Profile native Windows CPU InternVLA through Omni

2026-09-24 02:05:26 +08:00
1 parent 48646da78e
commit 752357d318
@@ -26,17 +26,26 @@ async def main() -> None:
        parser.add_argument(f"--{name}", type=Path, required=True)
    parser.add_argument("--graph-file", type=Path)
    parser.add_argument("--placement", choices=("cpu", "radeon-cosmos"), required=True)
+    parser.add_argument("--capacity-gib", type=int, default=30)
+    parser.add_argument("--reserve-gib", type=int, default=16)
    args = parser.parse_args()
    if args.placement == "radeon-cosmos" and args.graph_file is None:
        parser.error("Radeon placement requires graph-file")
+    if args.reserve_gib < 1 or args.capacity_gib < args.reserve_gib:
+        parser.error("invalid explicit host-RAM budget")

    import numpy as np
+    import psutil
    import torch
    import yaml

    from vllm_omni.diffusion.models.internvla_a1_whole_pipeline import INTERNVLA_A1_WHOLE_POLICY_PIPELINE
    from vllm_omni.entrypoints.async_omni import AsyncOmni

+    host_available_bytes = psutil.virtual_memory().available
+    if args.capacity_gib << 30 > host_available_bytes:
+        raise RuntimeError("declared host-RAM capacity exceeds OS available RAM before load")
+
    model = args.model_dir.resolve(strict=True)
    cosmos = args.cosmos_dir.resolve(strict=True)
    processor = args.processor_dir.resolve(strict=True)
@@ -63,7 +72,9 @@ async def main() -> None:
        "start_timeout_s": 180, "request_timeout_s": 60,
    }
    report = {"entrypoint": "AsyncOmni.generate", "placement": args.placement,
-              "pipeline": INTERNVLA_A1_WHOLE_POLICY_PIPELINE.model_type}
+              "pipeline": INTERNVLA_A1_WHOLE_POLICY_PIPELINE.model_type,
+              "host_available_bytes_before": host_available_bytes,
+              "budget": {"capacity_gib": args.capacity_gib, "reserve_gib": args.reserve_gib}}
    engine = None
    try:
        with tempfile.TemporaryDirectory(prefix="omni-internvla-") as directory:
@@ -72,8 +83,8 @@ async def main() -> None:
                "pipeline": INTERNVLA_A1_WHOLE_POLICY_PIPELINE.model_type,
                "async_chunk": False,
                "stages": [{"stage_id": 0, "backend": backend,
-                            "resource_budget": {"capacities": {"host_ram": 30 << 30},
-                                                "demands": {"host_ram": 16 << 30}}}],
+                            "resource_budget": {"capacities": {"host_ram": args.capacity_gib << 30},
+                                                "demands": {"host_ram": args.reserve_gib << 30}}}],
            }), encoding="utf-8")
            started = time.perf_counter()
            engine = AsyncOmni(
@@ -46,6 +46,11 @@ async def main() -> None:
        parser.error("Radeon placement requires --graph-file")

    import numpy as np
+    import psutil
+
+    host_available_bytes = psutil.virtual_memory().available
+    if args.capacity_gib << 30 > host_available_bytes:
+        raise RuntimeError("declared host-RAM capacity exceeds OS available RAM before load")

    from vllm_omni.config.stage_config import DeployConfig, StageDeployConfig, merge_pipeline_deploy
    from vllm_omni.diffusion.models.internvla_a1_whole_pipeline import INTERNVLA_A1_WHOLE_POLICY_PIPELINE
@@ -90,6 +95,7 @@ async def main() -> None:
    report = {
        "scope": "real InternVLA Place_Markpen checkpoint via bounded Omni whole-policy graph stage; synthetic patterned observations/noise; no robot-task quality claim",
        "placement": args.placement, "artifact_sha256": hashes, "budget": budget,
+        "host_available_bytes_before": host_available_bytes,
        "warmups": args.warmups, "repeats": args.repeats,
    }
    try:
@@ -4,6 +4,8 @@ This extends the [60-cell device × model matrix](../e2e_profiling_20260922/evid

 InternVLA also has a separate [admitted Omni whole-policy synthetic profile](evidence/internvla_omni_policy/README.md) on WSL CPU and WSL CPU+Radeon 890M: one warmup plus 20 serial complete requests per placement, explicit 16 GiB shared-RAM reservation, pre-launch refusal, hybrid in-flight abort, and a public `AsyncOmni.generate` action output. The two complete-request profiles are 3.175/3.219 s CPU and 3.120/3.197 s hybrid p50/p95; they use a patterned fixture distinct from the older direct hybrid run. Action relative L2 difference is 2.305% without a task tolerance. This advances integration depth only; robot-task E2E remains unqualified.

+A further [native Windows CPU Omni policy profile](evidence/internvla_omni_windows_cpu/README.md) ran the same patterned synthetic fixture with the real Place_Markpen checkpoint: one warmup plus 20/20 identical finite action chunks, nearest-rank complete-request wall p50/p95 4.878/4.902 s, and 135.60 s startup. The public action entrypoint, pre-launch 8 GiB refusal and in-flight abort passed with a 16 GiB host-RAM reservation and zero ledger residue. The native action chunk differed from the WSL CPU output by 2.1795% relative L2; no robot-task tolerance exists. The earlier one-forward native CPU result used a zero-input fixture and must not be compared as the same profile.
+
 | Device / model | Verified depth | Observed output and time | Remaining gate |
 |---|---|---|---|
 | Native Windows HX370 Radeon 890M / InternVLA-A1 Cosmos CI8x8 encoder | Component C; unchanged Omni encoder source in separate DirectML/PyTorch 2.4 environment, not an Omni full-policy stage | One warmup + 20 serial patterned 256×256 image encodes, including upload/readback: Radeon p50/p95 26.56/27.49 ms versus CPU 156.49/158.71 ms; relative L2 error 4.73e-07. Encoder weights and output device checked. | Complete policy/action and real-observation parity; policy/controller runtime compatibility, bounded handoff, shared-RAM admission and end-to-end benefit. [Raw evidence](evidence/internvla_cosmos_radeon890m/README.md). |
@@ -0,0 +1,77 @@
+# InternVLA-A1 admitted Omni whole-policy stage on native Windows CPU
+
+**Scoped synthetic action-policy pass, not robot-task E2E.** The real
+Place_Markpen checkpoint (revision
+`0a557c6a503e6545b369bff2f34c76a410bbd190`) strict-loaded in the
+existing Omni InternVLA diffusion policy. One `StageRuntime`/`StagePool` graph
+stage owns an isolated whole-policy worker, bounded one-request capacity,
+terminal action event, shared-RAM reservation and cancellation. The model
+parameters and output actions were checked on CPU; the CUDA-capable wheel does
+not imply GPU execution. This is the native Windows counterpart of the
+[WSL CPU/Radeon Omni policy record](../internvla_omni_policy/README.md).
