeypatch, _handler, client_kwargs)
-
-    image = await video_api_utils.decode_image_url("https://example.com/redirect.png")
-
-    assert image.size == (2, 1)
-    assert image.mode == "RGB"
-    assert requested_paths == ["/redirect.png", "/image.png"]
-    assert client_kwargs == [{"timeout": 60, "follow_redirects": True}]
-
-
-@pytest.mark.asyncio
-async def test_decode_image_url_rejects_redirects_when_disabled(monkeypatch):
-    requested_paths = []
-    client_kwargs = []
-
-    def _handler(request):
-        requested_paths.append(request.url.path)
-        return httpx.Response(302, headers={"location": "/image.png"})
-
-    monkeypatch.setattr(envs, "VLLM_MEDIA_URL_ALLOW_REDIRECTS", False)
-    _install_http_transport(monkeypatch, _handler, client_kwargs)
-
-    with pytest.raises(InvalidInputReferenceError, match="redirect.*VLLM_MEDIA_URL_ALLOW_REDIRECTS"):
-        await video_api_utils.decode_image_url("https://example.com/redirect.png")
-
-    assert requested_paths == ["/redirect.png"]
-    assert client_kwargs == [{"timeout": 60, "follow_redirects": False}]
-
-
-@pytest.mark.asyncio
-async def test_decode_image_url_reports_http_status(monkeypatch):
-    client_kwargs = []
-
-    def _handler(request):
-        return httpx.Response(404)
-
-    monkeypatch.setattr(envs, "VLLM_MEDIA_URL_ALLOW_REDIRECTS", True)
-    _install_http_transport(monkeypatch, _handler, client_kwargs)
-
-    with pytest.raises(InvalidInputReferenceError, match="server returned HTTP 404"):
-        await video_api_utils.decode_image_url("https://example.com/missing.png")
-
-
-@pytest.mark.asyncio
-async def test_decode_image_url_reports_connection_failure(monkeypatch):
-    client_kwargs = []
-
-    def _handler(request):
-        raise httpx.ConnectError("connection refused", request=request)
-
-    monkeypatch.setattr(envs, "VLLM_MEDIA_URL_ALLOW_REDIRECTS", True)
-    _install_http_transport(monkeypatch, _handler, client_kwargs)
-
-    with pytest.raises(InvalidInputReferenceError, match="failed to download image"):
-        await video_api_utils.decode_image_url("https://example.com/unreachable.png")
-
-
-@pytest.mark.asyncio
-async def test_decode_image_url_keeps_data_urls_local(monkeypatch):
-    def _unexpected_http_client(*args, **kwargs):
-        pytest.fail("data URLs must not create an HTTP client")
-
-    monkeypatch.setattr(video_api_utils.httpx, "AsyncClient", _unexpected_http_client)
-    encoded_image = base64.b64encode(_png_bytes()).decode()
-
-    image = await video_api_utils.decode_image_url(f"data:image/png;base64,{encoded_image}")
-
-    assert image.size == (2, 1)
-    assert image.mode == "RGB"
-
-
 def _install_fake_video_mux(monkeypatch, mux_calls):
     def _fake_mux_video_audio_bytes(frames, audio, fps, audio_sample_rate, video_codec_options=None):
         mux_calls.append(

tests/entrypoints/openai_api/test_video_server.py

@@ -16,17 +16,15 @@ from pathlib import Path
 from types import SimpleNamespace
 
 import av
-import httpx
 import numpy as np
 import pytest
 from fastapi import FastAPI, HTTPException
 from fastapi.testclient import TestClient
 from PIL import Image
 from pytest_mock import MockerFixture
-from vllm import envs
 
 from vllm_omni.diffusion.utils.media_utils import mux_video_audio_bytes
-from vllm_omni.entrypoints.openai import api_server, video_api_utils
+from vllm_omni.entrypoints.openai import api_server
 from vllm_omni.entrypoints.openai.api_server import router
 from vllm_omni.entrypoints.openai.protocol.videos import (
     VideoGenerationRequest,
@@ -601,45 +599,6 @@ def test_i2v_video_generation_with_image_reference_form(test_client, mocker: Moc
     assert input_image.size == (40, 24)
 
