update news

Signed-off-by: Roger Wang <hey@rogerw.io>

.buildkite/pipeline.yml

@@ -29,57 +29,7 @@ steps:
     depends_on: image-build
     commands:
       - pytest -s -v tests/e2e/offline_inference/test_t2i_model.py
-    agents:
-      queue: "gpu_1_queue" # g6.4xlarge instance on AWS, has 1 L4 GPU
-    plugins:
-      - docker#v5.2.0:
-          image: public.ecr.aws/q9t5s3a7/vllm-ci-test-repo:$BUILDKITE_COMMIT
-          always-pull: true
-          propagate-environment: true
-          environment:
-            - "HF_HOME=/fsx/hf_cache"
-          volumes:
-            - "/fsx/hf_cache:/fsx/hf_cache"
-
-  - label: "Diffusion Images API LoRA E2E"
-    timeout_in_minutes: 20
-    depends_on: image-build
-    commands:
-      - pytest -s -v tests/e2e/online_serving/test_images_generations_lora.py
-    agents:
-      queue: "gpu_1_queue" # g6.4xlarge instance on AWS, has 1 L4 GPU
-    plugins:
-      - docker#v5.2.0:
-          image: public.ecr.aws/q9t5s3a7/vllm-ci-test-repo:$BUILDKITE_COMMIT
-          always-pull: true
-          propagate-environment: true
-          environment:
-            - "HF_HOME=/fsx/hf_cache"
-          volumes:
-            - "/fsx/hf_cache:/fsx/hf_cache"
-
-  - label: "Diffusion Model CPU offloading Test"
-    timeout_in_minutes: 20
-    depends_on: image-build
-    commands:
-      - pytest -s -v tests/e2e/offline_inference/test_diffusion_cpu_offload.py
-    agents:
-      queue: "gpu_1_queue" # g6.4xlarge instance on AWS, has 1 L4 GPU
-    plugins:
-      - docker#v5.2.0:
-          image: public.ecr.aws/q9t5s3a7/vllm-ci-test-repo:$BUILDKITE_COMMIT
-          always-pull: true
-          propagate-environment: true
-          environment:
-            - "HF_HOME=/fsx/hf_cache"
-          volumes:
-            - "/fsx/hf_cache:/fsx/hf_cache"
-
-  - label: "Audio Generation Model Test"
-    timeout_in_minutes: 20
-    depends_on: image-build
-    commands:
-      - pytest -s -v tests/e2e/offline_inference/test_stable_audio_model.py
+      - pytest -s -v tests/e2e/offline_inference/test_ovis_image.py
     agents:
       queue: "gpu_1_queue" # g6.4xlarge instance on AWS, has 1 L4 GPU
     plugins:
@@ -109,8 +59,8 @@ steps:
           volumes:
             - "/fsx/hf_cache:/fsx/hf_cache"
 
-  - label: "Diffusion Sequence Parallelism Test"
-    timeout_in_minutes: 20
+  - label: "Diffusion Parallelism Test"
+    timeout_in_minutes: 15
     depends_on: image-build
     commands:
       - pytest -s -v tests/e2e/offline_inference/test_sequence_parallel.py
@@ -127,29 +77,11 @@ steps:
           volumes:
             - "/fsx/hf_cache:/fsx/hf_cache"
 
-  - label: "Diffusion Tensor Parallelism Test"
-    timeout_in_minutes: 20
-    depends_on: image-build
-    commands:
-      - pytest -s -v tests/e2e/offline_inference/test_zimage_tensor_parallel.py
-    agents:
-      queue: "gpu_4_queue" # g6.12xlarge instance on AWS, has 4 L4 GPU
-    plugins:
-      - docker#v5.2.0:
-          image: public.ecr.aws/q9t5s3a7/vllm-ci-test-repo:$BUILDKITE_COMMIT
-          always-pull: true
-          propagate-environment: true
-          shm-size: "8gb"
-          environment:
-            - "HF_HOME=/fsx/hf_cache"
-          volumes:
-            - "/fsx/hf_cache:/fsx/hf_cache"
-
   - label: "Diffusion GPU Worker Test"
     timeout_in_minutes: 20
     depends_on: image-build
     commands:
-      - pytest -s -v tests/diffusion/test_gpu_diffusion_worker.py
+      - pytest -s -v tests/diffusion/test_gpu_worker.py
     agents:
       queue: "gpu_4_queue" # g6.12xlarge instance on AWS, has 4 L4 GPU
     plugins:
@@ -187,10 +119,7 @@ steps:
     depends_on: image-build
     commands:
       - export VLLM_WORKER_MULTIPROC_METHOD=spawn
-      - export VLLM_TEST_CLEAN_GPU_MEMORY="1"
-      - pytest -s -v tests/e2e/offline_inference/test_qwen3_omni.py
-      - pytest -s -v tests/e2e/online_serving/test_qwen3_omni.py
-      - pytest -s -v tests/e2e/online_serving/test_async_omni.py
+      - pytest -s -v tests/e2e/offline_inference/test_qwen3_omni.py tests/e2e/online_serving/test_qwen3_omni.py tests/e2e/online_serving/test_async_omni.py
     agents:
       queue: "mithril-h100-pool"
     plugins:

.buildkite/scripts/docker_login_ecr_public.sh

@@ -1,73 +0,0 @@
-#!/bin/bash
-# Helper function to safely login to ECR Public with per-job config isolation
-# Uses DOCKER_CONFIG environment variable to prevent race conditions
-#
-# This script prevents the "device or resource busy" error by giving each
-# Buildkite job its own isolated Docker config directory.
-#
-# Usage:
-#   source docker_login_ecr_public.sh && safe_docker_login_ecr_public
-
-set -euo pipefail
-
-# Configuration
-ECR_REGISTRY="public.ecr.aws"
-
-setup_isolated_docker_config() {
-    # Use BUILDKITE_JOB_ID for job-specific isolation
-    # Fallback to PID if running outside Buildkite
-    local job_id="${BUILDKITE_JOB_ID:-$$}"
-
-    # Set Docker config to job-specific directory
-    export DOCKER_CONFIG="/tmp/docker-config-${job_id}"
-
-    # Create directory if it doesn't exist
-    mkdir -p "$DOCKER_CONFIG"
-
-    echo "[docker-config] Using isolated Docker config: $DOCKER_CONFIG"
-}
-
-check_docker_auth() {
-    # Check if already authenticated to the given registry
-    # Returns 0 if authenticated, 1 if not
-    local registry="$1"
-
-    # Check if credentials exist in the isolated config
-    if [[ -f "$DOCKER_CONFIG/config.json" ]]; then
-        # Check if registry is present in config
-        if grep -q "$registry" "$DOCKER_CONFIG/config.json" 2>/dev/null; then
-            return 0
-        fi
-    fi
-
-    return 1
-}
-
-safe_docker_login_ecr_public() {
-    # Setup isolated config first
-    setup_isolated_docker_config
-
-    local registry="$ECR_REGISTRY"
-
-    # Check if already authenticated (within this job)
-    if check_docker_auth "$registry"; then
-        echo "[docker-login] Already authenticated to $registry in this job"
-        return 0
-    fi
-
-    # Perform login to isolated config directory
-    echo "[docker-login] Logging in to $ECR_REGISTRY (isolated config)..."
-    if aws ecr-public get-login-password --region us-east-1 | docker login --username AWS --password-stdin "$ECR_REGISTRY"; then
-        echo "[docker-login] Login successful (config: $DOCKER_CONFIG)"
-        return 0
-    else
-        local exit_code=$?
-        echo "[docker-login] ERROR: Login failed with exit code $exit_code" >&2
-        return $exit_code
-    fi
-}
-
-# Execute if run as script (not sourced)
-if [[ "${BASH_SOURCE[0]}" == "${0}" ]]; then
-    safe_docker_login_ecr_public
-fi

.buildkite/scripts/hardware_ci/run-amd-test.sh

@@ -66,14 +66,6 @@ done
 echo "--- Pulling container"
 image_name="public.ecr.aws/q9t5s3a7/vllm-ci-test-repo:${BUILDKITE_COMMIT}-rocm-omni"
 container_name="rocm_${BUILDKITE_COMMIT}_$(tr -dc A-Za-z0-9 < /dev/urandom | head -c 10; echo)"
-
-# Install AWS CLI to authenticate to ECR Public Gallery to get higher rate limit for pulling images
-sudo apt-get update && sudo apt-get install -y awscli
-# Use safe docker login helper to prevent race conditions
-source "$(dirname "${BASH_SOURCE[0]}")/../docker_login_ecr_public.sh"
-safe_docker_login_ecr_public
-# Pull the container from ECR Public Gallery
-
 docker pull "${image_name}"
 
 remove_docker_container() {

.buildkite/scripts/simple_test.sh

@@ -47,12 +47,8 @@ fi
 VENV_PYTHON="${VENV_DIR}/bin/python"
 [[ -x "${VENV_PYTHON}" ]] || { echo "Python not found in ${VENV_DIR}"; exit 1; }
 
-"${UV_BIN}" pip install --python "${VENV_PYTHON}" vllm==0.14.0
+"${UV_BIN}" pip install --python "${VENV_PYTHON}" vllm==0.12.0
 "${UV_BIN}" pip install --python "${VENV_PYTHON}" -e ".[dev]"
-
 "${VENV_PYTHON}" -m pytest -v -s tests/entrypoints/
 "${VENV_PYTHON}" -m pytest -v -s tests/diffusion/cache/
-"${VENV_PYTHON}" -m pytest -v -s tests/diffusion/lora/
 "${VENV_PYTHON}" -m pytest -v -s tests/model_executor/models/qwen2_5_omni/test_audio_length.py
-"${VENV_PYTHON}" -m pytest -v -s tests/worker/
-"${VENV_PYTHON}" -m pytest -v -s tests/distributed/omni_connectors/test_kv_flow.py

.buildkite/test-amd.yaml

@@ -1,10 +1,10 @@
 steps:
 
 - label: "Diffusion Model Test"
-  timeout_in_minutes: 20
+  timeout_in_minutes: 15
   agent_pool: mi325_2
   depends_on: amd-build
-  mirror_hardwares: [amdproduction]
+  mirror_hardwares: [amdexperimental, amdproduction, amdtentative]
   grade: Blocking
   commands:
     - export GPU_ARCHS=gfx942
@@ -20,7 +20,7 @@ steps:
   timeout_in_minutes: 15
   agent_pool: mi325_1
   depends_on: amd-build
-  mirror_hardwares: [amdproduction]
+  mirror_hardwares: [amdexperimental, amdproduction, amdtentative]
   grade: Blocking
   commands:
     - export GPU_ARCHS=gfx942
@@ -34,33 +34,11 @@ steps:
     - export VLLM_ROCM_USE_AITER_RMSNORM=0
     - pytest -s -v tests/e2e/offline_inference/test_cache_dit.py tests/e2e/offline_inference/test_teacache.py
 
-- label: "Diffusion Parallelism Test"
-  timeout_in_minutes: 15
-  agent_pool: mi325_2
-  depends_on: amd-build
-  mirror_hardwares: [amdproduction]
-  grade: Blocking
-  commands:
-    - export MIOPEN_DEBUG_CONV_DIRECT=0
-    - export MIOPEN_DEBUG_CONV_GEMM=0
-    - pytest -s -v tests/e2e/offline_inference/test_sequence_parallel.py
-
-- label: "Diffusion GPU Worker Test"
-  timeout_in_minutes: 20
-  agent_pool: mi325_2
-  depends_on: amd-build
-  mirror_hardwares: [amdproduction]
-  grade: Blocking
-  commands:
-    - export MIOPEN_DEBUG_CONV_DIRECT=0
-    - export MIOPEN_DEBUG_CONV_GEMM=0
-    - pytest -s -v tests/diffusion/test_gpu_diffusion_worker.py
-
 - label: "Omni Model Test Qwen2-5-Omni"
   timeout_in_minutes: 15
   agent_pool: mi325_2
   depends_on: amd-build
-  mirror_hardwares: [amdproduction]
+  mirror_hardwares: [amdexperimental, amdproduction, amdtentative]
   grade: Blocking
   commands:
     - export GPU_ARCHS=gfx942
@@ -73,34 +51,3 @@ steps:
     - export VLLM_ROCM_USE_AITER_LINEAR=0
     - export VLLM_ROCM_USE_AITER_RMSNORM=0
     - pytest -s -v tests/e2e/offline_inference/test_qwen2_5_omni.py
-
-- label: "Omni Model Test Qwen3-Omni"
-  timeout_in_minutes: 15
-  agent_pool: mi325_2
-  depends_on: amd-build
-  mirror_hardwares: [amdproduction]
-  grade: Blocking
-  commands:
-    - export VLLM_LOGGING_LEVEL=DEBUG
-    - export VLLM_WORKER_MULTIPROC_METHOD=spawn
-    - export MIOPEN_DEBUG_CONV_DIRECT=0
-    - export MIOPEN_DEBUG_CONV_GEMM=0
-    - pytest -s -v tests/e2e/offline_inference/test_qwen3_omni.py tests/e2e/online_serving/test_qwen3_omni.py
-
-- label: "Diffusion Image Edit Test"
-  timeout_in_minutes: 15
-  agent_pool: mi325_1
-  depends_on: amd-build
-  mirror_hardwares: [amdproduction]
-  grade: Blocking
-  commands:
-    - export GPU_ARCHS=gfx942
-    - export VLLM_LOGGING_LEVEL=DEBUG
-    - export VLLM_WORKER_MULTIPROC_METHOD=spawn
-    - export MIOPEN_DEBUG_CONV_DIRECT=0
-    - export MIOPEN_DEBUG_CONV_GEMM=0
-    - export VLLM_ROCM_USE_AITER=1
-    - export VLLM_ROCM_USE_AITER_MHA=1
-    - export VLLM_ROCM_USE_AITER_LINEAR=0
-    - export VLLM_ROCM_USE_AITER_RMSNORM=0
-    - pytest -s -v tests/e2e/online_serving/test_i2i_multi_image_input.py

.buildkite/test-template-amd-omni.j2

@@ -13,8 +13,8 @@
         depends_on: ~
         soft_fail: false
         commands:
-          - "source .buildkite/scripts/docker_login_ecr_public.sh && safe_docker_login_ecr_public"
-          - "docker build -f docker/Dockerfile.rocm -t {{ docker_image_amd }} --target final --progress plain ."
+          - "aws ecr-public get-login-password --region us-east-1 | docker login --username AWS --password-stdin public.ecr.aws/q9t5s3a7"
+          - "docker build -f docker/Dockerfile.rocm -t {{ docker_image_amd }} --progress plain ."
           - "docker push {{ docker_image_amd }}"
         key: "amd-build"
         env:

.github/ISSUE_TEMPLATE/400-bug-report.yml

@@ -40,31 +40,6 @@ body:
       </details>
   validations:
     required: true
-
-- type: textarea
-  attributes:
-    label: Your code version
-    description: |
-      Please provide a description of which commit id or version the code is.
-
-    value: |
-      <details>
-      <summary>The commit id or version of vllm</summary>
-
-      ```text
-
-      ```
-      </details>
-      <details>
-      <summary>The commit id or version of vllm-omni</summary>
-
-      ```text
-
-      ```
-      </details>
-  validations:
-    required: true
-
 - type: textarea
   attributes:
     label: 🐛 Describe the bug

.github/ISSUE_TEMPLATE/750-RFC.yml

@@ -20,8 +20,6 @@ body:
     label: Proposed Change.
     description: >
       The proposed change of the RFC.
-    value: |
-      Please provide the detailed design document of the RFC using the [template](https://docs.google.com/document/d/12YxSsVeD1jvL-InClkeAEnZyWFDndz_65JmXvsamuV4/edit?tab=t.0#heading=h.4ef4szrsgspp).
   validations:
     required: true
 - type: textarea

.pre-commit-config.yaml

@@ -29,10 +29,7 @@ repos:
         # only for staged files
 
   - repo: https://github.com/rhysd/actionlint
-    # v1.7.8+ sets `go 1.24.0` in go.mod, which older Go toolchains (and most
-    # current CI images) cannot parse. Pin to v1.7.7 until actionlint fixes the
-    # go.mod directive.
-    rev: v1.7.7
+    rev: v1.7.9
     hooks:
       - id: actionlint
         files: ^\.github/workflows/.*\.ya?ml$

README.md

@@ -9,13 +9,14 @@ Easy, fast, and cheap omni-modality model serving for everyone
 </h3>
 
 <p align="center">
-| <a href="https://vllm-omni.readthedocs.io/en/latest/"><b>Documentation</b></a> | <a href="https://discuss.vllm.ai"><b>User Forum</b></a> | <a href="https://slack.vllm.ai"><b>Developer Slack</b></a> | <a href="docs/assets/WeChat.jpg"><b>WeChat</b></a> |
+| <a href="https://vllm-omni.readthedocs.io/en/latest/"><b>Documentation</b></a> | <a href="https://discuss.vllm.ai"><b>User Forum</b></a> | <a href="https://slack.vllm.ai"><b>Developer Slack</b></a> |
 </p>
 
 ---
 
 *Latest News* 🔥
-- [2026/01] We released [0.12.0rc1](https://github.com/vllm-project/vllm-omni/releases/tag/v0.12.0rc1) - a major RC milestone focused on maturing the diffusion stack, strengthening OpenAI-compatible serving, expanding omni-model coverage, and improving stability across platforms (GPU/NPU/ROCm), please check our latest [design](https://docs.google.com/presentation/d/1qv4qMW1rKAqDREMXiUDLIgqqHQe7TDPj/edit?usp=sharing&ouid=110473603432222024453&rtpof=true&sd=true).
+
+- [2026/01] We released [0.12.0rc1](https://github.com/vllm-project/vllm-omni/releases/tag/v0.12.0rc1) - a major RC milestone focused on maturing the diffusion stack, strengthening OpenAI-compatible serving, expanding omni-model coverage, and improving stability across platforms (GPU/NPU/ROCm).
 - [2025/11] vLLM community officially released [vllm-project/vllm-omni](https://github.com/vllm-project/vllm-omni) in order to support omni-modality models serving.
 
 ---
@@ -69,10 +70,6 @@ Please check out [Contributing to vLLM-Omni](https://vllm-omni.readthedocs.io/en
 ## Join the Community
 Feel free to ask questions, provide feedbacks and discuss with fellow users of vLLM-Omni in `#sig-omni` slack channel at [slack.vllm.ai](https://slack.vllm.ai) or vLLM user forum at [discuss.vllm.ai](https://discuss.vllm.ai).
 
-## Star History
-
-[![Star History Chart](https://api.star-history.com/svg?repos=vllm-project/vllm-omni&type=date&legend=top-left)](https://www.star-history.com/#vllm-project/vllm-omni&type=date&legend=top-left)
-
 ## License
 
 Apache License 2.0, as found in the [LICENSE](./LICENSE) file.

benchmarks/diffusion/README.md

@@ -1,117 +0,0 @@
-
-# Diffusion Serving Benchmark (Image/Video)
-
-This folder contains an online-serving benchmark script for diffusion models.
-It sends requests to a vLLM OpenAI-compatible endpoint and reports throughput,
-latency percentiles, and optional SLO attainment.
-
-The main entrypoint is:
-
-- `benchmarks/diffusion/diffusion_benchmark_serving.py`
-
-## 1. Quick Start
-
-1. Start the server:
-
-```bash
-vllm serve Qwen/Qwen-Image --omni --port 8099
-```
-
-2. Run a minimal benchmark:
-
-```bash
-python3 benchmarks/diffusion/diffusion_benchmark_serving.py \
-	--base-url http://localhost:8099 \
-	--model Qwen/Qwen-Image \
-	--task t2i \
-	--dataset vbench \
-	--num-prompts 5
-```
-
-**Notes**
-
-- The benchmark talks to `http://<host>:<port>/v1/chat/completions`.
-- If you run the server on another host or port, pass `--base-url` accordingly.
-
-## 2. Supported Datasets
-
-The benchmark supports three dataset modes via `--dataset`:
-
-- `vbench`: Built-in prompt/data loader.
-- `trace`: Heterogeneous request traces (each request can have different resolution/frames/steps).
-- `random`: Synthetic prompts for quick smoke tests.
-
-### VBench dataset
-
-If you use i2v/i2i bench datasets and need auto-download support, you may need:
-
-```bash
-uv pip install gdown
-```
-
-### Trace dataset
-
-Use `--dataset trace` to replay a trace file. The trace can specify per-request fields such as:
-
-- `width`, `height`
-- `num_frames` (video)
-- `num_inference_steps`
-- `seed`, `fps`
-- optional `slo_ms` (per-request SLO target)
-
-By default (when `--dataset-path` is not provided), the script downloads a default trace from
-the HuggingFace dataset repo `asukaqaqzz/Dit_Trace`. The default filename can depend on `--task`
-(e.g., `t2v` uses a video trace).
-
-Current defaults:
-
-- `--task t2i` -> `sd3_trace.txt`
-- `--task t2v` -> `cogvideox_trace.txt`
-
-You can point to your own trace using `--dataset-path`.
-
-## 3. Benchmark Parameters
-
-### Basic flags
-
-- `--base-url`: Server address (the script calls `.../v1/chat/completions`).
-- `--model`: The OpenAI-compatible `model` field.
-- `--task`: Task type (e.g., `t2i`, `t2v`, `i2i`, `i2v`).
-- `--dataset`: Dataset mode (`vbench` / `trace` / `random`).
-- `--num-prompts`: Number of requests to send.
-
-Common optional flags:
-
-- `--output-file`: Write metrics to a JSON file.
-- `--disable-tqdm`: Disable the progress bar.
-
-### Resolution / frames / steps: CLI defaults vs dataset fields
-
-Related flags: `--width`, `--height`, `--num-frames`, `--fps`, `--num-inference-steps`.
-
-- For `vbench` / `random`: these CLI flags act as global defaults for all generated requests.
-- For `trace`: each request can carry its own fields (e.g., `width/height/num_frames/num_inference_steps`).
-
-Precedence rules for `trace` (i.e., what actually gets sent):
-
-- `width/height`: if either `--width` or `--height` is explicitly set, it overrides per-request values from the trace; otherwise per-request values are used when present.
-- `num_frames`: per-request `num_frames` takes precedence; otherwise fall back to `--num-frames`.
-- `num_inference_steps`: per-request `num_inference_steps` takes precedence; otherwise fall back to `--num-inference-steps`.
-
-### SLO, warmup, and max concurrency
-
-Enable SLO evaluation with `--slo`.
-
-- If a request in the trace already has `slo_ms`, that value is used.
-- Otherwise, the script runs warmup requests to infer a base unit time, estimates `expected_ms` by linearly scaling with area/frames/steps, and then sets `slo_ms = expected_ms * --slo-scale`.
-
-Warmup flags:
-
-- `--warmup-requests`: Number of warmup requests.
-- `--warmup-num-inference-steps`: Steps used during warmup.
-- For `--task t2v`: warmup requests are forced to use `num_frames=1` to make warmup faster and less noisy.
-
-Traffic / concurrency flags:
-
-- `--request-rate`: Target request rate (requests/second). If set to `inf`, the script sends all requests immediately.
-- `--max-concurrency`: Max number of in-flight requests (default: `1`). This can hard-cap the achieved QPS: if it is too small, requests will queue behind the semaphore, and both achieved throughput and observed SLO attainment can be skewed.

benchmarks/diffusion_benchmark_serving.py

@@ -9,29 +9,28 @@ If you want to use i2v, i2i dataset, you should `uv pip install gdown` first
 Usage:
     # Video
     t2v:
-    python3 benchmarks/diffusion/diffusion_benchmark_serving.py \
+    python3 benchmarks/diffusion_benchmark_serving.py \
         --dataset vbench --task t2v --num-prompts 10 \
         --height 480 --width 640 --fps 16 --num-frames 80
 
     i2v:
-    python3 benchmarks/diffusion/diffusion_benchmark_serving.py \
+    python3 benchmarks/diffusion_benchmark_serving.py \
         --dataset vbench --task i2v --num-prompts 10
 
 
     # Image
     t2i:
-    python3 benchmarks/diffusion/diffusion_benchmark_serving.py \
+    python3 benchmarks/diffusion_benchmark_serving.py \
         --dataset vbench --task t2i --num-prompts 10 \
         --height 1024 --width 1024
 
     i2i:
-    python3 benchmarks/diffusion/diffusion_benchmark_serving.py \
+    python3 benchmarks/diffusion_benchmark_serving.py \
         --dataset vbench --task i2i --num-prompts 10
 
 """
 
 import argparse
-import ast
 import asyncio
 import base64
 import glob
@@ -41,7 +40,6 @@ import os
 import time
 import uuid
 from abc import ABC, abstractmethod
-from collections.abc import AsyncGenerator
 from dataclasses import dataclass, field, replace
 from typing import Any
 
@@ -62,8 +60,6 @@ class RequestFuncInput:
     num_inference_steps: int | None = None
     seed: int | None = None
     fps: int | None = None
-    timestamp: float | None = None
-    slo_ms: float | None = None
     extra_body: dict[str, Any] = field(default_factory=dict)
     image_paths: list[str] | None = None
     request_id: str = field(default_factory=lambda: str(uuid.uuid4()))
@@ -77,7 +73,6 @@ class RequestFuncOutput:
     start_time: float = 0.0
     response_body: dict[str, Any] = field(default_factory=dict)
     peak_memory_mb: float = 0.0
-    slo_achieved: bool | None = None
 
 
 class BaseDataset(ABC):
@@ -315,226 +310,6 @@ class VBenchDataset(BaseDataset):
         return [self[i] for i in range(len(self))]
 
 
-class TraceDataset(BaseDataset):
-    """Trace-based dataset loader for heterogeneous diffusion requests."""
-
-    DEFAULT_REPO_ID = "asukaqaqzz/Dit_Trace"
-    DEFAULT_FILENAME = "sd3_trace.txt"
-    DEFAULT_FILENAME_BY_TASK: dict[str, str] = {
-        # Text-to-image traces (e.g., SD3)
-        "t2i": "sd3_trace.txt",
-        # Text-to-video traces (e.g., CogVideoX)
-        "t2v": "cogvideox_trace.txt",
-    }
-
-    def __init__(self, args, api_url: str, model: str):
-        super().__init__(args, api_url, model)
-        self.cache_dir = os.path.join(os.path.expanduser("~"), ".cache", "vllm-omni", "trace")
-        self.default_filename = self.DEFAULT_FILENAME_BY_TASK.get(getattr(args, "task", ""), self.DEFAULT_FILENAME)
-        dataset_root = args.dataset_path
-        if not dataset_root:
-            dataset_root = self._download_default_trace()
-        self.items = self._load_items(dataset_root)
-
-    @staticmethod
-    def _coerce_int(x: Any) -> int | None:
-        if x is None:
-            return None
-        if isinstance(x, bool):
-            return None
-        if isinstance(x, int):
-            return x
-        try:
-            s = str(x).strip()
-            if not s:
-                return None
-            return int(float(s))
-        except Exception:
-            return None
-
-    @staticmethod
-    def _coerce_float(x: Any) -> float | None:
-        if x is None:
-            return None
-        if isinstance(x, float):
-            return x
-        if isinstance(x, int):
-            return float(x)
-        try:
-            s = str(x).strip()
-            if not s:
-                return None
-            return float(s)
-        except Exception:
-            return None
-
-    def _download_default_trace(self) -> str:
-        """Download default trace file from HuggingFace Hub if not provided."""
-
-        try:
-            from huggingface_hub import hf_hub_download
-        except ImportError as exc:
-            raise ImportError(
-                "huggingface_hub is required to download the default trace dataset. "
-                "Install via `pip install huggingface_hub`."
-            ) from exc
-
-        os.makedirs(self.cache_dir, exist_ok=True)
-        return hf_hub_download(
-            repo_id=self.DEFAULT_REPO_ID,
-            filename=self.default_filename,
-            repo_type="dataset",
-            local_dir=self.cache_dir,
-            local_dir_use_symlinks=False,
-        )
-
-    def _expand_paths(self, dataset_path: str | None) -> list[str]:
-        if not dataset_path:
-            return []
-
-        parts = [p.strip() for p in str(dataset_path).split(",") if p.strip()]
-        paths: list[str] = []
-        for p in parts:
-            if any(ch in p for ch in ["*", "?", "["]):
-                paths.extend(sorted(glob.glob(p)))
-            elif os.path.isdir(p):
-                paths.extend(sorted(glob.glob(os.path.join(p, "**", "*.txt"), recursive=True)))
-            else:
-                paths.append(p)
-
-        seen = set()
-        unique_paths = []
-        for p in paths:
-            if p not in seen:
-                seen.add(p)
-                unique_paths.append(p)
-        return unique_paths
-
-    def _parse_trace_file(self, path: str) -> list[dict[str, Any]]:
-        rows: list[dict[str, Any]] = []
-
-        def parse_request_repr_line(line: str) -> dict[str, Any] | None:
-            text = line.strip()
-            if not text:
-                return None
-            if not (text.startswith("Request(") and text.endswith(")")):
-                return None
-            inner = text[len("Request(") : -1]
-            try:
-                expr = ast.parse(f"f({inner})", mode="eval")
-                if not isinstance(expr.body, ast.Call):
-                    return None
-                call = expr.body
-                out: dict[str, Any] = {}
-                for kw in call.keywords:
-                    if kw.arg is None:
-                        continue
-                    out[kw.arg] = ast.literal_eval(kw.value)
-                return out
-            except Exception:
-                return None
-
-        # detect first non-empty line to pick parser
-        first_non_empty = None
-        with open(path, encoding="utf-8") as f:
-            for _ in range(50):
-                pos = f.tell()
-                line = f.readline()
-                if not line:
-                    break
-                if line.strip():
-                    first_non_empty = line.strip()
-                    f.seek(pos)
-                    break
-
-        if first_non_empty is None:
-            return rows
-
-        if first_non_empty.startswith("Request("):
-            with open(path, encoding="utf-8") as f:
-                for line in f:
-                    parsed = parse_request_repr_line(line)
-                    if isinstance(parsed, dict):
-                        rows.append(parsed)
-            return rows
-
-        # txt fallback: parse Request(...) lines only
-        with open(path, encoding="utf-8") as f:
-            for line in f:
-                parsed = parse_request_repr_line(line)
-                if isinstance(parsed, dict):
-                    rows.append(parsed)
-        return rows
-
-    def _load_items(self, dataset_root: str) -> list[dict[str, Any]]:
-        paths = self._expand_paths(dataset_root)
-        if not paths:
-            raise ValueError("No trace files found. Provide --dataset-path or rely on default HuggingFace download.")
-
-        items: list[dict[str, Any]] = []
-        for p in paths:
-            if not os.path.exists(p):
-                continue
-            for row in self._parse_trace_file(p):
-                if isinstance(row, dict):
-                    row = dict(row)
-                    row.setdefault("_source", p)
-                    items.append(row)
-
-        if not items:
-            raise ValueError("Trace dataset is empty after parsing provided paths.")
-
-        if self.args.num_prompts is not None:
-            items = items[: self.args.num_prompts]
-
-        return items
-
-    def __len__(self) -> int:
-        return len(self.items)
-
-    def __getitem__(self, idx: int) -> RequestFuncInput:
-        row = self.items[idx]
-        prompt = row.get("prompt") or row.get("text") or ""
-
-        row_height = self._coerce_int(row.get("height"))
-        row_width = self._coerce_int(row.get("width"))
-        num_frames = self._coerce_int(row.get("num_frames"))
-        num_steps = self._coerce_int(row.get("num_inference_steps"))
-        seed = self._coerce_int(row.get("seed"))
-        fps = self._coerce_int(row.get("fps"))
-        timestamp = self._coerce_float(row.get("timestamp"))
-        slo_ms = self._coerce_float(row.get("slo_ms"))
-        image_paths = row.get("image_paths")
-
-        override_w = self.args.width
-        override_h = self.args.height
-        if override_w is not None or override_h is not None:
-            width = override_w
-            height = override_h
-        else:
-            width = row_width
-            height = row_height
-
-        return RequestFuncInput(
-            prompt=str(prompt),
-            api_url=self.api_url,
-            model=self.model,
-            width=width,
-            height=height,
-            num_frames=num_frames if num_frames is not None else self.args.num_frames,
-            num_inference_steps=num_steps if num_steps is not None else self.args.num_inference_steps,
-            seed=seed if seed is not None else self.args.seed,
-            fps=fps if fps is not None else self.args.fps,
-            timestamp=timestamp,
-            slo_ms=slo_ms,
-            image_paths=image_paths,
-            request_id=str(row.get("request_id")) if row.get("request_id") is not None else str(uuid.uuid4()),
-        )
-
-    def get_requests(self) -> list[RequestFuncInput]:
-        return [self[i] for i in range(len(self))]
-
-
 class RandomDataset(BaseDataset):
     def __init__(self, args, api_url: str, model: str):
         self.args = args
@@ -562,128 +337,6 @@ class RandomDataset(BaseDataset):
         return [self[i] for i in range(len(self))]
 
 
-def _compute_expected_latency_ms_from_base(req: RequestFuncInput, args, base_time_ms: float | None) -> float | None:
-    """Compute expected execution time (ms) based on a base per-step-per-frame unit time.
-
-    Assumes linear scaling with pixel area, frame count, and num_inference_steps.
-    The base unit represents latency for a 16x16 resolution, single frame, single step.
-    """
-
-    if base_time_ms is None:
-        return None
-
-    width = req.width if req.width is not None else args.width
-    height = req.height if req.height is not None else args.height
-    if width is None or height is None:
-        return None
-
-    frames = req.num_frames if req.num_frames is not None else args.num_frames
-    steps = req.num_inference_steps if req.num_inference_steps is not None else args.num_inference_steps
-
-    frame_scale = frames if isinstance(frames, int) and frames > 0 else 1
-    step_scale = steps if isinstance(steps, int) and steps > 0 else 1
-
-    area_units = max((float(width) * float(height)) / float(16 * 16), 1.0)
-    return float(base_time_ms) * area_units * frame_scale * step_scale
-
-
-def _infer_slo_base_time_ms_from_warmups(
-    warmup_pairs: list[tuple[RequestFuncInput, RequestFuncOutput]],
-    args,
-) -> float | None:
-    """Infer base SLO unit time from warmup requests.
-
-    Returns the median base latency (ms) for a 16x16 resolution, single-frame,
-    single-step request. Only uses warmups that succeeded and have resolvable
-    width/height.
-    """
-
-    candidates_ms: list[float] = []
-    for req, out in warmup_pairs:
-        if not out.success or out.latency <= 0:
-            continue
-
-        width = req.width if req.width is not None else args.width
-        height = req.height if req.height is not None else args.height
-        if width is None or height is None:
-            continue
-
-        frames = req.num_frames if req.num_frames is not None else args.num_frames
-        steps = req.num_inference_steps if req.num_inference_steps is not None else args.num_inference_steps
-
-        frame_scale = int(frames) if isinstance(frames, int) and frames > 0 else 1
-        step_scale = int(steps) if isinstance(steps, int) and steps > 0 else 1
-
-        area_units = max((float(width) * float(height)) / float(16 * 16), 1.0)
-        denom = area_units * float(frame_scale) * float(step_scale)
-        if denom <= 0:
-            continue
-
-        candidates_ms.append((out.latency * 1000.0) / denom)
-
-    if not candidates_ms:
-        return None
-    return float(np.median(candidates_ms))
-
-
-def _populate_slo_ms_from_warmups(
-    requests_list: list[RequestFuncInput],
-    warmup_pairs: list[tuple[RequestFuncInput, RequestFuncOutput]],
-    args,
-) -> list[RequestFuncInput]:
-    """Populate missing RequestFuncInput.slo_ms using warmup outputs.
-
-    - If a request already has slo_ms (e.g., trace-provided), it is kept as-is.
-    - If any request has slo_ms is None and we can infer base time from warmups,
-      we estimate each missing request's expected execution time and set:
-        req.slo_ms = expected_latency_ms * args.slo_scale
-
-    Returns updated requests_list.
-    """
-
-    if not any(req.slo_ms is None for req in requests_list):
-        return requests_list
-
-    base_time_ms = _infer_slo_base_time_ms_from_warmups(warmup_pairs, args)
-    if base_time_ms is None:
-        return requests_list
-
-    slo_scale = float(getattr(args, "slo_scale", 3.0))
-    if slo_scale <= 0:
-        raise ValueError(f"slo_scale must be positive, got {slo_scale}.")
-
-    updated: list[RequestFuncInput] = []
-    for req in requests_list:
-        if req.slo_ms is not None:
-            updated.append(req)
-            continue
-        expected_ms = _compute_expected_latency_ms_from_base(req, args, base_time_ms)
-        updated.append(replace(req, slo_ms=(expected_ms * slo_scale) if expected_ms is not None else None))
-
-    return updated
-
-
-async def iter_requests(
-    requests_list: list[RequestFuncInput],
-    request_rate: float,
-) -> AsyncGenerator[RequestFuncInput, None]:
-    """Yield requests using a fixed interval if request_rate is set.
-
-    - If request_rate is inf, all requests are yielded immediately (no sleep).
-    - Otherwise, requests are emitted at a fixed cadence of 1 / request_rate seconds.
-    """
-
-    if request_rate != float("inf"):
-        if request_rate <= 0:
-            raise ValueError(f"request_rate must be positive or inf, got {request_rate}.")
-        interval_s = 1.0 / float(request_rate)
-
-    for i, req in enumerate(requests_list):
-        if request_rate != float("inf") and i > 0:
-            await asyncio.sleep(interval_s)
-        yield req
-
-
 def _guess_mime_type(path: str) -> str:
     mime, _ = mimetypes.guess_type(path)
     return mime or "application/octet-stream"
@@ -762,21 +415,12 @@ async def async_request_chat_completions(
 
     output.latency = time.perf_counter() - output.start_time
 
-    if output.success and input.slo_ms is not None:
-        output.slo_achieved = (output.latency * 1000.0) <= float(input.slo_ms)
-
     if pbar:
         pbar.update(1)
     return output
 
 
-def calculate_metrics(
-    outputs: list[RequestFuncOutput],
-    total_duration: float,
-    requests_list: list[RequestFuncInput],
-    args,
-    slo_enabled: bool,
-):
+def calculate_metrics(outputs: list[RequestFuncOutput], total_duration: float):
     success_outputs = [o for o in outputs if o.success]
     error_outputs = [o for o in outputs if not o.success]
 
@@ -798,29 +442,6 @@ def calculate_metrics(
         "peak_memory_mb_median": np.median(peak_memories) if peak_memories else 0,
     }
 
-    if slo_enabled:
-        slo_defined_total = 0
-        slo_met_success = 0
-
-        for req, out in zip(requests_list, outputs):
-            if req.slo_ms is None:
-                continue
-            slo_defined_total += 1
-            if out.slo_achieved is None:
-                continue
-            if out.slo_achieved:
-                slo_met_success += 1
-
-        slo_attain_all = (slo_met_success / slo_defined_total) if slo_defined_total > 0 else 0.0
-
-        metrics.update(
-            {
-                "slo_attainment_rate": slo_attain_all,
-                "slo_met_success": slo_met_success,
-                "slo_scale": getattr(args, "slo_scale", 3.0),
-            }
-        )
-
     return metrics
 
 
@@ -854,8 +475,6 @@ async def benchmark(args):
 
     if args.dataset == "vbench":
         dataset = VBenchDataset(args, api_url, args.model)
-    elif args.dataset == "trace":
-        dataset = TraceDataset(args, api_url, args.model)
     elif args.dataset == "random":
         dataset = RandomDataset(args, api_url, args.model)
     else:
@@ -882,7 +501,6 @@ async def benchmark(args):
     pbar = tqdm(total=len(requests_list), disable=args.disable_tqdm)
 
     async with aiohttp.ClientSession() as session:
-        warmup_pairs: list[tuple[RequestFuncInput, RequestFuncOutput]] = []
         if args.warmup_requests and requests_list:
             print(
                 f"Running {args.warmup_requests} warmup request(s) \
@@ -895,20 +513,16 @@ async def benchmark(args):
                         warm_req,
                         num_inference_steps=args.warmup_num_inference_steps,
                     )
-                warm_out = await limited_request_func(warm_req, session, None)
-                warmup_pairs.append((warm_req, warm_out))
-
-        if args.slo:
-            # Prefer trace-provided per-request slo_ms. Only populate when missing.
-            requests_list = _populate_slo_ms_from_warmups(
-                requests_list=requests_list,
-                warmup_pairs=warmup_pairs,
-                args=args,
-            )
+                await limited_request_func(warm_req, session, None)
 
         start_time = time.perf_counter()
         tasks = []
-        async for req in iter_requests(requests_list=requests_list, request_rate=args.request_rate):
+        for req in requests_list:
+            if args.request_rate != float("inf"):
+                # Poisson process: inter-arrival times follow exponential distribution
+                interval = np.random.exponential(1.0 / args.request_rate)
+                await asyncio.sleep(interval)
+
             task = asyncio.create_task(limited_request_func(req, session, pbar))
             tasks.append(task)
 
@@ -918,7 +532,7 @@ async def benchmark(args):
     pbar.close()
 
     # Calculate metrics
-    metrics = calculate_metrics(outputs, total_duration, requests_list, args, args.slo)
+    metrics = calculate_metrics(outputs, total_duration)
 
     print("\n{s:{c}^{n}}".format(s=" Serving Benchmark Result ", n=60, c="="))
 
@@ -947,12 +561,6 @@ async def benchmark(args):
     print("{:<40} {:<15.4f}".format("Latency Median (s):", metrics["latency_median"]))
     print("{:<40} {:<15.4f}".format("Latency P99 (s):", metrics["latency_p99"]))
 
-    if args.slo:
-        print(f"{'-' * 50}")
-        print("{:<40} {:<15.2%}".format("SLO Attainment Rate (all):", metrics.get("slo_attainment_rate", 0.0)))
-        print("{:<40} {:<15}".format("SLO Met (success count):", str(metrics.get("slo_met_success", 0))))
-        print("{:<40} {:<15}".format("SLO Scale:", str(metrics.get("slo_scale", 3.0))))
-
     if metrics["peak_memory_mb_max"] > 0:
         print(f"{'-' * 50}")
         print("{:<40} {:<15.2f}".format("Peak Memory Max (MB):", metrics["peak_memory_mb_max"]))
@@ -973,16 +581,16 @@ if __name__ == "__main__":
         "--base-url",
         type=str,
         default=None,
-        help="Base URL of the server (e.g., http://localhost:8091). Overrides host/port.",
+        help="Base URL of the server (e.g., http://localhost:30000). Overrides host/port.",
     )
     parser.add_argument("--host", type=str, default="localhost", help="Server host.")
-    parser.add_argument("--port", type=int, default=8091, help="Server port.")
+    parser.add_argument("--port", type=int, default=30000, help="Server port.")
     parser.add_argument("--model", type=str, default="default", help="Model name.")
     parser.add_argument(
         "--dataset",
         type=str,
         default="vbench",
-        choices=["vbench", "trace", "random"],
+        choices=["vbench", "random"],
         help="Dataset to use.",
     )
     parser.add_argument(
@@ -1037,7 +645,7 @@ if __name__ == "__main__":
     parser.add_argument(
         "--num-inference-steps",
         type=int,
-        default=50,
+        default=None,
         help="Number of inference steps (for diffusion models).",
     )
     parser.add_argument(
@@ -1048,21 +656,6 @@ if __name__ == "__main__":
     )
     parser.add_argument("--fps", type=int, default=None, help="FPS (for video).")
     parser.add_argument("--output-file", type=str, default=None, help="Output JSON file for metrics.")
-    parser.add_argument(
-        "--slo",
-        action="store_true",
-        help=(
-            "Enable SLO calculation and reporting. If trace provides per-request slo_ms, it is used. "
-            "Otherwise, warmup request(s) are used to infer expected execution time assuming linear "
-            "scaling by resolution, frames, and steps, then slo_ms = expected_time * --slo-scale."
-        ),
-    )
-    parser.add_argument(
-        "--slo-scale",
-        type=float,
-        default=3.0,
-        help="SLO target multiplier: slo_ms = estimated_exec_time_ms * slo_scale (default: 3).",
-    )
     parser.add_argument("--disable-tqdm", action="store_true", help="Disable progress bar.")
 
     args = parser.parse_args()

collect_env.py

@@ -57,7 +57,6 @@ SystemEnv = namedtuple(
         "cpu_info",
         "rocm_version",  # vllm specific field
         "vllm_version",  # vllm specific field
-        "vllm_omni_version",  # vllm-omni specific field
         "vllm_build_flags",  # vllm specific field
         "gpu_topo",  # vllm specific field
         "env_vars",
@@ -290,31 +289,6 @@ def get_vllm_version():
     return __version__
 
 
-def get_vllm_omni_version(run_lambda):
-    try:
-        import vllm_omni
-        from vllm_omni import __version__, __version_tuple__
-
-        version_str = __version_tuple__[-1]
-        if isinstance(version_str, str) and version_str.startswith("g"):
-            if "." in version_str:
-                git_sha = version_str.split(".")[0][1:]
-                date = version_str.split(".")[-1][1:]
-                return f"{__version__} (git sha: {git_sha}, date: {date})"
-            else:
-                git_sha = version_str[1:]
-                return f"{__version__} (git sha: {git_sha})"
-
-        package_dir = os.path.dirname(os.path.abspath(vllm_omni.__file__))
-        git_sha = run_and_read_all(run_lambda, f"git -C {package_dir} rev-parse --short HEAD")
-        if git_sha:
-            return f"{__version__} (git sha: {git_sha})"
-
-        return __version__
-    except ImportError:
-        return "N/A (vllm_omni not installed)"
-
-
 def summarize_vllm_build_flags():
     # This could be a static method if the flags are constant, or dynamic if you need to check environment variables, etc.
     return "CUDA Archs: {}; ROCm: {}".format(
@@ -550,7 +524,6 @@ def get_env_info():
 
     rocm_version = get_rocm_version(run_lambda)
     vllm_version = get_vllm_version()
-    vllm_omni_version = get_vllm_omni_version(run_lambda)
     vllm_build_flags = summarize_vllm_build_flags()
     gpu_topo = get_gpu_topo(run_lambda)
 
@@ -582,7 +555,6 @@ def get_env_info():
         cpu_info=get_cpu_info(run_lambda),
         rocm_version=rocm_version,
         vllm_version=vllm_version,
-        vllm_omni_version=vllm_omni_version,
         vllm_build_flags=vllm_build_flags,
         gpu_topo=gpu_topo,
         env_vars=get_env_vars(),
@@ -649,7 +621,6 @@ env_info_fmt += """
 ==============================
 ROCM Version                 : {rocm_version}
 vLLM Version                 : {vllm_version}
-vLLM-Omni Version            : {vllm_omni_version}
 vLLM Build Flags:
   {vllm_build_flags}
 GPU Topology:

docker/Dockerfile.ci

@@ -1,17 +1,11 @@
 ARG VLLM_BASE_IMAGE=vllm/vllm-openai
-ARG VLLM_BASE_TAG=v0.14.0
+ARG VLLM_BASE_TAG=v0.12.0
 FROM ${VLLM_BASE_IMAGE}:${VLLM_BASE_TAG}
 ARG APP_DIR=/workspace/vllm-omni
 WORKDIR ${APP_DIR}
 
 COPY . .
 