+
+The host was Ryzen AI 9 HX 370 running Windows 11 build 26200. The controller
+and worker used Python 3.12.10, PyTorch 2.13.0+cu130 on CPU, vLLM 0.29.0 and
+the current Omni source. The real checkpoint, processor and Cosmos files were
+read over the local WSL UNC path. Their SHA256 values, interpreter hash and
+worker properties are in the [20-request report](cpu_profile20.json). The
+fixture comprises three normalized float32 `[1,2,3,224,224]` patterned camera
+histories, valid masks, zero `[1,32]` state, the task text “Place the marker
+pen in its holder.” and zero `[1,50,32]` diffusion noise. Each response is a
+finite float32 `[1,50,32]` action chunk with an epoch/sequence-tagged terminal
+event and observation/generation timestamps. The checkpoint declares
+`action_mode=delta`; physical units, joint order and per-action step time are
+unverified and `control_ready=false`.
+
+| Check | Result | Evidence |
+|---|---|---|
+| Resident serial profile | One warmup excluded; 20/20 requests had the same action hash. Nearest-rank complete-request wall p50/p95 **4.878/4.902 s**, excluding **135.60 s** startup. Loaded worker-tree RSS was 9.00 GB, not loading peak. | [report](cpu_profile20.json), [driver](cpu_profile20_driver.log), [worker](cpu_profile20_worker.log), [actions](cpu_profile20_actions.npy) |
+| WSL CPU numerical comparison | Same pinned patterned fixture: max absolute action difference `0.06504`, relative L2 `2.1795%`, cosine `0.999764`. This is numerical comparison without a task tolerance. | [profile report](cpu_profile20.json), [WSL reference](../internvla_omni_policy/cpu_profile20_actions.npy) |
+| Public entrypoint | `AsyncOmni.generate` returned one terminal `action` event with the same action hash and finite shape `[1,50,32]`. Startup **134.61 s**, request wall **4.979 s**; OS-available RAM before launch was 23.67 GB. | [report](public.json), [driver](public_driver.log), [worker](public_worker.log) |
+| Admission and abort | A 16 GiB stage demand/capacity ran within the Windows available-RAM check. An 8 GiB demand was refused before model-worker launch. In-flight abort produced no stale output; the ledger had zero reserved and quarantined bytes after abort and shutdown. | [profile report](cpu_profile20.json), [refusal](refusal.json), [refusal log](refusal_driver.log) |
+
+The [first smoke](cpu_smoke.json) used a 24 GiB declared capacity when Windows
+reported about 20 GiB free, so it is retained as exploratory evidence only;
+the measured run uses a conservative 16 GiB capacity and demand. A first
+[public smoke](public_overdeclared_capacity.json) also used a 30 GiB ceiling;
+its valid action does not qualify memory admission. The public probe now
+checks that its declared ceiling fits OS-available RAM before loading, and
+the corrected [public rerun](public.json) passed. The probe's declared memory
+ceiling is a controller admission budget, not a measured loading peak or a
+claim that Windows could admit another stage concurrently. No real
+Place_Markpen observations or reference actions, robot-task tolerance,
+post-cancel restart, concurrency or sustained power/thermal behavior were
+verified.
+
+Reproduce from the repository root in PowerShell, using the pinned files and
+native environment already present on this host:
+
+```powershell
+$root = '\\wsl.localhost\Ubuntu\home\zhout\project\edge_infer'
+$repo = Join-Path $root 'vllm-omni-edge'
+$evidence = Join-Path $repo 'benchmarks\edge_harness\results\e2e_expansion_20260923\evidence\internvla_omni_windows_cpu'
+$py = 'C:\Users\zhout\w2\omni029venv\Scripts\python.exe'
+$env:PYTHONUTF8 = '1'
+$env:PYTHONPATH = $repo
+$env:VLLM_ENABLE_V1_MULTIPROCESSING = '0'
+$common = @(
+  '--model-dir', (Join-Path $root 'models\InternVLA-A1-3B-FT-Place_Markpen'),
+  '--processor-dir', (Join-Path $root 'models\Qwen3-VL-2B-Instruct-processor'),
+  '--cosmos-dir', (Join-Path $root 'models\Cosmos-Tokenizer-CI8x8'),
+  '--python-bin', $py, '--placement', 'cpu',
+  '--capacity-gib', '16', '--reserve-gib', '16'
+)
+& $py (Join-Path $repo 'benchmarks\edge_harness\experiments\probe_omni_internvla_policy.py') @common `
+  --warmups 1 --repeats 20 --abort-check `
+  --reference-actions (Join-Path $repo 'benchmarks\edge_harness\results\e2e_expansion_20260923\evidence\internvla_omni_policy\cpu_profile20_actions.npy') `
+  --log-file (Join-Path $evidence 'cpu_profile20_worker.log') `
+  --output-report (Join-Path $evidence 'cpu_profile20.json') `
+  --output-actions (Join-Path $evidence 'cpu_profile20_actions.npy')
+& $py (Join-Path $repo 'benchmarks\edge_harness\experiments\probe_omni_internvla_entrypoint.py') @common `
+  --log-file (Join-Path $evidence 'public_worker.log') `
+  --output-report (Join-Path $evidence 'public.json')
+```
+
+Next gate: real observation/reference-action agreement and physical action
+metadata, then loading peak, post-cancel restart, concurrent admission and
+sustained deadline/power/thermal behavior.
@@ -0,0 +1,277 @@
+INFO 09-24 01:48:55 [patch.py:259] NVFP4 W4A4 weight_scale NaN-clamp: installed.
+INFO 09-24 01:48:55 [patch.py:510] inductor factorable-divisibility patch: installed.