 
-def test_i2v_video_generation_follows_allowed_image_redirect(test_client, mocker: MockerFixture, monkeypatch):
-    requested_paths = []
-    async_client = httpx.AsyncClient
-
-    def _handler(request):
-        requested_paths.append(request.url.path)
-        if request.url.path == "/redirect.png":
-            return httpx.Response(302, headers={"location": "/image.png"})
-        return httpx.Response(200, content=_make_test_image_bytes((40, 24)))
-
-    def _client_factory(*args, **kwargs):
-        return async_client(*args, transport=httpx.MockTransport(_handler), **kwargs)
-
-    monkeypatch.setattr(envs, "VLLM_MEDIA_URL_ALLOW_REDIRECTS", True)
-    monkeypatch.setattr(video_api_utils.httpx, "AsyncClient", _client_factory)
-    mocker.patch(
-        "vllm_omni.entrypoints.openai.serving_video._encode_video_bytes",
-        return_value=b"fake-video",
-    )
-
-    response = test_client.post(
-        "/v1/videos",
-        data={
-            "prompt": "A fox running through snow.",
-            "image_reference": json.dumps({"image_url": "https://example.com/redirect.png"}),
-        },
-    )
-
-    assert response.status_code == 200
-    video_id = response.json()["id"]
-    _wait_for_status(test_client, video_id, VideoGenerationStatus.COMPLETED.value)
-    assert requested_paths == ["/redirect.png", "/image.png"]
-
-    engine = test_client.app.state.openai_serving_video._engine_client
-    input_image = engine.captured_prompt["multi_modal_data"]["image"]
-    assert isinstance(input_image, Image.Image)
-    assert input_image.size == (40, 24)
-
-
 def test_v2v_video_generation_form(test_client, mocker: MockerFixture):
     video_bytes = _make_test_video_bytes((32, 24), num_frames=3)
 
@@ -2168,31 +2127,6 @@ def test_sync_sampling_params_pass_through(test_client, mocker: MockerFixture):
     assert captured.quality == "high"
 
 
-def test_sync_sana_wm_extra_params_payload_passes_to_engine_prompt(test_client, mocker: MockerFixture):
-    _mock_encode_video_bytes(mocker)
-    response = test_client.post(
-        "/v1/videos/sync",
-        data={
-            "prompt": "drive forward",
-            "extra_params": json.dumps(
-                {
-                    "sana_wm": {"action": "d-4", "rotation_speed_deg": 1.5},
-                    "sana_wm_native_max_tokens": 30000,
-                }
-            ),
-        },
-    )
-
-    assert response.status_code == 200
-    engine = test_client.app.state.openai_serving_video._engine_client
-    captured_params = engine.captured_sampling_params_list[0]
-    # The camera block reaches the model through extra_args; the Sana preprocess
-    # hook is what lifts it onto the prompt.
-    assert captured_params.extra_args["sana_wm"]["action"] == "d-4"
-    assert captured_params.extra_args["sana_wm"]["rotation_speed_deg"] == 1.5
-    assert captured_params.extra_args["sana_wm_native_max_tokens"] == 30000
-
-
 def test_sync_frame_interpolation_params_pass_to_sampling_params(test_client, mocker: MockerFixture):
     """Frame interpolation parameters should be forwarded on the sync path."""
     encode_mock = _mock_encode_video_bytes(mocker)

tests/entrypoints/test_async_omni_diffusion_config.py

@@ -58,23 +58,6 @@ def test_default_stage_devices_from_sequence_parallel():
     assert stage_cfg["runtime"]["devices"] == "0,1,2,3"
 
 
-def test_default_stage_devices_and_dp_from_num_gpus():
-    """Resolve omitted DP before deriving devices from an explicit GPU count."""
-    stage_cfg = AsyncOmniEngine._create_default_diffusion_stage_cfg(
-        {
-            "num_gpus": 8,
-            "tensor_parallel_size": 2,
-            "ulysses_degree": 2,
-        }
-    )[0]
-
-    parallel_config = stage_cfg["engine_args"]["parallel_config"]
-    assert parallel_config.data_parallel_size == 2
-    assert parallel_config.world_size == 8
-    assert stage_cfg["engine_args"]["num_gpus"] == 8
-    assert stage_cfg["runtime"]["devices"] == "0,1,2,3,4,5,6,7"
-
-
 def test_default_stage_config_uses_parallel_size_kwargs():
     """Ensure default diffusion parallel_config uses CLI/API parallel sizes."""
     stage_cfg = AsyncOmniEngine._create_default_diffusion_stage_cfg(
@@ -93,8 +76,8 @@ def test_default_stage_config_uses_parallel_size_kwargs():
     assert parallel_config.enable_expert_parallel is True
 
 
-def test_default_stage_config_preserves_omitted_dp_for_runtime_inference():
-    """Keep omitted DP unresolved until the runtime WORLD size is known."""
+def test_default_stage_config_defaults_nullified_parallel_size_kwargs():
+    """Ensure nullified diffusion parallel-size kwargs fall back to defaults."""
     stage_cfg = AsyncOmniEngine._create_default_diffusion_stage_cfg(
         {
             "pipeline_parallel_size": None,
@@ -109,7 +92,7 @@ def test_default_stage_config_preserves_omitted_dp_for_runtime_inference():
 
     parallel_config = stage_cfg["engine_args"]["parallel_config"]
     assert parallel_config.pipeline_parallel_size == 1
-    assert parallel_config.data_parallel_size is None
+    assert parallel_config.data_parallel_size == 1
     assert parallel_config.tensor_parallel_size == 1
     assert parallel_config.enable_expert_parallel is False
     assert stage_cfg["engine_args"]["enforce_eager"] is False