-# Install system dependencies
-RUN apt-get update && \
-    apt-get install -y ffmpeg && \
-    apt-get clean && \
-    rm -rf /var/lib/apt/lists/*
-
 # Install vllm-omni into the same uv-managed Python environment used by the base image.
 RUN uv pip install --python "$(python3 -c 'import sys; print(sys.executable)')" --no-cache-dir ".[dev]"
 

docker/Dockerfile.rocm

@@ -1,7 +1,9 @@
-ARG BASE_IMAGE=vllm/vllm-openai-rocm:v0.14.0
-FROM ${BASE_IMAGE} AS final
+ARG BASE_IMAGE=rocm/vllm-dev:nightly_main_20251205
+FROM ${BASE_IMAGE}
 
 ARG COMMON_WORKDIR=/app
+ARG VLLM_VERSION=v0.12.0
+ARG PYTORCH_ROCM_ARCH="gfx942;gfx950"
 
 WORKDIR ${COMMON_WORKDIR}
 
@@ -11,22 +13,30 @@ RUN apt-get update && \
     apt-get clean && \
     rm -rf /var/lib/apt/lists/*
 
+# Step 2: Reinstall vllm from source
+RUN python3 -m pip uninstall -y vllm && rm -rf vllm &&\
+    git clone https://github.com/vllm-project/vllm.git && \
+    cd vllm && \
+    git checkout ${VLLM_VERSION} && \
+    python3 -m pip install -r requirements/rocm.txt && \
+    python3 setup.py clean --all && \
+    PYTORCH_ROCM_ARCH=${PYTORCH_ROCM_ARCH} python3 setup.py develop && \
+    cd ../ && \
+    rm -rf vllm/.git
+
 RUN mkdir -p ${COMMON_WORKDIR}/vllm-omni
 
-# Step 2: Copy vllm-omni code and install without uv
+# Step 3: Copy vllm-omni code and install without uv
 COPY . ${COMMON_WORKDIR}/vllm-omni
-RUN cd ${COMMON_WORKDIR}/vllm-omni && uv pip install --python "$(python3 -c 'import sys; print(sys.executable)')" --no-cache-dir ".[dev]"
-
-# When we are installing onnxruntime-rocm, we need to uninstall the system-installed onnxruntime first.
-# These are the dependencies of Qwen3-TTS.
-RUN uv pip uninstall onnxruntime --system && uv pip install --no-cache-dir onnxruntime-rocm sox --system
+RUN cd ${COMMON_WORKDIR}/vllm-omni && python3 -m pip install --no-cache-dir ".[dev]"
 
+# Create python symlink
+# `GPU_ARCHS` is an environment variable that is used to set the GPU archs for the AITER.
+# This is needed to prevent the AITER automatic GPU arch detection from failing on MI325X.
+# The AITER version used in this dockerfile has issues with handling
+# the GPU archs of MI325X (CI machine) correctly. So we manually set the GPU archs here.
+# We reuse AITER_ROCM_ARCH from the base image to avoid duplication.
+ENV GPU_ARCHS=${AITER_ROCM_ARCH}
 RUN ln -sf /usr/bin/python3 /usr/bin/python
 
-CMD ["/bin/bash"]
-
 ENTRYPOINT []
-
-#Set entrypoint for vllm-openai official images
-FROM final AS vllm-openai
-ENTRYPOINT ["vllm", "serve", "--omni"]

docs/.nav.yml

@@ -3,7 +3,7 @@ nav:
 - User Guide:
   - Getting Started:
     - getting_started/quickstart.md
-    - getting_started/installation/*
+    - getting_started/installation
   - Serving:
     - OpenAI-Compatible API:
       - Image Generation: serving/image_generation_api.md
@@ -11,10 +11,8 @@ nav:
     - examples/README.md
     - Offline Inference:
       - Image-To-Image: user_guide/examples/offline_inference/image_to_image.md
-      - Image-To-Video: user_guide/examples/offline_inference/image_to_video.md
       - Qwen2.5-Omni: user_guide/examples/offline_inference/qwen2_5_omni.md
       - Qwen3-Omni: user_guide/examples/offline_inference/qwen3_omni.md
-      - Qwen3-TTS Offline Inference: user_guide/examples/offline_inference/qwen3_tts.md
       - Text-To-Image: user_guide/examples/offline_inference/text_to_image.md
       - Text-To-Video: user_guide/examples/offline_inference/text_to_video.md
     - Online Serving:
@@ -27,16 +25,14 @@ nav:
   - Configuration:
     - configuration/README.md
     - configuration/*
+  - Diffusion Acceleration:
+    - Overview: user_guide/diffusion_acceleration.md
+    - Acceleration Methods:
+      - TeaCache: user_guide/acceleration/teacache.md
+      - Cache-DiT: user_guide/acceleration/cache_dit_acceleration.md
+      - Parallelism Acceleration: user_guide/acceleration/parallelism_acceleration.md
   - Models:
     - models/supported_models.md
-  - Features:
-    - Sleep Mode: features/sleep_mode.md
-    - Diffusion Features:
-      - Overview: user_guide/diffusion_acceleration.md
-      - TeaCache: user_guide/diffusion/teacache.md
-      - Cache-DiT: user_guide/diffusion/cache_dit_acceleration.md
-      - Parallelism Acceleration: user_guide/diffusion/parallelism_acceleration.md
-      - CPU Offloading: user_guide/diffusion/cpu_offload_diffusion.md
 - Developer Guide:
   - General:
     - contributing/README.md
@@ -47,6 +43,7 @@ nav:
     - contributing/model/adding_omni_model.md
     - contributing/model/adding_diffusion_model.md
   - CI: contributing/ci
+  - Tests: contributing/tests
   - Design Documents:
     - design/index.md
     - design/architecture_overview.md

docs/README.md

@@ -60,5 +60,5 @@ vLLM-Omni seamlessly supports most popular open-source models on HuggingFace, in
 
 For more information, checkout the following:
 
-- [vllm-omni architecture design and recent roadmaps](https://docs.google.com/presentation/d/1qv4qMW1rKAqDREMXiUDLIgqqHQe7TDPj/edit?usp=sharing&ouid=110473603432222024453&rtpof=true&sd=true)
+- [vllm-omni architecture design and recent roadmaps](https://docs.google.com/presentation/d/1Y7t2Zm3BIISPN-_X_sDpxZMNCWir2rAx/edit?usp=drive_link&ouid=110473603432222024453&rtpof=true&sd=true)
 - [vllm-omni announcement blogpost](https://blog.vllm.ai/2025/11/30/vllm-omni.html)

docs/api/README.md

@@ -36,6 +36,7 @@ Input data structures for multi-modal inputs.
 
 Engine classes for offline and online inference.
 
+- [vllm_omni.diffusion.diffusion_engine.BackgroundResources][]
 - [vllm_omni.diffusion.diffusion_engine.DiffusionEngine][]
 - [vllm_omni.engine.AdditionalInformationEntry][]
 - [vllm_omni.engine.AdditionalInformationPayload][]
@@ -53,16 +54,40 @@ Engine classes for offline and online inference.
 
 Core scheduling and caching components.
 
-- [vllm_omni.core.sched.omni_ar_scheduler.KVCacheTransferData][]
 - [vllm_omni.core.sched.omni_ar_scheduler.OmniARScheduler][]
 - [vllm_omni.core.sched.omni_generation_scheduler.OmniGenerationScheduler][]
-- [vllm_omni.core.sched.output.OmniCachedRequestData][]
 - [vllm_omni.core.sched.output.OmniNewRequestData][]
-- [vllm_omni.model_executor.models.qwen3_tts.tokenizer_25hz.vq.core_vq.DistributedGroupResidualVectorQuantization][]
-- [vllm_omni.model_executor.models.qwen3_tts.tokenizer_25hz.vq.core_vq.DistributedResidualVectorQuantization][]
-- [vllm_omni.model_executor.models.qwen3_tts.tokenizer_25hz.vq.core_vq.EuclideanCodebook][]
-- [vllm_omni.model_executor.models.qwen3_tts.tokenizer_25hz.vq.core_vq.VectorQuantization][]
-- [vllm_omni.model_executor.models.qwen3_tts.tokenizer_25hz.vq.core_vq.preprocess][]
+
+## Model Executor
+
+Model execution components.
+
+- [vllm_omni.model_executor.custom_process_mixin.CustomProcessMixin][]
+- [vllm_omni.model_executor.models.output_templates.OmniOutput][]
+- [vllm_omni.model_executor.models.qwen2_5_omni.qwen2_5_omni.Qwen2_5OmniForConditionalGeneration][]
+- [vllm_omni.model_executor.models.qwen2_5_omni.qwen2_5_omni_talker.Qwen2_5OmniTalkerForConditionalGeneration][]
+- [vllm_omni.model_executor.models.qwen2_5_omni.qwen2_5_omni_thinker.Qwen2_5OmniConditionalGenerationMixin][]
+- [vllm_omni.model_executor.models.qwen2_5_omni.qwen2_5_omni_thinker.Qwen2_5OmniThinkerForConditionalGeneration][]
+- [vllm_omni.model_executor.models.qwen2_5_omni.qwen2_5_omni_token2wav.Qwen2_5OmniToken2WavBigVGANModel][]
+- [vllm_omni.model_executor.models.qwen2_5_omni.qwen2_5_omni_token2wav.Qwen2_5OmniToken2WavDiTModel][]
+- [vllm_omni.model_executor.models.qwen2_5_omni.qwen2_5_omni_token2wav.Qwen2_5OmniToken2WavForConditionalGenerationVLLM][]
+- [vllm_omni.model_executor.models.qwen2_5_omni.qwen2_5_omni_token2wav.Qwen2_5OmniToken2WavModel][]
+- [vllm_omni.model_executor.models.qwen2_5_omni.qwen2_old.Qwen2EmbeddingModel][]
+- [vllm_omni.model_executor.models.qwen2_5_omni.qwen2_old.Qwen2ForCausalLM][]
+- [vllm_omni.model_executor.models.qwen2_5_omni.qwen2_old.Qwen2Model][]
+- [vllm_omni.model_executor.models.qwen3_omni.qwen3_moe.Qwen3MoeForCausalLM][]
+- [vllm_omni.model_executor.models.qwen3_omni.qwen3_omni.Qwen3OmniMoeForConditionalGeneration][]
+- [vllm_omni.model_executor.models.qwen3_omni.qwen3_omni_code2wav.Qwen3OmniMoeCode2Wav][]
+- [vllm_omni.model_executor.models.qwen3_omni.qwen3_omni_moe_code_predictor_mtp.Qwen3OmniCodePredictorBaseModel][]
+- [vllm_omni.model_executor.models.qwen3_omni.qwen3_omni_moe_code_predictor_mtp.Qwen3OmniMoeTalkerCodePredictor][]
+- [vllm_omni.model_executor.models.qwen3_omni.qwen3_omni_moe_talker.Qwen3OmniMoeModel][]
+- [vllm_omni.model_executor.models.qwen3_omni.qwen3_omni_moe_talker.Qwen3OmniMoeTalkerForConditionalGeneration][]
+- [vllm_omni.model_executor.models.qwen3_omni.qwen3_omni_moe_thinker.Qwen3MoeLLMForCausalLM][]
+- [vllm_omni.model_executor.models.qwen3_omni.qwen3_omni_moe_thinker.Qwen3MoeLLMModel][]
+- [vllm_omni.model_executor.models.qwen3_omni.qwen3_omni_moe_thinker.Qwen3OmniMoeConditionalGenerationMixin][]
+- [vllm_omni.model_executor.models.qwen3_omni.qwen3_omni_moe_thinker.Qwen3OmniMoeThinkerForConditionalGeneration][]
+- [vllm_omni.model_executor.models.qwen3_omni.qwen3_omni_moe_thinker.Qwen3OmniMoeThinkerMultiModalProcessor][]
+- [vllm_omni.model_executor.models.qwen3_omni.qwen3_omni_moe_thinker.Qwen3OmniMoeThinkerProcessingInfo][]
 
 ## Configuration
 
@@ -72,25 +97,13 @@ Configuration classes.
 - [vllm_omni.diffusion.cache.teacache.config.TeaCacheConfig][]
 - [vllm_omni.distributed.omni_connectors.utils.config.ConnectorSpec][]
 - [vllm_omni.distributed.omni_connectors.utils.config.OmniTransferConfig][]
-- [vllm_omni.model_executor.models.qwen3_tts.configuration_qwen3_tts.Qwen3TTSConfig][]
-- [vllm_omni.model_executor.models.qwen3_tts.configuration_qwen3_tts.Qwen3TTSSpeakerEncoderConfig][]
-- [vllm_omni.model_executor.models.qwen3_tts.configuration_qwen3_tts.Qwen3TTSTalkerCodePredictorConfig][]
-- [vllm_omni.model_executor.models.qwen3_tts.configuration_qwen3_tts.Qwen3TTSTalkerConfig][]
-- [vllm_omni.model_executor.models.qwen3_tts.tokenizer_12hz.configuration_qwen3_tts_tokenizer_v2.Qwen3TTSTokenizerV2Config][]
-- [vllm_omni.model_executor.models.qwen3_tts.tokenizer_12hz.configuration_qwen3_tts_tokenizer_v2.Qwen3TTSTokenizerV2DecoderConfig][]
-- [vllm_omni.model_executor.models.qwen3_tts.tokenizer_25hz.configuration_qwen3_tts_tokenizer_v1.Qwen3TTSTokenizerV1Config][]
-- [vllm_omni.model_executor.models.qwen3_tts.tokenizer_25hz.configuration_qwen3_tts_tokenizer_v1.Qwen3TTSTokenizerV1DecoderBigVGANConfig][]
-- [vllm_omni.model_executor.models.qwen3_tts.tokenizer_25hz.configuration_qwen3_tts_tokenizer_v1.Qwen3TTSTokenizerV1DecoderConfig][]
-- [vllm_omni.model_executor.models.qwen3_tts.tokenizer_25hz.configuration_qwen3_tts_tokenizer_v1.Qwen3TTSTokenizerV1DecoderDiTConfig][]
-- [vllm_omni.model_executor.models.qwen3_tts.tokenizer_25hz.configuration_qwen3_tts_tokenizer_v1.Qwen3TTSTokenizerV1EncoderConfig][]
 
 ## Workers
 
 Worker classes and model runners for distributed inference.
 
-- [vllm_omni.diffusion.worker.gpu_diffusion_model_runner.GPUDiffusionModelRunner][]
-- [vllm_omni.diffusion.worker.gpu_diffusion_worker.GPUDiffusionWorker][]
-- [vllm_omni.diffusion.worker.gpu_diffusion_worker.WorkerProc][]
+- [vllm_omni.diffusion.worker.gpu_worker.GPUWorker][]
+- [vllm_omni.diffusion.worker.gpu_worker.WorkerProc][]
 - [vllm_omni.diffusion.worker.npu.npu_worker.NPUWorker][]
 - [vllm_omni.diffusion.worker.npu.npu_worker.NPUWorkerProc][]
 - [vllm_omni.worker.gpu_ar_model_runner.ExecuteModelState][]

docs/configuration/README.md

@@ -2,7 +2,7 @@
 
 This section lists the most common options for running vLLM-Omni.
 
-For options within a vLLM Engine. Please refer to [vLLM Configuration](https://docs.vllm.ai/en/v0.14.0/configuration/index.html)
+For options within a vLLM Engine. Please refer to [vLLM Configuration](https://docs.vllm.ai/en/v0.12.0/configuration/index.html)
 
 Currently, the main options are maintained by stage configs for each model.
 
@@ -16,6 +16,6 @@ For introduction, please check [Introduction for stage config](./stage_configs.m
 
 ## Optimization Features
 
-- **[TeaCache Configuration](../user_guide/diffusion/teacache.md)** - Enable TeaCache adaptive caching for DiT models to achieve 1.5x-2.0x speedup with minimal quality loss
-- **[Cache-DiT Configuration](../user_guide/diffusion/cache_dit_acceleration.md)** - Enable Cache-DiT as cache acceleration backends for DiT models
-- **[Parallelism Configuration](../user_guide/diffusion/parallelism_acceleration.md)** - Enable parallelism (e.g., sequence parallelism) for for DiT models
+- **[TeaCache Configuration](../user_guide/acceleration/teacache.md)** - Enable TeaCache adaptive caching for DiT models to achieve 1.5x-2.0x speedup with minimal quality loss
+- **[Cache-DiT Configuration](../user_guide/acceleration/cache_dit_acceleration.md)** - Enable Cache-DiT as cache acceleration backends for DiT models
+- **[Parallelism Configuration](../user_guide/acceleration/parallelism_acceleration.md)** - Enable parallelism (e.g., sequence parallelism) for for DiT models

docs/contributing/ci/tests_markers.md

@@ -1,160 +0,0 @@
-# Markers for Tests
-
-By adding markers before test functions, tests can later be executed uniformly by simply declaring the corresponding marker type.
-
-## Current Markers
-Defined in `pyproject.toml`:
-
-| Marker             | Description                                             |
-| ------------------ | ------------------------------------------------------- |
-| `core_model`       | Core model tests (run in each PR)                       |
-| `diffusion`        | Diffusion model tests                                   |
-| `omni`             | Omni model tests                                        |
-| `cache`            | Cache backend tests                                     |
-| `parallel`         | Parallelism/distributed tests                           |
-| `cpu`              | Tests that run on CPU                                   |
-| `gpu`              | Tests that run on GPU (auto-added)                      |
-| `cuda`             | Tests that run on CUDA (auto-added)                     |
-| `rocm`             | Tests that run on AMD/ROCm (auto-added)                 |
-| `npu`              | Tests that run on NPU/Ascend (auto-added)               |
-| `H100`             | Tests that require H100 GPU                             |
-| `L4`               | Tests that require L4 GPU                               |
-| `MI325`            | Tests that require MI325 GPU (AMD/ROCm)                 |
-| `A2`               | Tests that require A2 NPU                               |
-| `A3`               | Tests that require A3 NPU                               |
-| `distributed_cuda` | Tests that require multi cards on CUDA platform         |
-| `distributed_rocm` | Tests that require multi cards on ROCm platform         |
-| `distributed_npu`  | Tests that require multi cards on NPU platform          |
-| `skipif_cuda`      | Skip if the num of CUDA cards is less than the required |
-| `skipif_rocm`      | Skip if the num of ROCm cards is less than the required |
-| `skipif_npu`       | Skip if the num of NPU cards is less than the required  |
-| `slow`             | Slow tests (may skip in quick CI)                       |
-| `benchmark`        | Benchmark tests                                         |
-
-For those markers shown as auto-added, they will be added by the `@hardware_test` decorator.
-
-### Example usage for markers
-
-```python
-from tests.utils import hardware_test
-
-@pytest.mark.core_model
-@pytest.mark.omni
-@hardware_test(
-   res={"cuda": "L4", "rocm": "MI325", "npu": "A2"},
-   num_cards=2,
-)
-@pytest.mark.parametrize("omni_server", test_params, indirect=True)
-def test_video_to_audio()
-    ...
-```
-### Decorator: `@hardware_test`
-
-This decorator is intended to make hardware-aware, cross-platform test authoring easier and more robust for CI/CD environments. The `hardware_test` decorator in `vllm-omni/tests/utils.py` performs the following actions:
-
-1. **Applies platform and resource markers**  
-   Adds the appropriate pytest markers for each specified hardware platform (e.g., `cuda`, `rocm`, `npu`) and resource type (e.g., `L4`, `H100`, `MI325`, `A2`, `A3`).
-   ```
-   @pytest.mark.cuda
-   @pytest.mark.L4
-   ```
-2. **Handles multi-card (distributed) scenarios**  
-   For tests requiring multiple cards, it automatically adds distributed markers such as `distributed_cuda`, `distributed_rocm`, or `distributed_npu`.
-   ```
-   @pytest.mark.distributed_cuda(num_cards=num_cards)
-   ```
-3. **Supports flexible card requirements**  
-   Accepts `num_cards` as either a single integer for all platforms or as a dictionary with per-platform values. If not specified, defaults to 1 card per platform.
-
-4. **Integrates resource validation**  
-   On CUDA, adds a skip marker (`skipif_cuda`) if the system does not have the required number of devices.
-   Support for `skipif_rocm` and `skipif_npu` will be implemented later.
-
-
-5. **Runs each test in a new process**  
-   Automatically wraps the distributed test with a decorator (`@create_new_process_for_each_test`) to ensure isolation and compatibility with multi-process hardware backends.
-
-6. **Works with pytest filtering**  
-   Allows tests to be filtered and selected at runtime using standard pytest marker expressions (e.g., `-m "distributed_cuda and L4"`).
-
-#### Example usage for decorator
-- Single call for multiple platforms:
-    ```python
-    @hardware_test(
-        res={"cuda": "L4", "rocm": "MI325", "npu": "A2"},
-        num_cards={"cuda": 2, "rocm": 2, "npu": 2},
-    )
-    ```
-    or
-    ```python
-    @hardware_test(
-        res={"cuda": "L4", "rocm": "MI325", "npu": "A2"},
-        num_cards=2,
-    )
-    ```
-- `res` must be a dict; supported resources: CUDA (L4/H100), ROCm (MI325), NPU (A2/A3)
-- `num_cards` can be int (all platforms) or dict (per platform); defaults to 1 when missing
-- `hardware_test` automatically applies `@create_new_process_for_each_test` for distributed tests.
-- Distributed markers (`distributed_cuda`, `distributed_rocm`, `distributed_npu`) are auto-added for multi-card cases
-- Filtering examples:
-    - CUDA only: `pytest -m "distributed_cuda and L4"`
-    - ROCm only: `pytest -m "distributed_rocm and MI325"`
-    - NPU only: `pytest -m "distributed_npu"`
-
-## Add Support for a New Platform
-
-If you want to add support for a new platform (e.g., "tpu" for a new accelerator), follow these steps:
-
-1. **Extend the marker list in your pytest config** so that platform/resource markers are defined:
-   ```toml
-   # In pyproject.toml or pytest.ini
-   [tool.pytest.ini_options]
-   markers = [
-       # ... existing markers ...
-       "tpu: Tests that require TPU device",
-       "TPU_V3: Tests that require TPU v3 hardware",
-       "distributed_tpu: Tests that require multiple TPU devices",
-   ]
-   ```
-2. **Implement a marker construction function for your platform** in `vllm-omni/tests/utils.py`:
-   ```python
-   # In vllm-omni/tests/utils.py
-
-   def tpu_marks(*, res: str, num_cards: int):
-       test_platform = pytest.mark.tpu
-       if res == "TPU_V3":
-           test_resource = pytest.mark.TPU_V3
-       else:
-           raise ValueError(
-               f"Invalid TPU resource type: {res}. Supported: TPU_V3")
-
-       if num_cards == 1:
-           return [test_platform, test_resource]
-       else:
-           test_distributed = pytest.mark.distributed_tpu(num_cards=num_cards)
-           # Optionally: add skipif_tpu when implemented
-           return [test_platform, test_resource, test_distributed]
-   ```
-3. **Update `hardware_test` to recognize your new platform**:
-    In the relevant place (see the `hardware_test` implementation), add:
-    ```python
-    if platform == "tpu":
-        marks = tpu_marks(res=resource, num_cards=cards)
-    ```
-4. **(Recommended) Add a test using your new markers**:
-   ```python
-   @hardware_test(
-       res={"tpu": "TPU_V3"},
-       num_cards=2,
-   )
-   def test_my_tpu_feature():
-       ...
-   ```
-
-**Summary**:  
-- Add pytest markers for your new platform/resources  
-- Implement a marker function (`xxx_marks`)  
-- Plug into `hardware_test`  
-- You're done: tests decorated with `@hardware_test` using your platform now automatically get the correct markers, distribution, and isolation!
-
-See code in `vllm-omni/tests/utils.py` for existing examples (`cuda_marks`, `rocm_marks`, `npu_marks`).

docs/contributing/model/adding_diffusion_model.md

@@ -44,7 +44,7 @@ vllm_omni/
 
 # Step-by-step-implementation
 ## Step 1: Model Implementation
-The diffusion pipeline’s implementation follows **HuggingFace Diffusers**.
+The diffusion pipeline’s implementation follows **HuggingFace Diffusers**, and components that do not need modification can be imported directly.
 ### 1.1 Define the Pipeline Class
 Define the pipeline class, e.g., `QwenImageEditPipeline`, and initialize all required submodules, either from HuggingFace `diffusers` or custom implementations. In `QwenImageEditPipeline`, only `QwenImageTransformer2DModel` is re-implemented to support optimizations such as Ulysses-SP. When adding new models in the future, you can either reuse this re-implemented `QwenImageTransformer2DModel` or extend it as needed.
 
@@ -73,75 +73,10 @@ def get_qwen_image_edit_post_process_func(
 The forward function of `QwenImageEditPipeline` follows the HuggingFace `diffusers` design for the most part. The key differences are:
 + As described in the overview, arguments are passed through `OnniDiffusionRequest`, so we need to get user parameters from it accordingly.
 ```python
-prompt = req.prompt
+prompt = req.prompt if req.prompt is not None else prompt
 ```
 + pre/post-processing are handled by the framework elsewhere, so skip them.
 
-### 1.4 Replace some ops or layers in DiT component
-
-vLLM-Omni provides a set of optimized operators with better performance and built-in support for parallelism, including attention, rotary embeddings (RoPE), and linear layers.
-
-Below is an example showing how to replace standard Transformer attention and FFN layers with vLLM-Omni implementations:
-
-```python
-from vllm_omni.diffusion.attention.layer import Attention
-from vllm_omni.diffusion.layers.rope import RotaryEmbedding
-
-class MyAttention(nn.Module):
-    def __init__(self):
-        super().__init__()
-        self.attn = Attention()
-        self.to_qkv = QKVParallelLinear()
-        self.to_out = RowParallelLinear()
-        self.rope = RotaryEmbedding(is_neox_style=False)
-
-    def forward(self, hidden_states):
-        qkv, _ = self.to_qkv(hidden_states)
-        q, k, v = qkv.split(...)
-        q, k = self.rope(...)
-        attn_output = self.attn(q, k, v)
-        output = self.to_out(attn_output)
-
-class MyFFN(nn.Module):
-    def __init__(self):
-        super().__init__()
-        self.fc1 = ColumnParallelLinear()
-        self.fc2 = RowParallelLinear()
-        self.act = F.gelu
-
-    def forward(self, hidden_states):
-        hidden, _ = self.fc1(hidden_states)
-        hidden = self.act(hidden)
-        output = self.fc2(hidden)
-        return output
-```
-
-In this example:
-
-+ Attention uses vLLM-Omni’s optimized attention kernel together with parallel QKV projection and RoPE.
-
-+ Linear layers are replaced with column- and row-parallel variants to enable tensor parallelism.
-
-+ The FFN follows a standard two-layer structure and can be further optimized (e.g., using fused or merged projections) if needed.
-
-
-### 1.5 Provide a `_repeated_blocks` in DiT model
-`_repeated_blocks` is the small and frequently-repeated block(s) of a model -- typically a transformer layer.
-
-It's used for torch compile optimizations.
-```python
-_repeated_blocks = ["QwenImageTransformerBlock"]
-```
-
-
-### 1.6 (Optional) implement sequence parallelism
-vLLM-Omni has a non-intrusive `_sp_plan` that enable sequence parallel without modifying `forward()` logic.
-You can refer to [How to parallelize a new model](../../user_guide/diffusion/parallelism_acceleration.md)
-
-
-### 1.7 (Optional) integrate with Cache-Dit
-vLLM-Omni supports acceleration via [Cache-Dit](../../user_guide/diffusion/cache_dit_acceleration.md). Most models compatible with Diffusers can use Cache-Dit seamlessly. For new models, you can extend support by modifying`cache_dit_backend.py`
-
 ## Step 2: Extend OmniDiffusionRequest Fields
 User-provided inputs are ultimately passed to the model’s forward method through OmniDiffusionRequest, so we add the required fields here to support the new model.
 ```python
@@ -204,23 +139,8 @@ Key point for writing the example:
 
 + Save or display the generated results so users can validate the integration.
 
-## Step 5: Open a Pull Request
-
-When submitting a pull request to add support for a new model, please include the following information in the PR description:
-
-+ Output verification: provide generation outputs to verify correctness and model behavior.
-
-+ Inference speed: provide a comparison with the corresponding implementation in Diffusers.
-
-+ Parallelism support: specify the supported parallel sizes and any relevant limitations.
-
-+ Cache acceleration: check whether the model can be accelerated using Cache-Dit or not.
-
-
-Providing these details helps reviewers evaluate correctness, performance improvements, and parallel scalability of the new model integration.
-
 # Testing
-For comprehensive testing guidelines, please refer to the [Test File Structure and Style Guide](../ci/tests_style.md).
+For comprehensive testing guidelines, please refer to the [Test File Structure and Style Guide](../tests/tests_style.md).
 
 
 ## Adding a Model Recipe

docs/contributing/model/adding_omni_model.md

@@ -572,7 +572,7 @@ def talker2code2wav(
 
 ## Testing
 
-For comprehensive testing guidelines, please refer to the [Test File Structure and Style Guide](../ci/tests_style.md).
+For comprehensive testing guidelines, please refer to the [Test File Structure and Style Guide](../tests/tests_style.md).
 
 ## Adding a Model Recipe
 

docs/contributing/profiling.md

@@ -1,30 +1,48 @@
 # Profiling vLLM-Omni
 
-> **Warning:** Profiling incurs significant overhead. Use only for development and debugging, never in production.
+Performance Profiling Guidelines Profiling capabilities in vLLM-Omni are reserved for development and maintenance tasks aimed at temporal analysis of the codebase. Production use is **strongly discouraged**; enabling the profiler incurs a substantial overhead that negatively impacts inference latency.
 
-vLLM-Omni uses the PyTorch Profiler to analyze performance across both **Multi-Stage LLMs** and **Diffusion Models**.
+**Mechanism**: vLLM-Omni implements cross-stage profiling via the PyTorch Profiler. To accommodate the architecture—where stages operate as distinct engine instances in separate processes—the profiling interface supports both holistic capturing (all stages) and targeted capturing (specific stages).
 
-### 1. Set the Output Directory
-Before running any script, set this environment variable. The system detects this and automatically saves traces here.
+**1. Enabling the Profiler**
 
-```bash
-export VLLM_TORCH_PROFILER_DIR=./profiles
+Before running your script, you must set the ```VLLM_TORCH_PROFILER_DIR``` environment
+variable.
+
+```Bash
+export VLLM_TORCH_PROFILER_DIR=/path/to/save/traces
 ```
 
-### 2. Start Profiling
+**Highly Recommended: Limit Profiling to a Single Iteration**  
+For most use cases (especially when profiling audio stages), you should limit the profiler to just **one iteration** to keep trace files small and readable.
 
-It is best to limit profiling to one iteration to keep trace files manageable.
 
 ```bash
 export VLLM_PROFILER_MAX_ITERS=1
 ```
 
+**2. Offline Inference**
+
+For offlinie processing using ```OmniLLM```, you can wrap your ```generate``` calls with ```start_profile``` and ```stop_profile()```.
+
+Basic Usage(All Stages)
+```Python
+from vllm_omni import OmniLLM
+
+omni_llm = OmniLLM.from_engine_args(engine_args)
+
+# Start profiling all active stages
+omni_llm.start_profile()
+
+outputs = omni_llm.generate(prompts, sampling_params)
+
+# Stop profiling and save traces
+omni_llm.stop_profile()
+```
+
 **Selective Stage Profiling**
 The profiler is default to function across all stages. But It is highly recommended to profile specific stages by passing the stages list, preventing from producing too large trace files:
 ```python
-# Profile all stages
-omni_llm.start_profile()
-
 # Only profile Stage 1
 omni_llm.start_profile(stages=[1])
 ```
@@ -34,105 +52,13 @@ omni_llm.start_profile(stages=[1])
 omni_llm.start_profile(stages=[0, 2])
 ```
 
-**Python Usage**: Wrap your generation logic with `start_profile()` and `stop_profile()`.
-
-```python
-from vllm_omni import omni_llm
-
-profiler_enabled = bool(os.getenv("VLLM_TORCH_PROFILER_DIR"))
-
-# 1. Start profiling if enabled
-if profiler_enabled:
-    omni_llm.start_profile(stages=[0])
-
-# Initialize generator
-omni_generator = omni_llm.generate(prompts, sampling_params_list, py_generator=args.py_generator)
-
-total_requests = len(prompts)
-processed_count = 0
-
-# Main Processing Loop
-for stage_outputs in omni_generator:
-
-    # ... [Output processing logic for text/audio would go here] ...
-
-    # Update count to track when to stop profiling
-    processed_count += len(stage_outputs.request_output)
-
-    # 2. Check if all requests are done to stop the profiler safely
-    if profiler_enabled and processed_count >= total_requests:
-        print(f"[Info] Processed {processed_count}/{total_requests}. Stopping profiler inside active loop...")
-
-        # Stop the profiler while workers are still active
-        omni_llm.stop_profile()
-
-        # Wait for traces to flush to disk
-        print("[Info] Waiting 30s for workers to write trace files to disk...")
-        time.sleep(30)
-        print("[Info] Trace export wait time finished.")
-
-omni_llm.close()
-```
-
-
 **Examples**:
 
 1. **Qwen-omni 2.5**:  [https://github.com/vllm-project/vllm-omni/blob/main/examples/offline_inference/qwen2_5_omni/end2end.py](https://github.com/vllm-project/vllm-omni/blob/main/examples/offline_inference/qwen2_5_omni/end2end.py)
 
 2. **Qwen-omni 3.0**:   [https://github.com/vllm-project/vllm-omni/blob/main/examples/offline_inference/qwen3_omni/end2end.py](https://github.com/vllm-project/vllm-omni/blob/main/examples/offline_inference/qwen3_omni/end2end.py)
 
-
-**For Diffusion Models as a single stage**
-
-Diffusion profiling is End-to-End, capturing encoding, denoising loops, and decoding.
-
-**CLI Usage:**
-```python
-
-python image_to_video.py \
-    --model Wan-AI/Wan2.2-I2V-A14B-Diffusers \
-    --image qwen-bear.png \
-    --prompt "A cat playing with yarn, smooth motion" \
-    \
-    # Minimize Spatial Dimensions (Optional but helpful):
-    #    Drastically reduces memory usage so the profiler doesn't
-    #    crash due to overhead, though for accurate performance
-    #    tuning you often want target resolutions.
-    --height 48 \
-    --width 64 \
-    \
-    # Minimize Temporal Dimension (Frames):
-    #    Video models process 3D tensors (Time, Height, Width).
-    #    Reducing frames to the absolute minimum (2) keeps the
-    #    tensor size small, ensuring the trace file doesn't become
-    #    multi-gigabytes in size.
-    --num_frames 2 \
-    \
-    # Minimize Iteration Loop (Steps):
-    #    This is the most critical setting for profiling.
-    #    Diffusion models run the same loop X times.
-    #    Profiling 2 steps gives you the exact same performance
-    #    data as 50 steps, but saves minutes of runtime and
-    #    prevents the trace viewer from freezing.
-    --num_inference_steps 2 \
-    \
-    --guidance_scale 5.0 \
-    --guidance_scale_high 6.0 \
-    --boundary_ratio 0.875 \
-    --flow_shift 12.0 \
-    --fps 16 \
-    --output i2v_output.mp4
-
-```
-
-**Examples**:
-
-1. **Qwen image edit**:  [https://github.com/vllm-project/vllm-omni/blob/main/examples/offline_inference/image_to_image/image_edit.py](https://github.com/vllm-project/vllm-omni/blob/main/examples/offline_inference/image_to_image/image_edit.py)
-
-2. **Wan-AI/Wan2.2-I2V-A14B-Diffusers**:   [https://github.com/vllm-project/vllm-omni/tree/main/examples/offline_inference/image_to_video](https://github.com/vllm-project/vllm-omni/tree/main/examples/offline_inference/image_to_video)
-
-
-**Online Inference(Async)**
+**3. Online Inference(Async)**
 
 For online serving using AsyncOmni, the methods are asynchronous. This allows you to toggle profiling dynamically without restarting the server.
 
@@ -151,16 +77,23 @@ async for output in async_omni.generate(prompt, sampling_params, request_id):
 await async_omni.stop_profile()
 ```
 
-### 3. Analyzing Omni Traces
+**4. Analyzing Omni Traces**
 
-Output files are saved to your configured ```VLLM_TORCH_PROFILER_DIR```.
+After ``stop_profile()`` completes (and the file write wait time has passed), the directory specified in ```VLLM_TORCH_PROFILER_DIR``` will contain the trace files.
 
-**Output**
-**Chrome Trace** (```.json.gz```): Visual timeline of kernels and stages. Open in Perfetto UI.
+```
+Output/
+│── ...rank-0.pt.trace.json.gz   # GPU 0 trace (TP=2 Example)
+│── ...rank-1.pt.trace.json.gz   # GPU 1 trace (TP=2 Example)
+│       # Load these into Perfetto to visualize synchronization
+│
+│── profiler_out_.txt            # Summary tables (CPU/CUDA time %)
+```
 
 **Viewing Tools:**
+     - [Perfetto](https://ui.perfetto.dev/): (Recommended): Best for handling large audio trace files.
+     - ```chrome://tracing```: Good for smaller text-only traces.
 
-- [Perfetto](https://ui.perfetto.dev/)(recommended)
-- ```chrome://tracing```(Chrome only)
 
-**Note**: vLLM-Omni reuses the PyTorch Profiler infrastructure from vLLM. See the official vLLM profiler documentation:  [vLLM Profiling Guide](https://docs.vllm.ai/en/latest/dev/profiling.html)
+**Note**: vLLM-Omni reuses the PyTorch Profiler infrastructure from vLLM.  
+For more advanced configuration options (memory profiling, custom activities, etc.), see the official vLLM profiler documentation:  [vLLM Profiling Guide](https://docs.vllm.ai/en/latest/dev/profiling.html)

docs/contributing/tests/tests_style.md

@@ -113,7 +113,7 @@ vllm_omni/                          tests/
 
 ### Naming Conventions
 
-- **Unit Tests**: Use `test_<module_name>.py` format. Example: `omni_llm.py` → `test_omni_llm.py`
+- **Unit/System Tests**: Use `test_<module_name>.py` format. Example: `omni_llm.py` → `test_omni_llm.py`
 
 - **E2E Tests**: Place in `tests/e2e/offline_inference/` or `tests/e2e/online_serving/` with descriptive names. Example: `tests/e2e/offline_inference/test_qwen3_omni.py`, `tests/e2e/offline_inference/test_diffusion_model.py`
 
@@ -139,7 +139,7 @@ vllm_omni/                          tests/
 4. **Documentation**: Add docstrings to all test functions
 5. **Environment variables**: Set uniformly in `conftest.py` or at the top of files
 6. **Type annotations**: Add type annotations to all test function parameters
-7. **Pytest Markers**: Add necessary markers like `@pytest.mark.core_model` and use `@hardware_test` to declare hardware requirements (check detailed in [Markers for Tests](../ci/tests_markers.md)).
+7. **Resources**, Using pytest tag to specify the computation resources the test required.
 
 ### Template
 #### E2E - Online serving
@@ -155,7 +155,6 @@ from pathlib import Path
 import pytest
 import openai
 
-from tests.utils import hardware_test
 
 # Optional: set process start method for workers
 os.environ["VLLM_WORKER_MULTIPROC_METHOD"] = "spawn"
@@ -185,12 +184,6 @@ def base64_encoded_video() -> str:
 def dummy_messages_from_video_data(video_data_url: str, content_text: str) -> str:
     xxx
 
-@pytest.mark.core_model
-@pytest.mark.omni
-@hardware_test(
-    res={"cuda": "L4", "rocm": "MI325", "npu": "A2"},
-    num_cards={"cuda": 2, "rocm": 2, "npu": 4},
-)
 @pytest.mark.parametrize("omni_server", test_params, indirect=True)
 def test_video_to_audio(
     client: openai.OpenAI,
@@ -233,7 +226,6 @@ from pathlib import Path
 import pytest
 from vllm.assets.video import VideoAsset
 
-from tests.utils import hardware_test
 from ..multi_stages.conftest import OmniRunner
 
 # Optional: set process start method for workers
@@ -247,12 +239,7 @@ test_params = [(model, stage_config) for model in models for stage_config in sta
 
 # function name: test_{input_modality}_to_{output_modality}
 # modality candidate: text, image, audio, video, mixed_modalities
-@pytest.mark.core_model
-@pytest.mark.omni
-@hardware_test(
-    res={"cuda": "L4", "rocm": "MI325", "npu": "A2"},
-    num_cards=2,
-)
+@pytest.mark.gpu_mem_high  # requires high-memory GPU node
 @pytest.mark.parametrize("test_config", test_params)
 def test_video_to_audio(omni_runner: type[OmniRunner], model: str) -> None:
     """Offline inference: video input, audio output."""
@@ -276,5 +263,4 @@ def test_video_to_audio(omni_runner: type[OmniRunner], model: str) -> None:
 
 1. The file is saved in an appropriate place and the file name is clear.
 2. The coding style follows the requirements outlined above.
-3. **All test functions have appropriate pytest markers**
-4. For tests that need run in CI, please ensure the test is configured under the `./buildkite/` folder.
+3. For e2e model test, please ensure the test is configured under the `./buildkite/` folder.

docs/design/architecture_overview.md

@@ -64,13 +64,13 @@ According to analysis for current popular open-source models, most of them have
 
 ## Key Components
 
-| Component         | Description                                                                                                                              |
-| ----------------- | ---------------------------------------------------------------------------------------------------------------------------------------- |
-| **OmniRouter**    | provide an intelligent router for Omni-modality requests dispatch                                                                        |
-| **EntryPoints**   | define the APIs for offline/online serving (APIServer, Omni/AsyncOmni) and provide the OmniStage abstraction for different AR/DiT stages |
-| **AR**            | adapted for omni-modality models while inheriting efficient features from vLLM, such as cache management                                 |
-| **Diffusion**     | natively implemented and optimized using acceleration components                                                                         |
-| **OmniConnector** | supports fully disaggregation based on E/P/D/G (Encoding/Processing/Decoding/Generation) disaggregation across stages                    |
+| Component | Description |
+|-----------|-------------|
+| **OmniRouter** | provide an intelligent router for Omni-modality requests dispatch |
+| **EntryPoints** | define the APIs for offline/online serving (APIServer, Omni/AsyncOmni) and provide the OmniStage abstraction for different AR/DiT stages |
+| **AR** | adapted for omni-modality models while inheriting efficient features from vLLM, such as cache management |
+| **Diffusion** | natively implemented and optimized using acceleration components |
+| **OmniConnector** | supports fully disaggregation based on E/P/D/G (Encoding/Processing/Decoding/Generation) disaggregation across stages |
 
 Disaggregated stages are managed through configuration, such as in the Qwen3-Omni example, where stages like Thinker, Talker, and Code2wav are defined as separate OmniStage instances with specific resources and input/output type.
 