+INFO 09-24 01:49:34 [stage_init_utils.py:768] [stage_init] Set VLLM_WORKER_MULTIPROC_METHOD=spawn
+WARNING 09-24 01:49:34 [stage_init_utils.py:1887] [stage_init] Failed to load transfer config: [Errno 13] Permission denied: 'C:\\Users\\zhout'
+{
+  "scope": "real InternVLA Place_Markpen checkpoint via bounded Omni whole-policy graph stage; synthetic patterned observations/noise; no robot-task quality claim",
+  "placement": "cpu",
+  "artifact_sha256": {
+    "python": "0b471133e110cfb53a061cad528ce8e517d7b9ac41a0a396c39ad795a487fc14",
+    "model": "586799539888c69c35f84d083e559cc5f0ba887801d000e70e65f58d3f85c313",
+    "model_config": "1272ca8fb93cb42f48b039bb298688b2bd50a66f4faa0afcd9e538e9762f48aa",
+    "train_config": "e50ec42e40ba1925983e0e8db36558d3738bea64e1698045ff7cc08a130e1053",
+    "stats": "354710650bb91fc0e84a461987b3554b30bef9d5de47567480624484cdfbf3a2",
+    "cosmos_encoder": "520a3b878e72ee872faccb1211b5d622d194c0fdd177f2f6df097586b8fa98e5",
+    "cosmos_decoder": "3b7791e6c8f6f92caacddf48ebb314f044bcb8302d0c7f47501d384a7a80f738",
+    "processor_tokenizer": "a5d85b6dcc535e6b93115a9ef287e6132fdbf30270da6218194ba742261173c7",
+    "processor_config": "27225450ac9c6529872ee1924fcb0962ff5634834f817040f444118116f4e516"
+  },
+  "budget": {
+    "capacities": {
+      "host_ram": 17179869184
+    },
+    "demands": {
+      "host_ram": 17179869184
+    }
+  },
+  "warmups": 1,
+  "repeats": 20,
+  "startup_s": 135.59662149997894,
+  "execution_plan_start": {
+    "backend": "external.internvla.policy.v1",
+    "stage_id": 0,
+    "worker_generation": "ee62715e4eb442d3bd951ce3646dcaa4",
+    "worker_pid": 13452,
+    "worker_script_sha256": "55ffe4ce194eee7e330a4e5def8a2bfec0bfbb98b182885e094f71ea57e4ac9a",
+    "artifact_sha256": {
+      "python": "0b471133e110cfb53a061cad528ce8e517d7b9ac41a0a396c39ad795a487fc14",
+      "model": "586799539888c69c35f84d083e559cc5f0ba887801d000e70e65f58d3f85c313",
+      "model_config": "1272ca8fb93cb42f48b039bb298688b2bd50a66f4faa0afcd9e538e9762f48aa",
+      "train_config": "e50ec42e40ba1925983e0e8db36558d3738bea64e1698045ff7cc08a130e1053",
+      "stats": "354710650bb91fc0e84a461987b3554b30bef9d5de47567480624484cdfbf3a2",
+      "cosmos_encoder": "520a3b878e72ee872faccb1211b5d622d194c0fdd177f2f6df097586b8fa98e5",
+      "cosmos_decoder": "3b7791e6c8f6f92caacddf48ebb314f044bcb8302d0c7f47501d384a7a80f738",
+      "processor_tokenizer": "a5d85b6dcc535e6b93115a9ef287e6132fdbf30270da6218194ba742261173c7",
+      "processor_config": "27225450ac9c6529872ee1924fcb0962ff5634834f817040f444118116f4e516"
+    },
+    "placement": "cpu",
+    "worker_props": {
+      "model_dir": "\\\\wsl.localhost\\Ubuntu\\home\\zhout\\project\\edge_infer\\models\\InternVLA-A1-3B-FT-Place_Markpen",
+      "processor_dir": "\\\\wsl.localhost\\Ubuntu\\home\\zhout\\project\\edge_infer\\models\\Qwen3-VL-2B-Instruct-processor",
+      "cosmos_dir": "\\\\wsl.localhost\\Ubuntu\\home\\zhout\\project\\edge_infer\\models\\Cosmos-Tokenizer-CI8x8",
+      "placement": "cpu",
+      "torch": "2.13.0+cu130",
+      "policy_dtype": "bfloat16",
+      "cosmos_dtype": "bfloat16",
+      "policy_device": "cpu",
+      "runtime_mode": "real_checkpoint_loaded",
+      "policy_load_s": 77.1901350999251,
+      "action_shape": [
+        1,
+        50,
+        32
+      ],
+      "image_shape": [
+        1,
+        2,
+        3,
+        224,
+        224
+      ],
+      "state_shape": [
+        1,
+        32
+      ],
+      "action_mode": "delta",
+      "action_units": "unverified",
+      "action_joint_order": "unverified",
+      "action_step_s": null,
+      "control_ready": false,
+      "external_load": null
+    },
+    "loaded_rss_bytes": 8997961728,
+    "loaded_process_tree": [
+      {
+        "pid": 13452,
+        "rss_bytes": 4386816,
+        "private_bytes": 790528
+      },
+      {
+        "pid": 56984,
+        "rss_bytes": 7462912,
+        "private_bytes": 1228800
+      },
+      {
+        "pid": 39924,
+        "rss_bytes": 8986112000,
+        "private_bytes": 11148115968
+      }
+    ],
+    "reserved_bytes": {
+      "host_ram": 17179869184
+    },
+    "memory_overhead_bytes": 8589934592,
+    "max_input_bytes": 8388608,
+    "max_action_bytes": 1048576,
+    "request_capacity": 1,
+    "stateful_session": "InternVLA policy worker-owned; complete action chunk per request",
+    "evidence": "B"
+  },
+  "warmup_results": [
+    {
+      "wall_s": 4.982390600023791,
+      "action_sha256": "82dd03dc31833fda60bff6620030315bad4034089b0c6c34c9de74e1c97e8d36",
+      "stage_event": {
+        "request_id": "internvla-warmup-0",
+        "stage_id": 0,
+        "epoch": 1,
+        "seq": 1,
+        "kind": "action",
+        "worker_generation": "ee62715e4eb442d3bd951ce3646dcaa4",
+        "buffers": [
+          {
+            "object_id": "actions",
+            "owner": "0",
+            "generation": "ee62715e4eb442d3bd951ce3646dcaa4",
+            "dtype": "float32",
+            "shape": [
+              1,
+              50,
+              32
+            ],
+            "nbytes": 6400,