@@ -192,4 +192,4 @@ curl -sS -X POST http://localhost:8091/v1/chat/completions \
 }
 ```
 
-For more usages, please refer to [examples](https://github.com/vllm-project/vllm-omni/tree/main/examples).
+For more usages, please refer to [examples](../user_guide/examples/).

docs/design/feature/disaggregated_inference.md

@@ -1,35 +1,27 @@
 # Disaggregated Inference for Omni-Modality Models
 
-This guide explains how to configure and use distributed connectors
-(`vllm_omni/distributed/omni_connectors`) in vllm-omni for multi-stage pipelines.
+This guide explains how to configure and use distributed connectors (vllm_omni/distributed/connectors) in vllm-omni for multi-stage pipelines.
 
-Backend-specific setup lives in separate docs:
-
-- [SharedMemoryConnector](omni_connectors/shared_memory_connector.md)
-- [MooncakeConnector](omni_connectors/mooncake_connector.md)
-- [YuanrongConnector](omni_connectors/yuanrong_connector.md)
-
-## Overview
+## 1. Overview
 
 Connectors enable data transfer between pipeline stages (e.g., Thinker -> Talker).
-Current connectors operate in D2H2D (device to host to device) mode.
+Currently supported connectors operate in **D2H2D (Device-to-Host-to-Device)** mode:
+1. **SharedMemoryConnector**: Uses system shared memory.
+2. **MooncakeConnector**: Uses [Mooncake](https://github.com/kvcache-ai/Mooncake).
 
-## Connector Choices
+*   **SharedMemoryConnector (Default)**: Zero-copy (on host), lowest latency. Best for **single-node** deployments. Auto-configured if no connectors are specified.
+*   **MooncakeConnector**: TCP/RDMA based. Best for **multi-node** distributed deployments. Requires a Mooncake Master service.
 
-| Use Case | Recommended Connector | Notes |
-| :--- | :--- | :--- |
-| Single node | SharedMemoryConnector | Auto-configured if no connector is specified. |
-| Multi node (Mooncake) | MooncakeConnector | Requires Mooncake Master + metadata server. |
-| Multi node (Yuanrong) | YuanrongConnector | Requires Yuanrong Datasystem + etcd. |
+## 2. API Design
 
-## Core API
+The connector system is built around the `OmniConnectorBase` abstraction, which decouples data transport from stage logic.
 
-The connector system is built around `OmniConnectorBase`.
+### Core Interface
 
 ```python
 class OmniConnectorBase(ABC):
     @abstractmethod
-    def put(self, from_stage: str, to_stage: str, put_key: str, data: Any) -> tuple[bool, int, Optional[dict]]:
+    def put(self, from_stage: str, to_stage: str, request_id: str, data: Any) -> tuple[bool, int, Optional[dict]]:
         """
         Store data.
         Returns: (success, serialized_size, metadata)
@@ -37,23 +29,145 @@ class OmniConnectorBase(ABC):
         pass
 
     @abstractmethod
-    def get(self, from_stage: str, to_stage: str, get_key: str, metadata: Optional[dict] = None) -> Optional[tuple[Any, int]]:
+    def get(self, from_stage: str, to_stage: str, request_id: str, metadata: Optional[dict] = None) -> Optional[tuple[Any, int]]:
         """
         Retrieve data.
-        Args: metadata - transport-specific handles returned by put() (e.g., SHM name).
+        Args: metadata - Transport-specific handles returned by put() (e.g., SHM name).
         Returns: (object, serialized_size)
         """
         pass
 ```
 
-### Metadata Passing
+### Key Concept: Metadata Passing
+Unlike a pure key-value store, some connectors (like `SharedMemoryConnector`) generate transient resources (e.g., a shared memory block name) during `put()`. This `metadata` **must be passed via the control plane** (e.g., HTTP headers, queue messages) from the producer stage to the consumer stage so `get()` can locate the data.
 
-Some connectors (e.g., SharedMemoryConnector) generate transient resources during `put()`.
-This `metadata` must be passed through the control plane so `get()` can locate the data.
+## 3. Backends & Use Cases
 
-## Configuration Model
+### 3.1 SharedMemoryConnector
+**Best for:** Single-node, high-performance IPC.
 
-Define connectors in runtime:
+*   **Mechanism:**
+    *   **Small Payloads (< Threshold)**: Data is serialized and passed directly "inline" within the `metadata` dictionary. This avoids the overhead of creating SHM blocks for tiny messages.
+    *   **Large Payloads (>= Threshold)**: Data is written to a named System Shared Memory block. The block name is returned in `metadata`.
+*   **Configuration:**
+    *   `shm_threshold_bytes`: Size in bytes to switch from inline to SHM (default: 64KB).
+
+### 3.2 MooncakeConnector
+**Best for:** Multi-node distributed inference.
+
+*   **Mechanism:** Uses Mooncake's distributed KVCache store.
+    *   **Data Plane**: TCP or RDMA for high-bandwidth transfer.
+    *   **Control Plane**: Uses a centralized Mooncake Master and Metadata Server.
+    *   **Keying**: Deterministic keys based on `request_id/from_stage_to_stage`.
+*   **Requirements**: Requires a running Mooncake Master service.
+
+## 4. Relationship with vLLM
+
+vLLM provides specialized distributed mechanisms for specific artifacts:
+*   **KV Transfer** (`vllm.distributed.kv_transfer`): Optimized for transferring KV caches between prefill and decode instances (using NCCL, Mooncake, etc.).
+*   **EC Transfer** (`vllm.distributed.ec_transfer`): Optimized for sharing encoder embeddings.
+*   **Device Communicators** (`vllm.distributed.device_communicators`): Low-level primitives (NCCL, SHM) for Tensor/Pipeline Parallelism.
+
+`vllm-omni` complements this by introducing a **Generalized Connector Abstraction** (`OmniConnector`) for multimodal pipelines. While vLLM's connectors are artifact-specific, `vllm-omni`:
+
+1.  **Unifies Transport**: Provides a single API (`put`/`get`) to transport *any* stage artifact (Input Embeddings, Hidden States, Audio/Image Tensors, KV Cache, Final Output) between arbitrary pipeline stages (e.g., AudioEncoder -> LLM -> AudioGenerator).
+2.  **Extends Connectivity**: Enables flexible construction of complex DAGs (Directed Acyclic Graphs) where stages can run in the same process, same node, or across nodes, using the most appropriate backend (SHM, Mooncake, etc.) for each edge.
+3.  **Wraps & Adapts**: Can internally utilize vLLM's specialized `kv_transfer` for KV paths while using generic transports (SHM/Mooncake) for other data types, presenting a consistent interface to the application layer.
+
+## 5. Installation (Mooncake)
+
+If using `MooncakeConnector`, install the library first:
+
+```bash
+# For CUDA-enabled systems (Recommended)
+pip install mooncake-transfer-engine
+
+# For non-CUDA systems
+pip install mooncake-transfer-engine-non-cuda
+```
+
+## 6. Using MooncakeConnector
+
+### 6.1 Start Mooncake Master
+
+Start the master service on your primary node:
+
+```bash
+# if you use mooncake SSD storage
+mkdir -p ./mc_storage
+
+mooncake_master \
+  --rpc_port=50051 \
+  --enable_http_metadata_server=true \
+  --http_metadata_server_host=0.0.0.0 \
+  --http_metadata_server_port=8080 \
+  --metrics_port=9003 \
+  --root_fs_dir=./mc_storage/ \
+  --cluster_id=mc-local-1 &
+```
+
+### 6.2 Configuration (YAML)
+
+Edit your stage config (e.g., `qwen2_5_omni.yaml`).
+
+**Step 1: Define Connector in Global Runtime**
+
+```yaml
+runtime:
+  connectors:
+    connector_of_mooncake:
+      name: MooncakeConnector
+      extra:
+        host: "127.0.0.1"           # Local Worker IP
+        metadata_server: "http://<MASTER_IP>:8080/metadata"
+        master: "<MASTER_IP>:50051"
+        segment: 512000000          # 512MB segment
+        localbuf: 64000000          # 64MB buffer
+        proto: "tcp"                # "tcp" or "rdma"
+    ```
+
+**Mooncake Configuration Parameters:**
+
+*   **host**: The hostname or IP address of the local machine (worker). Mooncake uses this to register itself in the metadata server so other nodes can find it.
+*   **metadata_server**: The URL of the metadata server. This is used for service discovery and connection establishment (e.g., exchanging QP information for RDMA).
+*   **master**: The address of the Mooncake Master Server (e.g., `<MASTER_IP>:50051`). This is used for global state management and control plane operations.
+*   **segment**: The size of the global memory segment in bytes (default: ~512MB). This defines the shared memory region accessible by Mooncake for data transfer.
+*   **localbuf**: The size of the local buffer in bytes (default: ~64MB). Used for local data buffering during transfer operations.
+*   **proto**: The transport protocol to use. Options:
+    *   `tcp`: Standard TCP/IP (easier setup, universal compatibility).
+    *   `rdma`: Remote Direct Memory Access (higher performance, requires RDMA-capable hardware).
+
+For more details, refer to the [Mooncake Repository](https://github.com/kvcache-ai/Mooncake).
+
+    **Step 2: Reference in Stages**
+
+Explicitly link stages using `input_connectors` and `output_connectors`:
+
+```yaml
+stage_args:
+  - stage_id: 0
+    # ...
+    output_connectors:
+      to_stage_1: connector_of_mooncake
+
+  - stage_id: 1
+    # ...
+    input_connectors:
+      from_stage_0: connector_of_mooncake
+```
+
+## 7. Using SharedMemoryConnector (Auto-Mode)
+
+**Best for single-node.**
+
+The system will automatically create `SharedMemoryConnector`s for any pipeline edge that does not have an explicit connector defined. This is inferred from:
+1.  `runtime.edges` list in the config.
+2.  `engine_input_source` dependencies defined in `stage_args`.
+
+### Threshold Configuration
+By default, payloads larger than **64KB** (default threshold) are transferred via shared memory, while smaller ones use the control queue (inline).
+
+To adjust this threshold (e.g., to 1GB), add the following to your `runtime.connectors`:
 
 ```yaml
 runtime:
@@ -61,46 +175,26 @@ runtime:
     connector_of_shared_memory:
       name: SharedMemoryConnector
       extra:
-        shm_threshold_bytes: 65536
+        shm_threshold_bytes: 1024 # 1KB threshold
 ```
 
-Wire stages to connectors:
+## 8. Summary
 
-```yaml
-stage_args:
-  - stage_id: 0
-    output_connectors:
-      to_stage_1: connector_of_shared_memory
+| Use Case | Recommended Connector | Configuration |
+| :--- | :--- | :--- |
+| **Single Node** | `SharedMemoryConnector` | **None** (Automatic) or Custom Threshold |
+| **Multi Node** | `MooncakeConnector` | Explicit YAML + Mooncake Master |
 
-  - stage_id: 1
-    input_connectors:
-      from_stage_0: connector_of_shared_memory
-```
+## 9. Operational Notes (important)
 
-If a pipeline edge has no explicit connector, the system auto-creates a
-SharedMemoryConnector for that edge.
+- **Fail-fast config validation**: the loader raises if any expected edge is missing a connector. Define `input_connectors`/`output_connectors` or rely on auto-SHM filling; otherwise startup aborts.
+- **Missing payloads halt the stage**: workers expect connector payloads; if metadata or connector config is missing, the stage raises and stops. Verify connector wiring and metadata propagation before production.
 
-## Relationship with vLLM
+## 10. Future Roadmap: Device-to-Device (D2D) Transport
 
-vLLM provides specialized distributed mechanisms for specific artifacts:
+The current implementations (`SharedMemoryConnector`, `MooncakeConnector`) utilize a **D2H2D (Device-to-Host-to-Device)** data path. Tensors are moved to CPU memory (Host) for transport, which incurs PCIe overhead.
 
-- KV Transfer (`vllm.distributed.kv_transfer`): optimized for KV caches.
-- EC Transfer (`vllm.distributed.ec_transfer`): optimized for encoder embeddings.
-- Device Communicators (`vllm.distributed.device_communicators`): low-level primitives (NCCL, SHM).
+As outlined in the design RFC, future versions will introduce **D2D (Device-to-Device)** connectors:
 
-vllm-omni complements this with a generalized connector abstraction:
-
-1. Unifies transport via a single `put`/`get` API for any stage artifact.
-2. Enables DAG-style pipelines across processes or nodes with per-edge transports.
-3. Can wrap vLLM-specific transfers for KV paths while keeping a consistent interface.
-
-## Operational Notes
-
-- Fail-fast config validation: missing expected edges cause startup failures.
-- Missing payloads halt stages: verify connector wiring and metadata propagation.
-
-## Future Roadmap: D2D Transport
-
-Current connectors use D2H2D paths. Future versions will introduce direct
-device-to-device connectors (NCCL, UCX, IPC) to reduce latency for large
-tensor payloads.
+*   **Goal**: Direct GPU-to-GPU transfer (via NCCL, UCX, or IPC) to minimize latency for large tensor payloads.
+*   **Mechanism**: The `OmniConnector` API allows `put()` to initiate a transfer and return a lightweight handle (metadata) via the control plane, while the heavy payload flows directly between devices.

docs/design/feature/omni_connectors/mooncake_connector.md

@@ -1,74 +0,0 @@
-# MooncakeConnector
-
-## When to Use
-
-Best for multi-node distributed inference using Mooncake.
-
-## Installation
-
-```bash
-# For CUDA-enabled systems (recommended)
-pip install mooncake-transfer-engine
-
-# For non-CUDA systems
-pip install mooncake-transfer-engine-non-cuda
-```
-
-## Start Mooncake Master
-
-```bash
-# If you use Mooncake SSD storage
-mkdir -p ./mc_storage
-
-mooncake_master \
-  --rpc_port=50051 \
-  --enable_http_metadata_server=true \
-  --http_metadata_server_host=0.0.0.0 \
-  --http_metadata_server_port=8080 \
-  --metrics_port=9003 \
-  --root_fs_dir=./mc_storage/ \
-  --cluster_id=mc-local-1 &
-```
-
-## Configuration
-
-Define the connector in runtime:
-
-```yaml
-runtime:
-  connectors:
-    connector_of_mooncake:
-      name: MooncakeConnector
-      extra:
-        host: "127.0.0.1"
-        metadata_server: "http://<MASTER_IP>:8080/metadata"
-        master: "<MASTER_IP>:50051"
-        segment: 512000000
-        localbuf: 64000000
-        proto: "tcp"
-```
-
-Wire stages to the connector:
-
-```yaml
-stage_args:
-  - stage_id: 0
-    output_connectors:
-      to_stage_1: connector_of_mooncake
-
-  - stage_id: 1
-    input_connectors:
-      from_stage_0: connector_of_mooncake
-```
-
-Parameters:
-
-- host: local worker IP registered in the metadata server.
-- metadata_server: metadata server URL for discovery and setup.
-- master: Mooncake Master address.
-- segment: global memory segment size in bytes.
-- localbuf: local buffer size in bytes.
-- proto: transport protocol ("tcp" or "rdma").
-
-For more details, refer to the
-[Mooncake repository](https://github.com/kvcache-ai/Mooncake).

docs/design/feature/omni_connectors/shared_memory_connector.md

@@ -1,27 +0,0 @@
-# SharedMemoryConnector
-
-## When to Use
-
-Best for single-node deployments where stages run on the same host. It is
-auto-configured when no explicit connector is specified for an edge.
-
-## How It Works
-
-- Small payloads (< threshold): serialized and passed inline in metadata.
-- Large payloads (>= threshold): stored in shared memory; the SHM name is
-  returned in metadata.
-
-## Configuration
-
-```yaml
-runtime:
-  connectors:
-    connector_of_shared_memory:
-      name: SharedMemoryConnector
-      extra:
-        shm_threshold_bytes: 65536
-```
-
-## Notes
-
-- Auto-mode uses SharedMemoryConnector if no connector is declared for an edge.

docs/design/feature/omni_connectors/yuanrong_connector.md

@@ -1,101 +0,0 @@
-# YuanrongConnector
-
-## When to Use
-
-Best for multi-node distributed inference using Yuanrong Datasystem.
-
-## Mechanism
-
-Uses Yuanrong Datasystem's distributed KV store (`datasystem.kv_client`).
-
-- Data Plane: TCP or RDMA for high-bandwidth transfer.
-- Control Plane: Yuanrong Datasystem workers and etcd.
-- Keying: deterministic keys based on `put_key` (often composed as `request_id:fromStage_toStage`).
-
-## Installation
-
-```bash
-pip install openyuanrong-datasystem
-```
-
-## Start etcd
-
-```bash
-# Download and install etcd (v3.5.12 or higher)
-ETCD_VERSION="v3.5.12"
-ETCD_ARCH="linux-arm64"
-wget https://github.com/etcd-io/etcd/releases/download/${ETCD_VERSION}/etcd-${ETCD_VERSION}-${ETCD_ARCH}.tar.gz
-tar -xvf etcd-${ETCD_VERSION}-${ETCD_ARCH}.tar.gz
-cd etcd-${ETCD_VERSION}-${ETCD_ARCH}
-sudo cp etcd etcdctl /usr/local/bin/
-
-# Start etcd
-etcd \
-  --name etcd-single \
-  --data-dir /tmp/etcd-data \
-  --listen-client-urls http://0.0.0.0:2379 \
-  --advertise-client-urls http://0.0.0.0:2379 \
-  --listen-peer-urls http://0.0.0.0:2380 \
-  --initial-advertise-peer-urls http://0.0.0.0:2380 \
-  --initial-cluster etcd-single=http://0.0.0.0:2380 &
-
-# Verify etcd is running
-etcdctl --endpoints "127.0.0.1:2379" put key "value"
-etcdctl --endpoints "127.0.0.1:2379" get key
-```
-
-For production environments, refer to the
-[official etcd clustering documentation](https://etcd.io/docs/current/op-guide/clustering/).
-
-## Start Datasystem Worker
-
-```bash
-# Replace ${ETCD_IP} with etcd node IP, ${WORKER_IP} with local node IP
-dscli start -w \
-  --worker_address "${WORKER_IP}:31501" \
-  --etcd_address "${ETCD_IP}:2379" \
-  --shared_memory_size_mb 20480
-```
-
-To stop the worker:
-
-```bash
-dscli stop --worker_address "${WORKER_IP}:31501"
-```
-
-## Configuration
-
-Define the connector in runtime:
-
-```yaml
-runtime:
-  connectors:
-    connector_of_yuanrong:
-      name: YuanrongConnector
-      extra:
-        host: "127.0.0.1"
-        port: 31501
-        get_sub_timeout_ms: 1000
-```
-
-Wire stages to the connector:
-
-```yaml
-stage_args:
-  - stage_id: 0
-    output_connectors:
-      to_stage_1: connector_of_yuanrong
-
-  - stage_id: 1
-    input_connectors:
-      from_stage_0: connector_of_yuanrong
-```
-
-Parameters:
-
-- host: datasystem worker host.
-- port: datasystem worker port.
-- get_sub_timeout_ms: get timeout in milliseconds (0 for no timeout).
-
-For more details, refer to the
-[Yuanrong Datasystem repository](https://atomgit.com/openeuler/yuanrong-datasystem).

docs/features/sleep_mode.md

@@ -1,39 +0,0 @@
-# Sleep Mode
-
-vLLM-Omni’s **Sleep Mode** allows you to temporarily release most GPU memory used by a model—such as model weights and key-value (KV) caches (for autoregressive models)—**without stopping the server or unloading the Docker container**.
-
-This feature is inherited from [vLLM’s Sleep Mode](https://blog.vllm.ai/2025/10/26/sleep-mode.html), which provides zero-reload model switching for multi-model serving.  
-
-It is especially useful in **RLHF**, **training**, or **cost-saving scenarios**, where GPU resources must be freed between inference workloads.
-
----
-
-## Omni Model
-
-Omni model inherit the feature from vLLM' Sleep Mode
-
-This means:
-
-- Support both Level 1 and Level 2 sleep, allow to release and reset both model weights and KV Cache
-
-## Diffusion Model Extension
-
-We added Sleep Mode support for **diffusion models**, which previously lacked this functionality.  
-In diffusion pipelines, this currently only offloads **model weight memory**, as these models typically do not use KV caches.
-
-This means:
-
-- Diffusion models can now enter Level 1 sleep.
-- Pipeline states (e.g., noise schedulers, buffers) remain intact after waking.
-- Useful for releasing VRAM between image generation or training cycles.
-
----
-
-## Enable sleep mode
-To enable sleep mode, set the `enable_sleep_mode` in `engine_args` to `True`
-
-
-Example:
-```python
-omni = Omni(model=...,enable_sleep_mode=True)
-```

docs/getting_started/installation/gpu.md

@@ -49,18 +49,15 @@ vLLM-Omni is a Python library that supports the following GPU variants. The libr
 
 ## Set up using Docker
 
-### Pre-built images
-
-=== "NVIDIA CUDA"
-
-    --8<-- "docs/getting_started/installation/gpu/cuda.inc.md:pre-built-images"
-
-=== "AMD ROCm"
-
-    --8<-- "docs/getting_started/installation/gpu/rocm.inc.md:pre-built-images"
 
 ### Build your own docker image
 
 === "AMD ROCm"
 
     --8<-- "docs/getting_started/installation/gpu/rocm.inc.md:build-docker"
+
+### Pre-built images
+
+=== "NVIDIA CUDA"
+
+    --8<-- "docs/getting_started/installation/gpu/cuda.inc.md:pre-built-images"

docs/getting_started/installation/gpu/cuda.inc.md

@@ -17,12 +17,12 @@ Therefore, it is recommended to install vLLM and vLLM-Omni with a **fresh new**
 # --8<-- [start:pre-built-wheels]
 
 #### Installation of vLLM
-Note: Pre-built wheels are currently only available for vLLM-Omni 0.11.0rc1, 0.12.0rc1, 0.14.0rc1. For the latest version, please [build from source](https://docs.vllm.ai/projects/vllm-omni/en/latest/getting_started/installation/gpu/#build-wheel-from-source).
+Note: Pre-built wheels are currently only available for vLLM-Omni 0.11.0rc1 and 0.12.0rc1. For the latest version, please [build from source](https://docs.vllm.ai/projects/vllm-omni/en/latest/getting_started/installation/gpu/#build-wheel-from-source).
 
 
 vLLM-Omni is built based on vLLM. Please install it with command below.
 ```bash
-uv pip install vllm==0.14.0 --torch-backend=auto
+uv pip install vllm==0.12.0 --torch-backend=auto
 ```
 
 #### Installation of vLLM-Omni
@@ -36,13 +36,13 @@ uv pip install vllm-omni
 # --8<-- [start:build-wheel-from-source]
 
 #### Installation of vLLM
-If you do not need to modify source code of vLLM, you can directly install the stable 0.14.0 release version of the library
+If you do not need to modify source code of vLLM, you can directly install the stable 0.12.0 release version of the library
 
 ```bash
-uv pip install vllm==0.14.0 --torch-backend=auto
+uv pip install vllm==0.12.0 --torch-backend=auto
 ```
 
-The release 0.14.0 of vLLM is based on PyTorch 2.9.0 which requires CUDA 12.9 environment.
+The release 0.12.0 of vLLM is based on PyTorch 2.9.0 which requires CUDA 12.9 environment.
 
 #### Installation of vLLM-Omni
 Since vllm-omni is rapidly evolving, it's recommended to install it from source
@@ -58,11 +58,11 @@ If you want to check, modify or debug with source code of vLLM, install the libr
 ```bash
 git clone https://github.com/vllm-project/vllm.git
 cd vllm
-git checkout v0.14.0
+git checkout v0.12.0
 ```
 Set up environment variables to get pre-built wheels. If there are internet problems, just download the whl file manually. And set `VLLM_PRECOMPILED_WHEEL_LOCATION` as your local absolute path of whl file.
 ```bash
-export VLLM_PRECOMPILED_WHEEL_LOCATION=https://github.com/vllm-project/vllm/releases/download/v0.14.0/vllm-0.14.0-cp38-abi3-manylinux_2_31_x86_64.whl
+export VLLM_PRECOMPILED_WHEEL_LOCATION=https://github.com/vllm-project/vllm/releases/download/v0.12.0/vllm-0.12.0-cp38-abi3-manylinux_2_31_x86_64.whl
 ```
 Install vllm with command below (If you have no existing PyTorch).
 ```bash
@@ -93,7 +93,7 @@ docker run --runtime nvidia --gpus 2 \
     --env "HF_TOKEN=$HF_TOKEN" \
     -p 8091:8091 \
     --ipc=host \
-    vllm/vllm-omni:v0.14.0rc1 \
+    vllm/vllm-omni:v0.12.0rc1 \
     --model Qwen/Qwen3-Omni-30B-A3B-Instruct --port 8091
 ```
 

docs/getting_started/installation/gpu/rocm.inc.md

@@ -23,30 +23,19 @@ vLLM-Omni current recommends the steps in under setup through Docker Images.
 DOCKER_BUILDKIT=1 docker build -f docker/Dockerfile.rocm -t vllm-omni-rocm .
 ```
 
+If you want to specify which GPU Arch to build for to cutdown build time:
+
+```bash
+DOCKER_BUILDKIT=1 docker build \
+  -f docker/Dockerfile.rocm \
+  --build-arg PYTORCH_ROCM_ARCH="gfx942;gfx950" \
+  -t vllm-omni-rocm .
+```
+
 #### Launch the docker image
 
-##### Launch with OpenAI API Server
-
 ```
-docker run --rm \
---group-add=video \
---ipc=host \
---cap-add=SYS_PTRACE \
---security-opt seccomp=unconfined \
---device /dev/kfd \
---device /dev/dri \
--v ~/.cache/huggingface:/root/.cache/huggingface \
---env "HF_TOKEN=$HF_TOKEN" \
--p 8091:8091 \
---ipc=host \
-vllm-omni-rocm \
---model Qwen/Qwen3-Omni-30B-A3B-Instruct --port 8091
-```
-
-##### Launch with interactive session for development
-
-```
-docker run --rm -it \
+docker run -it \
 --network=host \
 --group-add=video \
 --ipc=host \
@@ -55,51 +44,12 @@ docker run --rm -it \
 --device /dev/kfd \
 --device /dev/dri \
 -v <path/to/model>:/app/model \
--v ~/.cache/huggingface:/root/.cache/huggingface \
---entrypoint bash \
-vllm-omni-rocm
+vllm-omni-rocm \
+bash
 ```
 
 # --8<-- [end:build-docker]
 
 # --8<-- [start:pre-built-images]
 
-vLLM-Omni offers an official docker image for deployment. These images are built on top of vLLM docker images and available on Docker Hub as [vllm/vllm-omni-rocm](https://hub.docker.com/r/vllm/vllm-omni-rocm/tags). The version of vLLM-Omni indicates which release of vLLM it is based on.
-
-#### Launch vLLM-Omni Server
-Here's an example deployment command that has been verified on 2 x MI300's:
-```bash
-docker run --rm \
-  --group-add=video \
-  --ipc=host \
-  --cap-add=SYS_PTRACE \
-  --security-opt seccomp=unconfined \
-  --device /dev/kfd \
-  --device /dev/dri \
-  -v <path/to/model>:/app/model \
-  -v ~/.cache/huggingface:/root/.cache/huggingface \
-  --env "HF_TOKEN=$HF_TOKEN" \
-  -p 8091:8091 \
-  vllm/vllm-omni-rocm:v0.14.0rc1 \
-  --model Qwen/Qwen3-Omni-30B-A3B-Instruct --omni --port 8091
-```
-
-#### Launch an interactive terminal with prebuilt docker image.
-If you want to run in dev environment you can launch the docker image as follows:
-```bash
-docker run --rm -it \
-  --network=host \
-  --group-add=video \
-  --ipc=host \
-  --cap-add=SYS_PTRACE \
-  --security-opt seccomp=unconfined \
-  --device /dev/kfd \
-  --device /dev/dri \
-  -v <path/to/model>:/app/model \
-  -v ~/.cache/huggingface:/root/.cache/huggingface \
-  --env "HF_TOKEN=$HF_TOKEN" \
-  --entrypoint bash \
-  vllm/vllm-omni-rocm:v0.14.0rc1
-```
-
 # --8<-- [end:pre-built-images]

docs/getting_started/installation/npu/npu.inc.md

@@ -13,10 +13,10 @@ export DEVICE0=/dev/davinci0
 export DEVICE1=/dev/davinci1
 # Update the vllm-ascend image
 # Atlas A2:
-# export IMAGE=quay.io/ascend/vllm-ascend:v0.14.0
+# export IMAGE=quay.io/ascend/vllm-ascend:v0.12.0rc1
 # Atlas A3:
-# export IMAGE=quay.io/ascend/vllm-ascend:v0.14.0-a3
-export IMAGE=quay.io/ascend/vllm-ascend:v0.14.0
+# export IMAGE=quay.io/ascend/vllm-ascend:v0.12.0rc1-a3
+export IMAGE=quay.io/ascend/vllm-ascend:v0.12.0rc1
 docker run --rm \
     --name vllm-omni-npu \
     --shm-size=1g \
@@ -42,7 +42,7 @@ source ~/.bashrc
 
 # Inside the container, install vLLM-Omni from source
 cd /vllm-workspace
-git clone -b v0.14.0 https://github.com/vllm-project/vllm-omni.git
+git clone -b v0.12.0rc1 https://github.com/vllm-project/vllm-omni.git
 cd vllm-omni
 pip install -v -e .
 export VLLM_WORKER_MULTIPROC_METHOD=spawn

docs/getting_started/quickstart.md

@@ -17,13 +17,7 @@ For installation on GPU from source:
 ```bash
 uv venv --python 3.12 --seed
 source .venv/bin/activate
-
-# On CUDA
-uv pip install vllm==0.14.0 --torch-backend=auto
-
-# On ROCm
-uv pip install vllm==0.14.0 --extra-index-url https://wheels.vllm.ai/rocm/0.14.0/rocm700
-
+uv pip install vllm==0.12.0 --torch-backend=auto
 git clone https://github.com/vllm-project/vllm-omni.git
 cd vllm-omni
 uv pip install -e .
@@ -46,46 +40,6 @@ if __name__ == "__main__":
     images[0].save("coffee.png")
 ```
 
-You can pass a list of prompts and wait for them to process altogether, shown below.
-
-!!! info
-
-    However, it is not currently recommended to do so
-    because not all models support batch inference,
-    and batch requesting mostly does not provide significant performance improvement (despite the impression that it does).
-    This feature is primarily for the sake of interface compatibility with vLLM and to allow for future improvements.
-
-```python
-from vllm_omni.entrypoints.omni import Omni
-
-if __name__ == "__main__":
-    omni = Omni(
-        model="Tongyi-MAI/Z-Image-Turbo",
-        # stage_configs_path="./stage-config.yaml",  # See below
-    )
-    prompts = [
-        "a cup of coffee on a table",
-        "a toy dinosaur on a sandy beach",
-        "a fox waking up in bed and yawning",
-    ]
-    omni_outputs = omni.generate(prompts)
-    for i_prompt, prompt_output in enumerate(omni_outputs):
-        this_request_output = prompt_output.request_output[0]
-        this_images = this_request_output.images
-        for i_image, image in enumerate(this_images):
-            image.save(f"p{i_prompt}-img{i_image}.jpg")
-            print("saved to", f"p{i_prompt}-img{i_image}.jpg")
-            # saved to p0-img0.jpg
-            # saved to p1-img0.jpg
-            # saved to p2-img0.jpg
-```
-
-!!! info
-
-    For diffusion pipelines, the stage config field `stage_args.[].runtime.max_batch_size` is 1 by default, and the input
-    list is sliced into single-item requests before feeding into the diffusion pipeline. For models that do internally support
-    batched inputs, you can [modify this configuration](../configuration/stage_configs.md) to let the model accept a longer batch of prompts.
-
 For more usages, please refer to [offline inference](../user_guide/examples/offline_inference/qwen2_5_omni.md)
 
 ## Online Serving with OpenAI-Completions API

docs/mkdocs/hooks/generate_api_readme.py

@@ -34,10 +34,10 @@ CATEGORIES = {
         "name": "Core",
         "description": "Core scheduling and caching components.",
     },
-    # "model_executor": {
-    #     "name": "Model Executor",
-    #     "description": "Model execution components.",
-    # },
+    "model_executor": {
+        "name": "Model Executor",
+        "description": "Model execution components.",
+    },
     "config": {
         "name": "Configuration",
         "description": "Configuration classes.",

docs/models/supported_models.md

@@ -21,7 +21,6 @@ th {
 | `Qwen2_5OmniForConditionalGeneration` | Qwen2.5-Omni | `Qwen/Qwen2.5-Omni-7B`, `Qwen/Qwen2.5-Omni-3B` |
 | `BagelForConditionalGeneration` | BAGEL (DiT-only) | `ByteDance-Seed/BAGEL-7B-MoT` |
 | `QwenImagePipeline` | Qwen-Image | `Qwen/Qwen-Image` |
-| `QwenImagePipeline` | Qwen-Image-2512 | `Qwen/Qwen-Image-2512` |
 | `QwenImageEditPipeline` | Qwen-Image-Edit | `Qwen/Qwen-Image-Edit` |
 | `QwenImageEditPlusPipeline` | Qwen-Image-Edit-2509 | `Qwen/Qwen-Image-Edit-2509` |
 | `QwenImageLayeredPipeline` | Qwen-Image-Layered | `Qwen/Qwen-Image-Layered` |
@@ -32,12 +31,6 @@ th {
 |`LongcatImagePipeline` | LongCat-Image | `meituan-longcat/LongCat-Image` |
 |`LongCatImageEditPipeline` | LongCat-Image-Edit | `meituan-longcat/LongCat-Image-Edit` |
 |`StableDiffusion3Pipeline` | Stable-Diffusion-3 | `stabilityai/stable-diffusion-3.5-medium` |
-|`Flux2KleinPipeline` | FLUX.2-klein | `black-forest-labs/FLUX.2-klein-4B`, `black-forest-labs/FLUX.2-klein-9B` |
-|`FluxPipeline` | FLUX.1-dev | `black-forest-labs/FLUX.1-dev` |
-|`StableAudioPipeline` | Stable-Audio-Open | `stabilityai/stable-audio-open-1.0` |
-|`Qwen3TTSForConditionalGeneration` | Qwen3-TTS-12Hz-1.7B-CustomVoice | `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-0.6B-Base` |
 
 
 ## List of Supported Models for NPU

docs/user_guide/acceleration/cache_dit_acceleration.md

@@ -18,7 +18,6 @@ Enable cache-dit acceleration by simply setting `cache_backend="cache_dit"`. Cac
 
 ```python
 from vllm_omni.entrypoints.omni import Omni
-from vllm_omni.inputs.data import OmniDiffusionSamplingParams
 
 # Simplest way: just enable cache-dit with default parameters
 omni = Omni(
@@ -28,7 +27,7 @@ omni = Omni(
 
 images = omni.generate(
     "a beautiful landscape",
-    OmniDiffusionSamplingParams(num_inference_steps=50),
+    num_inference_steps=50,
 )
 ```
 
@@ -70,7 +69,7 @@ vllm serve Qwen/Qwen-Image --omni --port 8091 \
 
 ## Acceleration Methods
 
-For comprehensive illustration, please view cache-dit [User_Guide](https://cache-dit.readthedocs.io/en/latest/user_guide/OVERVIEWS/)
+For comprehensive illustration, please view cache-dit [User_Guide](https://github.com/vipshop/cache-dit/blob/main/docs/User_Guide.md)
 
 ### 1. DBCache (Dual Block Cache)
 
@@ -223,6 +222,6 @@ To test another model, you can modify `--model` with the target model identifier
 
 ## Additional Resources
 
-- [Cache-DiT User Guide](https://cache-dit.readthedocs.io/en/latest/user_guide/OVERVIEWS/)
-- [Cache-DiT Benchmark](https://cache-dit.readthedocs.io/en/latest/benchmark/HYBRID_CACHE/)
-- [DBCache Technical Details](https://cache-dit.readthedocs.io/en/latest/user_guide/CACHE_API/)
+- [Cache-DiT User Guide](https://github.com/vipshop/cache-dit/blob/main/docs/User_Guide.md)
+- [Cache-DiT Benchmark](https://github.com/vipshop/cache-dit/tree/main/bench)
+- [DBCache Technical Details](https://github.com/vipshop/cache-dit/blob/main/docs/DBCache.md)

docs/user_guide/acceleration/parallelism_acceleration.md

@@ -0,0 +1,274 @@
+# Parallelism Acceleration Guide
+
+This guide includes how to use parallelism methods in vLLM-Omni to speed up diffusion model inference as well as reduce the memory requirement on each device.
+
+## Overview
+
+The following parallelism methods are currently supported in vLLM-Omni:
+
+1. DeepSpeed Ulysses Sequence Parallel (DeepSpeed Ulysses-SP) ([arxiv paper](https://arxiv.org/pdf/2309.14509)): Ulysses-SP splits the input along the sequence dimension and uses all-to-all communication to allow each device to compute only a subset of attention heads.
+
+2. [Ring-Attention](#ring-attention) - splits the input along the sequence dimension and uses ring-based P2P communication to accumulate attention results, keeping the sequence dimension sharded
+
+
+The following table shows which models are currently supported by parallelism method:
+
+### ImageGen
+
+| Model | Model Identifier | Ulysses-SP | Ring-SP |
+|-------|------------------|-----------|---------|
+| **LongCat-Image** | `meituan-longcat/LongCat-Image` | ❌ | ❌ |
+| **LongCat-Image-Edit** | `meituan-longcat/LongCat-Image-Edit` | ❌ | ❌ |
+| **Ovis-Image** | `OvisAI/Ovis-Image` | ❌ | ❌ |
+| **Qwen-Image** | `Qwen/Qwen-Image` | ✅ | ✅ |
+| **Qwen-Image-Edit** | `Qwen/Qwen-Image-Edit` | ✅ | ✅ |
+| **Qwen-Image-Edit-2509** | `Qwen/Qwen-Image-Edit-2509` | ✅ | ✅ |
+| **Qwen-Image-Layered** | `Qwen/Qwen-Image-Layered` | ✅ | ✅ |
+| **Z-Image** | `Tongyi-MAI/Z-Image-Turbo` | ❌ | ❌ |
+
+### VideoGen
+
+| Model | Model Identifier | Ulysses-SP | Ring-SP |
+|-------|------------------|-----------|---------|
+| **Wan2.2** | `Wan-AI/Wan2.2-T2V-A14B-Diffusers` | ❌ | ❌ |
+
+### Sequence Parallelism
+
+#### Ulysses-SP
+
+##### Offline Inference
+
+An example of offline inference script using [Ulysses-SP](https://arxiv.org/pdf/2309.14509) is shown below:
+```python
+from vllm_omni import Omni
+from vllm_omni.diffusion.data import DiffusionParallelConfig
+ulysses_degree = 2
+
+omni = Omni(
+    model="Qwen/Qwen-Image",
+    parallel_config=DiffusionParallelConfig(ulysses_degree=2)
+)
+
+outputs = omni.generate(prompt="A cat sitting on a windowsill", num_inference_steps=50, width=2048, height=2048)
+```
+
+See `examples/offline_inference/text_to_image/text_to_image.py` for a complete working example.
+
+##### Online Serving
+
+You can enable Ulysses-SP in online serving for diffusion models via `--usp`:
+
+```bash
+# Text-to-image (requires >= 2 GPUs)
+vllm serve Qwen/Qwen-Image --omni --port 8091 --usp 2
+```
+
+##### Benchmarks
+!!! note "Benchmark Disclaimer"
+    These benchmarks are provided for **general reference only**. The configurations shown use default or common parameter settings and have not been exhaustively optimized for maximum performance. Actual performance may vary based on:
+
+    - Specific model and use case
+    - Hardware configuration
+    - Careful parameter tuning
+    - Different inference settings (e.g., number of steps, image resolution)
+
+
+To measure the parallelism methods, we run benchmarks with **Qwen/Qwen-Image** model generating images (**2048x2048** as long sequence input) with 50 inference steps. The hardware devices are NVIDIA H800 GPUs. `sdpa` is the attention backends.
+
+| Configuration | Ulysses degree |Generation Time | Speedup |
+|---------------|----------------|---------|---------|
+| **Baseline (diffusers)** | - | 112.5s | 1.0x |
+| Ulysses-SP  |  2  |  65.2s | 1.73x |
+| Ulysses-SP  |  4  | 39.6s | 2.84x |
+| Ulysses-SP  |  8  | 30.8s | 3.65x |
+
+#### Ring-Attention
+
+Ring-Attention ([arxiv paper](https://arxiv.org/abs/2310.01889)) splits the input along the sequence dimension and uses ring-based P2P communication to accumulate attention results. Unlike Ulysses-SP which uses all-to-all communication, Ring-Attention keeps the sequence dimension sharded throughout the computation and circulates Key/Value blocks through a ring topology.
+
+##### Offline Inference
+
+An example of offline inference script using Ring-Attention is shown below:
+```python
+from vllm_omni import Omni
+from vllm_omni.diffusion.data import DiffusionParallelConfig
+ring_degree = 2
+
+omni = Omni(
+    model="Qwen/Qwen-Image",
+    parallel_config=DiffusionParallelConfig(ring_degree=2)
+)
+
+outputs = omni.generate(prompt="A cat sitting on a windowsill", num_inference_steps=50, width=2048, height=2048)
+```
+
+See `examples/offline_inference/text_to_image/text_to_image.py` for a complete working example.
+
+
+##### Online Serving
+
+You can enable Ring-Attention in online serving for diffusion models via `--ring`:
+
+```bash
+# Text-to-image (requires >= 2 GPUs)
+vllm serve Qwen/Qwen-Image --omni --port 8091 --ring 2
+```
+
+##### Benchmarks
+!!! note "Benchmark Disclaimer"
+    These benchmarks are provided for **general reference only**. The configurations shown use default or common parameter settings and have not been exhaustively optimized for maximum performance. Actual performance may vary based on:
+
+    - Specific model and use case
+    - Hardware configuration
+    - Careful parameter tuning
+    - Different inference settings (e.g., number of steps, image resolution)
+
+
+To measure the parallelism methods, we run benchmarks with **Qwen/Qwen-Image** model generating images (**1024x1024** as long sequence input) with 50 inference steps. The hardware devices are NVIDIA A100 GPUs. `flash_attn` is the attention backends.
+
+| Configuration | Ring degree |Generation Time | Speedup |
+|---------------|----------------|---------|---------|
+| **Baseline (diffusers)** | - | 45.2s | 1.0x |
+| Ring-Attention  |  2  |  29.9s | 1.51x |
+| Ring-Attention  |  4  | 23.3s | 1.94x |
+
+
+#### Hybrid Ulysses + Ring
+
+You can combine both Ulysses-SP and Ring-Attention for larger scale parallelism. The total sequence parallel size equals `ulysses_degree × ring_degree`.
+
+##### Offline Inference
+
+```python
+from vllm_omni import Omni
+from vllm_omni.diffusion.data import DiffusionParallelConfig
+
+# Hybrid: 2 Ulysses × 2 Ring = 4 GPUs total
+omni = Omni(
+    model="Qwen/Qwen-Image",
+    parallel_config=DiffusionParallelConfig(ulysses_degree=2, ring_degree=2)
+)
+
+outputs = omni.generate(prompt="A cat sitting on a windowsill", num_inference_steps=50, width=2048, height=2048)
+```
+
+##### Online Serving
+
+```bash
+# Text-to-image (requires >= 4 GPUs)
+vllm serve Qwen/Qwen-Image --omni --port 8091 --usp 2 --ring 2
+```
+
+##### Benchmarks
+!!! note "Benchmark Disclaimer"
+    These benchmarks are provided for **general reference only**. The configurations shown use default or common parameter settings and have not been exhaustively optimized for maximum performance. Actual performance may vary based on:
+
+    - Specific model and use case
+    - Hardware configuration
+    - Careful parameter tuning
+    - Different inference settings (e.g., number of steps, image resolution)
+
+
+To measure the parallelism methods, we run benchmarks with **Qwen/Qwen-Image** model generating images (**1024x1024** as long sequence input) with 50 inference steps. The hardware devices are NVIDIA A100 GPUs. `flash_attn` is the attention backends.
+
+| Configuration | Ulysses degree | Ring degree | Generation Time | Speedup |
+|---------------|----------------|-------------|-----------------|---------|
+| **Baseline (diffusers)** | - | - | 45.2s | 1.0x |
+| Hybrid Ulysses + Ring  |  2  |  2  |  24.3s | 1.87x |
+
+
+##### How to parallelize a new model
+
+If a diffusion model has been deployed in vLLM-Omni and supports single-card inference, you can refer to the following instructions to parallelize it with [Ulysses-SP](https://arxiv.org/pdf/2309.14509).
+
+This section uses **Qwen-Image** (`QwenImageTransformer2DModel`) as the reference implementation. Qwen-Image is a **dual-stream** transformer (text + image) that performs **joint attention** across the concatenated sequences. Because of that, when enabling sequence parallel you typically:
+
+- Chunk **image tokens** (`hidden_states`) across SP ranks along the **sequence dimension**.
+- Keep **text embeddings** (`encoder_hidden_states`) **replicated** on all SP ranks for correctness (unless you implement full joint-SP semantics and explicitly split text too).
+- Chunk **image RoPE freqs** to match the chunked image tokens.
+- `all_gather` the final output back along the sequence dimension.
+
+First, add the sequence-parallel helpers and (for Qwen-Image) the forward-context flag. Then, in the transformer's `forward()`:
+
+- Chunk `hidden_states` by SP world size.
+- Set `get_forward_context().split_text_embed_in_sp = False` because Qwen-Image uses joint attention and we would use a full text embedding in each rank.
+- After `pos_embed`, chunk `img_freqs` on `dim=0` (token axis) to match the chunked image tokens; only chunk `txt_freqs` if you explicitly split text embeddings in SP.
+
+Taking `vllm_omni/diffusion/models/qwen_image/qwen_image_transformer.py` as an example:
+
+```diff
+from vllm_omni.diffusion.distributed.parallel_state import (
+    get_sequence_parallel_rank,
+    get_sequence_parallel_world_size,
+    get_sp_group,
+)
+from vllm_omni.diffusion.forward_context import get_forward_context
+
+class QwenImageTransformer2DModel(...):
+    def forward(
+        self,
+        hidden_states: torch.Tensor,
+        encoder_hidden_states: torch.Tensor = None,
+        encoder_hidden_states_mask: torch.Tensor = None,
+        timestep: torch.LongTensor = None,
+        img_shapes: list[tuple[int, int, int]] | None = None,
+        txt_seq_lens: list[int] | None = None,
+        ...
+    ):
++       if self.parallel_config.sequence_parallel_size > 1:
++           # Chunk image tokens along sequence dimension.
++           hidden_states = torch.chunk(
++               hidden_states,
++               get_sequence_parallel_world_size(),
++               dim=-2,
++           )[get_sequence_parallel_rank()]
++
++           # Qwen-Image uses *dual-stream* (text + image) and runs *joint attention*.
++           # Text embeddings should be replicated across SP ranks for correctness.
++           get_forward_context().split_text_embed_in_sp = False
+
+        hidden_states = self.img_in(hidden_states)
+
+        ...
+        image_rotary_emb = self.pos_embed(img_shapes, txt_seq_lens, device=hidden_states.device)
+
++       def get_rotary_emb_chunk(freqs):
++           return torch.chunk(freqs, get_sequence_parallel_world_size(), dim=0)[
++               get_sequence_parallel_rank()
++           ]
++
++       if self.parallel_config.sequence_parallel_size > 1:
++           img_freqs, txt_freqs = image_rotary_emb
++           img_freqs = get_rotary_emb_chunk(img_freqs)
++           if get_forward_context().split_text_embed_in_sp:
++               txt_freqs = get_rotary_emb_chunk(txt_freqs)
++           image_rotary_emb = (img_freqs, txt_freqs)
+```
+
+Next, at the end of the `forward()` function, call `get_sp_group().all_gather` to gather the chunked outputs across devices and concatenate them along the sequence dimension (matching the earlier `dim=-2` chunking):
+
+```diff
+class QwenImageTransformer2DModel(...):
+    def forward(...):
+        ...
+        hidden_states = self.norm_out(hidden_states, temb)
+        output = self.proj_out(hidden_states)
+
++       if self.parallel_config.sequence_parallel_size > 1:
++           output = get_sp_group().all_gather(output, dim=-2)
+        return Transformer2DModelOutput(sample=output)
+```
+
+Finally, you can set the parallel configuration and pass it to `Omni` and start parallel inference with:
+```diff
+from vllm_omni import Omni
++from vllm_omni.diffusion.data import DiffusionParallelConfig
+ulysses_degree = 2
+
+omni = Omni(
+    model="Qwen/Qwen-Image",
++    parallel_config=DiffusionParallelConfig(ulysses_degree=2)
+)
+
+outputs = omni.generate(prompt="A cat sitting on a windowsill", num_inference_steps=50)
+```

docs/user_guide/acceleration/teacache.md

@@ -8,7 +8,6 @@ Enable TeaCache by setting `cache_backend` to `"tea_cache"`:
 
 ```python
 from vllm_omni import Omni
-from vllm_omni.inputs.data import OmniDiffusionSamplingParams
 
 # Simple configuration - model_type is automatically extracted from pipeline.__class__.__name__
 omni = Omni(
@@ -18,12 +17,7 @@ omni = Omni(
         "rel_l1_thresh": 0.2  # Optional, defaults to 0.2
     }
 )
-outputs = omni.generate(
-    "A cat sitting on a windowsill",
-    OmniDiffusionSamplingParams(
-        num_inference_steps=50,
-    ),
-)
+outputs = omni.generate(prompt="A cat sitting on a windowsill", num_inference_steps=50)
 ```
 
 ### Using Environment Variable
@@ -74,19 +68,13 @@ Controls the balance between speed and quality. Lower values prioritize quality,
 
 ```python
 from vllm_omni import Omni
-from vllm_omni.inputs.data import OmniDiffusionSamplingParams
 
 omni = Omni(
     model="Qwen/Qwen-Image",
     cache_backend="tea_cache",
     cache_config={"rel_l1_thresh": 0.2}
 )
-outputs = omni.generate(
-    "A cat sitting on a windowsill",
-    OmniDiffusionSamplingParams(
-        num_inference_steps=50,
-    ),
-)
+outputs = omni.generate(prompt="A cat sitting on a windowsill", num_inference_steps=50)
 ```
 
 ## Performance Tuning
@@ -124,7 +112,6 @@ th {
 | `QwenImageEditPipeline` | Qwen-Image-Edit | `Qwen/Qwen-Image-Edit` |
 | `QwenImageEditPlusPipeline` | Qwen-Image-Edit-2509 | `Qwen/Qwen-Image-Edit-2509` |
 | `QwenImageLayeredPipeline` | Qwen-Image-Layered | `Qwen/Qwen-Image-Layered` |
-| `BagelForConditionalGeneration` | BAGEL (DiT-only) | `ByteDance-Seed/BAGEL-7B-MoT` |
 
 ### VideoGen
 

docs/user_guide/diffusion/cpu_offload_diffusion.md

@@ -1,29 +0,0 @@
-# CPU Offloading for Diffusion Model
-
-## Overview
-CPU offload lets the diffusion worker move large model components between GPU and CPU memory on demand. It keeps the DiT transformer resident on GPU only while it is actively running, and swaps it out when encoders modules need the device. This reduces peak VRAM usage so bigger checkpoints run on smaller GPUs, or multiple requests can share the same GPU.
-
-## Execution Model
-1. Text encoders run on GPU while the DiT transformer is offloaded to CPU.
-2. Before denoising, weights are prefetched back to GPU, honoring pinned-memory copies for speed.
-3. After the diffusion step, the transformer returns to CPU and the process repeats as needed.
-
-Transfers use pinned host buffers, and the worker coordinates swaps via mutex-style hooks so components never compete for memory.
-
-## Configuration
-You can enable CPU offload in two ways:
-
-- **Python API**: set `enable_cpu_offload=True`.
-
-```python
-from vllm_omni import Omni
-
-if __name__ == "__main__":
-
-    m = Omni(model="Qwen/Qwen-Image",enable_cpu_offload=True)
-```
-
-- **CLI**: pass `--enable-cpu-offload` to the diffusion service entrypoint.
-
-## Known Limitations
-- Cold start latency increases for over one minute for some models(e.g., Qwen-Image)

docs/user_guide/diffusion/parallelism_acceleration.md

@@ -1,473 +0,0 @@
-# Parallelism Acceleration Guide
-
-This guide includes how to use parallelism methods in vLLM-Omni to speed up diffusion model inference as well as reduce the memory requirement on each device.
-
-## Overview
-
-The following parallelism methods are currently supported in vLLM-Omni:
-
-1. DeepSpeed Ulysses Sequence Parallel (DeepSpeed Ulysses-SP) ([arxiv paper](https://arxiv.org/pdf/2309.14509)): Ulysses-SP splits the input along the sequence dimension and uses all-to-all communication to allow each device to compute only a subset of attention heads.
-
-2. [Ring-Attention](#ring-attention) - splits the input along the sequence dimension and uses ring-based P2P communication to accumulate attention results, keeping the sequence dimension sharded
-
-3. Classifier-Free-Guidance Parallel (CFG-Parallel): CFG-Parallel runs the positive/negative prompts of classifier-free guidance (CFG) on different devices, then merges on a single device to perform the scheduler step.
-
-4. [Tensor Parallelism](#tensor-parallelism): Tensor parallelism shards model weights across devices. This can reduce per-GPU memory usage. Note that for diffusion models we currently shard the majority of layers within the DiT.
-
-The following table shows which models are currently supported by parallelism method:
-
-### ImageGen
-
-| Model | Model Identifier | Ulysses-SP | Ring-SP | CFG-Parallel | Tensor-Parallel |
-|-------|------------------|------------|---------|--------------|--------------------------|
-| **LongCat-Image** | `meituan-longcat/LongCat-Image` | ✅ | ✅ | ❌ | ❌ |
-| **LongCat-Image-Edit** | `meituan-longcat/LongCat-Image-Edit` | ✅ | ✅ | ❌ | ❌ |
-| **Ovis-Image** | `OvisAI/Ovis-Image` | ❌ | ❌ | ❌ | ❌ |
-| **Qwen-Image** | `Qwen/Qwen-Image` | ✅ | ✅ | ✅ | ✅ |
-| **Qwen-Image-Edit** | `Qwen/Qwen-Image-Edit` | ✅ | ✅ | ✅ | ✅ |
-| **Qwen-Image-Edit-2509** | `Qwen/Qwen-Image-Edit-2509` | ✅ | ✅ | ✅ | ✅ |
-| **Qwen-Image-Layered** | `Qwen/Qwen-Image-Layered` | ✅ | ✅ | ✅ | ✅ |
-| **Z-Image** | `Tongyi-MAI/Z-Image-Turbo` | ✅ | ✅ | ❌ | ✅ (TP=2 only) |
-| **Stable-Diffusion3.5** | `stabilityai/stable-diffusion-3.5` | ❌ | ❌ | ❌ | ❌ |
-| **FLUX.2-klein** | `black-forest-labs/FLUX.2-klein-4B` | ❌ | ❌ | ❌ | ✅ |
-| **FLUX.1-dev** | `black-forest-labs/FLUX.1-dev` | ❌ | ❌ | ❌ | ✅ |
-
-
-!!! note "TP Limitations for Diffusion Models"
-    We currently implement Tensor Parallelism (TP) only for the DiT (Diffusion Transformer) blocks. This is because the `text_encoder` component in vLLM-Omni uses the original Transformers implementation, which does not yet support TP.
-
-    - Good news: The text_encoder typically has minimal impact on overall inference performance.
-    - Bad news: When TP is enabled, every TP process retains a full copy of the text_encoder weights, leading to significant GPU memory waste.
-
-    We are actively refactoring this design to address this. For details and progress, please refer to [Issue #771](https://github.com/vllm-project/vllm-omni/issues/771).
-
-
-!!! note "Why Z-Image is TP=2 only"
-    Z-Image Turbo is currently limited to `tensor_parallel_size` of **1 or 2** due to model shape divisibility constraints.
-    For example, the model has `n_heads=30` and a final projection out dimension of `64`, so valid TP sizes must divide both 30 and 64; the only common divisors are **1 and 2**.
-
-### VideoGen
-
-| Model | Model Identifier | Ulysses-SP | Ring-SP | Tensor-Parallel |
-|-------|------------------|------------|---------|--------------------------|
-| **Wan2.2** | `Wan-AI/Wan2.2-T2V-A14B-Diffusers` | ✅ | ✅ | ❌ |
-
-### Tensor Parallelism
-
-Tensor parallelism splits model parameters across GPUs. In vLLM-Omni, tensor parallelism is configured via `DiffusionParallelConfig.tensor_parallel_size`.
-
-#### Offline Inference
-
-```python
-from vllm_omni import Omni
-from vllm_omni.diffusion.data import DiffusionParallelConfig
-
-omni = Omni(
-    model="Tongyi-MAI/Z-Image-Turbo",
-    parallel_config=DiffusionParallelConfig(tensor_parallel_size=2),
-)
-
-outputs = omni.generate(
-    "a cat reading a book",
-    OmniDiffusionSamplingParams(
-        num_inference_steps=9,
-        width=512,
-        height=512,
-    ),
-)
-```
-
-### Sequence Parallelism
-
-#### Ulysses-SP
-
-##### Offline Inference
-
-An example of offline inference script using [Ulysses-SP](https://arxiv.org/pdf/2309.14509) is shown below:
-```python
-from vllm_omni import Omni
-from vllm_omni.inputs.data import OmniDiffusionSamplingParams
-from vllm_omni.diffusion.data import DiffusionParallelConfig
-ulysses_degree = 2
-
-omni = Omni(
-    model="Qwen/Qwen-Image",
-    parallel_config=DiffusionParallelConfig(ulysses_degree=2)
-)
-
-outputs = omni.generate(
-    "A cat sitting on a windowsill",
-    OmniDiffusionSamplingParams(num_inference_steps=50, width=2048, height=2048),
-)
-```
-
-See `examples/offline_inference/text_to_image/text_to_image.py` for a complete working example.
-
-##### Online Serving
-
-You can enable Ulysses-SP in online serving for diffusion models via `--usp`:
-
-```bash
-# Text-to-image (requires >= 2 GPUs)
-vllm serve Qwen/Qwen-Image --omni --port 8091 --usp 2
-```
-
-##### Benchmarks
-!!! note "Benchmark Disclaimer"
-    These benchmarks are provided for **general reference only**. The configurations shown use default or common parameter settings and have not been exhaustively optimized for maximum performance. Actual performance may vary based on:
-
-    - Specific model and use case
-    - Hardware configuration
-    - Careful parameter tuning
-    - Different inference settings (e.g., number of steps, image resolution)
-
-
-To measure the parallelism methods, we run benchmarks with **Qwen/Qwen-Image** model generating images (**2048x2048** as long sequence input) with 50 inference steps. The hardware devices are NVIDIA H800 GPUs. `sdpa` is the attention backends.
-
-| Configuration | Ulysses degree |Generation Time | Speedup |
-|---------------|----------------|---------|---------|
-| **Baseline (diffusers)** | - | 112.5s | 1.0x |
-| Ulysses-SP  |  2  |  65.2s | 1.73x |
-| Ulysses-SP  |  4  | 39.6s | 2.84x |
-| Ulysses-SP  |  8  | 30.8s | 3.65x |
-
-#### Ring-Attention
-
-Ring-Attention ([arxiv paper](https://arxiv.org/abs/2310.01889)) splits the input along the sequence dimension and uses ring-based P2P communication to accumulate attention results. Unlike Ulysses-SP which uses all-to-all communication, Ring-Attention keeps the sequence dimension sharded throughout the computation and circulates Key/Value blocks through a ring topology.
-
-##### Offline Inference
-
-An example of offline inference script using Ring-Attention is shown below:
-```python
-from vllm_omni import Omni
-from vllm_omni.inputs.data import OmniDiffusionSamplingParams
-from vllm_omni.diffusion.data import DiffusionParallelConfig
-ring_degree = 2
-
-omni = Omni(
-    model="Qwen/Qwen-Image",
-    parallel_config=DiffusionParallelConfig(ring_degree=2)
-)
-
-outputs = omni.generate(
-    "A cat sitting on a windowsill",
-    OmniDiffusionSamplingParams(num_inference_steps=50, width=2048, height=2048),
-)
-```
-
-See `examples/offline_inference/text_to_image/text_to_image.py` for a complete working example.
-
-
-##### Online Serving
-
-You can enable Ring-Attention in online serving for diffusion models via `--ring`:
-
-```bash
-# Text-to-image (requires >= 2 GPUs)
-vllm serve Qwen/Qwen-Image --omni --port 8091 --ring 2
-```
-
-##### Benchmarks
-!!! note "Benchmark Disclaimer"
-    These benchmarks are provided for **general reference only**. The configurations shown use default or common parameter settings and have not been exhaustively optimized for maximum performance. Actual performance may vary based on:
-
-    - Specific model and use case
-    - Hardware configuration
-    - Careful parameter tuning
-    - Different inference settings (e.g., number of steps, image resolution)
-
-
-To measure the parallelism methods, we run benchmarks with **Qwen/Qwen-Image** model generating images (**1024x1024** as long sequence input) with 50 inference steps. The hardware devices are NVIDIA A100 GPUs. `flash_attn` is the attention backends.
-
-| Configuration | Ring degree |Generation Time | Speedup |
-|---------------|----------------|---------|---------|
-| **Baseline (diffusers)** | - | 45.2s | 1.0x |
-| Ring-Attention  |  2  |  29.9s | 1.51x |
-| Ring-Attention  |  4  | 23.3s | 1.94x |
-
-
-#### Hybrid Ulysses + Ring
-
-You can combine both Ulysses-SP and Ring-Attention for larger scale parallelism. The total sequence parallel size equals `ulysses_degree × ring_degree`.
-
-##### Offline Inference
-
-```python
-from vllm_omni import Omni
-from vllm_omni.inputs.data import OmniDiffusionSamplingParams
-from vllm_omni.diffusion.data import DiffusionParallelConfig
-
-# Hybrid: 2 Ulysses × 2 Ring = 4 GPUs total
-omni = Omni(
-    model="Qwen/Qwen-Image",
-    parallel_config=DiffusionParallelConfig(ulysses_degree=2, ring_degree=2)
-)
-
-outputs = omni.generate(
-    "A cat sitting on a windowsill",
-    OmniDiffusionSamplingParams(num_inference_steps=50, width=2048, height=2048),
-)
-```
-
-##### Online Serving
-
-```bash
-# Text-to-image (requires >= 4 GPUs)
-vllm serve Qwen/Qwen-Image --omni --port 8091 --usp 2 --ring 2
-```
-
-##### Benchmarks
-!!! note "Benchmark Disclaimer"
-    These benchmarks are provided for **general reference only**. The configurations shown use default or common parameter settings and have not been exhaustively optimized for maximum performance. Actual performance may vary based on:
-
-    - Specific model and use case
-    - Hardware configuration
-    - Careful parameter tuning
-    - Different inference settings (e.g., number of steps, image resolution)
-
-
-To measure the parallelism methods, we run benchmarks with **Qwen/Qwen-Image** model generating images (**1024x1024** as long sequence input) with 50 inference steps. The hardware devices are NVIDIA A100 GPUs. `flash_attn` is the attention backends.
-
-| Configuration | Ulysses degree | Ring degree | Generation Time | Speedup |
-|---------------|----------------|-------------|-----------------|---------|
-| **Baseline (diffusers)** | - | - | 45.2s | 1.0x |
-| Hybrid Ulysses + Ring  |  2  |  2  |  24.3s | 1.87x |
-
-
-##### How to parallelize a new model
-
-NOTE: "Terminology: SP vs CP"
-    Our "Sequence Parallelism" (SP) corresponds to "Context Parallelism" (CP) in the [diffusers library](https://github.com/huggingface/diffusers/blob/main/src/diffusers/models/_modeling_parallel.py).
-    We use "Sequence Parallelism" to align with vLLM-Omni's terminology.
-
----
-
-###### Non-intrusive `_sp_plan` (Recommended)
-
-The `_sp_plan` mechanism allows SP without modifying `forward()` logic. The framework automatically registers hooks to shard inputs and gather outputs at module boundaries.
-
-**Requirements for `forward()` function:**
-
-- Tensor operations that need sharding/gathering must happen at **`nn.Module` boundaries** (not inline Python operations)
-- If your `forward()` contains inline tensor operations (e.g., `torch.cat`, `pad_sequence`) that need sharding, **extract them into a submodule**
-
-**When to create a submodule:**
-
-```python
-# ❌ BAD: Inline operations - hooks cannot intercept
-def forward(self, x, cap_feats):
-    unified = torch.cat([x, cap_feats], dim=1)  # Cannot be sharded via _sp_plan
-    ...
-
-# ✅ GOOD: Extract into a submodule
-class UnifiedPrepare(nn.Module):
-    def forward(self, x, cap_feats):
-        return torch.cat([x, cap_feats], dim=1)  # Now can be sharded via _sp_plan
-
-class MyModel(nn.Module):
-    def __init__(self):
-        self.unified_prepare = UnifiedPrepare()  # Submodule
-
-    def forward(self, x, cap_feats):
-        unified = self.unified_prepare(x, cap_feats)  # Hook can intercept here
-```
-
----
-
-###### Defining `_sp_plan`
-
-**Type definitions** (see [diffusers `_modeling_parallel.py`](https://github.com/huggingface/diffusers/blob/main/src/diffusers/models/_modeling_parallel.py) for reference):
-
-```python
-from vllm_omni.diffusion.distributed.sp_plan import (
-    SequenceParallelInput,   # Corresponds to diffusers' ContextParallelInput
-    SequenceParallelOutput,  # Corresponds to diffusers' ContextParallelOutput
-)
-```
-
-| Parameter | Description |
-|-----------|-------------|
-| `split_dim` | Dimension to split/gather (usually `1` for sequence) |
-| `expected_dims` | Expected tensor rank for validation (optional) |
-| `split_output` | `False`: shard **input** parameters; `True`: shard **output** tensors |
-| `auto_pad` | Auto-pad if sequence not divisible by world_size (Ulysses only) |
-
-**Key naming convention:**
-
-| Key | Meaning | Python equivalent |
-|-----|---------|-------------------|
-| `""` | Root model | `model` |
-| `"blocks.0"` | First element of ModuleList | `model.blocks[0]` |
-| `"blocks.*"` | All elements of ModuleList | `for b in model.blocks` |
-| `"outputs.main"` | ModuleDict entry | `model.outputs["main"]` |
-
-**Dictionary key types:**
-
-| Key type | `split_output` | Description |
-|----------|----------------|-------------|
-| `"param_name"` (str) | `False` | Shard **input parameter** by name |
-| `0`, `1` (int) | `True` | Shard **output tuple** by index |
-
-**Example** (similar to [diffusers `transformer_wan.py`](https://github.com/huggingface/diffusers/blob/main/src/diffusers/models/transformers/transformer_wan.py)):
-
-```python
-class MyTransformer(nn.Module):
-    _sp_plan = {
-        # Shard rope module OUTPUTS (returns tuple)
-        "rope": {
-            0: SequenceParallelInput(split_dim=1, expected_dims=4, split_output=True),  # cos
-            1: SequenceParallelInput(split_dim=1, expected_dims=4, split_output=True),  # sin
-        },
-        # Shard transformer block INPUT parameter
-        "blocks.0": {
-            "hidden_states": SequenceParallelInput(split_dim=1, expected_dims=3),
-        },
-        # Gather at final projection
-        "proj_out": SequenceParallelOutput(gather_dim=1, expected_dims=3),
-    }
-```
-
----
-
-###### Hook flow
-
-```
-Input → [SequenceParallelSplitHook: pre_forward] → Module.forward() → [post_forward] → ...
-                                                                              ↓
-... → [SequenceParallelGatherHook: post_forward] → Output
-```
-
-1. **SplitHook** shards tensors before/after the target module
-2. **Attention layers** handle Ulysses/Ring communication internally
-3. **GatherHook** collects sharded outputs
-
-The framework automatically applies these hooks when `sequence_parallel_size > 1`.
-
----
-
-###### Method 2: Intrusive modification (For complex cases)
-
-For models with dynamic sharding logic that cannot be expressed via `_sp_plan`:
-
-```python
-from vllm_omni.diffusion.distributed.sp_sharding import sp_shard, sp_gather
-
-def forward(self, hidden_states, ...):
-    if self.parallel_config.sequence_parallel_size > 1:
-        hidden_states = sp_shard(hidden_states, dim=1)
-        # ... computation ...
-        output = sp_gather(output, dim=1)
-    return output
-```
-
----
-
-###### Choosing the right approach
-
-| Scenario | Approach |
-|----------|----------|
-| Standard transformer | `_sp_plan` |
-| Inline tensor ops need sharding | Extract to submodule + `_sp_plan` |
-| Dynamic/conditional sharding | Intrusive modification |
-
-
-### CFG-Parallel
-
-##### Offline Inference
-
-CFG-Parallel is enabled through `DiffusionParallelConfig(cfg_parallel_size=...)`. The recommended configuration is `cfg_parallel_size=2` (one rank for the positive branch and one rank for the negative branch).
-
-An example of offline inference using CFG-Parallel (image-to-image) is shown below:
-
-```python
-from vllm_omni import Omni
-from vllm_omni.diffusion.data import DiffusionParallelConfig
-
-omni = Omni(
-    model="Qwen/Qwen-Image-Edit",
-    parallel_config=DiffusionParallelConfig(cfg_parallel_size=2),
-)
-
-outputs = omni.generate(
-    {
-        "prompt": "turn this cat to a dog",
-        "negative_prompt": "low quality, blurry",
-    },
-    OmniDiffusionSamplingParams(
-        true_cfg_scale=4.0,
-        pil_image=input_image,
-        num_inference_steps=50,
-    ),
-)
-```
-
-Notes:
-
-- CFG-Parallel is only effective when **true CFG** is enabled (i.e., `true_cfg_scale > 1` and a `negative_prompt` is provided).
-
-#### How to parallelize a pipeline
-
-This section describes how to add CFG-Parallel to a diffusion **pipeline**. We use the Qwen-Image pipeline (`vllm_omni/diffusion/models/qwen_image/pipeline_qwen_image.py`) as the reference implementation.
-
-In `QwenImagePipeline`, each diffusion step runs two denoiser forward passes sequentially:
-
-- positive (prompt-conditioned)
-- negative (negative-prompt-conditioned)
-
-CFG-Parallel assigns these two branches to different ranks in the **CFG group** and synchronizes the results.
-
-Below is an example of CFG-Parallel implementation:
-
-```python
-def diffuse(
-        self,
-        ...
-        ):
-    # Enable CFG-parallel: rank0 computes positive, rank1 computes negative.
-    cfg_parallel_ready = do_true_cfg and get_classifier_free_guidance_world_size() > 1
-
-    self.transformer.do_true_cfg = do_true_cfg
-
-    if cfg_parallel_ready:
-        cfg_group = get_cfg_group()
-        cfg_rank = get_classifier_free_guidance_rank()
-
-        if cfg_rank == 0:
-            local_pred = self.transformer(
-                hidden_states=latents,
-                timestep=timestep / 1000,
-                guidance=guidance,
-                encoder_hidden_states_mask=prompt_embeds_mask,
-                encoder_hidden_states=prompt_embeds,
-                img_shapes=img_shapes,
-                txt_seq_lens=txt_seq_lens,
-                attention_kwargs=self.attention_kwargs,
-                return_dict=False,
-            )[0]
-        else:
-            local_pred = self.transformer(
-                hidden_states=latents,
-                timestep=timestep / 1000,
-                guidance=guidance,
-                encoder_hidden_states_mask=negative_prompt_embeds_mask,
-                encoder_hidden_states=negative_prompt_embeds,
-                img_shapes=img_shapes,
-                txt_seq_lens=negative_txt_seq_lens,
-                attention_kwargs=self.attention_kwargs,
-                return_dict=False,
-            )[0]
-
-        gathered = cfg_group.all_gather(local_pred, separate_tensors=True)
-        if cfg_rank == 0:
-            noise_pred = gathered[0]
-            neg_noise_pred = gathered[1]
-            comb_pred = neg_noise_pred + true_cfg_scale * (noise_pred - neg_noise_pred)
-            cond_norm = torch.norm(noise_pred, dim=-1, keepdim=True)
-            noise_norm = torch.norm(comb_pred, dim=-1, keepdim=True)
-            noise_pred = comb_pred * (cond_norm / noise_norm)
-            latents = self.scheduler.step(noise_pred, t, latents, return_dict=False)[0]
-        cfg_group.broadcast(latents, src=0)
-    else:
-        # fallback: run positive then negative sequentially on one rank
-        ...
-```

docs/user_guide/diffusion_acceleration.md

@@ -6,21 +6,18 @@ vLLM-Omni supports various cache acceleration methods to speed up diffusion mode
 
 vLLM-Omni currently supports two main cache acceleration backends:
 
-1. **[TeaCache](diffusion/teacache.md)** - Hook-based adaptive caching that caches transformer computations when consecutive timesteps are similar
-2. **[Cache-DiT](diffusion/cache_dit_acceleration.md)** - Library-based acceleration using multiple techniques:
+1. **[TeaCache](acceleration/teacache.md)** - Hook-based adaptive caching that caches transformer computations when consecutive timesteps are similar
+2. **[Cache-DiT](acceleration/cache_dit_acceleration.md)** - Library-based acceleration using multiple techniques:
     - **DBCache** (Dual Block Cache): Caches intermediate transformer block outputs based on residual differences
     - **TaylorSeer**: Uses Taylor expansion-based forecasting for faster inference
     - **SCM** (Step Computation Masking): Selectively computes steps based on adaptive masking
 
 Both methods can provide significant speedups (typically **1.5x-2.0x**) while maintaining high output quality.
 
-vLLM-Omni also supports parallelism methods for diffusion models, including:
+vLLM-Omni also supports the sequence parallelism (SP) for the diffusion model, that includes:
 
-1. [Ulysses-SP](diffusion/parallelism_acceleration.md#ulysses-sp) - splits the input along the sequence dimension and uses all-to-all communication to allow each device to compute only a subset of attention heads.
-
-2. [Ring-Attention](diffusion/parallelism_acceleration.md#ring-attention) - splits the input along the sequence dimension and uses ring-based P2P communication to accumulate attention results, keeping the sequence dimension sharded.
-
-3. [CFG-Parallel](diffusion/parallelism_acceleration.md#cfg-parallel) - runs the positive/negative prompts of classifier-free guidance (CFG) on different devices, then merges on a single device to perform the scheduler step.
+1. [Ulysses-SP](acceleration/parallelism_acceleration.md#ulysses-sp) - splits the input along the sequence dimension and uses all-to-all communication to allow each device to compute only a subset of attention heads.
+2. [Ring-Attention](acceleration/parallelism_acceleration.md#ring-attention) - splits the input along the sequence dimension and uses ring-based P2P communication to accumulate attention results, keeping the sequence dimension sharded.
 
 ## Quick Comparison
 
@@ -37,25 +34,26 @@ The following table shows which models are currently supported by each accelerat
 
 ### ImageGen
 
-| Model | Model Identifier | TeaCache | Cache-DiT | Ulysses-SP | Ring-Attention | CFG-Parallel |
-|-------|------------------|:----------:|:-----------:|:-----------:|:----------------:|:----------------:|
-| **LongCat-Image** | `meituan-longcat/LongCat-Image` | ❌ | ✅ | ❌ | ❌ | ❌ |
-| **LongCat-Image-Edit** | `meituan-longcat/LongCat-Image-Edit` | ❌ | ✅ | ❌ | ❌ | ❌ |
-| **Ovis-Image** | `OvisAI/Ovis-Image` | ❌ | ✅ | ❌ | ❌ | ❌ |
-| **Qwen-Image** | `Qwen/Qwen-Image` | ✅ | ✅ | ✅ | ✅ | ✅ |
-| **Qwen-Image-2512** | `Qwen/Qwen-Image-2512` | ✅ | ✅ | ✅ | ✅ | ✅ |
-| **Qwen-Image-Edit** | `Qwen/Qwen-Image-Edit` | ✅ | ✅ | ✅ | ✅ | ✅ |
-| **Qwen-Image-Edit-2509** | `Qwen/Qwen-Image-Edit-2509` | ✅ | ✅ | ✅ | ✅ | ✅ |
-| **Qwen-Image-Layered** | `Qwen/Qwen-Image-Layered` | ❌ | ✅ | ✅ | ✅ | ✅ |
-| **Z-Image** | `Tongyi-MAI/Z-Image-Turbo` | ✅ | ✅ | ❌ | ❌ | ❌ |
-| **Stable-Diffusion3.5** | `stabilityai/stable-diffusion-3.5` | ❌ | ✅ | ❌ | ❌ | ❌ |
-| **Bagel** | `ByteDance-Seed/BAGEL-7B-MoT` | ✅ | ✅ | ❌ | ❌ | ❌ |
+| Model | Model Identifier | TeaCache | Cache-DiT | Ulysses-SP | Ring-Attention |
+|-------|------------------|----------|-----------|-----------|----------------|
+| **LongCat-Image** | `meituan-longcat/LongCat-Image` | ❌ | ✅ | ❌ | ❌ |
+| **LongCat-Image-Edit** | `meituan-longcat/LongCat-Image-Edit` | ❌ | ✅ | ❌ | ❌ |
+| **Ovis-Image** | `OvisAI/Ovis-Image` | ❌ | ✅ | ❌ | ❌ |
+| **Qwen-Image** | `Qwen/Qwen-Image` | ✅ | ✅ | ✅ | ✅ |
+| **Qwen-Image-Edit** | `Qwen/Qwen-Image-Edit` | ✅ | ✅ | ✅ | ✅ |
+| **Qwen-Image-Edit-2509** | `Qwen/Qwen-Image-Edit-2509` | ❌ | ✅ | ✅ | ✅ |
+| **Qwen-Image-Layered** | `Qwen/Qwen-Image-Layered` | ❌ | ✅ | ✅ | ✅ |
+| **Z-Image** | `Tongyi-MAI/Z-Image-Turbo` | ❌ | ✅ |❌ | ❌ |
+| **LongCat-Image** | `meituan-longcat/LongCat-Image` | ❌ | ✅ | ❌ | ❌ |
+| **LongCat-Image-Edit** | `meituan-longcat/LongCat-Image-Edit` | ❌ | ✅ | ❌ | ❌ |
+| **Ovis-Image** | `OvisAI/Ovis-Image` | ❌ | ✅ | ❌ | ❌ |
+| **Stable-Diffusion3.5** | `stabilityai/stable-diffusion-3.5` | ❌ | ✅ | ❌ | ❌ |
 
 ### VideoGen
 
-| Model | Model Identifier | TeaCache | Cache-DiT | Ulysses-SP | Ring-Attention |
-|-------|------------------|:--------:|:---------:|:----------:|:--------------:|
-| **Wan2.2** | `Wan-AI/Wan2.2-T2V-A14B-Diffusers` | ❌ | ✅ | ❌ | ❌ |
+| Model | Model Identifier | TeaCache | Cache-DiT | Ulysses-SP |
+|-------|------------------|----------|-----------|-----------|
+| **Wan2.2** | `Wan-AI/Wan2.2-T2V-A14B-Diffusers` | ❌ | ✅ | ❌ |
 
 
 ## Performance Benchmarks
@@ -96,7 +94,6 @@ To measure the parallelism methods, we run benchmarks with **Qwen/Qwen-Image** m
 
 ```python
 from vllm_omni import Omni
-from vllm_omni.inputs.data import OmniDiffusionSamplingParams
 
 omni = Omni(
     model="Qwen/Qwen-Image",
@@ -104,19 +101,13 @@ omni = Omni(
     cache_config={"rel_l1_thresh": 0.2}  # Optional, defaults to 0.2
 )
 
-outputs = omni.generate(
-    "A cat sitting on a windowsill",
-    OmniDiffusionSamplingParams(
-        num_inference_steps=50,
-    ),
-)
+outputs = omni.generate(prompt="A cat sitting on a windowsill", num_inference_steps=50)
 ```
 
 ### Using Cache-DiT
 
 ```python
 from vllm_omni import Omni
-from vllm_omni.inputs.data import OmniDiffusionSamplingParams
 
 omni = Omni(
     model="Qwen/Qwen-Image",
@@ -130,12 +121,7 @@ omni = Omni(
     }
 )
 
-outputs = omni.generate(
-    "A cat sitting on a windowsill",
-    OmniDiffusionSamplingParams(
-        num_inference_steps=50,
-    ),
-)
+outputs = omni.generate(prompt="A cat sitting on a windowsill", num_inference_steps=50)
 ```
 
 ### Using Ulysses-SP
@@ -143,41 +129,31 @@ outputs = omni.generate(
 Run text-to-image:
 ```python
 from vllm_omni import Omni
-from vllm_omni.inputs.data import OmniDiffusionSamplingParams
 from vllm_omni.diffusion.data import DiffusionParallelConfig
 ulysses_degree = 2
 
 omni = Omni(
     model="Qwen/Qwen-Image",
-    parallel_config=DiffusionParallelConfig(ulysses_degree=ulysses_degree)
+    parallel_config=DiffusionParallelConfig(ulysses_degree=2)
 )
 
-outputs = omni.generate(
-    "A cat sitting on a windowsill",
-    OmniDiffusionSamplingParams(num_inference_steps=50, width=2048, height=2048),
-)
+outputs = omni.generate(prompt="A cat sitting on a windowsill", num_inference_steps=50, width=2048, height=2048)
 ```
 
 
 Run image-to-image:
 ```python
 from vllm_omni import Omni
-from vllm_omni.inputs.data import OmniDiffusionSamplingParams
 from vllm_omni.diffusion.data import DiffusionParallelConfig
 ulysses_degree = 2
 
 omni = Omni(
     model="Qwen/Qwen-Image-Edit",
-    parallel_config=DiffusionParallelConfig(ulysses_degree=ulysses_degree)
+    parallel_config=DiffusionParallelConfig(ulysses_degree=2)
 )
 
-outputs = omni.generate(
-    {
-        "prompt": "turn this cat to a dog",
-        "multi_modal_data": {"image": input_image}
-    },
-    OmniDiffusionSamplingParams(num_inference_steps=50),
-)
+outputs = omni.generate(prompt="turn this cat to a dog",
+        pil_image=input_image, num_inference_steps=50)
 ```
 
 ### Using Ring-Attention
@@ -185,7 +161,6 @@ outputs = omni.generate(
 Run text-to-image:
 ```python
 from vllm_omni import Omni
-from vllm_omni.inputs.data import OmniDiffusionSamplingParams
 from vllm_omni.diffusion.data import DiffusionParallelConfig
 ring_degree = 2
 
@@ -194,43 +169,13 @@ omni = Omni(
     parallel_config=DiffusionParallelConfig(ring_degree=2)
 )
 
-outputs = omni.generate(
-    "A cat sitting on a windowsill",
-    OmniDiffusionSamplingParams(num_inference_steps=50, width=2048, height=2048),
-)
-```
-
-### Using CFG-Parallel
-
-Run image-to-image:
-
-CFG-Parallel splits the CFG positive/negative branches across GPUs. Use it when you set a non-trivial `true_cfg_scale`.
-
-```python
-from vllm_omni import Omni
-from vllm_omni.inputs.data import OmniDiffusionSamplingParams
-from vllm_omni.diffusion.data import DiffusionParallelConfig
-cfg_parallel_size = 2
-
-omni = Omni(
-    model="Qwen/Qwen-Image-Edit",
-    parallel_config=DiffusionParallelConfig(cfg_parallel_size=cfg_parallel_size)
-)
-
-outputs = omni.generate(
-    {
-        "prompt": "turn this cat to a dog",
-        "multi_modal_data": {"image": input_image}
-    },
-    OmniDiffusionSamplingParams(num_inference_steps=50, true_cfg_scale=4.0),
-)
+outputs = omni.generate(prompt="A cat sitting on a windowsill", num_inference_steps=50, width=2048, height=2048)
 ```
 
 ## Documentation
 
 For detailed information on each acceleration method:
 
-- **[TeaCache Guide](diffusion/teacache.md)** - Complete TeaCache documentation, configuration options, and best practices
-- **[Cache-DiT Acceleration Guide](diffusion/cache_dit_acceleration.md)** - Comprehensive Cache-DiT guide covering DBCache, TaylorSeer, SCM, and configuration parameters
-- **[Sequence Parallelism](diffusion/parallelism_acceleration.md#sequence-parallelism)** - Guidance on how to set sequence parallelism with configuration.
-- **[CFG-Parallel](diffusion/parallelism_acceleration.md#cfg-parallel)** - Guidance on how to set CFG-Parallel to run positive/negative branches across ranks.
+- **[TeaCache Guide](acceleration/teacache.md)** - Complete TeaCache documentation, configuration options, and best practices
+- **[Cache-DiT Acceleration Guide](acceleration/cache_dit_acceleration.md)** - Comprehensive Cache-DiT guide covering DBCache, TaylorSeer, SCM, and configuration parameters
+- **[Sequence Parallelism](acceleration/parallelism_acceleration.md#sequence-parallelism)** - Guidance on how to set sequence parallelism with configuration.

docs/user_guide/examples/offline_inference/image_to_video.md

@@ -1,66 +0,0 @@
-# Image-To-Video
-
-Source <https://github.com/vllm-project/vllm-omni/tree/main/examples/offline_inference/image_to_video>.
-
-
-This example demonstrates how to generate videos from images using Wan2.2 Image-to-Video models with vLLM-Omni's offline inference API.
-
-## Local CLI Usage
-
-### Wan2.2-I2V-A14B-Diffusers (MoE)
-```bash
-python image_to_video.py \
-  --model Wan-AI/Wan2.2-I2V-A14B-Diffusers \
-  --image input.png \
-  --prompt "A cat playing with yarn, smooth motion" \
-  --negative_prompt "<optional quality filter>" \
-  --height 480 \
-  --width 832 \
-  --num_frames 48 \
-  --guidance_scale 5.0 \
-  --guidance_scale_high 6.0 \
-  --num_inference_steps 40 \
-  --boundary_ratio 0.875 \
-  --flow_shift 12.0 \
-  --fps 16 \
-  --output i2v_output.mp4
-```
-
-### Wan2.2-TI2V-5B-Diffusers (Unified)
-```bash
-python image_to_video.py \
-  --model Wan-AI/Wan2.2-TI2V-5B-Diffusers \
-  --image input.png \
-  --prompt "A cat playing with yarn, smooth motion" \
-  --negative_prompt "<optional quality filter>" \
-  --height 480 \
-  --width 832 \
-  --num_frames 48 \
-  --guidance_scale 4.0 \
-  --num_inference_steps 40 \
-  --flow_shift 12.0 \
-  --fps 16 \
-  --output i2v_output.mp4
-```
-
-Key arguments:
-
-- `--model`: Model ID (I2V-A14B for MoE, TI2V-5B for unified T2V+I2V).
-- `--image`: Path to input image (required).
-- `--prompt`: Text description of desired motion/animation.
-- `--height/--width`: Output resolution (auto-calculated from image if not set). Dimensions should be multiples of 16.
-- `--num_frames`: Number of frames (default 81).
-- `--guidance_scale` and `--guidance_scale_high`: CFG scale (applied to low/high-noise stages for MoE).
-- `--negative_prompt`: Optional list of artifacts to suppress.
-- `--boundary_ratio`: Boundary split ratio for two-stage MoE models.
-- `--flow_shift`: Scheduler flow shift (5.0 for 720p, 12.0 for 480p).
-- `--num_inference_steps`: Number of denoising steps (default 50).
-- `--fps`: Frames per second for the saved MP4 (requires `diffusers` export_to_video).
-- `--output`: Path to save the generated video.
-
-## Example materials
-
-??? abstract "image_to_video.py"
-    ``````py
-    --8<-- "examples/offline_inference/image_to_video/image_to_video.py"
-    ``````

docs/user_guide/examples/offline_inference/lora_inference.md

@@ -1,107 +0,0 @@
-# LoRA-Inference
-
-Source <https://github.com/vllm-project/vllm-omni/tree/main/examples/offline_inference/lora_inference>.
-
-This contains examples for using LoRA (Low-Rank Adaptation) adapters with vLLM-omni diffusion models for offline inference.
-The example uses the  `stabilityai/stable-diffusion-3.5-medium` as the default model, but you can replace it with other models in vLLM-omni.
-
-## Overview
-
-Similar to vLLM, vLLM-omni uses a unified LoRA handling mechanism:
-
-- **Pre-loaded LoRA**: Loaded at initialization via `--lora-path` (pre-loaded into cache)
-- **Per-request LoRA**: Loaded on-demand. In the example, the LoRA is loaded via `--lora-request-path` in each request
-
-Both approaches use the same underlying mechanism - all LoRA adapters are handled uniformly through `set_active_adapter()`. If no LoRA request is provided in a request, all adapters are deactivated.
-
-## Usage
-
-### Pre-loaded LoRA (via --lora-path)
-
-Load a LoRA adapter at initialization. This adapter is pre-loaded into the cache and can be activated by requests:
-
-```bash
-python -m examples.offline_inference.lora_inference.lora_inference \
-    --prompt "A piece of cheesecake" \
-    --lora-path /path/to/lora/ \
-    --lora-scale 1.0 \
-    --num_inference_steps 50 \
-    --height 1024 \
-    --width 1024 \
-    --output output_preloaded.png
-```
-
-**Note**: When using `--lora-path`, the adapter is loaded at init time with a stable ID derived from the adapter path. This example activates it automatically for the request.
-
-### Per-request LoRA (via --lora-request-path)
-
-Load a LoRA adapter on-demand for each request:
-
-```bash
-python -m examples.offline_inference.lora_inference.lora_inference \
-    --prompt "A piece of cheesecake" \
-    --lora-request-path /path/to/lora/ \
-    --lora-scale 1.0 \
-    --num_inference_steps 50 \
-    --height 1024 \
-    --width 1024 \
-    --output output_per_request.png
-```
-
-### No LoRA
-
-If no LoRA request is provided, we will use the base model without any LoRA adapters:
-
-```bash
-python -m examples.offline_inference.lora_inference.lora_inference \
-    --prompt "A piece of cheesecake" \
-    --num_inference_steps 50 \
-    --height 1024 \
-    --width 1024 \
-    --output output_no_lora.png
-```
-
-## Parameters
-
-### LoRA Parameters
-
-- `--lora-path`: Path to LoRA adapter folder to pre-load at initialization (loads into cache with a stable ID derived from the path)
-- `--lora-request-path`: Path to LoRA adapter folder for per-request loading
-- `--lora-request-id`: Integer ID for the LoRA adapter (optional). If not provided and `--lora-request-path` is set, will derive a stable ID from the path.
-- `--lora-scale`: Scale factor for LoRA weights (default: 1.0). Higher values increase the influence of the LoRA adapter.
-
-### Standard Parameters
-
-- `--prompt`: Text prompt for image generation (required)
-- `--seed`: Random seed for reproducibility (default: 42)
-- `--height`: Image height in pixels (default: 1024)
-- `--width`: Image width in pixels (default: 1024)
-- `--num_inference_steps`: Number of denoising steps (default: 50)
-- `--output`: Output file path (default: `lora_output.png`)
-
-## How LoRA Works
-
-All LoRA adapters are handled uniformly:
-
-1. **Initialization**: If `--lora-path` is provided, the adapter is loaded into cache with a stable ID derived from the adapter path
-2. **Per-request**: If `--lora-request-path` is provided, the adapter is loaded/activated for that request
-3. **No LoRA**: If no LoRA request is provided (`req.lora_request` is None), all adapters are deactivated
-
-The system uses LRU cache management - adapters are cached and evicted when the cache is full (unless pinned).
-
-## LoRA Adapter Format
-
-LoRA adapters must be in PEFT (Parameter-Efficient Fine-Tuning) format. A typical LoRA adapter directory structure:
-
-```
-lora_adapter/
-├── adapter_config.json
-└── adapter_model.safetensors
-```
-
-## Example materials
-
-??? abstract "lora_inference.py"
-    ``````py
-    --8<-- "examples/offline_inference/lora_inference/lora_inference.py"
-    ``````

docs/user_guide/examples/offline_inference/qwen2_5_omni.md

@@ -2,7 +2,6 @@
 
 Source <https://github.com/vllm-project/vllm-omni/tree/main/examples/offline_inference/qwen2_5_omni>.
 
-
 ## Setup
 Please refer to the [stage configuration documentation](https://docs.vllm.ai/projects/vllm-omni/en/latest/configuration/stage_configs/) to configure memory allocation appropriately for your hardware setup.
 

docs/user_guide/examples/offline_inference/qwen3_omni.md

@@ -2,7 +2,6 @@
 
 Source <https://github.com/vllm-project/vllm-omni/tree/main/examples/offline_inference/qwen3_omni>.
 
-
 ## Setup
 Please refer to the [stage configuration documentation](https://docs.vllm.ai/projects/vllm-omni/en/latest/configuration/stage_configs/) to configure memory allocation appropriately for your hardware setup.
 

docs/user_guide/examples/offline_inference/qwen3_tts.md

@@ -1,94 +0,0 @@
-# Qwen3-TTS Offline Inference
-
-Source <https://github.com/vllm-project/vllm-omni/tree/main/examples/offline_inference/qwen3_tts>.
-
-
-This directory contains an offline demo for running Qwen3 TTS models with vLLM Omni. It builds task-specific inputs and generates WAV files locally.
-
-## Model Overview
-
-Qwen3 TTS provides multiple task variants for speech generation:
-
-- **CustomVoice**: Generate speech with a known speaker identity (speaker ID) and optional instruction.
-- **VoiceDesign**: Generate speech from text plus a descriptive instruction that designs a new voice.
-- **Base**: Voice cloning using a reference audio + reference transcript, with optional mode selection.
-
-## Setup
-Please refer to the [stage configuration documentation](https://docs.vllm.ai/projects/vllm-omni/en/latest/configuration/stage_configs/) to configure memory allocation appropriately for your hardware setup.
-
-## Quick Start
-
-Run a single sample for a task:
-
-```
-python end2end.py --query-type CustomVoice
-```
-
-Generated audio files are saved to `output_audio/` by default.
-
-## Task Usage
-
-### CustomVoice
-
-Single sample:
-
-```
-python end2end.py --query-type CustomVoice
-```
-
-Batch sample (multiple prompts in one run):
-
-```
-python end2end.py --query-type CustomVoice --use-batch-sample
-```
-
-### VoiceDesign
-
-Single sample:
-
-```
-python end2end.py --query-type VoiceDesign
-```
-
-Batch sample:
-
-```
-python end2end.py --query-type VoiceDesign --use-batch-sample
-```
-
-### Base (Voice Clone)
-
-Single sample:
-
-```
-python end2end.py --query-type Base
-```
-
-Batch sample:
-
-```
-python end2end.py --query-type Base --use-batch-sample
-```
-
-Mode selection for Base:
-
-- `--mode-tag icl` (default): standard mode
-- `--mode-tag xvec_only`: enable `x_vector_only_mode` in the request
-
-Examples:
-
-```
-python end2end.py --query-type Base --mode-tag icl
-```
-
-## Notes
-
-- The script uses the model paths embedded in `end2end.py`. Update them if your local cache path differs.
-- Use `--output-dir` (preferred) or `--output-wav` to change the output folder.
-
-## Example materials
-
-??? abstract "end2end.py"
-    ``````py
-    --8<-- "examples/offline_inference/qwen3_tts/end2end.py"
-    ``````

docs/user_guide/examples/offline_inference/text_to_image.md

@@ -23,7 +23,7 @@ if __name__ == "__main__":
     images[0].save("coffee.png")
 ```
 