+            "memory_domain": "host-copy",
+            "layout": "C",
+            "ready": true
+          }
+        ],
+        "terminal": true,
+        "error": null
+      },
+      "action_metadata": {
+        "observation_timestamp_ns": 1790185909964825100,
+        "generation_timestamp_ns": 1790185914940833500,
+        "observation_age_s": 4.9760084,
+        "action_mode": "delta",
+        "action_shape": [
+          1,
+          50,
+          32
+        ],
+        "action_units": "unverified",
+        "joint_order": "unverified",
+        "action_step_s": null,
+        "control_ready": false
+      },
+      "metrics": {
+        "policy_wall_s": 4.9746512999990955,
+        "worker_wall_s": 4.942081900080666
+      }
+    }
+  ],
+  "nearest_rank_p50_wall_s": 4.8775576000334695,
+  "nearest_rank_p95_wall_s": 4.902349299984053,
+  "unique_action_hashes": [
+    "82dd03dc31833fda60bff6620030315bad4034089b0c6c34c9de74e1c97e8d36"
+  ],
+  "output_actions": "\\\\wsl.localhost\\Ubuntu\\home\\zhout\\project\\edge_infer\\vllm-omni-edge\\benchmarks\\edge_harness\\results\\e2e_expansion_20260923\\evidence\\internvla_omni_windows_cpu\\cpu_profile20_actions.npy",
+  "reference_comparison": {
+    "max_abs": 0.0650382936000824,
+    "relative_l2": 0.021795250475406647,
+    "cosine": 0.9997638463973999
+  },
+  "execution_plan_end": {
+    "backend": "external.internvla.policy.v1",
+    "stage_id": 0,
+    "worker_generation": "ee62715e4eb442d3bd951ce3646dcaa4",
+    "worker_pid": 13452,
+    "worker_script_sha256": "55ffe4ce194eee7e330a4e5def8a2bfec0bfbb98b182885e094f71ea57e4ac9a",
+    "artifact_sha256": {
+      "python": "0b471133e110cfb53a061cad528ce8e517d7b9ac41a0a396c39ad795a487fc14",
+      "model": "586799539888c69c35f84d083e559cc5f0ba887801d000e70e65f58d3f85c313",
+      "model_config": "1272ca8fb93cb42f48b039bb298688b2bd50a66f4faa0afcd9e538e9762f48aa",
+      "train_config": "e50ec42e40ba1925983e0e8db36558d3738bea64e1698045ff7cc08a130e1053",
+      "stats": "354710650bb91fc0e84a461987b3554b30bef9d5de47567480624484cdfbf3a2",
+      "cosmos_encoder": "520a3b878e72ee872faccb1211b5d622d194c0fdd177f2f6df097586b8fa98e5",
+      "cosmos_decoder": "3b7791e6c8f6f92caacddf48ebb314f044bcb8302d0c7f47501d384a7a80f738",
+      "processor_tokenizer": "a5d85b6dcc535e6b93115a9ef287e6132fdbf30270da6218194ba742261173c7",
+      "processor_config": "27225450ac9c6529872ee1924fcb0962ff5634834f817040f444118116f4e516"
+    },
+    "placement": "cpu",
+    "worker_props": {
+      "model_dir": "\\\\wsl.localhost\\Ubuntu\\home\\zhout\\project\\edge_infer\\models\\InternVLA-A1-3B-FT-Place_Markpen",
+      "processor_dir": "\\\\wsl.localhost\\Ubuntu\\home\\zhout\\project\\edge_infer\\models\\Qwen3-VL-2B-Instruct-processor",
+      "cosmos_dir": "\\\\wsl.localhost\\Ubuntu\\home\\zhout\\project\\edge_infer\\models\\Cosmos-Tokenizer-CI8x8",
+      "placement": "cpu",
+      "torch": "2.13.0+cu130",
+      "policy_dtype": "bfloat16",
+      "cosmos_dtype": "bfloat16",
+      "policy_device": "cpu",
+      "runtime_mode": "real_checkpoint_loaded",
+      "policy_load_s": 77.1901350999251,
+      "action_shape": [
+        1,
+        50,
+        32
+      ],
+      "image_shape": [
+        1,
+        2,
+        3,
+        224,
+        224
+      ],
+      "state_shape": [
+        1,
+        32
+      ],
+      "action_mode": "delta",
+      "action_units": "unverified",
+      "action_joint_order": "unverified",
+      "action_step_s": null,
+      "control_ready": false,
+      "external_load": null
+    },
+    "loaded_rss_bytes": 8997961728,
+    "loaded_process_tree": [
+      {
+        "pid": 13452,
+        "rss_bytes": 4386816,
+        "private_bytes": 790528
+      },
+      {
+        "pid": 56984,
+        "rss_bytes": 7462912,
+        "private_bytes": 1228800
+      },
+      {
+        "pid": 39924,
+        "rss_bytes": 8986112000,
+        "private_bytes": 11148115968
+      }
+    ],
+    "reserved_bytes": {
+      "host_ram": 17179869184
+    },
+    "memory_overhead_bytes": 8589934592,
+    "max_input_bytes": 8388608,
+    "max_action_bytes": 1048576,
+    "request_capacity": 1,
+    "stateful_session": "InternVLA policy worker-owned; complete action chunk per request",
+    "evidence": "B"
+  },
+  "abort_check": {
+    "stale_output": false,
+    "ledger": {
+      "capacities": {
+        "host_ram": 17179869184
+      },
+      "reserved": {
+        "host_ram": 0
+      },
+      "owners": [],
+      "quarantined": []
+    }
+  },
+  "status": "passed",
+  "ledger_after_shutdown": {
+    "capacities": {
+      "host_ram": 17179869184
+    },
+    "reserved": {
+      "host_ram": 0
+    },
+    "owners": [],
+    "quarantined": []
+  }
+}
@@ -0,0 +1,28 @@
+W0924 01:50:24.677000 39924 Lib\site-packages\torch\utils\cpp_extension.py:178] No CUDA runtime is found, using CUDA_HOME='C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.4'
+INFO 09-24 01:50:27 [patch.py:259] NVFP4 W4A4 weight_scale NaN-clamp: installed.
+INFO 09-24 01:50:27 [patch.py:510] inductor factorable-divisibility patch: installed.
+INFO 09-24 01:50:32 [pipeline_internvla_a1.py:164] Loading InternVLA-A1 weights from \\wsl.localhost\Ubuntu\home\zhout\project\edge_infer\models\InternVLA-A1-3B-FT-Place_Markpen
+internvla-worker ready port=62205 props={"model_dir": "\\\\wsl.localhost\\Ubuntu\\home\\zhout\\project\\edge_infer\\models\\InternVLA-A1-3B-FT-Place_Markpen", "processor_dir": "\\\\wsl.localhost\\Ubuntu\\home\\zhout\\project\\edge_infer\\models\\Qwen3-VL-2B-Instruct-processor", "cosmos_dir": "\\\\wsl.localhost\\Ubuntu\\home\\zhout\\project\\edge_infer\\models\\Cosmos-Tokenizer-CI8x8", "placement": "cpu", "torch": "2.13.0+cu130", "policy_dtype": "bfloat16", "cosmos_dtype": "bfloat16", "policy_device": "cpu", "runtime_mode": "real_checkpoint_loaded", "policy_load_s": 77.1901350999251, "action_shape": [1, 50, 32], "image_shape": [1, 2, 3, 224, 224], "state_shape": [1, 32], "action_mode": "delta", "action_units": "unverified", "action_joint_order": "unverified", "action_step_s": null, "control_ready": false, "external_load": null}
+127.0.0.1 - - [24/Sep/2026 01:51:49] "GET /health HTTP/1.1" 200 -
+127.0.0.1 - - [24/Sep/2026 01:51:49] "GET /props HTTP/1.1" 200 -
+127.0.0.1 - - [24/Sep/2026 01:51:54] "POST /v1/actions HTTP/1.1" 200 -
+127.0.0.1 - - [24/Sep/2026 01:51:59] "POST /v1/actions HTTP/1.1" 200 -
+127.0.0.1 - - [24/Sep/2026 01:52:04] "POST /v1/actions HTTP/1.1" 200 -
+127.0.0.1 - - [24/Sep/2026 01:52:09] "POST /v1/actions HTTP/1.1" 200 -
+127.0.0.1 - - [24/Sep/2026 01:52:14] "POST /v1/actions HTTP/1.1" 200 -
+127.0.0.1 - - [24/Sep/2026 01:52:19] "POST /v1/actions HTTP/1.1" 200 -
+127.0.0.1 - - [24/Sep/2026 01:52:24] "POST /v1/actions HTTP/1.1" 200 -
+127.0.0.1 - - [24/Sep/2026 01:52:29] "POST /v1/actions HTTP/1.1" 200 -
+127.0.0.1 - - [24/Sep/2026 01:52:33] "POST /v1/actions HTTP/1.1" 200 -
+127.0.0.1 - - [24/Sep/2026 01:52:38] "POST /v1/actions HTTP/1.1" 200 -
+127.0.0.1 - - [24/Sep/2026 01:52:43] "POST /v1/actions HTTP/1.1" 200 -
+127.0.0.1 - - [24/Sep/2026 01:52:48] "POST /v1/actions HTTP/1.1" 200 -
+127.0.0.1 - - [24/Sep/2026 01:52:53] "POST /v1/actions HTTP/1.1" 200 -
+127.0.0.1 - - [24/Sep/2026 01:52:58] "POST /v1/actions HTTP/1.1" 200 -
+127.0.0.1 - - [24/Sep/2026 01:53:03] "POST /v1/actions HTTP/1.1" 200 -
+127.0.0.1 - - [24/Sep/2026 01:53:08] "POST /v1/actions HTTP/1.1" 200 -
+127.0.0.1 - - [24/Sep/2026 01:53:12] "POST /v1/actions HTTP/1.1" 200 -
+127.0.0.1 - - [24/Sep/2026 01:53:17] "POST /v1/actions HTTP/1.1" 200 -
+127.0.0.1 - - [24/Sep/2026 01:53:22] "POST /v1/actions HTTP/1.1" 200 -
+127.0.0.1 - - [24/Sep/2026 01:53:27] "POST /v1/actions HTTP/1.1" 200 -
+127.0.0.1 - - [24/Sep/2026 01:53:32] "POST /v1/actions HTTP/1.1" 200 -
@@ -0,0 +1,256 @@
+{
+  "scope": "real InternVLA Place_Markpen checkpoint via bounded Omni whole-policy graph stage; synthetic patterned observations/noise; no robot-task quality claim",
+  "placement": "cpu",
+  "artifact_sha256": {
+    "python": "0b471133e110cfb53a061cad528ce8e517d7b9ac41a0a396c39ad795a487fc14",
+    "model": "586799539888c69c35f84d083e559cc5f0ba887801d000e70e65f58d3f85c313",
+    "model_config": "1272ca8fb93cb42f48b039bb298688b2bd50a66f4faa0afcd9e538e9762f48aa",
+    "train_config": "e50ec42e40ba1925983e0e8db36558d3738bea64e1698045ff7cc08a130e1053",
+    "stats": "354710650bb91fc0e84a461987b3554b30bef9d5de47567480624484cdfbf3a2",
+    "cosmos_encoder": "520a3b878e72ee872faccb1211b5d622d194c0fdd177f2f6df097586b8fa98e5",
+    "cosmos_decoder": "3b7791e6c8f6f92caacddf48ebb314f044bcb8302d0c7f47501d384a7a80f738",
+    "processor_tokenizer": "a5d85b6dcc535e6b93115a9ef287e6132fdbf30270da6218194ba742261173c7",
+    "processor_config": "27225450ac9c6529872ee1924fcb0962ff5634834f817040f444118116f4e516"
+  },
+  "budget": {
+    "capacities": {
+      "host_ram": 25769803776
+    },
+    "demands": {
+      "host_ram": 17179869184
+    }
+  },
+  "warmups": 0,
+  "repeats": 1,
+  "startup_s": 133.42465160007123,
+  "execution_plan_start": {
+    "backend": "external.internvla.policy.v1",
+    "stage_id": 0,
+    "worker_generation": "45be61d394b54deaa7b9f6ab67631b5b",
+    "worker_pid": 52696,
+    "worker_script_sha256": "55ffe4ce194eee7e330a4e5def8a2bfec0bfbb98b182885e094f71ea57e4ac9a",
+    "artifact_sha256": {
+      "python": "0b471133e110cfb53a061cad528ce8e517d7b9ac41a0a396c39ad795a487fc14",
+      "model": "586799539888c69c35f84d083e559cc5f0ba887801d000e70e65f58d3f85c313",
+      "model_config": "1272ca8fb93cb42f48b039bb298688b2bd50a66f4faa0afcd9e538e9762f48aa",
+      "train_config": "e50ec42e40ba1925983e0e8db36558d3738bea64e1698045ff7cc08a130e1053",
+      "stats": "354710650bb91fc0e84a461987b3554b30bef9d5de47567480624484cdfbf3a2",
+      "cosmos_encoder": "520a3b878e72ee872faccb1211b5d622d194c0fdd177f2f6df097586b8fa98e5",
+      "cosmos_decoder": "3b7791e6c8f6f92caacddf48ebb314f044bcb8302d0c7f47501d384a7a80f738",
+      "processor_tokenizer": "a5d85b6dcc535e6b93115a9ef287e6132fdbf30270da6218194ba742261173c7",
+      "processor_config": "27225450ac9c6529872ee1924fcb0962ff5634834f817040f444118116f4e516"
+    },
+    "placement": "cpu",
+    "worker_props": {
+      "model_dir": "\\\\wsl.localhost\\Ubuntu\\home\\zhout\\project\\edge_infer\\models\\InternVLA-A1-3B-FT-Place_Markpen",
+      "processor_dir": "\\\\wsl.localhost\\Ubuntu\\home\\zhout\\project\\edge_infer\\models\\Qwen3-VL-2B-Instruct-processor",
+      "cosmos_dir": "\\\\wsl.localhost\\Ubuntu\\home\\zhout\\project\\edge_infer\\models\\Cosmos-Tokenizer-CI8x8",
+      "placement": "cpu",
+      "torch": "2.13.0+cu130",
+      "policy_dtype": "bfloat16",
+      "cosmos_dtype": "bfloat16",
+      "policy_device": "cpu",
+      "runtime_mode": "real_checkpoint_loaded",
+      "policy_load_s": 75.62602760002483,
+      "action_shape": [
+        1,
+        50,
+        32
+      ],
+      "image_shape": [
+        1,
+        2,
+        3,
+        224,
+        224
+      ],
+      "state_shape": [
+        1,
+        32
+      ],
+      "action_mode": "delta",
+      "action_units": "unverified",