-Or put more than one prompt in a request.
+Or put more than one prompt in a request, processing them sequentially.
 
 ```python
 from vllm_omni.entrypoints.omni import Omni
@@ -40,41 +40,6 @@ if __name__ == "__main__":
       image = output.request_output[0].images[0].save(f"{i}.jpg")
 ```
 
-!!! info
-
-    However, it is not currently recommended to do so
-    because not all models support batch inference,
-    and batch requesting mostly does not provide significant performance improvement (despite the impression that it does).
-    This feature is primarily for the sake of interface compatibility with vLLM and to allow for future improvements.
-
-!!! info
-
-    For diffusion pipelines, the stage config field `stage_args.[].runtime.max_batch_size` is 1 by default, and the input
-    list is sliced into single-item requests before feeding into the diffusion pipeline. For models that do internally support
-    batched inputs, you can [modify this configuration](../../../configuration/stage_configs.md) to let the model accept a longer batch of prompts.
-
-Apart from string prompt, vLLM-Omni also supports dictionary prompts in the same style as vLLM.
-This is useful for models that support negative prompts.
-
-```python
-from vllm_omni.entrypoints.omni import Omni
-
-if __name__ == "__main__":
-    omni = Omni(model="Qwen/Qwen-Image")
-    outputs = omni.generate([
-      {
-        "prompt": "a cup of coffee on a table"，
-        "negative_prompt": "low resolution"
-      },
-      {
-        "prompt": "a toy dinosaur on a sandy beach"，
-        "negative_prompt": "cinematic, realistic"
-      }
-    ])
-    for i, output in enumerate(outputs):
-      image = output.request_output[0].images[0].save(f"{i}.jpg")
-```
-
 ## Local CLI Usage
 
 ```bash

docs/user_guide/examples/online_serving/gradio_demo.md

@@ -0,0 +1,7 @@
+# Gradio Demo
+
+Source <https://github.com/vllm-project/vllm-omni/blob/main/examples/online_serving/gradio_demo.py>.
+
+``````py
+--8<-- "examples/online_serving/gradio_demo.py"
+``````

docs/user_guide/examples/online_serving/image_to_image.md

@@ -230,10 +230,6 @@ Provide multiple images in `content` (order matters):
     ``````py
     --8<-- "examples/online_serving/image_to_image/openai_chat_client.py"
     ``````
-??? abstract "run_curl_image_edit.sh"
-    ``````sh
-    --8<-- "examples/online_serving/image_to_image/run_curl_image_edit.sh"
-    ``````
 ??? abstract "run_server.sh"
     ``````sh
     --8<-- "examples/online_serving/image_to_image/run_server.sh"

docs/user_guide/examples/online_serving/lora_inference.md

@@ -1,69 +0,0 @@
-# LoRA-Inference
-
-Source <https://github.com/vllm-project/vllm-omni/tree/main/examples/online_serving/lora_inference>.
-
-This example shows how to use **per-request LoRA** with vLLM-Omni diffusion models via the OpenAI-compatible Chat Completions API.
-
-> Note: The LoRA adapter path must be readable on the **server** machine (usually a local path or a mounted directory).
-> Note: This example uses `/v1/chat/completions`. LoRA payloads for other OpenAI endpoints are not implemented here.
-
-## Start Server
-
-```bash
-# Pick a diffusion model (examples)
-# export MODEL=stabilityai/stable-diffusion-3.5-medium
-# export MODEL=Qwen/Qwen-Image
-
-bash run_server.sh
-```
-
-## Call API (curl)
-
-```bash
-# Required: local LoRA folder on the server
-export LORA_PATH=/path/to/lora_adapter
-
-# Optional
-export SERVER=http://localhost:8091
-export PROMPT="A piece of cheesecake"
-export LORA_NAME=my_lora
-export LORA_SCALE=1.0
-# Optional: if omitted, the server derives a stable id from LORA_PATH.
-# export LORA_INT_ID=123
-
-bash run_curl_lora_inference.sh
-```
-
-## Call API (Python)
-
-```bash
-python openai_chat_client.py \
-  --prompt "A piece of cheesecake" \
-  --lora-path /path/to/lora_adapter \
-  --lora-name my_lora \
-  --lora-scale 1.0 \
-  --output output.png
-```
-
-## LoRA Format
-
-LoRA adapters should be in PEFT format, for example:
-
-```
-lora_adapter/
-├── adapter_config.json
-└── adapter_model.safetensors
-```
-
-??? abstract "openai_chat_client.py"
-    ``````py
-    --8<-- "examples/online_serving/lora_inference/openai_chat_client.py"
-    ``````
-??? abstract "run_curl_lora_inference.sh"
-    ``````py
-    --8<-- "examples/online_serving/lora_inference/run_curl_lora_inference.sh"
-    ``````
-??? abstract "run_server.sh"
-    ``````py
-    --8<-- "examples/online_serving/lora_inference/run_server.sh"
-    ``````

docs/user_guide/examples/online_serving/openai_chat_completion_client_for_multimodal_generation.md

@@ -0,0 +1,7 @@
+# OpenAI Chat Completion Client For Multimodal Generation
+
+Source <https://github.com/vllm-project/vllm-omni/blob/main/examples/online_serving/openai_chat_completion_client_for_multimodal_generation.py>.
+
+``````py
+--8<-- "examples/online_serving/openai_chat_completion_client_for_multimodal_generation.py"
+``````

docs/user_guide/examples/online_serving/qwen2_5_omni.md

@@ -137,7 +137,6 @@ response = client.chat.completions.create(
 # Response contains two choices: one with text, one with audio
 print(response.choices[0].message.content)  # Text response
 print(response.choices[1].message.audio)    # Audio response
-```
 
 ## Streaming Output
 If you want to enable streaming output, please set the argument as below. The final output will be obtained just after generated by corresponding stage. Now we only support text streaming output. Other modalities can output normally.

docs/user_guide/examples/online_serving/qwen3_omni.md

@@ -15,12 +15,6 @@ Please refer to [README.md](https://github.com/vllm-project/vllm-omni/tree/main/
 vllm serve Qwen/Qwen3-Omni-30B-A3B-Instruct --omni --port 8091
 ```
 
-If you want to open async chunking for qwen3-omni, launch the server with command below
-
-```bash
-vllm serve Qwen/Qwen3-Omni-30B-A3B-Instruct --omni --port 8091 --stage-configs-path /vllm_omni/model_executor/stage_configs/qwen3_omni_moe_async_chunk.yaml
-```
-
 If you have custom stage configs file, launch the server with command below
 ```bash
 vllm serve Qwen/Qwen3-Omni-30B-A3B-Instruct --omni --port 8091 --stage-configs-path /path/to/stage_configs_file
@@ -151,7 +145,6 @@ response = client.chat.completions.create(
 # Response contains two choices: one with text, one with audio
 print(response.choices[0].message.content)  # Text response
 print(response.choices[1].message.audio)    # Audio response
-```
 
 ## Streaming Output
 If you want to enable streaming output, please set the argument as below. The final output will be obtained just after generated by corresponding stage. Now we only support text streaming output. Other modalities can output normally.

docs/user_guide/examples/online_serving/text_to_image.md

@@ -12,8 +12,6 @@ This example demonstrates how to deploy Qwen-Image model for online image genera
 ```bash
 vllm serve Qwen/Qwen-Image --omni --port 8091
 ```
-!!! note
-    If you encounter Out-of-Memory (OOM) issues or have limited GPU memory, you can enable VAE slicing and tiling to reduce memory usage, --vae-use-slicing --vae-use-tiling
 
 ### Start with Parameters
 
@@ -45,7 +43,7 @@ curl -s http://localhost:8091/v1/chat/completions \
       "true_cfg_scale": 4.0,
       "seed": 42
     }
-  }' | jq -r '.choices[0].message.content[0].image_url.url' | cut -d',' -f2- | base64 -d > output.png
+  }' | jq -r '.choices[0].message.content[0].image_url.url' | cut -d',' -f2 | base64 -d > output.png
 ```
 
 ### Method 2: Using Python Client
@@ -119,7 +117,6 @@ Use `extra_body` to pass generation parameters:
 | `seed`                   | int   | None    | Random seed (reproducible)     |
 | `negative_prompt`        | str   | None    | Negative prompt                |
 | `num_outputs_per_prompt` | int   | 1       | Number of images to generate   |
-| `--cfg-parallel-size`.   | int   | 1       | Number of GPUs for CFG parallelism |
 
 ## Response Format
 
@@ -149,7 +146,7 @@ Use `extra_body` to pass generation parameters:
 
 ```bash
 # Extract base64 from response and decode to image
-cat response.json | jq -r '.choices[0].message.content[0].image_url.url' | cut -d',' -f2- | base64 -d > output.png
+cat response.json | jq -r '.choices[0].message.content[0].image_url.url' | cut -d',' -f2 | base64 -d > output.png
 ```
 
 ## File Description

examples/offline_inference/bagel/end2end.py

@@ -1,189 +0,0 @@
-import argparse
-import os
-from typing import cast
-
-from vllm_omni.inputs.data import OmniDiffusionSamplingParams, OmniPromptType
-
-
-def parse_args():
-    parser = argparse.ArgumentParser()
-    parser.add_argument(
-        "--model",
-        default="ByteDance-Seed/BAGEL-7B-MoT",
-        help="Path to merged model directory.",
-    )
-    parser.add_argument("--prompts", nargs="+", default=None, help="Input text prompts.")
-    parser.add_argument(
-        "--txt-prompts",
-        type=str,
-        default=None,
-        help="Path to a .txt file with one prompt per line (preferred).",
-    )
-    parser.add_argument("--prompt_type", default="text", choices=["text"])
-
-    parser.add_argument(
-        "--modality",
-        default="text2img",
-        choices=["text2img", "img2img", "img2text", "text2text"],
-        help="Modality mode to control stage execution.",
-    )
-
-    parser.add_argument(
-        "--image-path",
-        type=str,
-        default=None,
-        help="Path to input image for img2img.",
-    )
-
-    # OmniLLM init args
-    parser.add_argument("--enable-stats", action="store_true", default=False)
-    parser.add_argument("--init-sleep-seconds", type=int, default=20)
-    parser.add_argument("--batch-timeout", type=int, default=5)
-    parser.add_argument("--init-timeout", type=int, default=300)
-    parser.add_argument("--shm-threshold-bytes", type=int, default=65536)
-    parser.add_argument("--worker-backend", type=str, default="process", choices=["process", "ray"])
-    parser.add_argument("--ray-address", type=str, default=None)
-    parser.add_argument("--stage-configs-path", type=str, default=None)
-    parser.add_argument("--steps", type=int, default=50, help="Number of inference steps.")
-
-    args = parser.parse_args()
-    return args
-
-
-def main():
-    args = parse_args()
-    model_name = args.model
-    prompts: list[OmniPromptType] = []
-    try:
-        # Preferred: load from txt file (one prompt per line)
-        if getattr(args, "txt_prompts", None) and args.prompt_type == "text":
-            with open(args.txt_prompts, encoding="utf-8") as f:
-                lines = [ln.strip() for ln in f.readlines()]
-            prompts = [ln for ln in lines if ln != ""]
-            print(f"[Info] Loaded {len(prompts)} prompts from {args.txt_prompts}")
-        else:
-            prompts = args.prompts
-    except Exception as e:
-        print(f"[Error] Failed to load prompts: {e}")
-        raise
-
-    if not prompts:
-        # Default prompt for text2img test if none provided
-        prompts = ["<|im_start|>A cute cat<|im_end|>"]
-        print(f"[Info] No prompts provided, using default: {prompts}")
-    omni_outputs = []
-
-    from PIL import Image
-
-    if args.modality == "img2img":
-        from PIL import Image
-
-        from vllm_omni.entrypoints.omni_diffusion import OmniDiffusion
-
-        print("[Info] Running in img2img mode (Stage 1 only)")
-        client = OmniDiffusion(model=model_name)
-
-        if args.image_path:
-            if os.path.exists(args.image_path):
-                loaded_image = Image.open(args.image_path).convert("RGB")
-                prompts = [
-                    {
-                        "prompt": cast(str, p),
-                        "multi_modal_data": {"image": loaded_image},
-                    }
-                    for p in prompts
-                ]
-            else:
-                print(f"[Warning] Image path {args.image_path} does not exist.")
-
-        result = client.generate(
-            prompts,
-            OmniDiffusionSamplingParams(
-                seed=52,
-                need_kv_receive=False,
-                num_inference_steps=args.steps,
-            ),
-        )
-
-        # Ensure result is a list for iteration
-        if not isinstance(result, list):
-            omni_outputs = [result]
-        else:
-            omni_outputs = result
-
-    else:
-        from vllm_omni.entrypoints.omni import Omni
-
-        omni_kwargs = {}
-        if args.stage_configs_path:
-            omni_kwargs["stage_configs_path"] = args.stage_configs_path
-
-        omni_kwargs.update(
-            {
-                "log_stats": args.enable_stats,
-                "init_sleep_seconds": args.init_sleep_seconds,
-                "batch_timeout": args.batch_timeout,
-                "init_timeout": args.init_timeout,
-                "shm_threshold_bytes": args.shm_threshold_bytes,
-                "worker_backend": args.worker_backend,
-                "ray_address": args.ray_address,
-            }
-        )
-
-        omni = Omni(model=model_name, **omni_kwargs)
-
-        formatted_prompts = []
-        for p in args.prompts:
-            if args.modality == "img2text":
-                if args.image_path:
-                    loaded_image = Image.open(args.image_path).convert("RGB")
-                    final_prompt_text = f"<|im_start|>user\n<|image_pad|>\n{p}<|im_end|>\n<|im_start|>assistant\n"
-                    prompt_dict = {
-                        "prompt": final_prompt_text,
-                        "multi_modal_data": {"image": loaded_image},
-                        "modalities": ["text"],
-                    }
-                    formatted_prompts.append(prompt_dict)
-            elif args.modality == "text2text":
-                final_prompt_text = f"<|im_start|>user\n{p}<|im_end|>\n<|im_start|>assistant\n"
-                prompt_dict = {"prompt": final_prompt_text, "modalities": ["text"]}
-                formatted_prompts.append(prompt_dict)
-            else:
-                # text2img
-                final_prompt_text = f"<|im_start|>{p}<|im_end|>"
-                prompt_dict = {"prompt": final_prompt_text, "modalities": ["image"]}
-                formatted_prompts.append(prompt_dict)
-
-        params_list = omni.default_sampling_params_list
-        if args.modality == "text2img":
-            params_list[0].max_tokens = 1  # type: ignore # The first stage is a SamplingParam (vllm)
-            if len(params_list) > 1:
-                params_list[1].num_inference_steps = args.steps  # type: ignore # The second stage is an OmniDiffusionSamplingParam
-
-        omni_outputs = list(omni.generate(prompts=formatted_prompts, sampling_params_list=params_list))
-
-    for i, req_output in enumerate(omni_outputs):
-        images = getattr(req_output, "images", None)
-        if not images and hasattr(req_output, "output"):
-            if isinstance(req_output.output, list):
-                images = req_output.output
-            else:
-                images = [req_output.output]
-
-        if images:
-            for j, img in enumerate(images):
-                img.save(f"output_{i}_{j}.png")
-
-        if hasattr(req_output, "request_output") and req_output.request_output:
-            for stage_out in req_output.request_output:
-                if hasattr(stage_out, "images") and stage_out.images:
-                    for k, img in enumerate(stage_out.images):
-                        save_path = f"output_{i}_stage_{getattr(stage_out, 'stage_id', '?')}_{k}.png"
-                        img.save(save_path)
-                        print(f"[Info] Saved stage output image to {save_path}")
-
-    print(omni_outputs)
-
-
-if __name__ == "__main__":
-    main()

examples/offline_inference/bagel/run_single_prompt.sh

@@ -1,6 +0,0 @@
-prompt="<|im_start|>user\nWhat is the capital of France?<|im_end|>\n<|im_start|>assistant\n"
-
-python end2end.py --model ByteDance-Seed/BAGEL-7B-MoT \
-                                 --prompt_type text \
-                                 --init-sleep-seconds 0 \
-                                 --prompts ${prompt}

examples/offline_inference/image_to_image/image_edit.py

@@ -49,22 +49,6 @@ Usage (layered):
         --layers 4 \
         --color-format "RGBA"
 
-Usage (with CFG Parallel):
-    python image_edit.py \
-        --image input.png \
-        --prompt "Edit description" \
-        --cfg_parallel_size 2 \
-        --num_inference_steps 50 \
-        --cfg_scale 4.0
-
-Usage (disable torch.compile):
-    python image_edit.py \
-        --image input.png \
-        --prompt "Edit description" \
-        --enforce_eager \
-        --num_inference_steps 50 \
-        --cfg_scale 4.0
-
 For more options, run:
     python image_edit.py --help
 """
@@ -79,7 +63,6 @@ from PIL import Image
 
 from vllm_omni.diffusion.data import DiffusionParallelConfig, logger
 from vllm_omni.entrypoints.omni import Omni
-from vllm_omni.inputs.data import OmniDiffusionSamplingParams
 from vllm_omni.outputs import OmniRequestOutput
 from vllm_omni.utils.platform_utils import detect_device_type, is_npu
 
@@ -111,7 +94,7 @@ def parse_args() -> argparse.Namespace:
     parser.add_argument(
         "--negative_prompt",
         type=str,
-        default=None,
+        default=" ",
         required=False,
     )
     parser.add_argument(
@@ -182,12 +165,6 @@ def parse_args() -> argparse.Namespace:
         default=1,
         help="Number of GPUs used for ring sequence parallelism.",
     )
-    parser.add_argument(
-        "--tensor_parallel_size",
-        type=int,
-        default=1,
-        help="Number of GPUs used for tensor parallelism (TP) inside the DiT.",
-    )
     parser.add_argument("--layers", type=int, default=4, help="Number of layers to decompose the input image into.")
     parser.add_argument(
         "--resolution",
@@ -268,23 +245,7 @@ def parse_args() -> argparse.Namespace:
         default=0.2,
         help="[tea_cache] Threshold for accumulated relative L1 distance.",
     )
-    parser.add_argument(
-        "--cfg_parallel_size",
-        type=int,
-        default=1,
-        choices=[1, 2],
-        help="Number of GPUs used for classifier free guidance parallel size.",
-    )
-    parser.add_argument(
-        "--enforce_eager",
-        action="store_true",
-        help="Disable torch.compile and force eager execution.",
-    )
-    parser.add_argument(
-        "--enable-cpu-offload",
-        action="store_true",
-        help="Enable CPU offloading for diffusion models.",
-    )
+
     return parser.parse_args()
 
 
@@ -312,13 +273,7 @@ def main():
     # Enable VAE memory optimizations on NPU
     vae_use_slicing = is_npu()
     vae_use_tiling = is_npu()
-    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,
-    )
-
+    parallel_config = DiffusionParallelConfig(ulysses_degree=args.ulysses_degree, ring_degree=args.ring_degree)
     # Configure cache based on backend type
     cache_config = None
     if args.cache_backend == "cache_dit":
@@ -349,14 +304,9 @@ def main():
         cache_backend=args.cache_backend,
         cache_config=cache_config,
         parallel_config=parallel_config,
-        enforce_eager=args.enforce_eager,
-        enable_cpu_offload=args.enable_cpu_offload,
     )
     print("Pipeline loaded")
 
-    # Check if profiling is requested via environment variable
-    profiler_enabled = bool(os.getenv("VLLM_TORCH_PROFILER_DIR"))
-
     # Time profiling for generation
     print(f"\n{'=' * 60}")
     print("Generation Configuration:")
@@ -369,57 +319,31 @@ def main():
             print(f"    Image {idx + 1} size: {img.size}")
     else:
         print(f"  Input image size: {input_image.size}")
-    print(
-        f"  Parallel configuration: ulysses_degree={args.ulysses_degree}, ring_degree={args.ring_degree}, cfg_parallel_size={args.cfg_parallel_size}, tensor_parallel_size={args.tensor_parallel_size}"
-    )
+    print(f"  Parallel configuration: ulysses_degree={args.ulysses_degree}, ring_degree={args.ring_degree}")
     print(f"{'=' * 60}\n")
 
     generation_start = time.perf_counter()
-
-    if profiler_enabled:
-        print("[Profiler] Starting profiling...")
-        omni.start_profile()
-
     # Generate edited image
-    outputs = omni.generate(
-        {
-            "prompt": args.prompt,
-            "negative_prompt": args.negative_prompt,
-            "multi_modal_data": {"image": input_image},
-        },
-        OmniDiffusionSamplingParams(
-            generator=generator,
-            true_cfg_scale=args.cfg_scale,
-            guidance_scale=args.guidance_scale,
-            num_inference_steps=args.num_inference_steps,
-            num_outputs_per_prompt=args.num_outputs_per_prompt,
-            layers=args.layers,
-            resolution=args.resolution,
-        ),
-    )
+    generate_kwargs = {
+        "prompt": args.prompt,
+        "pil_image": input_image,
+        "negative_prompt": args.negative_prompt,
+        "generator": generator,
+        "true_cfg_scale": args.cfg_scale,
+        "guidance_scale": args.guidance_scale,
+        "num_inference_steps": args.num_inference_steps,
+        "num_outputs_per_prompt": args.num_outputs_per_prompt,
+        "layers": args.layers,
+        "resolution": args.resolution,
+    }
+
+    outputs = omni.generate(**generate_kwargs)
     generation_end = time.perf_counter()
     generation_time = generation_end - generation_start
 
     # Print profiling results
     print(f"Total generation time: {generation_time:.4f} seconds ({generation_time * 1000:.2f} ms)")
 
-    if profiler_enabled:
-        print("\n[Profiler] Stopping profiler and collecting results...")
-        profile_results = omni.stop_profile()
-        if profile_results and isinstance(profile_results, dict):
-            traces = profile_results.get("traces", [])
-            print("\n" + "=" * 60)
-            print("PROFILING RESULTS:")
-            for rank, trace in enumerate(traces):
-                print(f"\nRank {rank}:")
-                if trace:
-                    print(f"  • Trace: {trace}")
-            if not traces:
-                print("  No traces collected.")
-            print("=" * 60)
-        else:
-            print("[Profiler] No valid profiling data returned.")
-
     if not outputs:
         raise ValueError("No output generated from omni.generate()")
     logger.info("Outputs: %s", outputs)

examples/offline_inference/image_to_image/image_to_image.md

@@ -47,4 +47,3 @@ Key arguments:
 - `--guidance_scale`: guidance scale for guidance-distilled models (default: 1.0, disabled). Unlike classifier-free guidance (--cfg_scale), guidance-distilled models take the guidance scale directly as an input parameter. Enabled when guidance_scale > 1. Ignored when not using guidance-distilled models.
 - `--num_inference_steps`: diffusion sampling steps (more steps = higher quality, slower).
 - `--output`: path to save the generated PNG.
-- `--enable-cpu-offload`: enable CPU offloading for diffusion models.

examples/offline_inference/image_to_video/README.md

@@ -54,4 +54,3 @@ Key arguments:
 - `--num_inference_steps`: Number of denoising steps (default 50).
 - `--fps`: Frames per second for the saved MP4 (requires `diffusers` export_to_video).
 - `--output`: Path to save the generated video.
-- `--enable-cpu-offload`: enable CPU offloading for diffusion models.

examples/offline_inference/image_to_video/image_to_video.py

@@ -19,7 +19,6 @@ Usage:
 """
 
 import argparse
-import os
 from pathlib import Path
 
 import numpy as np
@@ -27,7 +26,6 @@ import PIL.Image
 import torch
 
 from vllm_omni.entrypoints.omni import Omni
-from vllm_omni.inputs.data import OmniDiffusionSamplingParams
 from vllm_omni.outputs import OmniRequestOutput
 from vllm_omni.utils.platform_utils import detect_device_type, is_npu
 
@@ -59,11 +57,6 @@ def parse_args() -> argparse.Namespace:
     )
     parser.add_argument("--output", type=str, default="i2v_output.mp4", help="Path to save the video (mp4).")
     parser.add_argument("--fps", type=int, default=16, help="Frames per second for the output video.")
-    parser.add_argument(
-        "--enable-cpu-offload",
-        action="store_true",
-        help="Enable CPU offloading for diffusion models.",
-    )
     return parser.parse_args()
 
 
@@ -102,38 +95,26 @@ def main():
     vae_use_slicing = is_npu()
     vae_use_tiling = is_npu()
 
-    # Check if profiling is requested via environment variable
-    profiler_enabled = bool(os.getenv("VLLM_TORCH_PROFILER_DIR"))
-
     omni = Omni(
         model=args.model,
         vae_use_slicing=vae_use_slicing,
         vae_use_tiling=vae_use_tiling,
         boundary_ratio=args.boundary_ratio,
         flow_shift=args.flow_shift,
-        enable_cpu_offload=args.enable_cpu_offload,
     )
 
-    if profiler_enabled:
-        print("[Profiler] Starting profiling...")
-        omni.start_profile()
-
     # omni.generate() returns Generator[OmniRequestOutput, None, None]
     frames = omni.generate(
-        {
-            "prompt": args.prompt,
-            "negative_prompt": args.negative_prompt,
-            "multi_modal_data": {"image": image},
-        },
-        OmniDiffusionSamplingParams(
-            height=height,
-            width=width,
-            generator=generator,
-            guidance_scale=args.guidance_scale,
-            guidance_scale_2=args.guidance_scale_high,
-            num_inference_steps=args.num_inference_steps,
-            num_frames=args.num_frames,
-        ),
+        args.prompt,
+        negative_prompt=args.negative_prompt,
+        pil_image=image,
+        height=height,
+        width=width,
+        generator=generator,
+        guidance_scale=args.guidance_scale,
+        guidance_scale_2=args.guidance_scale_high,
+        num_inference_steps=args.num_inference_steps,
+        num_frames=args.num_frames,
     )
 
     # Extract video frames from OmniRequestOutput
@@ -197,23 +178,6 @@ def main():
     export_to_video(video_array, str(output_path), fps=args.fps)
     print(f"Saved generated video to {output_path}")
 
-    if profiler_enabled:
-        print("\n[Profiler] Stopping profiler and collecting results...")
-        profile_results = omni.stop_profile()
-        if profile_results and isinstance(profile_results, dict):
-            traces = profile_results.get("traces", [])
-            print("\n" + "=" * 60)
-            print("PROFILING RESULTS:")
-            for rank, trace in enumerate(traces):
-                print(f"\nRank {rank}:")
-                if trace:
-                    print(f"  • Trace: {trace}")
-            if not traces:
-                print("  No traces collected.")
-            print("=" * 60)
-        else:
-            print("[Profiler] No valid profiling data returned.")
-
 
 if __name__ == "__main__":
     main()

examples/offline_inference/lora_inference/README.md

@@ -1,98 +0,0 @@
-# LoRA Inference Examples
-
-This directory contains examples for using LoRA (Low-Rank Adaptation) adapters with vLLM-omni diffusion models for offline inference.
-The example uses the  `stabilityai/stable-diffusion-3.5-medium` as the default model, but you can replace it with other models in vLLM-omni.
-
-## Overview
-
-Similar to vLLM, vLLM-omni uses a unified LoRA handling mechanism:
-
-- **Pre-loaded LoRA**: Loaded at initialization via `--lora-path` (pre-loaded into cache)
-- **Per-request LoRA**: Loaded on-demand. In the example, the LoRA is loaded via `--lora-request-path` in each request
-
-Both approaches use the same underlying mechanism - all LoRA adapters are handled uniformly through `set_active_adapter()`. If no LoRA request is provided in a request, all adapters are deactivated.
-
-## Usage
-
-### Pre-loaded LoRA (via --lora-path)
-
-Load a LoRA adapter at initialization. This adapter is pre-loaded into the cache and can be activated by requests:
-
-```bash
-python -m examples.offline_inference.lora_inference.lora_inference \
-    --prompt "A piece of cheesecake" \
-    --lora-path /path/to/lora/ \
-    --lora-scale 1.0 \
-    --num_inference_steps 50 \
-    --height 1024 \
-    --width 1024 \
-    --output output_preloaded.png
-```
-
-**Note**: When using `--lora-path`, the adapter is loaded at init time with a stable ID derived from the adapter path. This example activates it automatically for the request.
-
-### Per-request LoRA (via --lora-request-path)
-
-Load a LoRA adapter on-demand for each request:
-
-```bash
-python -m examples.offline_inference.lora_inference.lora_inference \
-    --prompt "A piece of cheesecake" \
-    --lora-request-path /path/to/lora/ \
-    --lora-scale 1.0 \
-    --num_inference_steps 50 \
-    --height 1024 \
-    --width 1024 \
-    --output output_per_request.png
-```
-
-### No LoRA
-
-If no LoRA request is provided, we will use the base model without any LoRA adapters:
-
-```bash
-python -m examples.offline_inference.lora_inference.lora_inference \
-    --prompt "A piece of cheesecake" \
-    --num_inference_steps 50 \
-    --height 1024 \
-    --width 1024 \
-    --output output_no_lora.png
-```
-
-## Parameters
-
-### LoRA Parameters
-
-- `--lora-path`: Path to LoRA adapter folder to pre-load at initialization (loads into cache with a stable ID derived from the path)
-- `--lora-request-path`: Path to LoRA adapter folder for per-request loading
-- `--lora-request-id`: Integer ID for the LoRA adapter (optional). If not provided and `--lora-request-path` is set, will derive a stable ID from the path.
-- `--lora-scale`: Scale factor for LoRA weights (default: 1.0). Higher values increase the influence of the LoRA adapter.
-
-### Standard Parameters
-
-- `--prompt`: Text prompt for image generation (required)
-- `--seed`: Random seed for reproducibility (default: 42)
-- `--height`: Image height in pixels (default: 1024)
-- `--width`: Image width in pixels (default: 1024)
-- `--num_inference_steps`: Number of denoising steps (default: 50)
-- `--output`: Output file path (default: `lora_output.png`)
-
-## How LoRA Works
-
-All LoRA adapters are handled uniformly:
-
-1. **Initialization**: If `--lora-path` is provided, the adapter is loaded into cache with a stable ID derived from the adapter path
-2. **Per-request**: If `--lora-request-path` is provided, the adapter is loaded/activated for that request
-3. **No LoRA**: If no LoRA request is provided (`req.lora_request` is None), all adapters are deactivated
-
-The system uses LRU cache management - adapters are cached and evicted when the cache is full (unless pinned).
-
-## LoRA Adapter Format
-
-LoRA adapters must be in PEFT (Parameter-Efficient Fine-Tuning) format. A typical LoRA adapter directory structure:
-
-```
-lora_adapter/
-├── adapter_config.json
-└── adapter_model.safetensors
-```

examples/offline_inference/lora_inference/lora_inference.py

@@ -1,147 +0,0 @@
-# SPDX-License-Identifier: Apache-2.0
-# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
-
-import argparse
-from pathlib import Path
-
-from vllm_omni.entrypoints.omni import Omni
-from vllm_omni.inputs.data import OmniDiffusionSamplingParams
-from vllm_omni.lora.request import LoRARequest
-from vllm_omni.lora.utils import stable_lora_int_id
-
-
-def parse_args() -> argparse.Namespace:
-    parser = argparse.ArgumentParser(description="Generate images with LoRA adapters.")
-    parser.add_argument("--model", default="stabilityai/stable-diffusion-3.5-medium", help="Model name or path.")
-    parser.add_argument("--prompt", required=True, help="Text prompt for image generation.")
-    parser.add_argument("--seed", type=int, default=42, help="Random seed for deterministic results.")
-    parser.add_argument("--height", type=int, default=1024, help="Height of generated image.")
-    parser.add_argument("--width", type=int, default=1024, help="Width of generated image.")
-    parser.add_argument(
-        "--num_inference_steps",
-        type=int,
-        default=50,
-        help="Number of denoising steps for the diffusion sampler.",
-    )
-    parser.add_argument(
-        "--output",
-        type=str,
-        default="lora_output.png",
-        help="Path to save the generated image (PNG).",
-    )
-    parser.add_argument(
-        "--lora-path",
-        type=str,
-        default=None,
-        help="Path to LoRA adapter folder to pre-load at initialization (PEFT format). "
-        "Note: pre-loading populates the cache; you still need to pass a lora_request to activate it.",
-    )
-    parser.add_argument(
-        "--lora-request-path",
-        type=str,
-        default=None,
-        help="Path to LoRA adapter folder for per-request activation (dynamic LoRA). "
-        "If --lora-request-id is not provided, a stable ID will be derived from this path.",
-    )
-    parser.add_argument(
-        "--lora-request-id",
-        type=int,
-        default=None,
-        help="Integer ID for the LoRA adapter (for dynamic LoRA). "
-        "If not provided and --lora-request-path is set, will derive a stable ID from the path.",
-    )
-    parser.add_argument(
-        "--lora-scale",
-        type=float,
-        default=1.0,
-        help="Scale factor for LoRA weights (default: 1.0).",
-    )
-    return parser.parse_args()
-
-
-def main():
-    args = parse_args()
-
-    model = args.model
-
-    omni_kwargs = {}
-
-    if args.lora_path:
-        omni_kwargs["lora_path"] = args.lora_path
-        print(f"Using static LoRA from: {args.lora_path}")
-
-    omni = Omni(model=model, **omni_kwargs)
-
-    lora_request = None
-    if args.lora_request_path:
-        if args.lora_request_id is None:
-            lora_request_id = stable_lora_int_id(args.lora_request_path)
-        else:
-            lora_request_id = args.lora_request_id
-
-        lora_name = Path(args.lora_request_path).stem
-        lora_request = LoRARequest(
-            lora_name=lora_name,
-            lora_int_id=lora_request_id,
-            lora_path=args.lora_request_path,
-        )
-        print(f"Using per-request LoRA: name={lora_name}, id={lora_request_id}, scale={args.lora_scale}")
-    elif args.lora_path:
-        # pre-loaded LoRA
-        lora_request_id = stable_lora_int_id(args.lora_path)
-        lora_request = LoRARequest(
-            lora_name="preloaded",
-            lora_int_id=lora_request_id,
-            lora_path=args.lora_path,
-        )
-        print(f"Activating pre-loaded LoRA: id={lora_request_id}, scale={args.lora_scale}")
-
-    sampling_params = OmniDiffusionSamplingParams(
-        height=args.height,
-        width=args.width,
-        num_inference_steps=args.num_inference_steps,
-    )
-
-    if lora_request:
-        sampling_params.lora_request = lora_request
-        sampling_params.lora_scale = args.lora_scale
-
-    outputs = omni.generate(args.prompt, sampling_params)
-
-    if not outputs or len(outputs) == 0:
-        raise ValueError("No output generated from omni.generate()")
-
-    if isinstance(outputs, list):
-        first_output = outputs[0]
-    else:
-        first_output = outputs
-
-    images = None
-    if hasattr(first_output, "images") and first_output.images:
-        images = first_output.images
-    elif hasattr(first_output, "request_output") and first_output.request_output:
-        req_out = first_output.request_output
-        if isinstance(req_out, list) and len(req_out) > 0:
-            req_out = req_out[0]
-        if hasattr(req_out, "images") and req_out.images:
-            images = req_out.images
-
-    if not images:
-        raise ValueError("No images found in request_output")
-
-    output_path = Path(args.output)
-    output_path.parent.mkdir(parents=True, exist_ok=True)
-    suffix = output_path.suffix or ".png"
-    stem = output_path.stem or "lora_output"
-    if len(images) <= 1:
-        images[0].save(output_path)
-        print(f"Saved generated image to {output_path}")
-    else:
-        for idx, img in enumerate(images):
-            save_path = output_path.parent / f"{stem}_{idx}{suffix}"
-            img.save(save_path)
-            print(f"Saved generated image to {save_path}")
-
-
-if __name__ == "__main__":
-    main()

examples/offline_inference/qwen2_5_omni/end2end.py

@@ -278,9 +278,9 @@ def get_audio_query(question: str = None, audio_path: str | None = None, samplin
 
 
 query_map = {
-    "use_mixed_modalities": get_mixed_modalities_query,
+    "mixed_modalities": get_mixed_modalities_query,
     "use_audio_in_video": get_use_audio_in_video_query,
-    "use_multi_audios": get_multi_audios_query,
+    "multi_audios": get_multi_audios_query,
     "use_image": get_image_query,
     "use_video": get_video_query,
     "use_audio": get_audio_query,
@@ -434,7 +434,7 @@ def parse_args():
         "--query-type",
         "-q",
         type=str,
-        default="use_mixed_modalities",
+        default="mixed_modalities",
         choices=query_map.keys(),
         help="Query type.",
     )

examples/offline_inference/qwen2_5_omni/run_single_prompt.sh

@@ -1,2 +1,2 @@
 python end2end.py --output-wav output_audio \
-                  --query-type use_mixed_modalities
+                  --query-type mixed_modalities

examples/offline_inference/qwen3_omni/end2end.py

@@ -12,7 +12,6 @@ from typing import NamedTuple
 import librosa
 import numpy as np
 import soundfile as sf
-import vllm
 from PIL import Image
 from vllm import SamplingParams
 from vllm.assets.audio import AudioAsset
@@ -226,76 +225,18 @@ def get_multi_audios_query() -> QueryResult:
     )
 
 
-# def get_use_audio_in_video_query(video_path: str | None = None) -> QueryResult:
-# question = (
-#     "Describe the content of the video in details, then convert what the "
-#     "baby say into text."
-# )
-# prompt = (
-#     f"<|im_start|>system\n{default_system}<|im_end|>\n"
-#     "<|im_start|>user\n<|vision_start|><|video_pad|><|vision_end|>"
-#     f"{question}<|im_end|>\n"
-#     f"<|im_start|>assistant\n"
-# )
-# if video_path:
-#     if not os.path.exists(video_path):
-#         raise FileNotFoundError(f"Video file not found: {video_path}")
-#     video_frames = video_to_ndarrays(video_path, num_frames=16)
-# else:
-#     video_frames = VideoAsset(name="baby_reading", num_frames=16).np_ndarrays
-# audio = extract_video_audio(video_path, sampling_rate=16000)
-# return QueryResult(
-#     inputs={
-#         "prompt": prompt,
-#         "multi_modal_data": {
-#             "video": video_frames,
-#             "audio": audio,
-#         },
-#         "mm_processor_kwargs": {
-#             "use_audio_in_video": True,
-#         },
-#     },
-#     limit_mm_per_prompt={"audio": 1, "video": 1},
-# )
-def get_use_audio_in_video_query() -> QueryResult:
-    question = "Describe the content of the video in details, then convert what the baby say into text."
-    prompt = (
-        f"<|im_start|>system\n{default_system}<|im_end|>\n"
-        "<|im_start|>user\n<|vision_start|><|video_pad|><|vision_end|>"
-        f"{question}<|im_end|>\n"
-        f"<|im_start|>assistant\n"
-    )
-    asset = VideoAsset(name="baby_reading", num_frames=16)
-    audio = asset.get_audio(sampling_rate=16000)
-    return QueryResult(
-        inputs={
-            "prompt": prompt,
-            "multi_modal_data": {
-                "video": asset.np_ndarrays,
-                "audio": audio,
-            },
-            "mm_processor_kwargs": {
-                "use_audio_in_video": True,
-            },
-        },
-        limit_mm_per_prompt={"audio": 1, "video": 1},
-    )
-
-
 query_map = {
     "text": get_text_query,
     "use_audio": get_audio_query,
     "use_image": get_image_query,
     "use_video": get_video_query,
-    "use_multi_audios": get_multi_audios_query,
-    "use_mixed_modalities": get_mixed_modalities_query,
-    "use_audio_in_video": get_use_audio_in_video_query,
+    "multi_audios": get_multi_audios_query,
+    "mixed_modalities": get_mixed_modalities_query,
 }
 
 
 def main(args):
     model_name = "Qwen/Qwen3-Omni-30B-A3B-Instruct"
-    print("=" * 20, "\n", f"vllm version: {vllm.__version__}", "\n", "=" * 20)
 
     # Get paths from args
     video_path = getattr(args, "video_path", None)
@@ -318,10 +259,6 @@ def main(args):
             num_frames=getattr(args, "num_frames", 16),
             sampling_rate=getattr(args, "sampling_rate", 16000),
         )
-    elif args.query_type == "multi_audios":
-        query_result = query_func()
-    elif args.query_type == "use_audio_in_video":
-        query_result = query_func()
     else:
         query_result = query_func()
 
@@ -333,7 +270,7 @@ def main(args):
     )
 
     thinker_sampling_params = SamplingParams(
-        temperature=0.9,
+        temperature=0.4,
         top_p=0.9,
         top_k=-1,
         max_tokens=1200,
@@ -394,8 +331,6 @@ def main(args):
     total_requests = len(prompts)
     processed_count = 0
 