+      "action_joint_order": "unverified",
+      "action_step_s": null,
+      "control_ready": false,
+      "external_load": null
+    },
+    "loaded_rss_bytes": 9013301248,
+    "loaded_process_tree": [
+      {
+        "pid": 52696,
+        "rss_bytes": 4386816,
+        "private_bytes": 786432
+      },
+      {
+        "pid": 4048,
+        "rss_bytes": 7462912,
+        "private_bytes": 1232896
+      },
+      {
+        "pid": 64800,
+        "rss_bytes": 9001451520,
+        "private_bytes": 11164360704
+      }
+    ],
+    "reserved_bytes": {
+      "host_ram": 17179869184
+    },
+    "memory_overhead_bytes": 8589934592,
+    "max_input_bytes": 8388608,
+    "max_action_bytes": 1048576,
+    "request_capacity": 1,
+    "stateful_session": "InternVLA policy worker-owned; complete action chunk per request",
+    "evidence": "B"
+  },
+  "warmup_results": [],
+  "measured": [
+    {
+      "wall_s": 5.013468899996951,
+      "action_sha256": "82dd03dc31833fda60bff6620030315bad4034089b0c6c34c9de74e1c97e8d36",
+      "stage_event": {
+        "request_id": "internvla-measured-0",
+        "stage_id": 0,
+        "epoch": 1,
+        "seq": 1,
+        "kind": "action",
+        "worker_generation": "45be61d394b54deaa7b9f6ab67631b5b",
+        "buffers": [
+          {
+            "object_id": "actions",
+            "owner": "0",
+            "generation": "45be61d394b54deaa7b9f6ab67631b5b",
+            "dtype": "float32",
+            "shape": [
+              1,
+              50,
+              32
+            ],
+            "nbytes": 6400,
+            "memory_domain": "host-copy",
+            "layout": "C",
+            "ready": true
+          }
+        ],
+        "terminal": true,
+        "error": null
+      },
+      "action_metadata": {
+        "observation_timestamp_ns": 1790185682157368800,
+        "generation_timestamp_ns": 1790185687166017300,
+        "observation_age_s": 5.0086485,
+        "action_mode": "delta",
+        "action_shape": [
+          1,
+          50,
+          32
+        ],
+        "action_units": "unverified",
+        "joint_order": "unverified",
+        "action_step_s": null,
+        "control_ready": false
+      },
+      "metrics": {
+        "policy_wall_s": 4.99497799994424,
+        "worker_wall_s": 4.953044899972156
+      }
+    }
+  ],
+  "nearest_rank_p50_wall_s": 5.013468899996951,
+  "nearest_rank_p95_wall_s": 5.013468899996951,
+  "unique_action_hashes": [
+    "82dd03dc31833fda60bff6620030315bad4034089b0c6c34c9de74e1c97e8d36"
+  ],
+  "output_actions": "\\\\wsl.localhost\\Ubuntu\\home\\zhout\\project\\edge_infer\\vllm-omni-edge\\benchmarks\\edge_harness\\results\\e2e_expansion_20260923\\evidence\\internvla_omni_windows_cpu\\cpu_smoke_actions.npy",
+  "execution_plan_end": {
+    "backend": "external.internvla.policy.v1",
+    "stage_id": 0,
+    "worker_generation": "45be61d394b54deaa7b9f6ab67631b5b",
+    "worker_pid": 52696,
+    "worker_script_sha256": "55ffe4ce194eee7e330a4e5def8a2bfec0bfbb98b182885e094f71ea57e4ac9a",
+    "artifact_sha256": {
+      "python": "0b471133e110cfb53a061cad528ce8e517d7b9ac41a0a396c39ad795a487fc14",
+      "model": "586799539888c69c35f84d083e559cc5f0ba887801d000e70e65f58d3f85c313",
+      "model_config": "1272ca8fb93cb42f48b039bb298688b2bd50a66f4faa0afcd9e538e9762f48aa",
+      "train_config": "e50ec42e40ba1925983e0e8db36558d3738bea64e1698045ff7cc08a130e1053",
+      "stats": "354710650bb91fc0e84a461987b3554b30bef9d5de47567480624484cdfbf3a2",
+      "cosmos_encoder": "520a3b878e72ee872faccb1211b5d622d194c0fdd177f2f6df097586b8fa98e5",
+      "cosmos_decoder": "3b7791e6c8f6f92caacddf48ebb314f044bcb8302d0c7f47501d384a7a80f738",
+      "processor_tokenizer": "a5d85b6dcc535e6b93115a9ef287e6132fdbf30270da6218194ba742261173c7",
+      "processor_config": "27225450ac9c6529872ee1924fcb0962ff5634834f817040f444118116f4e516"
+    },
+    "placement": "cpu",
+    "worker_props": {
+      "model_dir": "\\\\wsl.localhost\\Ubuntu\\home\\zhout\\project\\edge_infer\\models\\InternVLA-A1-3B-FT-Place_Markpen",
+      "processor_dir": "\\\\wsl.localhost\\Ubuntu\\home\\zhout\\project\\edge_infer\\models\\Qwen3-VL-2B-Instruct-processor",
+      "cosmos_dir": "\\\\wsl.localhost\\Ubuntu\\home\\zhout\\project\\edge_infer\\models\\Cosmos-Tokenizer-CI8x8",
+      "placement": "cpu",
+      "torch": "2.13.0+cu130",
+      "policy_dtype": "bfloat16",
+      "cosmos_dtype": "bfloat16",
+      "policy_device": "cpu",
+      "runtime_mode": "real_checkpoint_loaded",
+      "policy_load_s": 75.62602760002483,
+      "action_shape": [
+        1,
+        50,
+        32
+      ],
+      "image_shape": [
+        1,
+        2,
+        3,
+        224,
+        224
+      ],
+      "state_shape": [
+        1,
+        32
+      ],
+      "action_mode": "delta",
+      "action_units": "unverified",
+      "action_joint_order": "unverified",
+      "action_step_s": null,
+      "control_ready": false,
+      "external_load": null
+    },
+    "loaded_rss_bytes": 9013301248,
+    "loaded_process_tree": [
+      {
+        "pid": 52696,
+        "rss_bytes": 4386816,
+        "private_bytes": 786432
+      },
+      {
+        "pid": 4048,
+        "rss_bytes": 7462912,
+        "private_bytes": 1232896
+      },
+      {
+        "pid": 64800,
+        "rss_bytes": 9001451520,
+        "private_bytes": 11164360704
+      }
+    ],
+    "reserved_bytes": {
+      "host_ram": 17179869184
+    },
+    "memory_overhead_bytes": 8589934592,
+    "max_input_bytes": 8388608,
+    "max_action_bytes": 1048576,
+    "request_capacity": 1,
+    "stateful_session": "InternVLA policy worker-owned; complete action chunk per request",
+    "evidence": "B"
+  },
+  "status": "passed",
+  "ledger_after_shutdown": {
+    "capacities": {
+      "host_ram": 25769803776
+    },
+    "reserved": {
+      "host_ram": 0
+    },
+    "owners": [],
+    "quarantined": []
+  }
+}
@@ -0,0 +1,208 @@
+INFO 09-24 01:45:09 [patch.py:259] NVFP4 W4A4 weight_scale NaN-clamp: installed.