-    print(f"query type: {args.query_type}")
-
     for stage_outputs in omni_generator:
         if stage_outputs.final_output_type == "text":
             for output in stage_outputs.request_output:
@@ -450,7 +385,7 @@ def parse_args():
         "--query-type",
         "-q",
         type=str,
-        default="use_mixed_modalities",
+        default="mixed_modalities",
         choices=query_map.keys(),
         help="Query type.",
     )

examples/offline_inference/qwen3_tts/README.md

@@ -1,93 +0,0 @@
-# Qwen3-TTS Offline Inference
-
-This directory contains an offline demo for running Qwen3 TTS models with vLLM Omni. It builds task-specific inputs and generates WAV files locally.
-
-## Model Overview
-
-Qwen3 TTS provides multiple task variants for speech generation:
-
-- **CustomVoice**: Generate speech with a known speaker identity (speaker ID) and optional instruction.
-- **VoiceDesign**: Generate speech from text plus a descriptive instruction that designs a new voice.
-- **Base**: Voice cloning using a reference audio + reference transcript, with optional mode selection.
-
-## Setup
-Please refer to the [stage configuration documentation](https://docs.vllm.ai/projects/vllm-omni/en/latest/configuration/stage_configs/) to configure memory allocation appropriately for your hardware setup.
-
-### ROCm Dependencies
-
-You will need to install these two dependencies `onnxruntime-rocm` and `sox`.
-
-```
-pip uninstall onnxruntime # should be removed before we can install onnxruntime-rocm
-pip install onnxruntime-rocm sox
-```
-
-## Quick Start
-
-Run a single sample for a task:
-
-```
-python end2end.py --query-type CustomVoice
-```
-
-Generated audio files are saved to `output_audio/` by default.
-
-## Task Usage
-
-### CustomVoice
-
-Single sample:
-
-```
-python end2end.py --query-type CustomVoice
-```
-
-Batch sample (multiple prompts in one run):
-
-```
-python end2end.py --query-type CustomVoice --use-batch-sample
-```
-
-### VoiceDesign
-
-Single sample:
-
-```
-python end2end.py --query-type VoiceDesign
-```
-
-Batch sample:
-
-```
-python end2end.py --query-type VoiceDesign --use-batch-sample
-```
-
-### Base (Voice Clone)
-
-Single sample:
-
-```
-python end2end.py --query-type Base
-```
-
-Batch sample:
-
-```
-python end2end.py --query-type Base --use-batch-sample
-```
-
-Mode selection for Base:
-
-- `--mode-tag icl` (default): standard mode
-- `--mode-tag xvec_only`: enable `x_vector_only_mode` in the request
-
-Examples:
-
-```
-python end2end.py --query-type Base --mode-tag icl
-```
-
-## Notes
-
-- The script uses the model paths embedded in `end2end.py`. Update them if your local cache path differs.
-- Use `--output-dir` (preferred) or `--output-wav` to change the output folder.

examples/offline_inference/qwen3_tts/end2end.py

@@ -1,381 +0,0 @@
-"""Offline inference demo for Qwen3 TTS via vLLM Omni.
-
-Provides single and batch sample inputs for CustomVoice, VoiceDesign, and Base
-tasks, then runs Omni generation and saves output wav files.
-"""
-
-import os
-from typing import NamedTuple
-
-import soundfile as sf
-
-os.environ["VLLM_WORKER_MULTIPROC_METHOD"] = "spawn"
-
-from vllm import SamplingParams
-from vllm.utils.argparse_utils import FlexibleArgumentParser
-
-from vllm_omni import Omni
-
-
-class QueryResult(NamedTuple):
-    """Container for a prepared Omni request."""
-
-    inputs: dict
-    model_name: str
-
-
-def get_custom_voice_query(use_batch_sample: bool = False) -> QueryResult:
-    """Build CustomVoice sample inputs.
-
-    Args:
-        use_batch_sample: When True, return a batch of prompts; otherwise a single prompt.
-
-    Returns:
-        QueryResult with Omni inputs and the CustomVoice model path.
-    """
-    task_type = "CustomVoice"
-    if use_batch_sample:
-        texts = ["其实我真的有发现，我是一个特别善于观察别人情绪的人。", "She said she would be here by noon."]
-        instructs = ["", "Very happy."]
-        languages = ["Chinese", "English"]
-        speakers = ["Vivian", "Ryan"]
-        inputs = []
-        for text, instruct, language, speaker in zip(texts, instructs, languages, speakers):
-            prompt = f"<|im_start|>assistant\n{text}<|im_end|>\n<|im_start|>assistant\n"
-            inputs.append(
-                {
-                    "prompt": prompt,
-                    "additional_information": {
-                        "task_type": [task_type],
-                        "text": [text],
-                        "instruct": [instruct],
-                        "language": [language],
-                        "speaker": [speaker],
-                        "max_new_tokens": [2048],
-                    },
-                }
-            )
-    else:
-        text = "其实我真的有发现，我是一个特别善于观察别人情绪的人。"
-        language = "Chinese"
-        speaker = "Vivian"
-        instruct = "用特别愤怒的语气说"
-        prompts = f"<|im_start|>assistant\n{text}<|im_end|>\n<|im_start|>assistant\n"
-        inputs = {
-            "prompt": prompts,
-            "additional_information": {
-                "task_type": [task_type],
-                "text": [text],
-                "language": [language],
-                "speaker": [speaker],
-                "instruct": [instruct],
-                "max_new_tokens": [2048],
-            },
-        }
-    return QueryResult(
-        inputs=inputs,
-        model_name="Qwen/Qwen3-TTS-12Hz-1.7B-CustomVoice",
-    )
-
-
-def get_voice_design_query(use_batch_sample: bool = False) -> QueryResult:
-    """Build VoiceDesign sample inputs.
-
-    Args:
-        use_batch_sample: When True, return a batch of prompts; otherwise a single prompt.
-
-    Returns:
-        QueryResult with Omni inputs and the VoiceDesign model path.
-    """
-    task_type = "VoiceDesign"
-    if use_batch_sample:
-        texts = [
-            "哥哥，你回来啦，人家等了你好久好久了，要抱抱！",
-            "It's in the top drawer... wait, it's empty? No way, that's impossible! I'm sure I put it there!",
-        ]
-        instructs = [
-            "体现撒娇稚嫩的萝莉女声，音调偏高且起伏明显，营造出黏人、做作又刻意卖萌的听觉效果。",
-            "Speak in an incredulous tone, but with a hint of panic beginning to creep into your voice.",
-        ]
-        languages = ["Chinese", "English"]
-        inputs = []
-        for text, instruct, language in zip(texts, instructs, languages):
-            prompt = f"<|im_start|>assistant\n{text}<|im_end|>\n<|im_start|>assistant\n"
-            inputs.append(
-                {
-                    "prompt": prompt,
-                    "additional_information": {
-                        "task_type": [task_type],
-                        "text": [text],
-                        "language": [language],
-                        "instruct": [instruct],
-                        "max_new_tokens": [2048],
-                        "non_streaming_mode": [True],
-                    },
-                }
-            )
-    else:
-        text = "哥哥，你回来啦，人家等了你好久好久了，要抱抱！"
-        instruct = "体现撒娇稚嫩的萝莉女声，音调偏高且起伏明显，营造出黏人、做作又刻意卖萌的听觉效果。"
-        language = "Chinese"
-        prompt = f"<|im_start|>assistant\n{text}<|im_end|>\n<|im_start|>assistant\n"
-        inputs = {
-            "prompt": prompt,
-            "additional_information": {
-                "task_type": [task_type],
-                "text": [text],
-                "language": [language],
-                "instruct": [instruct],
-                "max_new_tokens": [2048],
-                "non_streaming_mode": [True],
-            },
-        }
-    return QueryResult(
-        inputs=inputs,
-        model_name="Qwen/Qwen3-TTS-12Hz-1.7B-VoiceDesign",
-    )
-
-
-def get_base_query(use_batch_sample: bool = False, mode_tag: str = "icl") -> QueryResult:
-    """Build Base (voice clone) sample inputs.
-
-    Args:
-        use_batch_sample: When True, return a batch of prompts (Case 2).
-        mode_tag: "icl" or "xvec_only" to control x_vector_only_mode behavior.
-
-    Returns:
-        QueryResult with Omni inputs and the Base model path.
-    """
-    task_type = "Base"
-    ref_audio_path_1 = "https://qianwen-res.oss-cn-beijing.aliyuncs.com/Qwen3-TTS-Repo/clone_2.wav"
-    ref_audio_single = ref_audio_path_1
-    ref_text_single = (
-        "Okay. Yeah. I resent you. I love you. I respect you. But you know what? You blew it! And thanks to you."
-    )
-    syn_text_single = "Good one. Okay, fine, I'm just gonna leave this sock monkey here. Goodbye."
-    syn_lang_single = "Auto"
-    x_vector_only_mode = mode_tag == "xvec_only"
-    if use_batch_sample:
-        syn_text_batch = [
-            "Good one. Okay, fine, I'm just gonna leave this sock monkey here. Goodbye.",
-            "其实我真的有发现，我是一个特别善于观察别人情绪的人。",
-        ]
-        syn_lang_batch = ["Chinese", "English"]
-        inputs = []
-        for text, language in zip(syn_text_batch, syn_lang_batch):
-            prompt = f"<|im_start|>assistant\n{text}<|im_end|>\n<|im_start|>assistant\n"
-            inputs.append(
-                {
-                    "prompt": prompt,
-                    "additional_information": {
-                        "task_type": [task_type],
-                        "ref_audio": [ref_audio_single],
-                        "ref_text": [ref_text_single],
-                        "text": [text],
-                        "language": [language],
-                        "x_vector_only_mode": [x_vector_only_mode],
-                        "max_new_tokens": [2048],
-                    },
-                }
-            )
-    else:
-        prompt = f"<|im_start|>assistant\n{syn_text_single}<|im_end|>\n<|im_start|>assistant\n"
-        inputs = {
-            "prompt": prompt,
-            "additional_information": {
-                "task_type": [task_type],
-                "ref_audio": [ref_audio_single],
-                "ref_text": [ref_text_single],
-                "text": [syn_text_single],
-                "language": [syn_lang_single],
-                "x_vector_only_mode": [x_vector_only_mode],
-                "max_new_tokens": [2048],
-            },
-        }
-    return QueryResult(
-        inputs=inputs,
-        model_name="Qwen/Qwen3-TTS-12Hz-1.7B-Base",
-    )
-
-
-def main(args):
-    """Run offline inference with Omni using prepared sample inputs.
-
-    Args:
-        args: Parsed CLI args from parse_args().
-    """
-    query_func = query_map[args.query_type]
-    if args.query_type in {"CustomVoice", "VoiceDesign"}:
-        query_result = query_func(use_batch_sample=args.use_batch_sample)
-    elif args.query_type == "Base":
-        query_result = query_func(
-            use_batch_sample=args.use_batch_sample,
-            mode_tag=args.mode_tag,
-        )
-    else:
-        query_result = query_func()
-
-    model_name = query_result.model_name
-    omni = Omni(
-        model=model_name,
-        stage_configs_path=args.stage_configs_path,
-        log_stats=args.enable_stats,
-        stage_init_timeout=args.stage_init_timeout,
-    )
-
-    sampling_params = SamplingParams(
-        temperature=0.9,
-        top_p=1.0,
-        top_k=50,
-        max_tokens=2048,
-        seed=42,
-        detokenize=False,
-        repetition_penalty=1.05,
-    )
-
-    sampling_params_list = [
-        sampling_params,
-    ]
-
-    output_dir = args.output_dir if getattr(args, "output_dir", None) else args.output_wav
-    os.makedirs(output_dir, exist_ok=True)
-
-    omni_generator = omni.generate(query_result.inputs, sampling_params_list)
-    for stage_outputs in omni_generator:
-        for output in stage_outputs.request_output:
-            request_id = output.request_id
-            audio_tensor = output.multimodal_output["audio"]
-            output_wav = os.path.join(output_dir, f"output_{request_id}.wav")
-            audio_samplerate = output.multimodal_output["sr"].item()
-            # Convert to numpy array and ensure correct format
-            audio_numpy = audio_tensor.float().detach().cpu().numpy()
-
-            # Ensure audio is 1D (flatten if needed)
-            if audio_numpy.ndim > 1:
-                audio_numpy = audio_numpy.flatten()
-
-            # Save audio file with explicit WAV format
-            sf.write(output_wav, audio_numpy, samplerate=audio_samplerate, format="WAV")
-            print(f"Request ID: {request_id}, Saved audio to {output_wav}")
-
-
-def parse_args():
-    """Parse CLI arguments for offline TTS inference.
-
-    Returns:
-        argparse.Namespace with CLI options.
-    """
-    parser = FlexibleArgumentParser(description="Demo on using vLLM for offline inference with audio language models")
-    parser.add_argument(
-        "--query-type",
-        "-q",
-        type=str,
-        default="CustomVoice",
-        choices=query_map.keys(),
-        help="Query type.",
-    )
-    parser.add_argument(
-        "--enable-stats",
-        action="store_true",
-        default=False,
-        help="Enable writing detailed statistics (default: disabled)",
-    )
-    parser.add_argument(
-        "--stage-init-timeout",
-        type=int,
-        default=300,
-        help="Timeout for initializing a single stage in seconds (default: 300)",
-    )
-    parser.add_argument(
-        "--batch-timeout",
-        type=int,
-        default=5,
-        help="Timeout for batching in seconds (default: 5)",
-    )
-    parser.add_argument(
-        "--init-timeout",
-        type=int,
-        default=300,
-        help="Timeout for initializing stages in seconds (default: 300)",
-    )
-    parser.add_argument(
-        "--shm-threshold-bytes",
-        type=int,
-        default=65536,
-        help="Threshold for using shared memory in bytes (default: 65536)",
-    )
-    parser.add_argument(
-        "--output-wav",
-        default="output_audio",
-        help="[Deprecated] Output wav directory (use --output-dir).",
-    )
-    parser.add_argument(
-        "--num-prompts",
-        type=int,
-        default=1,
-        help="Number of prompts to generate.",
-    )
-    parser.add_argument(
-        "--txt-prompts",
-        type=str,
-        default=None,
-        help="Path to a .txt file with one prompt per line (preferred).",
-    )
-    parser.add_argument(
-        "--stage-configs-path",
-        type=str,
-        default=None,
-        help="Path to a stage configs file.",
-    )
-    parser.add_argument(
-        "--audio-path",
-        "-a",
-        type=str,
-        default=None,
-        help="Path to local audio file. If not provided, uses default audio asset.",
-    )
-    parser.add_argument(
-        "--sampling-rate",
-        type=int,
-        default=16000,
-        help="Sampling rate for audio loading (default: 16000).",
-    )
-    parser.add_argument(
-        "--log-dir",
-        type=str,
-        default="logs",
-        help="Log directory (default: logs).",
-    )
-    parser.add_argument(
-        "--py-generator",
-        action="store_true",
-        default=False,
-        help="Use py_generator mode. The returned type of Omni.generate() is a Python Generator object.",
-    )
-    parser.add_argument(
-        "--use-batch-sample",
-        action="store_true",
-        default=False,
-        help="Use batch input sample for CustomVoice/VoiceDesign/Base query.",
-    )
-    parser.add_argument(
-        "--mode-tag",
-        type=str,
-        default="icl",
-        choices=["icl", "xvec_only"],
-        help="Mode tag for Base query x_vector_only_mode (default: icl).",
-    )
-
-    return parser.parse_args()
-
-
-query_map = {
-    "CustomVoice": get_custom_voice_query,
-    "VoiceDesign": get_voice_design_query,
-    "Base": get_base_query,
-}
-
-
-if __name__ == "__main__":
-    args = parse_args()
-    main(args)

examples/offline_inference/text_to_audio/README.md

@@ -1,37 +0,0 @@
-# Text-To-Audio
-
-The `stabilityai/stable-audio-open-1.0` pipeline generates audio from text prompts.
-
-## Prerequisites
-
-If you use a gated model (e.g., `stabilityai/stable-audio-open-1.0`), ensure you have access:
-
-1. **Accept Model License**: Visit the model page on Hugging Face (e.g., [stabilityai/stable-audio-open-1.0]) and accept the user agreement.
-2. **Authenticate**: Log in to Hugging Face locally to access the gated model.
-   ```bash
-   huggingface-cli login
-   ```
-
-## Local CLI Usage
-
-```bash
-python text_to_audio.py \
-  --model stabilityai/stable-audio-open-1.0 \
-  --prompt "The sound of a hammer hitting a wooden surface" \
-  --negative_prompt "Low quality" \
-  --seed 42 \
-  --guidance_scale 7.0 \
-  --audio_length 10.0 \
-  --num_inference_steps 100 \
-  --output stable_audio_output.wav
-```
-
-Key arguments:
-
-- `--prompt`: text description (string).
-- `--negative_prompt`: negative prompt for classifier-free guidance.
-- `--seed`: integer seed for deterministic generation.
-- `--guidance_scale`: classifier-free guidance scale.
-- `--audio_length`: audio duration in seconds.
-- `--num_inference_steps`: diffusion sampling steps.(more steps = higher quality, slower).
-- `--output`: path to save the generated WAV file.

examples/offline_inference/text_to_audio/text_to_audio.py

@@ -1,220 +0,0 @@
-# SPDX-License-Identifier: Apache-2.0
-# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
-
-"""
-Example script for text-to-audio generation using Stable Audio Open.
-
-This script demonstrates how to generate audio from text prompts using
-the Stable Audio Open model with vLLM-Omni.
-
-Usage:
-    python text_to_audio.py --prompt "The sound of a dog barking"
-    python text_to_audio.py --prompt "A piano playing a gentle melody" --audio_length 10.0
-    python text_to_audio.py --prompt "Thunder and rain sounds" --negative_prompt "Low quality"
-"""
-
-import argparse
-import time
-from pathlib import Path
-
-import numpy as np
-import torch
-
-from vllm_omni.entrypoints.omni import Omni
-from vllm_omni.inputs.data import OmniDiffusionSamplingParams
-from vllm_omni.utils.platform_utils import detect_device_type
-
-
-def parse_args() -> argparse.Namespace:
-    parser = argparse.ArgumentParser(description="Generate audio with Stable Audio Open.")
-    parser.add_argument(
-        "--model",
-        default="stabilityai/stable-audio-open-1.0",
-        help="Stable Audio model name or local path.",
-    )
-    parser.add_argument(
-        "--prompt",
-        default="The sound of a hammer hitting a wooden surface.",
-        help="Text prompt for audio generation.",
-    )
-    parser.add_argument(
-        "--negative_prompt",
-        default="Low quality.",
-        help="Negative prompt for classifier-free guidance.",
-    )
-    parser.add_argument(
-        "--seed",
-        type=int,
-        default=42,
-        help="Random seed for deterministic results.",
-    )
-    parser.add_argument(
-        "--guidance_scale",
-        type=float,
-        default=7.0,
-        help="Classifier-free guidance scale.",
-    )
-    parser.add_argument(
-        "--audio_start",
-        type=float,
-        default=0.0,
-        help="Audio start time in seconds.",
-    )
-    parser.add_argument(
-        "--audio_length",
-        type=float,
-        default=10.0,
-        help="Audio length in seconds (max ~47s for stable-audio-open-1.0).",
-    )
-    parser.add_argument(
-        "--num_inference_steps",
-        type=int,
-        default=100,
-        help="Number of denoising steps for the diffusion sampler.",
-    )
-    parser.add_argument(
-        "--num_waveforms",
-        type=int,
-        default=1,
-        help="Number of audio waveforms to generate for the given prompt.",
-    )
-    parser.add_argument(
-        "--output",
-        type=str,
-        default="stable_audio_output.wav",
-        help="Path to save the generated audio (WAV format).",
-    )
-    parser.add_argument(
-        "--sample_rate",
-        type=int,
-        default=44100,
-        help="Sample rate for output audio (Stable Audio uses 44100 Hz).",
-    )
-    return parser.parse_args()
-
-
-def save_audio(audio_data: np.ndarray, output_path: str, sample_rate: int = 44100):
-    """Save audio data to a WAV file."""
-    try:
-        import soundfile as sf
-
-        sf.write(output_path, audio_data, sample_rate)
-    except ImportError:
-        try:
-            import scipy.io.wavfile as wav
-
-            # Ensure audio is in the correct format for scipy
-            if audio_data.dtype == np.float32 or audio_data.dtype == np.float64:
-                # Normalize to int16 range
-                audio_data = np.clip(audio_data, -1.0, 1.0)
-                audio_data = (audio_data * 32767).astype(np.int16)
-            wav.write(output_path, sample_rate, audio_data)
-        except ImportError:
-            raise ImportError(
-                "Either 'soundfile' or 'scipy' is required to save audio files. "
-                "Install with: pip install soundfile or pip install scipy"
-            )
-
-
-def main():
-    args = parse_args()
-    device = detect_device_type()
-    generator = torch.Generator(device=device).manual_seed(args.seed)
-
-    print(f"\n{'=' * 60}")
-    print("Stable Audio Open - Text-to-Audio Generation")
-    print(f"{'=' * 60}")
-    print(f"  Model: {args.model}")
-    print(f"  Prompt: {args.prompt}")
-    print(f"  Negative prompt: {args.negative_prompt}")
-    print(f"  Audio length: {args.audio_length}s")
-    print(f"  Inference steps: {args.num_inference_steps}")
-    print(f"  Guidance scale: {args.guidance_scale}")
-    print(f"  Seed: {args.seed}")
-    print(f"{'=' * 60}\n")
-
-    # Initialize Omni with Stable Audio model
-    omni = Omni(model=args.model)
-
-    # Calculate audio end time
-    audio_end_in_s = args.audio_start + args.audio_length
-
-    # Time profiling for generation
-    generation_start = time.perf_counter()
-
-    # Generate audio
-    outputs = omni.generate(
-        {
-            "prompt": args.prompt,
-            "negative_prompt": args.negative_prompt,
-        },
-        OmniDiffusionSamplingParams(
-            generator=generator,
-            guidance_scale=args.guidance_scale,
-            num_inference_steps=args.num_inference_steps,
-            num_outputs_per_prompt=args.num_waveforms,
-            extra_args={
-                "audio_start_in_s": args.audio_start,
-                "audio_end_in_s": audio_end_in_s,
-            },
-        ),
-    )
-
-    generation_end = time.perf_counter()
-    generation_time = generation_end - generation_start
-
-    print(f"Total generation time: {generation_time:.2f} seconds")
-
-    # Process and save audio
-    output_path = Path(args.output)
-    output_path.parent.mkdir(parents=True, exist_ok=True)
-    suffix = output_path.suffix or ".wav"
-    stem = output_path.stem or "stable_audio_output"
-
-    # Extract audio from omni.generate() outputs
-    if not outputs:
-        raise ValueError("No output generated from omni.generate()")
-
-    output = outputs[0]
-    if not hasattr(output, "request_output") or not output.request_output:
-        raise ValueError("No request_output found in OmniRequestOutput")
-    request_output = output.request_output[0]
-    if not hasattr(request_output, "multimodal_output"):
-        raise ValueError("No multimodal_output found in request_output")
-
-    audio = request_output.multimodal_output.get("audio")
-    if audio is None:
-        raise ValueError("No audio output found in request_output")
-
-    # Handle different output formats
-    if isinstance(audio, torch.Tensor):
-        audio = audio.cpu().float().numpy()
-
-    # Audio shape is typically [batch, channels, samples] or [channels, samples]
-    if audio.ndim == 3:
-        # [batch, channels, samples]
-        if args.num_waveforms <= 1:
-            audio_data = audio[0].T  # [samples, channels]
-            save_audio(audio_data, str(output_path), args.sample_rate)
-            print(f"Saved generated audio to {output_path}")
-        else:
-            for idx in range(audio.shape[0]):
-                audio_data = audio[idx].T  # [samples, channels]
-                save_path = output_path.parent / f"{stem}_{idx}{suffix}"
-                save_audio(audio_data, str(save_path), args.sample_rate)
-                print(f"Saved generated audio to {save_path}")
-    elif audio.ndim == 2:
-        # [channels, samples]
-        audio_data = audio.T  # [samples, channels]
-        save_audio(audio_data, str(output_path), args.sample_rate)
-        print(f"Saved generated audio to {output_path}")
-    else:
-        # [samples] - mono audio
-        save_audio(audio, str(output_path), args.sample_rate)
-        print(f"Saved generated audio to {output_path}")
-
-    print(f"\nGenerated {args.audio_length}s of audio at {args.sample_rate} Hz")
-
-
-if __name__ == "__main__":
-    main()

examples/offline_inference/text_to_image/README.md

@@ -20,7 +20,7 @@ if __name__ == "__main__":
     images[0].save("coffee.png")
 ```
 
-Or put more than one prompt in a request.
+Or put more than one prompt in a request, processing them sequentially.
 
 ```python
 from vllm_omni.entrypoints.omni import Omni
@@ -37,41 +37,6 @@ if __name__ == "__main__":
       image = output.request_output[0].images[0].save(f"{i}.jpg")
 ```
 
-!!! info
-
-    However, it is not currently recommended to do so
-    because not all models support batch inference,
-    and batch requesting mostly does not provide significant performance improvement (despite the impression that it does).
-    This feature is primarily for the sake of interface compatibility with vLLM and to allow for future improvements.
-
-!!! info
-
-    For diffusion pipelines, the stage config field `stage_args.[].runtime.max_batch_size` is 1 by default, and the input
-    list is sliced into single-item requests before feeding into the diffusion pipeline. For models that do internally support
-    batched inputs, you can [modify this configuration](../../../configuration/stage_configs.md) to let the model accept a longer batch of prompts.
-
-Apart from string prompt, vLLM-Omni also supports dictionary prompts in the same style as vLLM.
-This is useful for models that support negative prompts.
-
-```python
-from vllm_omni.entrypoints.omni import Omni
-
-if __name__ == "__main__":
-    omni = Omni(model="Qwen/Qwen-Image")
-    outputs = omni.generate([
-      {
-        "prompt": "a cup of coffee on a table"，
-        "negative_prompt": "low resolution"
-      },
-      {
-        "prompt": "a toy dinosaur on a sandy beach"，
-        "negative_prompt": "cinematic, realistic"
-      }
-    ])
-    for i, output in enumerate(outputs):
-      image = output.request_output[0].images[0].save(f"{i}.jpg")
-```
-
 ## Local CLI Usage
 
 ```bash
@@ -96,7 +61,6 @@ Key arguments:
 - `--num_inference_steps`: diffusion sampling steps (more steps = higher quality, slower).
 - `--height/--width`: output resolution (defaults 1024x1024).
 - `--output`: path to save the generated PNG.
-- `--enable-cpu-offload`: enable CPU offloading for diffusion models.
 
 > ℹ️ Qwen-Image currently publishes best-effort presets at `1328x1328`, `1664x928`, `928x1664`, `1472x1140`, `1140x1472`, `1584x1056`, and `1056x1584`. Adjust `--height/--width` accordingly for the most reliable outcomes.
 

examples/offline_inference/text_to_image/gradio_demo.py

@@ -5,7 +5,6 @@ import gradio as gr
 import torch
 
 from vllm_omni.entrypoints.omni import Omni
-from vllm_omni.inputs.data import OmniDiffusionSamplingParams
 from vllm_omni.outputs import OmniRequestOutput
 from vllm_omni.utils.platform_utils import detect_device_type, is_npu
 
@@ -102,14 +101,12 @@ def build_demo(args: argparse.Namespace) -> gr.Blocks:
         generator = torch.Generator(device=device).manual_seed(seed)
         outputs = omni.generate(
             prompt.strip(),
-            OmniDiffusionSamplingParams(
-                height=height,
-                width=width,
-                generator=generator,
-                true_cfg_scale=float(cfg_scale_value),
-                num_inference_steps=num_steps,
-                num_outputs_per_prompt=num_images,
-            ),
+            height=height,
+            width=width,
+            generator=generator,
+            true_cfg_scale=float(cfg_scale_value),
+            num_inference_steps=num_steps,
+            num_outputs_per_prompt=num_images,
         )
         images_outputs = []
         for output in outputs:

examples/offline_inference/text_to_image/text_to_image.py

@@ -2,7 +2,6 @@
 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project
 
 import argparse
-import os
 import time
 from pathlib import Path
 
@@ -10,7 +9,6 @@ import torch
 
 from vllm_omni.diffusion.data import DiffusionParallelConfig, logger
 from vllm_omni.entrypoints.omni import Omni
-from vllm_omni.inputs.data import OmniDiffusionSamplingParams
 from vllm_omni.outputs import OmniRequestOutput
 from vllm_omni.utils.platform_utils import detect_device_type, is_npu
 
@@ -25,9 +23,7 @@ def parse_args() -> argparse.Namespace:
     )
     parser.add_argument("--prompt", default="a cup of coffee on the table", help="Text prompt for image generation.")
     parser.add_argument(
-        "--negative_prompt",
-        default=None,
-        help="negative prompt for classifier-free conditional guidance.",
+        "--negative_prompt", default="", help="negative prompt for classifier-free conditional guidance."
     )
     parser.add_argument("--seed", type=int, default=142, help="Random seed for deterministic results.")
     parser.add_argument(
@@ -85,29 +81,6 @@ def parse_args() -> argparse.Namespace:
         default=1,
         help="Number of GPUs used for ring sequence parallelism.",
     )
-    parser.add_argument(
-        "--cfg_parallel_size",
-        type=int,
-        default=1,
-        choices=[1, 2],
-        help="Number of GPUs used for classifier free guidance parallel size.",
-    )
-    parser.add_argument(
-        "--enforce_eager",
-        action="store_true",
-        help="Disable torch.compile and force eager execution.",
-    )
-    parser.add_argument(
-        "--enable-cpu-offload",
-        action="store_true",
-        help="Enable CPU offloading for diffusion models.",
-    )
-    parser.add_argument(
-        "--tensor_parallel_size",
-        type=int,
-        default=1,
-        help="Number of GPUs used for tensor parallelism (TP) inside the DiT.",
-    )
     return parser.parse_args()
 
 
@@ -150,16 +123,7 @@ def main():
         }
 
     # assert args.ring_degree == 1, "Ring attention is not supported yet"
-    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,
-    )
-
-    # Check if profiling is requested via environment variable
-    profiler_enabled = bool(os.getenv("VLLM_TORCH_PROFILER_DIR"))
-
+    parallel_config = DiffusionParallelConfig(ulysses_degree=args.ulysses_degree, ring_degree=args.ring_degree)
     omni = Omni(
         model=args.model,
         vae_use_slicing=vae_use_slicing,
@@ -167,42 +131,29 @@ def main():
         cache_backend=args.cache_backend,
         cache_config=cache_config,
         parallel_config=parallel_config,
-        enforce_eager=args.enforce_eager,
-        enable_cpu_offload=args.enable_cpu_offload,
     )
 
-    if profiler_enabled:
-        print("[Profiler] Starting profiling...")
-        omni.start_profile()
-
     # Time profiling for generation
     print(f"\n{'=' * 60}")
     print("Generation Configuration:")
     print(f"  Model: {args.model}")
     print(f"  Inference steps: {args.num_inference_steps}")
     print(f"  Cache backend: {args.cache_backend if args.cache_backend else 'None (no acceleration)'}")
-    print(
-        f"  Parallel configuration: tensor_parallel_size={args.tensor_parallel_size}, "
-        f"ulysses_degree={args.ulysses_degree}, ring_degree={args.ring_degree}, cfg_parallel_size={args.cfg_parallel_size}"
-    )
+    print(f"  Parallel configuration: ulysses_degree={args.ulysses_degree}, ring_degree={args.ring_degree}")
     print(f"  Image size: {args.width}x{args.height}")
     print(f"{'=' * 60}\n")
 
     generation_start = time.perf_counter()
     outputs = omni.generate(
-        {
-            "prompt": args.prompt,
-            "negative_prompt": args.negative_prompt,
-        },
-        OmniDiffusionSamplingParams(
-            height=args.height,
-            width=args.width,
-            generator=generator,
-            true_cfg_scale=args.cfg_scale,
-            guidance_scale=args.guidance_scale,
-            num_inference_steps=args.num_inference_steps,
-            num_outputs_per_prompt=args.num_images_per_prompt,
-        ),
+        args.prompt,
+        negative_prompt=args.negative_prompt,
+        height=args.height,
+        width=args.width,
+        generator=generator,
+        true_cfg_scale=args.cfg_scale,
+        guidance_scale=args.guidance_scale,
+        num_inference_steps=args.num_inference_steps,
+        num_outputs_per_prompt=args.num_images_per_prompt,
     )
     generation_end = time.perf_counter()
     generation_time = generation_end - generation_start
@@ -210,23 +161,6 @@ def main():
     # Print profiling results
     print(f"Total generation time: {generation_time:.4f} seconds ({generation_time * 1000:.2f} ms)")
 
-    if profiler_enabled:
-        print("\n[Profiler] Stopping profiler and collecting results...")
-        profile_results = omni.stop_profile()
-        if profile_results and isinstance(profile_results, dict):
-            traces = profile_results.get("traces", [])
-            print("\n" + "=" * 60)
-            print("PROFILING RESULTS:")
-            for rank, trace in enumerate(traces):
-                print(f"\nRank {rank}:")
-                if trace:
-                    print(f"  • Trace: {trace}")
-            if not traces:
-                print("  No traces collected.")
-            print("=" * 60)
-        else:
-            print("[Profiler] No valid profiling data returned.")
-
     # Extract images from OmniRequestOutput
     # omni.generate() returns list[OmniRequestOutput], extract images from the first output
     if not outputs or len(outputs) == 0:

examples/offline_inference/text_to_video/text_to_video.md

@@ -9,11 +9,10 @@ python text_to_video.py \
   --prompt "Two anthropomorphic cats in comfy boxing gear and bright gloves fight intensely on a spotlighted stage." \
   --negative_prompt "<optional quality filter>" \
   --height 480 \
-  --width 832 \
-  --num_frames 33 \
+  --width 640 \
+  --num_frames 32 \
   --guidance_scale 4.0 \
   --guidance_scale_high 3.0 \
-  --flow_shift 12.0 \
   --num_inference_steps 40 \
   --fps 16 \
   --output t2v_out.mp4
@@ -22,11 +21,10 @@ python text_to_video.py \
 Key arguments:
 
 - `--prompt`: text description (string).
-- `--height/--width`: output resolution (defaults 480x832, i.e. 480P). Dimensions should align with Wan VAE downsampling (multiples of 8).
+- `--height/--width`: output resolution (defaults 720x1280). Dimensions should align with Wan VAE downsampling (multiples of 8).
 - `--num_frames`: Number of frames (Wan default is 81).
-- `--guidance_scale` and `--guidance_scale_high`: CFG scale (applied to low/high).
+- `--guidance_scale` and `--guidance_scale_high`: CFG scale (applied to low/high)..
 - `--negative_prompt`: optional list of artifacts to suppress (the PR demo used a long Chinese string).
 - `--boundary_ratio`: Boundary split ratio for low/high DiT.
 - `--fps`: frames per second for the saved MP4 (requires `diffusers` export_to_video).
 - `--output`: path to save the generated video.
-- `--enable-cpu-offload`: enable CPU offloading for diffusion models.

examples/offline_inference/text_to_video/text_to_video.py

@@ -2,16 +2,12 @@
 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project
 
 import argparse
-import os
-import time
 from pathlib import Path
 
 import numpy as np
 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.outputs import OmniRequestOutput
 from vllm_omni.utils.platform_utils import detect_device_type, is_npu
 
@@ -38,28 +34,6 @@ def parse_args() -> argparse.Namespace:
     )
     parser.add_argument("--output", type=str, default="wan22_output.mp4", help="Path to save the video (mp4).")
     parser.add_argument("--fps", type=int, default=24, help="Frames per second for the output video.")
-    parser.add_argument(
-        "--enable-cpu-offload",
-        action="store_true",
-        help="Enable CPU offloading for diffusion models.",
-    )
-    parser.add_argument(
-        "--ulysses_degree",
-        type=int,
-        default=1,
-        help="Number of GPUs used for ulysses sequence parallelism.",
-    )
-    parser.add_argument(
-        "--ring_degree",
-        type=int,
-        default=1,
-        help="Number of GPUs used for ring sequence parallelism.",
-    )
-    parser.add_argument(
-        "--enforce_eager",
-        action="store_true",
-        help="Disable torch.compile and force eager execution.",
-    )
     return parser.parse_args()
 
 
@@ -72,62 +46,25 @@ def main():
     vae_use_slicing = is_npu()
     vae_use_tiling = is_npu()
 
-    # Configure parallel settings (only SP is supported for Wan)
-    # Note: cfg_parallel and tensor_parallel are not implemented for Wan models
-    parallel_config = DiffusionParallelConfig(
-        ulysses_degree=args.ulysses_degree,
-        ring_degree=args.ring_degree,
-    )
-
-    # Check if profiling is requested via environment variable
-    profiler_enabled = bool(os.getenv("VLLM_TORCH_PROFILER_DIR"))
-
     omni = Omni(
         model=args.model,
         vae_use_slicing=vae_use_slicing,
         vae_use_tiling=vae_use_tiling,
         boundary_ratio=args.boundary_ratio,
         flow_shift=args.flow_shift,
-        enable_cpu_offload=args.enable_cpu_offload,
-        parallel_config=parallel_config,
-        enforce_eager=args.enforce_eager,
     )
 
-    if profiler_enabled:
-        print("[Profiler] Starting profiling...")
-        omni.start_profile()
-
-    # Print generation configuration
-    print(f"\n{'=' * 60}")
-    print("Generation Configuration:")
-    print(f"  Model: {args.model}")
-    print(f"  Inference steps: {args.num_inference_steps}")
-    print(f"  Frames: {args.num_frames}")
-    print(f"  Parallel configuration: ulysses_degree={args.ulysses_degree}, ring_degree={args.ring_degree}")
-    print(f"  Video size: {args.width}x{args.height}")
-    print(f"{'=' * 60}\n")
-
-    generation_start = time.perf_counter()
     frames = omni.generate(
-        {
-            "prompt": args.prompt,
-            "negative_prompt": args.negative_prompt,
-        },
-        OmniDiffusionSamplingParams(
-            height=args.height,
-            width=args.width,
-            generator=generator,
-            guidance_scale=args.guidance_scale,
-            guidance_scale_2=args.guidance_scale_high,
-            num_inference_steps=args.num_inference_steps,
-            num_frames=args.num_frames,
-        ),
+        args.prompt,
+        negative_prompt=args.negative_prompt,
+        height=args.height,
+        width=args.width,
+        generator=generator,
+        guidance_scale=args.guidance_scale,
+        guidance_scale_2=args.guidance_scale_high,
+        num_inference_steps=args.num_inference_steps,
+        num_frames=args.num_frames,
     )
-    generation_end = time.perf_counter()
-    generation_time = generation_end - generation_start
-
-    # Print profiling results
-    print(f"Total generation time: {generation_time:.4f} seconds ({generation_time * 1000:.2f} ms)")
 
     # Extract video frames from OmniRequestOutput
     if isinstance(frames, list) and len(frames) > 0:
@@ -189,23 +126,6 @@ def main():
     export_to_video(video_array, str(output_path), fps=args.fps)
     print(f"Saved generated video to {output_path}")
 