+INFO 09-24 01:45:09 [patch.py:510] inductor factorable-divisibility patch: installed.
+INFO 09-24 01:45:49 [stage_init_utils.py:768] [stage_init] Set VLLM_WORKER_MULTIPROC_METHOD=spawn
+WARNING 09-24 01:45:49 [stage_init_utils.py:1887] [stage_init] Failed to load transfer config: [Errno 13] Permission denied: 'C:\\Users\\zhout'
+{
+  "scope": "real InternVLA Place_Markpen checkpoint via bounded Omni whole-policy graph stage; synthetic patterned observations/noise; no robot-task quality claim",
+  "placement": "cpu",
+  "artifact_sha256": {
+    "python": "0b471133e110cfb53a061cad528ce8e517d7b9ac41a0a396c39ad795a487fc14",
+    "model": "586799539888c69c35f84d083e559cc5f0ba887801d000e70e65f58d3f85c313",
+    "model_config": "1272ca8fb93cb42f48b039bb298688b2bd50a66f4faa0afcd9e538e9762f48aa",
+    "train_config": "e50ec42e40ba1925983e0e8db36558d3738bea64e1698045ff7cc08a130e1053",
+    "stats": "354710650bb91fc0e84a461987b3554b30bef9d5de47567480624484cdfbf3a2",
+    "cosmos_encoder": "520a3b878e72ee872faccb1211b5d622d194c0fdd177f2f6df097586b8fa98e5",
+    "cosmos_decoder": "3b7791e6c8f6f92caacddf48ebb314f044bcb8302d0c7f47501d384a7a80f738",
+    "processor_tokenizer": "a5d85b6dcc535e6b93115a9ef287e6132fdbf30270da6218194ba742261173c7",
+    "processor_config": "27225450ac9c6529872ee1924fcb0962ff5634834f817040f444118116f4e516"
+  },
+  "budget": {
+    "capacities": {
+      "host_ram": 25769803776
+    },
+    "demands": {
+      "host_ram": 17179869184
+    }
+  },
+  "warmups": 0,
+  "repeats": 1,
+  "startup_s": 133.42465160007123,
+  "execution_plan_start": {
+    "backend": "external.internvla.policy.v1",
+    "stage_id": 0,
+    "worker_generation": "45be61d394b54deaa7b9f6ab67631b5b",
+    "worker_pid": 52696,
+    "worker_script_sha256": "55ffe4ce194eee7e330a4e5def8a2bfec0bfbb98b182885e094f71ea57e4ac9a",
+    "artifact_sha256": {
+      "python": "0b471133e110cfb53a061cad528ce8e517d7b9ac41a0a396c39ad795a487fc14",
+      "model": "586799539888c69c35f84d083e559cc5f0ba887801d000e70e65f58d3f85c313",
+      "model_config": "1272ca8fb93cb42f48b039bb298688b2bd50a66f4faa0afcd9e538e9762f48aa",
+      "train_config": "e50ec42e40ba1925983e0e8db36558d3738bea64e1698045ff7cc08a130e1053",
+      "stats": "354710650bb91fc0e84a461987b3554b30bef9d5de47567480624484cdfbf3a2",
+      "cosmos_encoder": "520a3b878e72ee872faccb1211b5d622d194c0fdd177f2f6df097586b8fa98e5",
+      "cosmos_decoder": "3b7791e6c8f6f92caacddf48ebb314f044bcb8302d0c7f47501d384a7a80f738",
+      "processor_tokenizer": "a5d85b6dcc535e6b93115a9ef287e6132fdbf30270da6218194ba742261173c7",
+      "processor_config": "27225450ac9c6529872ee1924fcb0962ff5634834f817040f444118116f4e516"
+    },
+    "placement": "cpu",
+    "worker_props": {
+      "model_dir": "\\\\wsl.localhost\\Ubuntu\\home\\zhout\\project\\edge_infer\\models\\InternVLA-A1-3B-FT-Place_Markpen",
+      "processor_dir": "\\\\wsl.localhost\\Ubuntu\\home\\zhout\\project\\edge_infer\\models\\Qwen3-VL-2B-Instruct-processor",
+      "cosmos_dir": "\\\\wsl.localhost\\Ubuntu\\home\\zhout\\project\\edge_infer\\models\\Cosmos-Tokenizer-CI8x8",
+      "placement": "cpu",
+      "torch": "2.13.0+cu130",
+      "policy_dtype": "bfloat16",
+      "cosmos_dtype": "bfloat16",
+      "policy_device": "cpu",
+      "runtime_mode": "real_checkpoint_loaded",
+      "policy_load_s": 75.62602760002483,
+      "action_shape": [
+        1,
+        50,
+        32
+      ],
+      "image_shape": [
+        1,
+        2,
+        3,
+        224,
+        224
+      ],
+      "state_shape": [
+        1,
+        32
+      ],
+      "action_mode": "delta",
+      "action_units": "unverified",
+      "action_joint_order": "unverified",
+      "action_step_s": null,
+      "control_ready": false,
+      "external_load": null
+    },
+    "loaded_rss_bytes": 9013301248,
+    "loaded_process_tree": [
+      {
+        "pid": 52696,
+        "rss_bytes": 4386816,
+        "private_bytes": 786432
+      },
+      {
+        "pid": 4048,
+        "rss_bytes": 7462912,
+        "private_bytes": 1232896
+      },
+      {
+        "pid": 64800,
+        "rss_bytes": 9001451520,
+        "private_bytes": 11164360704
+      }
+    ],
+    "reserved_bytes": {
+      "host_ram": 17179869184
+    },
+    "memory_overhead_bytes": 8589934592,
+    "max_input_bytes": 8388608,
+    "max_action_bytes": 1048576,
+    "request_capacity": 1,
+    "stateful_session": "InternVLA policy worker-owned; complete action chunk per request",
+    "evidence": "B"
+  },
+  "warmup_results": [],
+  "nearest_rank_p50_wall_s": 5.013468899996951,