-    if profiler_enabled:
-        print("\n[Profiler] Stopping profiler and collecting results...")
-        profile_results = omni.stop_profile()
-        if profile_results and isinstance(profile_results, dict):
-            traces = profile_results.get("traces", [])
-            print("\n" + "=" * 60)
-            print("PROFILING RESULTS:")
-            for rank, trace in enumerate(traces):
-                print(f"\nRank {rank}:")
-                if trace:
-                    print(f"  • Trace: {trace}")
-            if not traces:
-                print("  No traces collected.")
-            print("=" * 60)
-        else:
-            print("[Profiler] No valid profiling data returned.")
-
 
 if __name__ == "__main__":
     main()

examples/online_serving/bagel/openai_chat_client.py

@@ -1,184 +0,0 @@
-#!/usr/bin/env python3
-"""
-Bagel OpenAI-compatible chat client for image generation and multimodal tasks.
-
-Usage:
-    python openai_chat_client.py --prompt "A cute cat" --output output.png
-    python openai_chat_client.py --prompt "Describe this image" --image-url https://example.com/image.png
-"""
-
-import argparse
-import base64
-from pathlib import Path
-
-import requests
-
-
-def generate_image(
-    prompt: str,
-    server_url: str = "http://localhost:8091",
-    image_url: str | None = None,
-    height: int | None = None,
-    width: int | None = None,
-    steps: int | None = None,
-    seed: int | None = None,
-    negative_prompt: str | None = None,
-    modality: str = "text2img",  # "text2img" (default), "img2img", "img2text", "text2text"
-) -> bytes | str | None:
-    """Generate an image or text using the chat completions API.
-
-    Args:
-        prompt: Text description or prompt
-        server_url: Server URL
-        image_url: URL or path to input image (for img2img/img2text)
-        height: Image height in pixels
-        width: Image width in pixels
-        steps: Number of inference steps
-        seed: Random seed
-        negative_prompt: Negative prompt
-        modality: Task modality hint
-
-    Returns:
-        Image bytes (for image outputs) or Text string (for text outputs) or None if failed
-    """
-
-    # Construct Message Content
-    content = [{"type": "text", "text": f"<|im_start|>{prompt}<|im_end|>"}]
-
-    if image_url:
-        # Check if local file
-        if Path(image_url).exists():
-            with open(image_url, "rb") as f:
-                b64_data = base64.b64encode(f.read()).decode("utf-8")
-                final_image_url = f"data:image/jpeg;base64,{b64_data}"
-        else:
-            final_image_url = image_url
-
-        content.append({"type": "image_url", "image_url": {"url": final_image_url}})
-
-    messages = [{"role": "user", "content": content}]
-
-    # Build request payload with all parameters at top level
-    # Note: vLLM ignores "extra_body", so we put parameters directly in the payload
-    payload = {"messages": messages}
-
-    # Set output modalities at top level
-    if modality == "text2img" or modality == "img2img":
-        payload["modalities"] = ["image"]
-    elif modality == "img2text" or modality == "text2text":
-        payload["modalities"] = ["text"]
-
-    # Add generation parameters directly to payload
-    if height is not None:
-        payload["height"] = height
-    if width is not None:
-        payload["width"] = width
-    if steps is not None:
-        payload["num_inference_steps"] = steps
-    if seed is not None:
-        payload["seed"] = seed
-    if negative_prompt:
-        payload["negative_prompt"] = negative_prompt
-
-    # Send request
-    try:
-        print(f"Sending request to {server_url} with modality {modality}...")
-        response = requests.post(
-            f"{server_url}/v1/chat/completions",
-            headers={"Content-Type": "application/json"},
-            json=payload,
-            timeout=300,
-        )
-        response.raise_for_status()
-        data = response.json()
-
-        # Extract content - check ALL choices since server may return multiple
-        # (e.g., text in choices[0], image in choices[1])
-        choices = data.get("choices", [])
-
-        # First pass: look for image output in any choice
-        for choice in choices:
-            choice_content = choice.get("message", {}).get("content")
-
-            # Handle Image Output
-            if isinstance(choice_content, list) and len(choice_content) > 0:
-                first_item = choice_content[0]
-                if isinstance(first_item, dict) and "image_url" in first_item:
-                    img_url_str = first_item["image_url"].get("url", "")
-                    if img_url_str.startswith("data:image"):
-                        _, b64_data = img_url_str.split(",", 1)
-                        return base64.b64decode(b64_data)
-
-        # Second pass: look for text output if no image found
-        for choice in choices:
-            choice_content = choice.get("message", {}).get("content")
-            if isinstance(choice_content, str) and choice_content:
-                return choice_content
-
-        print(f"Unexpected response format: {choices}")
-        return None
-
-    except Exception as e:
-        print(f"Error: {e}")
-        return None
-
-
-def main():
-    parser = argparse.ArgumentParser(description="Bagel multimodal chat client")
-    parser.add_argument("--prompt", "-p", default="<|im_start|>A cute cat<|im_end|>", help="Text prompt")
-    parser.add_argument("--output", "-o", default="bagel_output.png", help="Output file (for image results)")
-    parser.add_argument("--server", "-s", default="http://localhost:8091", help="Server URL")
-
-    # Modality Control
-    parser.add_argument("--image-url", "-i", type=str, help="Input image URL or local path")
-    parser.add_argument(
-        "--modality",
-        "-m",
-        default="text2img",
-        choices=["text2img", "img2img", "img2text", "text2text"],
-        help="Task modality",
-    )
-
-    # Generation Params
-    parser.add_argument("--height", type=int, default=512, help="Image height")
-    parser.add_argument("--width", type=int, default=512, help="Image width")
-    parser.add_argument("--steps", type=int, default=25, help="Inference steps")
-    parser.add_argument("--seed", type=int, default=42, help="Random seed")
-    parser.add_argument("--negative", help="Negative prompt")
-
-    args = parser.parse_args()
-
-    print(f"Mode: {args.modality}")
-    if args.image_url:
-        print(f"Input Image: {args.image_url}")
-
-    result = generate_image(
-        prompt=args.prompt,
-        server_url=args.server,
-        image_url=args.image_url,
-        height=args.height,
-        width=args.width,
-        steps=args.steps,
-        seed=args.seed,
-        negative_prompt=args.negative,
-        modality=args.modality,
-    )
-
-    if result:
-        if isinstance(result, bytes):
-            # It's an image
-            output_path = Path(args.output)
-            output_path.write_bytes(result)
-            print(f"Image saved to: {output_path}")
-            print(f"Size: {len(result) / 1024:.1f} KB")
-        elif isinstance(result, str):
-            # It's text
-            print("Response:")
-            print(result)
-    else:
-        print("Failed to generate response")
-        exit(1)
-
-
-if __name__ == "__main__":
-    main()

examples/online_serving/bagel/run_server.sh

@@ -1,12 +0,0 @@
-#!/bin/bash
-# Bagel online serving startup script
-
-MODEL="${MODEL:-ByteDance-Seed/BAGEL-7B-MoT}"
-PORT="${PORT:-8091}"
-
-echo "Starting Bagel server..."
-echo "Model: $MODEL"
-echo "Port: $PORT"
-
-vllm serve "$MODEL" --omni \
-    --port "$PORT"

examples/online_serving/image_to_image/openai_chat_client.py

@@ -127,7 +127,7 @@ def main():
     parser.add_argument("--width", type=int, default=1024, help="Output image width")
     parser.add_argument("--steps", type=int, default=50, help="Inference steps")
     parser.add_argument("--guidance", type=float, default=7.5, help="Guidance scale")
-    parser.add_argument("--seed", type=int, default=0, help="Random seed")
+    parser.add_argument("--seed", type=int, help="Random seed")
     parser.add_argument("--negative", help="Negative prompt")
 
     args = parser.parse_args()

examples/online_serving/lora_inference/README.md

@@ -1,54 +0,0 @@
-# Online LoRA Inference (Diffusion)
-
-This example shows how to use **per-request LoRA** with vLLM-Omni diffusion models via the OpenAI-compatible Chat Completions API.
-
-> Note: The LoRA adapter path must be readable on the **server** machine (usually a local path or a mounted directory).
-> Note: This example uses `/v1/chat/completions`. LoRA payloads for other OpenAI endpoints are not implemented here.
-
-## Start Server
-
-```bash
-# Pick a diffusion model (examples)
-# export MODEL=stabilityai/stable-diffusion-3.5-medium
-# export MODEL=Qwen/Qwen-Image
-
-bash run_server.sh
-```
-
-## Call API (curl)
-
-```bash
-# Required: local LoRA folder on the server
-export LORA_PATH=/path/to/lora_adapter
-
-# Optional
-export SERVER=http://localhost:8091
-export PROMPT="A piece of cheesecake"
-export LORA_NAME=my_lora
-export LORA_SCALE=1.0
-# Optional: if omitted, the server derives a stable id from LORA_PATH.
-# export LORA_INT_ID=123
-
-bash run_curl_lora_inference.sh
-```
-
-## Call API (Python)
-
-```bash
-python openai_chat_client.py \
-  --prompt "A piece of cheesecake" \
-  --lora-path /path/to/lora_adapter \
-  --lora-name my_lora \
-  --lora-scale 1.0 \
-  --output output.png
-```
-
-## LoRA Format
-
-LoRA adapters should be in PEFT format, for example:
-
-```
-lora_adapter/
-├── adapter_config.json
-└── adapter_model.safetensors
-```

examples/online_serving/lora_inference/openai_chat_client.py

@@ -1,124 +0,0 @@
-#!/usr/bin/env python3
-"""
-OpenAI-compatible chat client for diffusion LoRA inference.
-
-Example:
-  python openai_chat_client.py \
-    --server http://localhost:8091 \
-    --prompt "A piece of cheesecake" \
-    --lora-path /path/to/lora_adapter \
-    --lora-name my_lora \
-    --lora-scale 1.0 \
-    --output output.png
-"""
-
-import argparse
-import base64
-from pathlib import Path
-
-import requests
-
-
-def generate_image(
-    *,
-    prompt: str,
-    server_url: str,
-    height: int | None,
-    width: int | None,
-    num_inference_steps: int | None,
-    seed: int | None,
-    lora_name: str | None,
-    lora_path: str | None,
-    lora_scale: float | None,
-    lora_int_id: int | None,
-) -> bytes | None:
-    messages = [{"role": "user", "content": prompt}]
-
-    extra_body: dict = {}
-    if height is not None:
-        extra_body["height"] = height
-    if width is not None:
-        extra_body["width"] = width
-    if num_inference_steps is not None:
-        extra_body["num_inference_steps"] = num_inference_steps
-    if seed is not None:
-        extra_body["seed"] = seed
-
-    if lora_path:
-        lora_body: dict = {
-            "local_path": lora_path,
-            "name": lora_name or Path(lora_path).stem,
-        }
-        if lora_scale is not None:
-            lora_body["scale"] = float(lora_scale)
-        if lora_int_id is not None:
-            lora_body["int_id"] = int(lora_int_id)
-        extra_body["lora"] = lora_body
-
-    payload = {"messages": messages}
-    if extra_body:
-        payload["extra_body"] = extra_body
-
-    response = requests.post(
-        f"{server_url}/v1/chat/completions",
-        headers={"Content-Type": "application/json"},
-        json=payload,
-        timeout=300,
-    )
-    response.raise_for_status()
-    data = response.json()
-
-    content = data["choices"][0]["message"]["content"]
-    if isinstance(content, list) and content:
-        image_url = content[0].get("image_url", {}).get("url", "")
-        if image_url.startswith("data:image"):
-            _, b64_data = image_url.split(",", 1)
-            return base64.b64decode(b64_data)
-
-    raise RuntimeError(f"Unexpected response format: {content!r}")
-
-
-def main() -> None:
-    parser = argparse.ArgumentParser(description="Diffusion LoRA OpenAI chat client")
-    parser.add_argument("--server", default="http://localhost:8091", help="Server URL")
-    parser.add_argument("--prompt", default="A piece of cheesecake", help="Text prompt")
-    parser.add_argument("--output", default="lora_online_output.png", help="Output image path")
-
-    parser.add_argument("--height", type=int, default=1024, help="Image height")
-    parser.add_argument("--width", type=int, default=1024, help="Image width")
-    parser.add_argument("--steps", type=int, default=50, help="num_inference_steps")
-    parser.add_argument("--seed", type=int, default=42, help="Random seed")
-
-    parser.add_argument("--lora-path", default=None, help="Server-local LoRA adapter folder (PEFT format)")
-    parser.add_argument("--lora-name", default=None, help="LoRA name (optional)")
-    parser.add_argument("--lora-scale", type=float, default=1.0, help="LoRA scale")
-    parser.add_argument(
-        "--lora-int-id",
-        type=int,
-        default=None,
-        help="LoRA integer id (cache key). If omitted, the server derives a stable id from lora_path.",
-    )
-
-    args = parser.parse_args()
-
-    image_bytes = generate_image(
-        prompt=args.prompt,
-        server_url=args.server,
-        height=args.height,
-        width=args.width,
-        num_inference_steps=args.steps,
-        seed=args.seed,
-        lora_name=args.lora_name,
-        lora_path=args.lora_path,
-        lora_scale=args.lora_scale if args.lora_path else None,
-        lora_int_id=args.lora_int_id if args.lora_path else None,
-    )
-
-    out_path = Path(args.output)
-    out_path.parent.mkdir(parents=True, exist_ok=True)
-    out_path.write_bytes(image_bytes)
-    print(f"Saved: {out_path} ({len(image_bytes) / 1024:.1f} KiB)")
-
-
-if __name__ == "__main__":
-    main()

examples/online_serving/lora_inference/run_curl_lora_inference.sh

@@ -1,61 +0,0 @@
-#!/bin/bash
-# Online diffusion LoRA inference via OpenAI-compatible chat API.
-
-SERVER="${SERVER:-http://localhost:8091}"
-PROMPT="${PROMPT:-A piece of cheesecake}"
-
-LORA_PATH="${LORA_PATH:-}"
-LORA_NAME="${LORA_NAME:-lora}"
-LORA_SCALE="${LORA_SCALE:-1.0}"
-LORA_INT_ID="${LORA_INT_ID:-}"
-
-HEIGHT="${HEIGHT:-1024}"
-WIDTH="${WIDTH:-1024}"
-NUM_INFERENCE_STEPS="${NUM_INFERENCE_STEPS:-50}"
-SEED="${SEED:-42}"
-
-CURRENT_TIME=$(date +%Y%m%d%H%M%S)
-OUTPUT="${OUTPUT:-lora_online_output_${CURRENT_TIME}.png}"
-
-if [ -z "$LORA_PATH" ]; then
-  echo "ERROR: LORA_PATH is required (must be a server-local path)."
-  exit 1
-fi
-
-echo "Generating image with LoRA..."
-echo "Server: $SERVER"
-echo "Prompt: $PROMPT"
-echo "LoRA: name=$LORA_NAME id=${LORA_INT_ID:-auto} scale=$LORA_SCALE path=$LORA_PATH"
-echo "Output: $OUTPUT"
-
-LORA_INT_ID_FIELD=""
-if [ -n "$LORA_INT_ID" ]; then
-  LORA_INT_ID_FIELD=", \"int_id\": $LORA_INT_ID"
-fi
-
-curl -s "$SERVER/v1/chat/completions" \
-  -H "Content-Type: application/json" \
-  -d "{
-    \"messages\": [
-      {\"role\": \"user\", \"content\": \"$PROMPT\"}
-    ],
-    \"extra_body\": {
-      \"height\": $HEIGHT,
-      \"width\": $WIDTH,
-      \"num_inference_steps\": $NUM_INFERENCE_STEPS,
-      \"seed\": $SEED,
-      \"lora\": {
-        \"name\": \"$LORA_NAME\",
-        \"local_path\": \"$LORA_PATH\",
-        \"scale\": $LORA_SCALE$LORA_INT_ID_FIELD
-      }
-    }
-  }" | jq -r '.choices[0].message.content[0].image_url.url' | sed 's/^data:image[^,]*,\s*//' | base64 -d > "$OUTPUT"
-
-if [ -f "$OUTPUT" ]; then
-  echo "Image saved to: $OUTPUT"
-  echo "Size: $(du -h "$OUTPUT" | cut -f1)"
-else
-  echo "Failed to generate image"
-  exit 1
-fi

examples/online_serving/lora_inference/run_server.sh

@@ -1,20 +0,0 @@
-#!/bin/bash
-# Online diffusion serving with vLLM-Omni (OpenAI-compatible API).
-
-MODEL="${MODEL:-stabilityai/stable-diffusion-3.5-medium}"
-PORT="${PORT:-8091}"
-
-echo "Starting vLLM-Omni diffusion server..."
-echo "Model: $MODEL"
-echo "Port: $PORT"
-
-if [ -z "${VLLM_BIN:-}" ]; then
-  if command -v vllm-omni >/dev/null 2>&1; then
-    VLLM_BIN="vllm-omni"
-  else
-    VLLM_BIN="vllm"
-  fi
-fi
-
-"$VLLM_BIN" serve "$MODEL" --omni \
-  --port "$PORT"

examples/online_serving/qwen2_5_omni/README.md

@@ -134,7 +134,6 @@ response = client.chat.completions.create(
 # Response contains two choices: one with text, one with audio
 print(response.choices[0].message.content)  # Text response
 print(response.choices[1].message.audio)    # Audio response
-```
 
 ## Streaming Output
 If you want to enable streaming output, please set the argument as below. The final output will be obtained just after generated by corresponding stage. Now we only support text streaming output. Other modalities can output normally.

examples/online_serving/qwen3_omni/README.md

@@ -12,12 +12,6 @@ Please refer to [README.md](../../../README.md)
 vllm serve Qwen/Qwen3-Omni-30B-A3B-Instruct --omni --port 8091
 ```
 
-If you want to open async chunking for qwen3-omni, launch the server with command below
-
-```bash
-vllm serve Qwen/Qwen3-Omni-30B-A3B-Instruct --omni --port 8091 --stage-configs-path /vllm_omni/model_executor/stage_configs/qwen3_omni_moe_async_chunk.yaml
-```
-
 If you have custom stage configs file, launch the server with command below
 ```bash
 vllm serve Qwen/Qwen3-Omni-30B-A3B-Instruct --omni --port 8091 --stage-configs-path /path/to/stage_configs_file
@@ -148,7 +142,6 @@ response = client.chat.completions.create(
 # Response contains two choices: one with text, one with audio
 print(response.choices[0].message.content)  # Text response
 print(response.choices[1].message.audio)    # Audio response
-```
 
 ## Streaming Output
 If you want to enable streaming output, please set the argument as below. The final output will be obtained just after generated by corresponding stage. Now we only support text streaming output. Other modalities can output normally.

examples/online_serving/qwen3_omni/openai_chat_completion_client_for_multimodal_generation.py

@@ -1,5 +1,4 @@
 import base64
-import concurrent.futures
 import os
 from typing import NamedTuple
 
@@ -306,26 +305,6 @@ def get_multi_audios_query(custom_prompt: str | None = None):
     }
 
 
-def get_use_audio_in_video_query(
-    video_path: str | None = None,
-    audio_path: str | None = None,
-    custom_prompt: str | None = None,
-):
-    question = custom_prompt or (
-        "Describe the content of the video in details, then convert what the baby say into text."
-    )
-    video_url = get_video_url_from_path(video_path)
-    audio_url = get_audio_url_from_path(audio_path)
-    return {
-        "role": "user",
-        "content": [
-            {"type": "video_url", "video_url": {"url": video_url}},
-            {"type": "audio_url", "audio_url": {"url": audio_url}},
-            {"type": "text", "text": question},
-        ],
-    }
-
-
 query_map = {
     "text": get_text_query,
     "use_audio": get_audio_query,
@@ -333,7 +312,6 @@ query_map = {
     "use_video": get_video_query,
     "use_mixed_modalities": get_mixed_modalities_query,
     "use_multi_audios": get_multi_audios_query,
-    "use_audio_in_video": get_use_audio_in_video_query,
 }
 
 
@@ -394,12 +372,6 @@ def run_multimodal_generation(args) -> None:
         prompt = query_func(audio_path=audio_path, custom_prompt=custom_prompt)
     elif args.query_type == "text":
         prompt = query_func(custom_prompt=custom_prompt)
-    elif args.query_type == "use_audio_in_video":
-        prompt = query_func(
-            video_path=video_path,
-            audio_path=audio_path,
-            custom_prompt=custom_prompt,
-        )
     else:
         prompt = query_func()
 
@@ -415,67 +387,51 @@ def run_multimodal_generation(args) -> None:
     else:
         output_modalities = None
 
-    # Test multiple concurrent completions
-    num_concurrent_requests = args.num_concurrent_requests
+    chat_completion = client.chat.completions.create(
+        messages=[
+            get_system_prompt(),
+            prompt,
+        ],
+        model=model_name,
+        modalities=output_modalities,
+        extra_body=extra_body,
+        stream=args.stream,
+    )
 
-    with concurrent.futures.ThreadPoolExecutor(max_workers=num_concurrent_requests) as executor:
-        # Submit multiple completion requests concurrently
-        futures = [
-            executor.submit(
-                client.chat.completions.create,
-                messages=[
-                    get_system_prompt(),
-                    prompt,
-                ],
-                model=model_name,
-                modalities=output_modalities,
-                extra_body=extra_body,
-                stream=args.stream,
-            )
-            for _ in range(num_concurrent_requests)
-        ]
-
-        # Wait for all requests to complete and collect results
-        chat_completions = [future.result() for future in concurrent.futures.as_completed(futures)]
-
-    assert len(chat_completions) == num_concurrent_requests
     count = 0
     if not args.stream:
-        # Verify all completions succeeded
-        for chat_completion in chat_completions:
-            for choice in chat_completion.choices:
-                if choice.message.audio:
-                    audio_data = base64.b64decode(choice.message.audio.data)
+        for choice in chat_completion.choices:
+            if choice.message.audio:
+                audio_data = base64.b64decode(choice.message.audio.data)
+                audio_file_path = f"audio_{count}.wav"
+                with open(audio_file_path, "wb") as f:
+                    f.write(audio_data)
+                print(f"Audio saved to {audio_file_path}")
+                count += 1
+            elif choice.message.content:
+                print("Chat completion output from text:", choice.message.content)
+    else:
+        printed_content = False
+        for chunk in chat_completion:
+            for choice in chunk.choices:
+                if hasattr(choice, "delta"):
+                    content = getattr(choice.delta, "content", None)
+                else:
+                    content = None
+
+                if getattr(chunk, "modality", None) == "audio" and content:
+                    audio_data = base64.b64decode(content)
                     audio_file_path = f"audio_{count}.wav"
                     with open(audio_file_path, "wb") as f:
                         f.write(audio_data)
-                    print(f"Audio saved to {audio_file_path}")
+                    print(f"\nAudio saved to {audio_file_path}")
                     count += 1
-                elif choice.message.content:
-                    print("Chat completion output from text:", choice.message.content)
-    else:
-        printed_content = False
-        for chat_completion in chat_completions:
-            for chunk in chat_completion:
-                for choice in chunk.choices:
-                    if hasattr(choice, "delta"):
-                        content = getattr(choice.delta, "content", None)
-                    else:
-                        content = None
 
-                    if getattr(chunk, "modality", None) == "audio" and content:
-                        audio_data = base64.b64decode(content)
-                        audio_file_path = f"audio_{count}.wav"
-                        with open(audio_file_path, "wb") as f:
-                            f.write(audio_data)
-                        print(f"\nAudio saved to {audio_file_path}")
-                        count += 1
-
-                    elif getattr(chunk, "modality", None) == "text":
-                        if not printed_content:
-                            printed_content = True
-                            print("\ncontent:", end="", flush=True)
-                        print(content, end="", flush=True)
+                elif getattr(chunk, "modality", None) == "text":
+                    if not printed_content:
+                        printed_content = True
+                        print("\ncontent:", end="", flush=True)
+                    print(content, end="", flush=True)
 
 
 def parse_args():
@@ -534,12 +490,6 @@ def parse_args():
         action="store_true",
         help="Stream the response.",
     )
-    parser.add_argument(
-        "--num-concurrent-requests",
-        type=int,
-        default=1,
-        help="Number of concurrent requests to send. Default is 1.",
-    )
 
     return parser.parse_args()
 

examples/online_serving/qwen3_tts/README.md

@@ -1,170 +0,0 @@
-# Qwen3-TTS Online Serving
-
-This directory contains examples for running Qwen3-TTS models with vLLM-Omni's online serving API.
-
-## Supported Models
-
-| Model | Task Type | Description |
-|-------|-----------|-------------|
-| `Qwen/Qwen3-TTS-12Hz-1.7B-CustomVoice` | CustomVoice | Predefined speaker voices with optional style control |
-| `Qwen/Qwen3-TTS-12Hz-1.7B-VoiceDesign` | VoiceDesign | Natural language voice style description |
-| `Qwen/Qwen3-TTS-12Hz-1.7B-Base` | Base | Voice cloning from reference audio |
-
-## Quick Start
-
-### 1. Start the Server
-
-```bash
-# CustomVoice model (default)
-./run_server.sh
-
-# Or specify task type
-./run_server.sh CustomVoice
-./run_server.sh VoiceDesign
-./run_server.sh Base
-```
-
-### 2. Run the Client
-
-```bash
-# CustomVoice: Use predefined speaker
-python openai_speech_client.py \
-    --text "你好，我是通义千问" \
-    --voice Vivian \
-    --language Chinese
-
-# CustomVoice with style instruction
-python openai_speech_client.py \
-    --text "今天天气真好" \
-    --voice Ryan \
-    --instructions "用开心的语气说"
-
-# VoiceDesign: Describe the voice style
-python openai_speech_client.py \
-    --task-type VoiceDesign \
-    --text "哥哥，你回来啦" \
-    --instructions "体现撒娇稚嫩的萝莉女声，音调偏高"
-
-# Base: Voice cloning
-python openai_speech_client.py \
-    --task-type Base \
-    --text "Hello, this is a cloned voice" \
-    --ref-audio /path/to/reference.wav \
-    --ref-text "Original transcript of the reference audio"
-```
-
-### 3. Using curl
-
-```bash
-# Simple TTS request
-curl -X POST http://localhost:8000/v1/audio/speech \
-    -H "Content-Type: application/json" \
-    -d '{
-        "input": "Hello, how are you?",
-        "voice": "Vivian",
-        "language": "English"
-    }' --output output.wav
-
-# With style instruction
-curl -X POST http://localhost:8000/v1/audio/speech \
-    -H "Content-Type: application/json" \
-    -d '{
-        "input": "I am so excited!",
-        "voice": "Vivian",
-        "instructions": "Speak with great enthusiasm"
-    }' --output excited.wav
-```
-
-## API Reference
-
-### Endpoint
-
-```
-POST /v1/audio/speech
-```
-
-This endpoint follows the [OpenAI Audio Speech API](https://platform.openai.com/docs/api-reference/audio/createSpeech) format with additional Qwen3-TTS parameters.
-
-### Request Body
-
-```json
-{
-    "input": "Text to synthesize",
-    "voice": "Vivian",
-    "response_format": "wav",
-    "task_type": "CustomVoice",
-    "language": "Auto",
-    "instructions": "Optional style instructions",
-    "ref_audio": "URL or base64 for voice cloning",
-    "ref_text": "Reference audio transcript",
-    "x_vector_only_mode": false,
-    "max_new_tokens": 2048
-}
-```
-
-> **Note:** The `model` field is optional when serving a single model, as the server already knows which model is loaded.
-
-### Response
-
-Returns audio data in the requested format (default: WAV).
-
-## Parameters
-
-### Standard OpenAI Parameters
-
-| Parameter | Type | Default | Description |
-|-----------|------|---------|-------------|
-| `input` | string | required | Text to synthesize |
-| `voice` | string | "Vivian" | Speaker/voice name |
-| `response_format` | string | "wav" | Audio format: wav, mp3, flac, pcm, aac, opus |
-| `speed` | float | 1.0 | Playback speed (0.25-4.0) |
-| `model` | string | optional | Model name (optional when serving single model) |
-
-### Qwen3-TTS Parameters
-
-| Parameter | Type | Default | Description |
-|-----------|------|---------|-------------|
-| `task_type` | string | "CustomVoice" | Task: CustomVoice, VoiceDesign, or Base |
-| `language` | string | "Auto" | Language: Auto, Chinese, English, Japanese, Korean |
-| `instructions` | string | "" | Voice style/emotion instructions |
-| `max_new_tokens` | int | 2048 | Maximum tokens to generate |
-
-### Voice Clone Parameters (Base task)
-
-| Parameter | Type | Required | Description |
-|-----------|------|----------|-------------|
-| `ref_audio` | string | Yes* | Reference audio (file path, URL, or base64) |
-| `ref_text` | string | No | Transcript of reference audio (for ICL mode) |
-| `x_vector_only_mode` | bool | false | Use speaker embedding only (no ICL) |
-
-## Python Usage
-
-```python
-import httpx
-
-# Simple request
-response = httpx.post(
-    "http://localhost:8000/v1/audio/speech",
-    json={
-        "model": "Qwen/Qwen3-TTS-12Hz-1.7B-CustomVoice",
-        "input": "Hello world",
-        "voice": "Vivian",
-    },
-    timeout=300.0,
-)
-
-with open("output.wav", "wb") as f:
-    f.write(response.content)
-```
-
-## Limitations
-
-- **No streaming**: Audio is generated completely before being returned. Streaming will be supported after the pipeline is disaggregated (see RFC #938).
-- **Single request**: Batch processing is not yet optimized for online serving.
-
-## Troubleshooting
-
-1. **Connection refused**: Make sure the server is running on the correct port
-2. **Out of memory**: Reduce `--gpu-memory-utilization` in run_server.sh
-3. **Unsupported speaker**: Check supported speakers via model documentation
-4. **Voice clone fails**: Ensure you're using the Base model variant for voice cloning

examples/online_serving/qwen3_tts/openai_speech_client.py

@@ -1,235 +0,0 @@
-"""OpenAI-compatible client for Qwen3-TTS via /v1/audio/speech endpoint.
-
-This script demonstrates how to use the OpenAI-compatible speech API
-to generate audio from text using Qwen3-TTS models.
-
-Examples:
-    # CustomVoice task (predefined speaker)
-    python openai_speech_client.py --text "Hello, how are you?" --voice Vivian
-
-    # CustomVoice with emotion instruction
-    python openai_speech_client.py --text "I'm so happy!" --voice Vivian \
-        --instructions "Speak with excitement"
-
-    # VoiceDesign task (voice from description)
-    python openai_speech_client.py --text "Hello world" \
-        --task-type VoiceDesign \
-        --instructions "A warm, friendly female voice"
-
-    # Base task (voice cloning)
-    python openai_speech_client.py --text "Hello world" \
-        --task-type Base \
-        --ref-audio "https://example.com/reference.wav" \
-        --ref-text "This is the reference transcript"
-"""
-
-import argparse
-import base64
-import os
-
-import httpx
-
-# Default server configuration
-DEFAULT_API_BASE = "http://localhost:8000"
-DEFAULT_API_KEY = "EMPTY"
-
-
-def encode_audio_to_base64(audio_path: str) -> str:
-    """Encode a local audio file to base64 data URL."""
-    if not os.path.exists(audio_path):
-        raise FileNotFoundError(f"Audio file not found: {audio_path}")
-
-    # Detect MIME type from extension
-    audio_path_lower = audio_path.lower()
-    if audio_path_lower.endswith(".wav"):
-        mime_type = "audio/wav"
-    elif audio_path_lower.endswith((".mp3", ".mpeg")):
-        mime_type = "audio/mpeg"
-    elif audio_path_lower.endswith(".flac"):
-        mime_type = "audio/flac"
-    elif audio_path_lower.endswith(".ogg"):
-        mime_type = "audio/ogg"
-    else:
-        mime_type = "audio/wav"  # Default
-
-    with open(audio_path, "rb") as f:
-        audio_bytes = f.read()
-    audio_b64 = base64.b64encode(audio_bytes).decode("utf-8")
-    return f"data:{mime_type};base64,{audio_b64}"
-
-
-def run_tts_generation(args) -> None:
-    """Run TTS generation via OpenAI-compatible /v1/audio/speech API."""
-
-    # Build request payload
-    payload = {
-        "model": args.model,
-        "input": args.text,
-        "voice": args.voice,
-        "response_format": args.response_format,
-    }
-
-    # Add optional parameters
-    if args.instructions:
-        payload["instructions"] = args.instructions
-    if args.task_type:
-        payload["task_type"] = args.task_type
-    if args.language:
-        payload["language"] = args.language
-    if args.max_new_tokens:
-        payload["max_new_tokens"] = args.max_new_tokens
-
-    # Voice clone parameters (Base task)
-    if args.ref_audio:
-        if args.ref_audio.startswith(("http://", "https://")):
-            payload["ref_audio"] = args.ref_audio
-        else:
-            payload["ref_audio"] = encode_audio_to_base64(args.ref_audio)
-    if args.ref_text:
-        payload["ref_text"] = args.ref_text
-    if args.x_vector_only:
-        payload["x_vector_only_mode"] = True
-
-    print(f"Task type: {args.task_type or 'CustomVoice'}")
-    print(f"Text: {args.text}")
-    print(f"Voice: {args.voice}")
-    print("Generating audio...")
-
-    # Make the API call
-    api_url = f"{args.api_base}/v1/audio/speech"
-    headers = {
-        "Content-Type": "application/json",
-        "Authorization": f"Bearer {args.api_key}",
-    }
-
-    with httpx.Client(timeout=300.0) as client:
-        response = client.post(api_url, json=payload, headers=headers)
-
-    if response.status_code != 200:
-        print(f"Error: {response.status_code}")
-        print(response.text)
-        return
-
-    # Save audio response
-    output_path = args.output or "tts_output.wav"
-    with open(output_path, "wb") as f:
-        f.write(response.content)
-    print(f"Audio saved to: {output_path}")
-
-
-def parse_args():
-    """Parse command line arguments."""
-    parser = argparse.ArgumentParser(
-        description="OpenAI-compatible client for Qwen3-TTS via /v1/audio/speech",
-        formatter_class=argparse.RawDescriptionHelpFormatter,
-        epilog=__doc__,
-    )
-
-    # Server configuration
-    parser.add_argument(
-        "--api-base",
-        type=str,
-        default=DEFAULT_API_BASE,
-        help=f"API base URL (default: {DEFAULT_API_BASE})",
-    )
-    parser.add_argument(
-        "--api-key",
-        type=str,
-        default=DEFAULT_API_KEY,
-        help="API key (default: EMPTY)",
-    )
-    parser.add_argument(
-        "--model",
-        "-m",
-        type=str,
-        default="Qwen/Qwen3-TTS-12Hz-1.7B-CustomVoice",
-        help="Model name/path",
-    )
-
-    # Task configuration
-    parser.add_argument(
-        "--task-type",
-        "-t",
-        type=str,
-        default=None,
-        choices=["CustomVoice", "VoiceDesign", "Base"],
-        help="TTS task type (default: CustomVoice)",
-    )
-
-    # Input text
-    parser.add_argument(
-        "--text",
-        type=str,
-        required=True,
-        help="Text to synthesize",
-    )
-
-    # Voice/speaker
-    parser.add_argument(
-        "--voice",
-        type=str,
-        default="Vivian",
-        help="Speaker/voice name (default: Vivian). Options: Vivian, Ryan, etc.",
-    )
-    parser.add_argument(
-        "--language",
-        type=str,
-        default=None,
-        help="Language: Auto, Chinese, English, etc.",
-    )
-    parser.add_argument(
-        "--instructions",
-        type=str,
-        default=None,
-        help="Voice style/emotion instructions",
-    )
-
-    # Base (voice clone) parameters
-    parser.add_argument(
-        "--ref-audio",
-        type=str,
-        default=None,
-        help="Reference audio file path or URL for voice cloning (Base task)",
-    )
-    parser.add_argument(
-        "--ref-text",
-        type=str,
-        default=None,
-        help="Reference audio transcript for voice cloning (Base task)",
-    )
-    parser.add_argument(
-        "--x-vector-only",
-        action="store_true",
-        help="Use x-vector only mode for voice cloning (no ICL)",
-    )
-
-    # Generation parameters
-    parser.add_argument(
-        "--max-new-tokens",
-        type=int,
-        default=None,
-        help="Maximum new tokens to generate",
-    )
-
-    # Output
-    parser.add_argument(
-        "--response-format",
-        type=str,
-        default="wav",
-        choices=["wav", "mp3", "flac", "pcm", "aac", "opus"],
-        help="Audio output format (default: wav)",
-    )
-    parser.add_argument(
-        "--output",
-        "-o",
-        type=str,
-        default=None,
-        help="Output audio file path (default: tts_output.wav)",
-    )
-
-    return parser.parse_args()
-
-
-if __name__ == "__main__":
-    args = parse_args()
-    run_tts_generation(args)

examples/online_serving/qwen3_tts/run_server.sh

@@ -1,40 +0,0 @@
-#!/bin/bash
-# Launch vLLM-Omni server for Qwen3-TTS models
-#
-# Usage:
-#   ./run_server.sh                           # Default: CustomVoice model
-#   ./run_server.sh CustomVoice               # CustomVoice model
-#   ./run_server.sh VoiceDesign               # VoiceDesign model
-#   ./run_server.sh Base                      # Base (voice clone) model
-
-set -e
-
-TASK_TYPE="${1:-CustomVoice}"
-
-case "$TASK_TYPE" in
-    CustomVoice)
-        MODEL="Qwen/Qwen3-TTS-12Hz-1.7B-CustomVoice"
-        ;;
-    VoiceDesign)
-        MODEL="Qwen/Qwen3-TTS-12Hz-1.7B-VoiceDesign"
-        ;;
-    Base)
-        MODEL="Qwen/Qwen3-TTS-12Hz-1.7B-Base"
-        ;;
-    *)
-        echo "Unknown task type: $TASK_TYPE"
-        echo "Supported: CustomVoice, VoiceDesign, Base"
-        exit 1
-        ;;
-esac
-
-echo "Starting Qwen3-TTS server with model: $MODEL"
-
-vllm-omni serve "$MODEL" \
-    --stage-configs-path vllm_omni/model_executor/stage_configs/qwen3_tts.yaml \
-    --host 0.0.0.0 \
-    --port 8000 \
-    --gpu-memory-utilization 0.9 \
-    --trust-remote-code \
-    --enforce-eager \
-    --omni

examples/online_serving/text_to_image/README.md

@@ -9,8 +9,6 @@ This example demonstrates how to deploy Qwen-Image model for online image genera
 ```bash
 vllm serve Qwen/Qwen-Image --omni --port 8091
 ```
-!!! note
-    If you encounter Out-of-Memory (OOM) issues or have limited GPU memory, you can enable VAE slicing and tiling to reduce memory usage, --vae-use-slicing --vae-use-tiling
 
 ### Start with Parameters
 
@@ -42,7 +40,7 @@ curl -s http://localhost:8091/v1/chat/completions \
       "true_cfg_scale": 4.0,
       "seed": 42
     }
-  }' | jq -r '.choices[0].message.content[0].image_url.url' | cut -d',' -f2- | base64 -d > output.png
+  }' | jq -r '.choices[0].message.content[0].image_url.url' | cut -d',' -f2 | base64 -d > output.png
 ```
 
 ### Method 2: Using Python Client
@@ -116,7 +114,6 @@ Use `extra_body` to pass generation parameters:
 | `seed`                   | int   | None    | Random seed (reproducible)     |
 | `negative_prompt`        | str   | None    | Negative prompt                |
 | `num_outputs_per_prompt` | int   | 1       | Number of images to generate   |
-| `--cfg-parallel-size`.   | int   | 1       | Number of GPUs for CFG parallelism |
 
 ## Response Format
 
@@ -146,7 +143,7 @@ Use `extra_body` to pass generation parameters:
 
 ```bash
 # Extract base64 from response and decode to image
-cat response.json | jq -r '.choices[0].message.content[0].image_url.url' | cut -d',' -f2- | base64 -d > output.png
+cat response.json | jq -r '.choices[0].message.content[0].image_url.url' | cut -d',' -f2 | base64 -d > output.png
 ```
 
 ## File Description

examples/online_serving/text_to_image/openai_chat_client.py

@@ -1,6 +1,6 @@
 #!/usr/bin/env python3
 """
-Qwen-Image OpenAI-compatible image generation client.
+Qwen-Image OpenAI-compatible chat client for image generation.
 
 Usage:
     python openai_chat_client.py --prompt "A beautiful landscape" --output output.png
@@ -25,14 +25,14 @@ def generate_image(
     negative_prompt: str | None = None,
     num_outputs_per_prompt: int = 1,
 ) -> bytes | None:
-    """Generate an image using the images generation API.
+    """Generate an image using the chat completions API.
 
     Args:
         prompt: Text description of the image
         server_url: Server URL
         height: Image height in pixels
         width: Image width in pixels
-        steps: Number of diffusion steps
+        steps: Number of inference steps
         true_cfg_scale: Qwen-Image CFG scale
         seed: Random seed
         negative_prompt: Negative prompt
@@ -41,31 +41,33 @@ def generate_image(
     Returns:
         Image bytes or None if failed
     """
-    payload: dict[str, object] = {
-        "prompt": prompt,
-        "response_format": "b64_json",
-        "n": num_outputs_per_prompt,
-    }
-
-    if width is not None and height is not None:
-        payload["size"] = f"{width}x{height}"
-    elif width is not None:
-        payload["size"] = f"{width}x{width}"
-    elif height is not None:
-        payload["size"] = f"{height}x{height}"
+    messages = [{"role": "user", "content": prompt}]
 
+    # Build extra_body with generation parameters
+    extra_body = {}
+    if height is not None:
+        extra_body["height"] = height
+    if width is not None:
+        extra_body["width"] = width
     if steps is not None:
-        payload["num_inference_steps"] = steps
+        extra_body["num_inference_steps"] = steps
     if true_cfg_scale is not None:
-        payload["true_cfg_scale"] = true_cfg_scale
-    if negative_prompt:
-        payload["negative_prompt"] = negative_prompt
+        extra_body["true_cfg_scale"] = true_cfg_scale
     if seed is not None:
-        payload["seed"] = seed
+        extra_body["seed"] = seed
+    if negative_prompt:
+        extra_body["negative_prompt"] = negative_prompt
+    extra_body["num_outputs_per_prompt"] = num_outputs_per_prompt
 
+    # Build request payload
+    payload = {"messages": messages}
+    if extra_body:
+        payload["extra_body"] = extra_body
+
+    # Send request
     try:
         response = requests.post(
-            f"{server_url}/v1/images/generations",
+            f"{server_url}/v1/chat/completions",
             headers={"Content-Type": "application/json"},
             json=payload,
             timeout=300,
@@ -73,13 +75,15 @@ def generate_image(
         response.raise_for_status()
         data = response.json()
 
-        items = data.get("data")
-        if isinstance(items, list) and items:
-            first = items[0].get("b64_json") if isinstance(items[0], dict) else None
-            if isinstance(first, str):
-                return base64.b64decode(first)
+        # Extract image from response
+        content = data["choices"][0]["message"]["content"]
+        if isinstance(content, list) and len(content) > 0:
+            image_url = content[0].get("image_url", {}).get("url", "")
+            if image_url.startswith("data:image"):
+                _, b64_data = image_url.split(",", 1)
+                return base64.b64decode(b64_data)
 
-        print(f"Unexpected response format: {data}")
+        print(f"Unexpected response format: {content}")
         return None
 
     except Exception as e:
@@ -96,7 +100,7 @@ def main():
     parser.add_argument("--width", type=int, default=1024, help="Image width")
     parser.add_argument("--steps", type=int, default=50, help="Inference steps")
     parser.add_argument("--cfg-scale", type=float, default=4.0, help="True CFG scale")
-    parser.add_argument("--seed", type=int, default=0, help="Random seed")
+    parser.add_argument("--seed", type=int, default=42, help="Random seed")
     parser.add_argument("--negative", help="Negative prompt")
 
     args = parser.parse_args()

examples/online_serving/text_to_image/run_curl_text_to_image.sh

@@ -1,10 +1,35 @@
 #!/bin/bash
 # Qwen-Image text-to-image curl example
 
-curl -X POST http://localhost:8091/v1/images/generations \
+SERVER="${SERVER:-http://localhost:8091}"
+PROMPT="${PROMPT:-a good boy in the ocean}"
+CURRENT_TIME=$(date +%Y%m%d%H%M%S)
+OUTPUT="${OUTPUT:-qwen_image_output_${CURRENT_TIME}.png}"
+
+echo "Generating image..."
+echo "Prompt: $PROMPT"
+echo "Output: $OUTPUT"
+
+curl -s "$SERVER/v1/chat/completions" \
   -H "Content-Type: application/json" \
-  -d '{
-    "prompt": "a dragon laying over the spine of the Green Mountains of Vermont",
-    "size": "1024x1024",
-    "seed": 42
-  }' | jq -r '.data[0].b64_json' | base64 -d > dragon.png
+  -d "{
+    \"messages\": [
+      {\"role\": \"user\", \"content\": \"$PROMPT\"}
+    ],
+    \"extra_body\": {
+      \"height\": 1024,
+      \"width\": 1024,
+      \"num_inference_steps\": 100,
+      \"true_cfg_scale\": 4.0,
+      \"seed\": 42,
+      \"num_outputs_per_prompt\": 1
+    }
+  }" | jq -r '.choices[0].message.content[0].image_url.url' | sed 's/^data:image[^,]*,\s*//' | base64 -d > "$OUTPUT"
+
+if [ -f "$OUTPUT" ]; then
+    echo "Image saved to: $OUTPUT"
+    echo "Size: $(du -h "$OUTPUT" | cut -f1)"
+else
+    echo "Failed to generate image"
+    exit 1
+fi

mkdocs.yml

@@ -60,10 +60,6 @@ hooks:
   - docs/mkdocs/hooks/url_schemes.py
   - docs/mkdocs/hooks/generate_examples.py
 
-# Exclude include files from navigation warnings
-exclude_docs: |
-  **/*.inc.md
-
 # Plugins
 plugins:
   - meta
@@ -90,7 +86,6 @@ plugins:
       api_root_uri: "api"
       nav_item_prefix: ""  # No prefix in navigation tree (clean names)
       show_full_namespace: false  # Show only module name, not full path
-      on_implicit_namespace_package: skip  # Skip directories without __init__.py (e.g., assets)
       exclude:
         - "re:vllm_omni\\._.*"  # Internal modules
         - "vllm_omni.diffusion.models.qwen_image"  # avoid importing vllm in mkdocs building

pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "vllm-omni"
-version = "0.14.0rc1"
+version = "0.12.0rc1"
 description = "A framework for efficient model inference with omni-modality models"
 readme = "README.md"
 requires-python = ">=3.10,<3.14"
@@ -36,15 +36,9 @@ dependencies = [
     "accelerate==1.12.0",
     "gradio==5.50",
     "soundfile>=0.13.1",
-    "cache-dit==1.2.0",
+    "cache-dit==1.1.8",
     "tqdm>=4.66.0",
-    "torchsde>=0.2.6",  # Required for Stable Audio scheduler
-    "fa3-fwd", # flash attention 3, maintained by @ZJY0516
-    "openai-whisper>=20250625",
-    "imageio[ffmpeg]>=2.37.2",
-    "onnxruntime>=1.19.0",
-    "sox>=1.5.0",
-    # "vllm==0.14.0",  # TODO: fix the entrypoints overwrite problem
+    # "vllm==0.12.0",  # TODO: fix the entrypoints overwrite problem
 ]
 
 [project.optional-dependencies]
@@ -55,11 +49,6 @@ dev = [
     "pytest-cov>=4.0.0",
     "mypy==1.11.1",
     "pre-commit==4.0.1",
-    "openai-whisper>=20250625",
-    "psutil>=7.2.0",
-    "soundfile>=0.13.1",
-    "imageio[ffmpeg]>=0.6.0",
-    "opencv-python>=4.12.0.88"
 ]
 
 docs = [
@@ -161,34 +150,11 @@ addopts = [
     "--cov-report=xml",
 ]
 markers = [
-    # ci/cd required
-    "core_model: Core model tests (run in each PR)",
-    # function module markers
-    "diffusion: Diffusion model tests",
-    "omni: Omni model tests",
-    "cache: Cache backend tests",
-    "parallel: Parallelism/distributed tests",
-    # platform markers
-    "cpu: Tests that run on CPU",
-    "gpu: Tests that run on GPU (auto-added)",
-    "cuda: Tests that run on CUDA (auto-added)",
-    "rocm: Tests that run on AMD/ROCm (auto-added)",
-    "npu: Tests that run on NPU/Ascend (auto-added)",
-    # specified computation resources marks (auto-added)
-    "H100: Tests that require H100 GPU",
-    "L4: Tests that require L4 GPU",
-    "MI325: Tests that require MI325 GPU (AMD/ROCm)",
-    "A2: Tests that require A2 NPU",
-    "A3: Tests that require A3 NPU",
-    "distributed_cuda: Tests that require multi cards on CUDA platform",
-    "distributed_rocm: Tests that require multi cards on ROCm platform",
-    "distributed_npu: Tests that require multi cards on NPU platform",
-    "skipif_cuda: Skip if the num of CUDA cards is less than the required",
-    "skipif_rocm: Skip if the num of ROCm cards is less than the required",
-    "skipif_npu: Skip if the num of NPU cards is less than the required",
-    # more detailed markers
-    "slow: Slow tests (may skip in quick CI)",
+    "unit: Unit tests",
+    "integration: Integration tests",
     "benchmark: Benchmark tests",
+    "slow: Slow tests",
+    "core_model: enable this model test in each PR instead of only nightly",
 ]
 
 [tool.typos.default]
@@ -202,5 +168,4 @@ extend-ignore-identifiers-re = [
     ".*nothink.*",
     ".*NOTHINK.*",
     ".*nin.*",
-    "Ono_Anna",
 ]

pytest.ini

@@ -0,0 +1,3 @@
+[pytest]
+markers =
+    gpu_mem_high: needs high VRAM

tests/conftest.py

@@ -1,990 +1,36 @@
-import base64
-import datetime
-import io
-import math
 import os
-import random
 
-os.environ["VLLM_WORKER_MULTIPROC_METHOD"] = "spawn"
-# Set CPU device for CI environments without GPU
-if "VLLM_TARGET_DEVICE" not in os.environ:
-    os.environ["VLLM_TARGET_DEVICE"] = "cpu"
-
-import socket
-import subprocess
-import sys
-import time
-from pathlib import Path
-from typing import Any
-
-import numpy as np
-import psutil
 import pytest
 import torch
-import yaml
-from vllm.distributed.parallel_state import cleanup_dist_env_and_memory
 from vllm.logger import init_logger
-from vllm.utils.network_utils import get_open_port
 
 logger = init_logger(__name__)
 
 
-@pytest.fixture(autouse=True)
-def default_vllm_config():
-    """Set a default VllmConfig for all tests.
-
-    This fixture is auto-used for all tests to ensure that any test
-    that directly instantiates vLLM CustomOps (e.g., RMSNorm, LayerNorm)
-    or model components has the required VllmConfig context.
-
-    This fixture is required for vLLM 0.14.0+ where CustomOp initialization
-    requires a VllmConfig context set via set_current_vllm_config().
-    """
-    from vllm.config import DeviceConfig, VllmConfig, set_current_vllm_config
-
-    # Use CPU device if no GPU is available (e.g., in CI environments)
-    has_gpu = torch.cuda.is_available() and torch.cuda.device_count() > 0
-    device = "cuda" if has_gpu else "cpu"
-    device_config = DeviceConfig(device=device)
-
-    with set_current_vllm_config(VllmConfig(device_config=device_config)):
-        yield
-
-
 @pytest.fixture(autouse=True)
 def clean_gpu_memory_between_tests():
-    print("\n=== PRE-TEST GPU CLEANUP ===")
-    _run_pre_test_cleanup()
-    yield
-    _run_post_test_cleanup()
-
-
-def _run_pre_test_cleanup():
     if os.getenv("VLLM_TEST_CLEAN_GPU_MEMORY", "0") != "1":
-        print("GPU cleanup disabled")
+        yield
         return
 
-    print("Pre-test GPU status:")
-    _print_simple_gpu_status()
+    # Wait for GPU memory to be cleared before starting the test
+    import gc
+
+    from tests.utils import wait_for_gpu_memory_to_clear
 
     num_gpus = torch.cuda.device_count()
     if num_gpus > 0:
         try:
-            from tests.utils import wait_for_gpu_memory_to_clear
-
             wait_for_gpu_memory_to_clear(
                 devices=list(range(num_gpus)),
                 threshold_ratio=0.1,
             )
-        except Exception as e:
-            print(f"Pre-test cleanup note: {e}")
+        except ValueError as e:
+            logger.info("Failed to clean GPU memory: %s", e)
 
+    yield
 
-def _run_post_test_cleanup():
-    if os.getenv("VLLM_TEST_CLEAN_GPU_MEMORY", "0") != "1":
-        return
-
-    import gc
-
+    # Clean up GPU memory after the test
     if torch.cuda.is_available():
-        gc.collect()
         torch.cuda.empty_cache()
-
-        print("Post-test GPU status:")
-        _print_simple_gpu_status()
-        _print_gpu_processes()
-
-
-def _print_simple_gpu_status():
-    """Print simple GPU memory status"""
-    if not torch.cuda.is_available():
-        print("  CUDA not available")
-        return
-
-    num_devices = torch.cuda.device_count()
-    for device_id in range(num_devices):
-        try:
-            torch.cuda.set_device(device_id)
-            allocated = torch.cuda.memory_allocated(device_id) / (1024**2)
-            reserved = torch.cuda.memory_reserved(device_id) / (1024**2)
-            print(f"  GPU {device_id}: Allocated: {allocated:.1f}MB, Reserved: {reserved:.1f}MB")
-        except Exception:
-            print(f"  GPU {device_id}: Error reading status")
-
-
-def _print_gpu_processes():
-    """Print GPU information including nvidia-smi and system processes"""
-
-    print("\n" + "=" * 80)
-    print("NVIDIA GPU Information (nvidia-smi)")
-    print("=" * 80)
-
-    try:
-        nvidia_result = subprocess.run(
-            ["nvidia-smi"],
-            capture_output=True,
-            text=True,
-            timeout=5,
-        )
-
-        if nvidia_result.returncode == 0:
-            lines = nvidia_result.stdout.strip().split("\n")
-            for line in lines[:20]:
-                print(line)
-
-            if len(lines) > 20:
-                print(f"... (showing first 20 of {len(lines)} lines)")
-        else:
-            print("nvidia-smi command failed")
-
-    except (subprocess.TimeoutExpired, FileNotFoundError):
-        print("nvidia-smi not available or timed out")
-    except Exception as e:
-        print(f"Error running nvidia-smi: {e}")
-
-    print("\n" + "=" * 80)
-    print("Detailed GPU Processes (nvidia-smi pmon)")
-    print("=" * 80)
-
-    try:
-        pmon_result = subprocess.run(
-            ["nvidia-smi", "pmon", "-c", "1"],
-            capture_output=True,
-            text=True,
-            timeout=3,
-        )
-
-        if pmon_result.returncode == 0 and pmon_result.stdout.strip():
-            print(pmon_result.stdout)
-        else:
-            print("No active GPU processes found via nvidia-smi pmon")
-
-    except Exception:
-        print("nvidia-smi pmon not available")
-
-    print("\n" + "=" * 80)
-    print("System Processes with GPU keywords")
-    print("=" * 80)
-
-
-def dummy_messages_from_mix_data(
-    system_prompt: dict[str, Any] = None,
-    video_data_url: Any = None,
-    audio_data_url: Any = None,
-    image_data_url: Any = None,
-    content_text: str = None,
-):
-    """Create messages with video、image、audio data URL for OpenAI API."""
-
-    if content_text is not None:
-        content = [{"type": "text", "text": content_text}]
-    else:
-        content = []
-
-    media_items = []
-    if isinstance(video_data_url, list):
-        for video_url in video_data_url:
-            media_items.append((video_url, "video"))
-    else:
-        media_items.append((video_data_url, "video"))
-
-    if isinstance(image_data_url, list):
-        for url in image_data_url:
-            media_items.append((url, "image"))
-    else:
-        media_items.append((image_data_url, "image"))
-
-    if isinstance(audio_data_url, list):
-        for url in audio_data_url:
-            media_items.append((url, "audio"))
-    else:
-        media_items.append((audio_data_url, "audio"))
-
-    content.extend(
-        {"type": f"{media_type}_url", f"{media_type}_url": {"url": url}}
-        for url, media_type in media_items
-        if url is not None
-    )
-    messages = [{"role": "user", "content": content}]
-    if system_prompt is not None:
-        messages = [system_prompt] + messages
-    return messages
-
-
-def generate_synthetic_audio(
-    duration: int,  # seconds
-    num_channels: int,  # 1：Mono，2：Stereo 5：5.1 surround sound
-    sample_rate: int = 48000,  # Default use 48000Hz.
-    save_to_file: bool = False,
-) -> dict[str, Any]:
-    """ "Generate synthetic audio with rain."""
-    import soundfile as sf
-
-    # Initialize audio data array
-    num_samples = int(sample_rate * duration)
-    audio_data = np.zeros((num_samples, num_channels), dtype=np.float32)
-
-    # Configure parameters based on rain intensity
-    drop_density = 10  # Number of raindrops per second
-    drop_volume = 0.15  # Volume of individual raindrops
-    background_volume = 0.02  # Volume of background rain noise
-
-    # Pink noise sounds more natural than white noise for rain
-    white_noise = np.random.randn(num_samples)
-    pink_noise = np.convolve(white_noise, np.ones(8) / 8, mode="same")
-    pink_noise = pink_noise / np.max(np.abs(pink_noise)) if np.max(np.abs(pink_noise)) > 0 else pink_noise
-    bg_noise = pink_noise * background_volume
-
-    # Add background noise to all channels
-    for ch in range(num_channels):
-        audio_data[:, ch] += bg_noise
-
-    # Total number of raindrops = density × duration × channels for stereo effect
-    total_drops = int(drop_density * duration * num_channels)
-
-    for _ in range(total_drops):
-        # Random timing for raindrop
-        drop_time = random.uniform(0, duration)
-
-        # Random duration of raindrop sound (0.01-0.05 seconds)
-        drop_duration = random.uniform(0.01, 0.05)
-
-        # Random frequency gives variation in raindrop pitch
-        drop_freq = random.uniform(500, 5000)  # Hz
-
-        # Random channel selection for stereo positioning
-        channel = random.randint(0, num_channels - 1)
-
-        # Calculate sample positions for this raindrop
-        start_sample = int(drop_time * sample_rate)
-        drop_samples = int(drop_duration * sample_rate)
-        end_sample = min(start_sample + drop_samples, num_samples)
-
-        if start_sample < end_sample:
-            # Generate the raindrop sound
-            num_drop_samples = end_sample - start_sample
-            t = np.arange(num_drop_samples) / sample_rate
-
-            # Basic sine wave for raindrop sound
-            drop_sound = drop_volume * np.sin(2 * math.pi * drop_freq * t)
-
-            # Apply envelope for natural attack and decay
-            envelope = np.ones(num_drop_samples)
-            attack_samples = int(num_drop_samples * 0.1)  # 10% of samples for attack
-            decay_samples = num_drop_samples - attack_samples
-
-            if attack_samples > 0:
-                # Linear attack: volume increases from 0 to 1
-                envelope[:attack_samples] = np.linspace(0, 1, attack_samples)
-
-            if decay_samples > 0:
-                # Exponential decay for natural sound fade
-                decay = np.exp(-8 * t[attack_samples:] / drop_duration)
-                envelope[attack_samples:] = decay
-
-            # Apply envelope to raindrop sound
-            drop_sound *= envelope
-
-            # Add raindrop sound to selected channel
-            audio_data[start_sample:end_sample, channel] += drop_sound
-
-    # Step 3: Add simple reverb effect for realism
-    # Reverb simulates sound reflections in environment
-    if duration > 2:
-        # Single delay reverb (100ms delay)
-        delay_samples = int(0.1 * sample_rate)
-        if delay_samples < num_samples:
-            for ch in range(num_channels):
-                delayed = np.zeros(num_samples)
-                delayed[delay_samples:] = audio_data[:-delay_samples, ch] * 0.3
-                audio_data[:, ch] += delayed
-
-    # Step 4: Normalize audio to prevent clipping
-    # Find maximum amplitude and scale to 80% of maximum volume
-    max_amp = np.max(np.abs(audio_data))
-    if max_amp > 0:
-        audio_data = audio_data / max_amp * 0.8
-
-    # Handle file saving
-    audio_bytes = None
-
-    if save_to_file:
-        timestamp = datetime.datetime.now().strftime("%Y%m%d_%H%M%S")
-        output_path = f"audio_{num_channels}ch_{timestamp}.wav"
-
-        try:
-            sf.write(output_path, audio_data, sample_rate, format="WAV", subtype="PCM_16")
-            print(f"Audio saved: {output_path}")
-
-            with open(output_path, "rb") as f:
-                audio_bytes = f.read()
-        except Exception as e:
-            print(f"Save failed: {e}")
-            save_to_file = False
-
-    # If not saving or save failed, create in memory
-    if not save_to_file or audio_bytes is None:
-        buffer = io.BytesIO()
-        sf.write(buffer, audio_data, sample_rate, format="WAV", subtype="PCM_16")
-        buffer.seek(0)
-        audio_bytes = buffer.read()
-
-    # Return result
-    base64_audio = base64.b64encode(audio_bytes).decode("utf-8")
-    result = {
-        "base64": base64_audio,
-    }
-    if save_to_file and output_path:
-        result["file_path"] = output_path
-
-    return result
-
-
-def generate_synthetic_video(width: int, height: int, num_frames: int, save_to_file: bool = False) -> str:
-    """Generate synthetic video with bouncing balls and return base64 string."""
-
-    import cv2
-    import imageio
-
-    # Create random balls
-    num_balls = random.randint(3, 8)
-    balls = []
-
-    for _ in range(num_balls):
-        radius = min(width, height) // 8
-        if radius < 1:
-            raise ValueError(f"Video dimensions ({width}x{height}) are too small for synthetic video generation")
-        x = random.randint(radius, width - radius)
-        y = random.randint(radius, height - radius)
-
-        speed = random.uniform(3.0, 8.0)
-        angle = random.uniform(0, 2 * math.pi)
-        vx = speed * math.cos(angle)
-        vy = speed * math.sin(angle)
-
-        # OpenCV uses BGR format, but imageio expects RGB
-        # We'll create in BGR first, then convert to RGB later
-        color_bgr = (random.randint(50, 255), random.randint(50, 255), random.randint(50, 255))
-
-        balls.append({"x": x, "y": y, "vx": vx, "vy": vy, "radius": radius, "color_bgr": color_bgr})
-
-    # Generate video frames
-    video_frames = []
-
-    for frame_idx in range(num_frames):
-        # Create black background (BGR format)
-        frame_bgr = np.zeros((height, width, 3), dtype=np.uint8)
-
-        for ball in balls:
-            # Update position
-            ball["x"] += ball["vx"]
-            ball["y"] += ball["vy"]
-
-            # Boundary collision detection
-            if ball["x"] - ball["radius"] <= 0 or ball["x"] + ball["radius"] >= width:
-                ball["vx"] = -ball["vx"]
-                ball["x"] = max(ball["radius"], min(width - ball["radius"], ball["x"]))
-
-            if ball["y"] - ball["radius"] <= 0 or ball["y"] + ball["radius"] >= height:
-                ball["vy"] = -ball["vy"]
-                ball["y"] = max(ball["radius"], min(height - ball["radius"], ball["y"]))
-
-            # Use cv2 to draw circle
-            x, y = int(ball["x"]), int(ball["y"])
-            radius = ball["radius"]
-
-            # Draw solid circle (main circle)
-            cv2.circle(frame_bgr, (x, y), radius, ball["color_bgr"], -1)
-
-            # Add simple 3D effect: draw a brighter center
-            if radius > 3:  # Only add highlight when radius is large enough
-                highlight_radius = max(1, radius // 2)
-                highlight_x = max(highlight_radius, min(x - radius // 4, width - highlight_radius))
-                highlight_y = max(highlight_radius, min(y - radius // 4, height - highlight_radius))
-
-                # Create highlight color (brighter)
-                highlight_color = tuple(min(c + 40, 255) for c in ball["color_bgr"])
-                cv2.circle(frame_bgr, (highlight_x, highlight_y), highlight_radius, highlight_color, -1)
-
-        # Convert BGR to RGB for imageio
-        frame_rgb = cv2.cvtColor(frame_bgr, cv2.COLOR_BGR2RGB)
-        video_frames.append(frame_rgb)
-
-    video_bytes = None
-    saved_file_path = None
-
-    buffer = io.BytesIO()
-    writer_kwargs = {
-        "format": "mp4",
-        "fps": 30,
-        "codec": "libx264",
-        "quality": 7,
-        "pixelformat": "yuv420p",
-        "macro_block_size": 16,
-        "ffmpeg_params": [
-            "-preset",
-            "medium",
-            "-crf",
-            "23",
-            "-movflags",
-            "+faststart",
-            "-pix_fmt",
-            "yuv420p",
-            "-vf",
-            f"scale={width}:{height}",
-        ],
-    }
-
-    if save_to_file:
-        import datetime
-
-        timestamp = datetime.datetime.now().strftime("%Y%m%d_%H%M%S")
-        output_path = f"video_{width}x{height}_{timestamp}.mp4"
-        try:
-            with imageio.get_writer(output_path, **writer_kwargs) as writer:
-                for frame in video_frames:
-                    writer.append_data(frame)
-
-            saved_file_path = output_path
-            print(f"Video saved to: {saved_file_path}")
-            with open(output_path, "rb") as f:
-                video_bytes = f.read()
-
-        except Exception as e:
-            print(f"Warning: Failed to save video to file {output_path}: {e}")
-            save_to_file = False
-
-    if not save_to_file or video_bytes is None:
-        with imageio.get_writer(buffer, **writer_kwargs) as writer:
-            for frame in video_frames:
-                writer.append_data(frame)
-
-        buffer.seek(0)
-        video_bytes = buffer.read()
-
-    base64_video = base64.b64encode(video_bytes).decode("utf-8")
-
-    result = {
-        "base64": base64_video,
-    }
-    if save_to_file and saved_file_path:
-        result["file_path"] = saved_file_path
-
-    return result
-
-
-def generate_synthetic_image(width: int, height: int, save_to_file: bool = False) -> Any:
-    """Generate synthetic image with randomly colored squares and return base64 string."""
-    from PIL import Image, ImageDraw
-
-    # Create white background
-    image = Image.new("RGB", (width, height), (255, 255, 255))
-    draw = ImageDraw.Draw(image)
-
-    # Generate random number of squares
-    num_squares = random.randint(3, 8)
-
-    for _ in range(num_squares):
-        # Random square size
-        square_size = random.randint(min(width, height) // 8, min(width, height) // 4)
-
-        # Random position
-        x = random.randint(0, width - square_size - 1)
-        y = random.randint(0, height - square_size - 1)
-
-        # Random color
-        color = (random.randint(0, 255), random.randint(0, 255), random.randint(0, 255))
-
-        # Random border width
-        border_width = random.randint(1, 5)
-
-        # Draw square
-        draw.rectangle([x, y, x + square_size, y + square_size], fill=color, outline=(0, 0, 0), width=border_width)
-
-    # Handle file saving
-    image_bytes = None
-    saved_file_path = None
-
-    if save_to_file:
-        timestamp = datetime.datetime.now().strftime("%Y%m%d_%H%M%S")
-        output_path = f"image_{width}x{height}_{timestamp}.jpg"
-
-        try:
-            # Save image to file
-            image.save(output_path, format="JPEG", quality=85, optimize=True)
-            saved_file_path = output_path
-            print(f"Image saved to: {saved_file_path}")
-
-            # Read file for base64 encoding
-            with open(output_path, "rb") as f:
-                image_bytes = f.read()
-
-        except Exception as e:
-            print(f"Warning: Failed to save image to file {output_path}: {e}")
-            save_to_file = False
-
-    # If not saving or save failed, create in memory
-    if not save_to_file or image_bytes is None:
-        buffer = io.BytesIO()
-        image.save(buffer, format="JPEG", quality=85, optimize=True)
-        buffer.seek(0)
-        image_bytes = buffer.read()
-
-    # Generate base64
-    base64_image = base64.b64encode(image_bytes).decode("utf-8")
-
-    # Return result
-    result = {
-        "base64": base64_image,
-    }
-    if save_to_file and saved_file_path:
-        result["file_path"] = saved_file_path
-
-    return result
-
-
-def preprocess_text(text):
-    import re
-
-    word_to_num = {
-        "zero": "0",
-        "one": "1",
-        "two": "2",
-        "three": "3",
-        "four": "4",
-        "five": "5",
-        "six": "6",
-        "seven": "7",
-        "eight": "8",
-        "nine": "9",
-        "ten": "10",
-    }
-
-    for word, num in word_to_num.items():
-        pattern = r"\b" + re.escape(word) + r"\b"
-        text = re.sub(pattern, num, text, flags=re.IGNORECASE)
-
-    text = re.sub(r"[^\w\s]", "", text)
-    text = re.sub(r"\s+", " ", text)
-    return text.lower().strip()
-
-
-def cosine_similarity_text(text1, text2, n: int = 3):
-    from collections import Counter
-
-    if not text1 or not text2:
-        return 0.0
-
-    text1 = preprocess_text(text1)
-    text2 = preprocess_text(text2)
-
-    ngrams1 = [text1[i : i + n] for i in range(len(text1) - n + 1)]
-    ngrams2 = [text2[i : i + n] for i in range(len(text2) - n + 1)]
-
-    counter1 = Counter(ngrams1)
-    counter2 = Counter(ngrams2)
-
-    all_ngrams = set(counter1.keys()) | set(counter2.keys())
-    vec1 = [counter1.get(ng, 0) for ng in all_ngrams]
-    vec2 = [counter2.get(ng, 0) for ng in all_ngrams]
-
-    dot_product = sum(a * b for a, b in zip(vec1, vec2))
-    norm1 = sum(a * a for a in vec1) ** 0.5
-    norm2 = sum(b * b for b in vec2) ** 0.5
-
-    if norm1 == 0 or norm2 == 0:
-        return 0.0
-    return dot_product / (norm1 * norm2)
-
-
-def convert_audio_to_text(audio_data):
-    """
-    Convert base64 encoded audio data to text using speech recognition.
-    """
-    import whisper
-
-    audio_data = base64.b64decode(audio_data)
-    output_path = f"./test_{int(time.time())}"
-    with open(output_path, "wb") as audio_file:
-        audio_file.write(audio_data)
-
-    print(f"audio data is saved: {output_path}")
-
-    model = whisper.load_model("base")
-    text = model.transcribe(
-        output_path,
-        temperature=0.0,
-        word_timestamps=True,
-        condition_on_previous_text=False,
-    )["text"]
-    if text:
-        return text
-    else:
-        return ""
-
-
-def modify_stage_config(
-    yaml_path: str,
-    updates: dict[str, Any],
-    deletes: dict[str, Any] = None,
-) -> str:
-    """
-    Modify configurations in a YAML file, supporting both top-level and stage-specific modifications,
-    including addition, modification, and deletion of configurations.
-
-    Args:
-        yaml_path: Path to the YAML configuration file.
-        updates: Dictionary containing both top-level and stage-specific modifications to add or update.
-                Format: {
-                    'async_chunk': True,
-                    'stage_args': {
-                        0: {'engine_args.max_model_len': 5800},
-                        1: {'runtime.max_batch_size': 2}
-                    }
-                }
-        deletes: Dictionary containing configurations to delete.
-                Format: {
-                    'old_config': None,  # Delete entire key
-                    'stage_args': {
-                        0: ['engine_args.old_param'],
-                        1: ['runtime.unused_setting']
-                    }
-                }
-
-    Returns:
-        str: Path to the newly created modified YAML file with timestamp suffix.
-    """
-    path = Path(yaml_path)
-    if not path.exists():
-        raise FileNotFoundError(f"yaml does not exist: {path}")
-
-    try:
-        with open(yaml_path, encoding="utf-8") as f:
-            config = yaml.safe_load(f) or {}
-    except Exception as e:
-        raise ValueError(f"Cannot parse YAML file: {e}")
-
-    # Helper function to apply update
-    def apply_update(config_dict: dict, key_path: str, value: Any) -> None:
-        """Apply update to dictionary using dot-separated path."""
-        # Handle direct list assignment (e.g., engine_input_source: [1, 2])
-        if "." not in key_path:
-            # Simple key, set directly
-            config_dict[key_path] = value
-            return
-
-        current = config_dict
-        keys = key_path.split(".")
-
-        for i in range(len(keys) - 1):
-            key = keys[i]
-
-            # Handle list indices
-            if key.isdigit() and isinstance(current, list):
-                index = int(key)
-                if index < 0:
-                    raise ValueError(f"Negative list index not allowed: {index}")
-                if index >= len(current):
-                    # Expand list if needed
-                    while len(current) <= index:
-                        # If we need to go deeper (more keys after this), create a dict
-                        # Otherwise, create None placeholder
-                        current.append({} if i < len(keys) - 2 else None)
-                current = current[index]
-            elif isinstance(current, dict):
-                # Handle dictionary keys
-                if key not in current:
-                    # If there are more keys after this, create appropriate structure
-                    if i < len(keys) - 1:
-                        # Check if next key is a digit (list index) or string (dict key)
-                        if keys[i + 1].isdigit():
-                            current[key] = []
-                        else:
-                            current[key] = {}
-                    else:
-                        # This is the last key, create based on value type
-                        current[key] = [] if isinstance(value, list) else {}
-                elif not isinstance(current[key], (dict, list)) and i < len(keys) - 1:
-                    # If current value is not dict/list but we need to go deeper, replace it
-                    if keys[i + 1].isdigit():
-                        current[key] = []
-                    else:
-                        current[key] = {}
-                current = current[key]
-            else:
-                # Current is not a dict or list, cannot traverse further
-                raise TypeError(
-                    f"Cannot access {'.'.join(keys[: i + 1])} as a dict/list. It's a {type(current).__name__}"
-                )
-
-        # Set the final value
-        last_key = keys[-1]
-        if isinstance(current, list) and last_key.isdigit():
-            # Setting a value in a list by index
-            index = int(last_key)
-            if index < 0:
-                raise ValueError(f"Negative list index not allowed: {index}")
-            if index >= len(current):
-                # Expand list if needed
-                while len(current) <= index:
-                    current.append(None)
-            current[index] = value
-        elif isinstance(current, dict):
-            # Special case: if the value is a list and we're setting a top-level key
-            # Example: updating engine_input_source with [1, 2]
-            current[last_key] = value
-        else:
-            # Current is not a dict, cannot set key
-            raise TypeError(f"Cannot set value at {key_path}. Current type is {type(current).__name__}, expected dict.")
-
-    # Helper function to delete by path
-    def delete_by_path(config_dict: dict, path: str) -> None:
-        """Delete configuration by dot-separated path."""
-        if not path:
-            return
-
-        current = config_dict
-        keys = path.split(".")
-
-        # Traverse to the parent
-        for i in range(len(keys) - 1):
-            key = keys[i]
-
-            # Handle list indices
-            if key.isdigit() and isinstance(current, list):
-                index = int(key)
-                if index < 0 or index >= len(current):
-                    raise KeyError(f"List index {index} out of bounds")
-                current = current[index]
-            elif isinstance(current, dict):
-                if key not in current:
-                    raise KeyError(f"Path {'.'.join(keys[: i + 1])} does not exist")
-                current = current[key]
-            else:
-                raise TypeError(
-                    f"Cannot access {'.'.join(keys[: i + 1])} as a dict/list. It's a {type(current).__name__}"
-                )
-
-        # Delete the item
-        last_key = keys[-1]
-
-        if isinstance(current, list) and last_key.isdigit():
-            index = int(last_key)
-            if index < 0 or index >= len(current):
-                raise KeyError(f"List index {index} out of bounds")
-            del current[index]
-        elif isinstance(current, dict) and last_key in current:
-            del current[last_key]
-        else:
-            raise KeyError(f"Path {path} does not exist")
-
-    # Apply deletions first
-    if deletes:
-        for key, value in deletes.items():
-            if key == "stage_args":
-                if value and isinstance(value, dict):
-                    stage_args = config.get("stage_args", [])
-                    if not stage_args:
-                        raise ValueError("stage_args does not exist in config")
-
-                    for stage_id, delete_paths in value.items():
-                        if not delete_paths:
-                            continue
-
-                        # Find stage by ID
-                        target_stage = None
-                        for stage in stage_args:
-                            if stage.get("stage_id") == stage_id:
-                                target_stage = stage
-                                break
-
-                        if target_stage is None:
-                            available_ids = [s.get("stage_id") for s in stage_args if "stage_id" in s]
-                            raise KeyError(f"Stage ID {stage_id} not found, available: {available_ids}")
-
-                        # Delete specified paths in this stage
-                        for path in delete_paths:
-                            if path:  # Skip empty paths
-                                delete_by_path(target_stage, path)
-            elif "." in key:
-                # Delete using dot-separated path
-                delete_by_path(config, key)
-            elif value is None and key in config:
-                # Delete entire key
-                del config[key]
-
-    # Apply updates
-    for key, value in updates.items():
-        if key == "stage_args":
-            if value and isinstance(value, dict):
-                stage_args = config.get("stage_args", [])
-                if not stage_args:
-                    raise ValueError("stage_args does not exist in config")
-
-                for stage_id, stage_updates in value.items():
-                    # Find stage by ID
-                    target_stage = None
-                    for stage in stage_args:
-                        if stage.get("stage_id") == stage_id:
-                            target_stage = stage
-                            break
-
-                    if target_stage is None:
-                        available_ids = [s.get("stage_id") for s in stage_args if "stage_id" in s]
-                        raise KeyError(f"Stage ID {stage_id} not found, available: {available_ids}")
-
-                    # Apply updates to this stage
-                    for path, val in stage_updates.items():
-                        # Check if this is a simple key (not dot-separated)
-                        # Example: 'engine_input_source' vs 'engine_args.max_model_len'
-                        if "." not in path:
-                            # Direct key assignment (e.g., updating a list value)
-                            target_stage[path] = val
-                        else:
-                            # Dot-separated path (e.g., nested dict access)
-                            apply_update(target_stage, path, val)
-        elif "." in key:
-            # Apply using dot-separated path
-            apply_update(config, key, value)
-        else:
-            # Direct top-level key
-            config[key] = value
-
-    # Save to new file with timestamp
-    timestamp = int(time.time())
-    base_name = yaml_path.rsplit(".", 1)[0] if "." in yaml_path else yaml_path
-    output_path = f"{base_name}_{timestamp}.yaml"
-
-    with open(output_path, "w", encoding="utf-8") as f:
-        yaml.dump(config, f, default_flow_style=None, sort_keys=False, allow_unicode=True, indent=2)
-
-    return output_path
-
-
-class OmniServer:
-    """Omniserver for vLLM-Omni tests."""
-
-    def __init__(
-        self,
-        model: str,
-        serve_args: list[str],
-        *,
-        env_dict: dict[str, str] | None = None,
-    ) -> None:
-        self.model = model
-        self.serve_args = serve_args
-        self.env_dict = env_dict
-        self.proc: subprocess.Popen | None = None
-        self.host = "127.0.0.1"
-        self.port = get_open_port()
-
-    def _start_server(self) -> None:
-        """Start the vLLM-Omni server subprocess."""
-        env = os.environ.copy()
-        env["VLLM_WORKER_MULTIPROC_METHOD"] = "spawn"
-        if self.env_dict is not None:
-            env.update(self.env_dict)
-
-        cmd = [
-            sys.executable,
-            "-m",
-            "vllm_omni.entrypoints.cli.main",
-            "serve",
-            self.model,
-            "--omni",
-            "--host",
-            self.host,
-            "--port",
-            str(self.port),
-        ] + self.serve_args
-
-        print(f"Launching OmniServer with: {' '.join(cmd)}")
-        self.proc = subprocess.Popen(
-            cmd,
-            env=env,
-            cwd=os.path.dirname(os.path.dirname(os.path.abspath(__file__))),  # Set working directory to vllm-omni root
-        )
-
-        # Wait for server to be ready
-        max_wait = 600  # 10 minutes
-        start_time = time.time()
-        while time.time() - start_time < max_wait:
-            try:
-                with socket.socket(socket.AF_INET, socket.SOCK_STREAM) as sock:
-                    sock.settimeout(1)
-                    result = sock.connect_ex((self.host, self.port))
-                    if result == 0:
-                        print(f"Server ready on {self.host}:{self.port}")
-                        return
-            except Exception:
-                pass
-            time.sleep(2)
-
-        raise RuntimeError(f"Server failed to start within {max_wait} seconds")
-
-    def _kill_process_tree(self, pid):
-        """kill process and its children with verification"""
-        try:
-            parent = psutil.Process(pid)
-            children = parent.children(recursive=True)
-
-            # Get all PIDs first
-            all_pids = [pid] + [child.pid for child in children]
-
-            # Terminate children
-            for child in children:
-                try:
-                    child.terminate()
-                except psutil.NoSuchProcess:
-                    pass
-
-            # Wait for children
-            gone, still_alive = psutil.wait_procs(children, timeout=10)
-
-            # Kill remaining children
-            for child in still_alive:
-                try:
-                    child.kill()
-                except psutil.NoSuchProcess:
-                    pass
-
-            # Terminate parent
-            try:
-                parent.terminate()
-                parent.wait(timeout=10)
-            except (psutil.NoSuchProcess, psutil.TimeoutExpired):
-                try:
-                    parent.kill()
-                except psutil.NoSuchProcess:
-                    pass
-
-            # VERIFICATION: Check if all processes are gone
-            time.sleep(1)  # Give system time
-            alive_processes = []
-            for check_pid in all_pids:
-                if psutil.pid_exists(check_pid):
-                    alive_processes.append(check_pid)
-
-            if alive_processes:
-                print(f"Warning: Processes still alive: {alive_processes}")
-                # Optional: Try system kill
-                import subprocess
-
-                for alive_pid in alive_processes:
-                    try:
-                        subprocess.run(["kill", "-9", str(alive_pid)], timeout=2)
-                    except Exception as e:
-                        print(f"Cleanup failed: {e}")
-
-        except psutil.NoSuchProcess:
-            pass
-
-    def __enter__(self):
-        self._start_server()
-        return self
-
-    def __exit__(self, exc_type, exc_val, exc_tb):
-        if self.proc:
-            self._kill_process_tree(self.proc.pid)
-        _run_post_test_cleanup()
-        cleanup_dist_env_and_memory()
+        gc.collect()

tests/diffusion/attention/test_flash_attn.py

@@ -1,290 +0,0 @@
-# SPDX-License-Identifier: Apache-2.0
-# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
-
-"""
-Test script for FlashAttention backend with padding handling.
-
-This script tests two main scenarios:
-1. Case 1: Comparing padded vs unpadded inputs for batch_size=1
-2. Case 2: Comparing FlashAttention and SDPA backends for batch_size=2 with padding
-"""
-
-import pytest
-import torch
-
-from vllm_omni.diffusion.attention.backends.abstract import AttentionMetadata
-from vllm_omni.diffusion.attention.backends.flash_attn import FlashAttentionImpl
-from vllm_omni.diffusion.attention.backends.sdpa import SDPAImpl
-
-
-def create_attention_mask(batch_size: int, seq_len: int, valid_len: int, device: torch.device) -> torch.Tensor:
-    """
-    Create attention mask where first valid_len tokens are valid (1) and rest are padding (0).
-
-    Args:
-        batch_size: Batch size
-        seq_len: Total sequence length (including padding)
-        valid_len: Number of valid (non-padded) tokens
-
-    Returns:
-        Attention mask of shape (batch_size, seq_len)
-    """
-    mask = torch.zeros(batch_size, seq_len, dtype=torch.bool, device=device)
-    mask[:, :valid_len] = True
-    return mask
-
-
-def pad_tensor(tensor: torch.Tensor, target_seq_len: int, pad_value: float = 0.0) -> torch.Tensor:
-    """
-    Pad tensor along sequence dimension (dim=1).
-
-    Args:
-        tensor: Input tensor of shape (batch_size, seq_len, num_heads, head_dim)
-        target_seq_len: Target sequence length after padding
-        pad_value: Value to use for padding
-
-    Returns:
-        Padded tensor of shape (batch_size, target_seq_len, num_heads, head_dim)
-    """
-    batch_size, seq_len, num_heads, head_dim = tensor.shape
-    if target_seq_len <= seq_len:
-        return tensor
-
-    padding = torch.full(
-        (batch_size, target_seq_len - seq_len, num_heads, head_dim), pad_value, dtype=tensor.dtype, device=tensor.device
-    )
-    return torch.cat([tensor, padding], dim=1)
-
-
-@pytest.mark.skipif(not torch.cuda.is_available(), reason="FlashAttention requires CUDA")
-def test_padding_equivalence():
-    """
-    Case 1: Test that padded and unpadded inputs produce similar outputs.
-
-    - Input A: batch_size=1, hidden_states (1, 48), encoder_hidden_states (1, 16)
-      Concatenated length: 64, NO attention_mask
-    - Input B: Same data but padded: hidden_states (1, 58), encoder_hidden_states (1, 26)
-      Concatenated length: 84, WITH attention_mask
-
-    Expected: Output A and Output B should be very close.
-    """
-    device = torch.device("cuda")
-    dtype = torch.bfloat16
-
-    # Configuration
-    batch_size = 1
-    hidden_seq_len = 48
-    encoder_seq_len = 16
-    pad_length = 10
-    num_heads = 8
-    head_dim = 64
-
-    # Initialize FlashAttention
-    fa_impl = FlashAttentionImpl(
-        num_heads=num_heads, head_size=head_dim, softmax_scale=1.0 / (head_dim**0.5), causal=False
-    )
-
-    # Create base tensors with random values (same for both A and B)
-    torch.manual_seed(42)
-    hidden_states_base = torch.randn(batch_size, hidden_seq_len, num_heads, head_dim, device=device, dtype=dtype)
-    encoder_hidden_states_base = torch.randn(
-        batch_size, encoder_seq_len, num_heads, head_dim, device=device, dtype=dtype
-    )
-
-    # ========== Input A: Unpadded, no attention mask ==========
-    query_a = torch.cat([hidden_states_base, encoder_hidden_states_base], dim=1)
-    key_a = query_a.clone()
-    value_a = query_a.clone()
-
-    attn_metadata_a = AttentionMetadata(attn_mask=None)
-
-    output_a = fa_impl.forward(query=query_a, key=key_a, value=value_a, attn_metadata=attn_metadata_a)
-
-    # ========== Input B: Padded with attention mask ==========
-    hidden_states_padded = pad_tensor(hidden_states_base, hidden_seq_len + pad_length)
-    encoder_hidden_states_padded = pad_tensor(encoder_hidden_states_base, encoder_seq_len + pad_length)
-
-    query_b = torch.cat([hidden_states_padded, encoder_hidden_states_padded], dim=1)
-    key_b = query_b.clone()
-    value_b = query_b.clone()
-
-    # Create attention mask
-    attn_mask_b = torch.cat(
-        [
-            create_attention_mask(batch_size, hidden_seq_len + pad_length, hidden_seq_len, device),
-            create_attention_mask(batch_size, encoder_seq_len + pad_length, encoder_seq_len, device),
-        ],
-        dim=1,
-    )
-
-    attn_metadata_b = AttentionMetadata(attn_mask=attn_mask_b)
-
-    output_b = fa_impl.forward(query=query_b, key=key_b, value=value_b, attn_metadata=attn_metadata_b)
-
-    # Extract non-padded portion from output_b
-    output_b_unpadded = torch.cat(
-        [
-            output_b[:, :hidden_seq_len, :, :],
-            output_b[:, hidden_seq_len + pad_length : hidden_seq_len + pad_length + encoder_seq_len, :, :],
-        ],
-        dim=1,
-    )
-
-    # Compare outputs
-    max_diff = torch.max(torch.abs(output_a - output_b_unpadded)).item()
-    mean_diff = torch.mean(torch.abs(output_a - output_b_unpadded)).item()
-
-    print("\n=== Case 1: Padding Equivalence Test ===")
-    print(f"Output A shape: {output_a.shape}")
-    print(f"Output B shape: {output_b.shape}")
-    print(f"Output B unpadded shape: {output_b_unpadded.shape}")
-    print(f"Max absolute difference: {max_diff:.6f}")
-    print(f"Mean absolute difference: {mean_diff:.6f}")
-
-    # Assert that outputs are close
-    # Using higher tolerance for bfloat16
-    assert max_diff < 0.1, f"Max difference {max_diff} exceeds threshold 0.1"
-    assert mean_diff < 0.01, f"Mean difference {mean_diff} exceeds threshold 0.01"
-
-    print("✓ Case 1 PASSED: Padded and unpadded outputs are very close!")
-
-
-@pytest.mark.skipif(not torch.cuda.is_available(), reason="FlashAttention requires CUDA")
-def test_fa_vs_sdpa():
-    """
-    Case 2: Compare FlashAttention and SDPA backends with padding.
-
-    - batch_size=2
-    - hidden_states: (2, 48) padded to (2, 58)
-    - encoder_hidden_states: (2, 16) padded to (2, 26)
-    - Concatenated length: 84
-    - Compare FA and SDPA outputs
-
-    Expected: FA and SDPA outputs should be very close.
-    """
-    device = torch.device("cuda")
-    dtype = torch.bfloat16
-
-    # Configuration
-    batch_size = 2
-    hidden_seq_len = 48
-    encoder_seq_len = 16
-    pad_length = 10
-    num_heads = 8
-    head_dim = 64
-
-    # Initialize both backends
-    fa_impl = FlashAttentionImpl(
-        num_heads=num_heads, head_size=head_dim, softmax_scale=1.0 / (head_dim**0.5), causal=False
-    )
-
-    sdpa_impl = SDPAImpl(num_heads=num_heads, head_size=head_dim, softmax_scale=1.0 / (head_dim**0.5), causal=False)
-
-    # Create base tensors
-    torch.manual_seed(123)
-    hidden_states_base = torch.randn(batch_size, hidden_seq_len, num_heads, head_dim, device=device, dtype=dtype)
-    encoder_hidden_states_base = torch.randn(
-        batch_size, encoder_seq_len, num_heads, head_dim, device=device, dtype=dtype
-    )
-
-    # Pad tensors
-    hidden_states_padded = pad_tensor(hidden_states_base, hidden_seq_len + pad_length)
-    encoder_hidden_states_padded = pad_tensor(encoder_hidden_states_base, encoder_seq_len + pad_length)
-
-    # Concatenate
-    query = torch.cat([hidden_states_padded, encoder_hidden_states_padded], dim=1)
-    key = query.clone()
-    value = query.clone()
-
-    # Create attention mask
-    attn_mask = torch.cat(
-        [
-            create_attention_mask(batch_size, hidden_seq_len + pad_length, hidden_seq_len, device),
-            create_attention_mask(batch_size, encoder_seq_len + pad_length, encoder_seq_len, device),
-        ],
-        dim=1,
-    )
-
-    attn_metadata = AttentionMetadata(attn_mask=attn_mask)
-
-    # Run FlashAttention
-    output_fa = fa_impl.forward(query=query.clone(), key=key.clone(), value=value.clone(), attn_metadata=attn_metadata)
-
-    # Run SDPA
-    # SDPA expects 4D attention mask: (batch_size, 1, seq_len, seq_len) or (batch_size, seq_len)
-    # For causal=False, we need to convert 2D mask to 4D
-    if attn_mask is not None:
-        # Expand mask for SDPA: (batch_size, seq_len) -> (batch_size, 1, 1, seq_len)
-        attn_mask_4d = attn_mask.unsqueeze(1).unsqueeze(2)
-        # Convert bool to float: True -> 0.0, False -> -inf
-        attn_mask_float = torch.zeros_like(attn_mask_4d, dtype=dtype)
-        attn_mask_float.masked_fill_(~attn_mask_4d, float("-inf"))
-        attn_metadata_sdpa = AttentionMetadata(attn_mask=attn_mask_float)
-    else:
-        attn_metadata_sdpa = AttentionMetadata(attn_mask=None)
-
-    output_sdpa = sdpa_impl.forward(
-        query=query.clone(), key=key.clone(), value=value.clone(), attn_metadata=attn_metadata_sdpa
-    )
-
-    # Compare outputs (only compare valid regions)
-    output_fa_valid = torch.cat(
-        [
-            output_fa[:, :hidden_seq_len, :, :],
-            output_fa[:, hidden_seq_len + pad_length : hidden_seq_len + pad_length + encoder_seq_len, :, :],
-        ],
-        dim=1,
-    )
-    output_sdpa_valid = torch.cat(
-        [
-            output_sdpa[:, :hidden_seq_len, :, :],
-            output_sdpa[:, hidden_seq_len + pad_length : hidden_seq_len + pad_length + encoder_seq_len, :, :],
-        ],
-        dim=1,
-    )
-
-    max_diff = torch.max(torch.abs(output_fa_valid - output_sdpa_valid)).item()
-    mean_diff = torch.mean(torch.abs(output_fa_valid - output_sdpa_valid)).item()
-
-    print("\n=== Case 2: FA vs SDPA Comparison ===")
-    print(f"Batch size: {batch_size}")
-    print(f"FA output shape: {output_fa.shape}")
-    print(f"SDPA output shape: {output_sdpa.shape}")
-    print(f"Max absolute difference (valid region): {max_diff:.6f}")
-    print(f"Mean absolute difference (valid region): {mean_diff:.6f}")
-
-    # Assert that outputs are close
-    # Using higher tolerance for bfloat16 and different implementations
-    assert max_diff < 0.01, f"Max difference {max_diff} exceeds threshold 0.01"
-    assert mean_diff < 0.001, f"Mean difference {mean_diff} exceeds threshold 0.001"
-
-    print("✓ Case 2 PASSED: FA and SDPA outputs are very close!")
-
-
-if __name__ == "__main__":
-    print("Running FlashAttention Padding Tests...")
-    print("=" * 60)
-
-    # Try to run CUDA tests
-    if torch.cuda.is_available():
-        try:
-            print("\n[Running Case 1: Padding Equivalence for FA]")
-            test_padding_equivalence()
-        except Exception as e:
-            print(f"✗ Case 1 failed: {e}")
-            import traceback
-
-            traceback.print_exc()
-
-        try:
-            print("\n[Running Case 2: FA vs SDPA]")
-            test_fa_vs_sdpa()
-        except Exception as e:
-            print(f"✗ Case 2 failed: {e}")
-            import traceback
-
-            traceback.print_exc()
-    else:
-        raise RuntimeError("CUDA is not available")
-    print("\n" + "=" * 60)
-    print("Test suite completed!")

tests/diffusion/attention/test_sequence_parallel.py

@@ -170,11 +170,6 @@ def test_sequence_parallel(
     """Test Ulysses attention by comparing with and without SP enabled."""
     sequence_parallel_size = ulysses_degree * ring_degree
 
-    # Skip if not enough GPUs available
-    available_gpus = torch_device.device_count()
-    if available_gpus < sequence_parallel_size:
-        pytest.skip(f"Test requires {sequence_parallel_size} GPUs but only {available_gpus} available")
-
     # Create temporary files to share results between processes
     with tempfile.NamedTemporaryFile(delete=False, suffix=".pkl") as f:
         baseline_output_file = f.name

tests/diffusion/distributed/test_comm.py

@@ -48,11 +48,6 @@ def test_4d_identity(
     use_sync: bool,
 ):
     """Test that two consecutive all-to-all operations return the original input."""
-    # Skip if not enough GPUs available
-    available_gpus = torch_device.device_count()
-    if available_gpus < world_size:
-        pytest.skip(f"Test requires {world_size} GPUs but only {available_gpus} available")
-
     # Ensure num_heads is divisible by world_size
     if num_heads % world_size != 0:
         pytest.skip(f"num_heads ({num_heads}) not divisible by world_size ({world_size})")
@@ -182,11 +177,6 @@ def test_5d_identity(
     use_sync: bool,
 ):
     """Test that two consecutive all-to-all operations return the original input."""
-    # Skip if not enough GPUs available
-    available_gpus = torch_device.device_count()
-    if available_gpus < world_size:
-        pytest.skip(f"Test requires {world_size} GPUs but only {available_gpus} available")
-
     # Ensure num_heads is divisible by world_size
     if num_heads % world_size != 0:
         pytest.skip(f"num_heads ({num_heads}) not divisible by world_size ({world_size})")
@@ -315,11 +305,6 @@ def test_ring_p2p(
     head_size: int,
 ):
     """Test Ring P2P communication (send_recv)."""
-    # Skip if not enough GPUs available
-    available_gpus = torch_device.device_count()
-    if available_gpus < world_size:
-        pytest.skip(f"Test requires {world_size} GPUs but only {available_gpus} available")
-
     torch.multiprocessing.spawn(
         _test_ring_p2p_worker,
         args=(world_size, dtype, batch_size, num_heads, head_size),