+  "nearest_rank_p95_wall_s": 5.013468899996951,
+  "unique_action_hashes": [
+    "82dd03dc31833fda60bff6620030315bad4034089b0c6c34c9de74e1c97e8d36"
+  ],
+  "output_actions": "\\\\wsl.localhost\\Ubuntu\\home\\zhout\\project\\edge_infer\\vllm-omni-edge\\benchmarks\\edge_harness\\results\\e2e_expansion_20260923\\evidence\\internvla_omni_windows_cpu\\cpu_smoke_actions.npy",
+  "execution_plan_end": {
+    "backend": "external.internvla.policy.v1",
+    "stage_id": 0,
+    "worker_generation": "45be61d394b54deaa7b9f6ab67631b5b",
+    "worker_pid": 52696,
+    "worker_script_sha256": "55ffe4ce194eee7e330a4e5def8a2bfec0bfbb98b182885e094f71ea57e4ac9a",
+    "artifact_sha256": {
+      "python": "0b471133e110cfb53a061cad528ce8e517d7b9ac41a0a396c39ad795a487fc14",
+      "model": "586799539888c69c35f84d083e559cc5f0ba887801d000e70e65f58d3f85c313",
+      "model_config": "1272ca8fb93cb42f48b039bb298688b2bd50a66f4faa0afcd9e538e9762f48aa",
+      "train_config": "e50ec42e40ba1925983e0e8db36558d3738bea64e1698045ff7cc08a130e1053",
+      "stats": "354710650bb91fc0e84a461987b3554b30bef9d5de47567480624484cdfbf3a2",
+      "cosmos_encoder": "520a3b878e72ee872faccb1211b5d622d194c0fdd177f2f6df097586b8fa98e5",
+      "cosmos_decoder": "3b7791e6c8f6f92caacddf48ebb314f044bcb8302d0c7f47501d384a7a80f738",
+      "processor_tokenizer": "a5d85b6dcc535e6b93115a9ef287e6132fdbf30270da6218194ba742261173c7",
+      "processor_config": "27225450ac9c6529872ee1924fcb0962ff5634834f817040f444118116f4e516"
+    },
+    "placement": "cpu",
+    "worker_props": {
+      "model_dir": "\\\\wsl.localhost\\Ubuntu\\home\\zhout\\project\\edge_infer\\models\\InternVLA-A1-3B-FT-Place_Markpen",
+      "processor_dir": "\\\\wsl.localhost\\Ubuntu\\home\\zhout\\project\\edge_infer\\models\\Qwen3-VL-2B-Instruct-processor",
+      "cosmos_dir": "\\\\wsl.localhost\\Ubuntu\\home\\zhout\\project\\edge_infer\\models\\Cosmos-Tokenizer-CI8x8",
+      "placement": "cpu",
+      "torch": "2.13.0+cu130",
+      "policy_dtype": "bfloat16",
+      "cosmos_dtype": "bfloat16",
+      "policy_device": "cpu",
+      "runtime_mode": "real_checkpoint_loaded",
+      "policy_load_s": 75.62602760002483,
+      "action_shape": [
+        1,
+        50,
+        32
+      ],
+      "image_shape": [
+        1,
+        2,
+        3,
+        224,
+        224
+      ],
+      "state_shape": [
+        1,
+        32
+      ],
+      "action_mode": "delta",
+      "action_units": "unverified",
+      "action_joint_order": "unverified",
+      "action_step_s": null,
+      "control_ready": false,
+      "external_load": null
+    },
+    "loaded_rss_bytes": 9013301248,
+    "loaded_process_tree": [
+      {
+        "pid": 52696,
+        "rss_bytes": 4386816,
+        "private_bytes": 786432
+      },
+      {
+        "pid": 4048,
+        "rss_bytes": 7462912,
+        "private_bytes": 1232896
+      },
+      {
+        "pid": 64800,
+        "rss_bytes": 9001451520,
+        "private_bytes": 11164360704
+      }
+    ],
+    "reserved_bytes": {
+      "host_ram": 17179869184
+    },
+    "memory_overhead_bytes": 8589934592,
+    "max_input_bytes": 8388608,
+    "max_action_bytes": 1048576,
+    "request_capacity": 1,
+    "stateful_session": "InternVLA policy worker-owned; complete action chunk per request",
+    "evidence": "B"
+  },
+  "status": "passed",
+  "ledger_after_shutdown": {
+    "capacities": {
+      "host_ram": 25769803776
+    },
+    "reserved": {
+      "host_ram": 0
+    },
+    "owners": [],
+    "quarantined": []
+  }
+}
@@ -0,0 +1,8 @@
+W0924 01:46:38.665000 64800 Lib\site-packages\torch\utils\cpp_extension.py:178] No CUDA runtime is found, using CUDA_HOME='C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.4'
+INFO 09-24 01:46:41 [patch.py:259] NVFP4 W4A4 weight_scale NaN-clamp: installed.
+INFO 09-24 01:46:41 [patch.py:510] inductor factorable-divisibility patch: installed.
+INFO 09-24 01:46:46 [pipeline_internvla_a1.py:164] Loading InternVLA-A1 weights from \\wsl.localhost\Ubuntu\home\zhout\project\edge_infer\models\InternVLA-A1-3B-FT-Place_Markpen
+internvla-worker ready port=61305 props={"model_dir": "\\\\wsl.localhost\\Ubuntu\\home\\zhout\\project\\edge_infer\\models\\InternVLA-A1-3B-FT-Place_Markpen", "processor_dir": "\\\\wsl.localhost\\Ubuntu\\home\\zhout\\project\\edge_infer\\models\\Qwen3-VL-2B-Instruct-processor", "cosmos_dir": "\\\\wsl.localhost\\Ubuntu\\home\\zhout\\project\\edge_infer\\models\\Cosmos-Tokenizer-CI8x8", "placement": "cpu", "torch": "2.13.0+cu130", "policy_dtype": "bfloat16", "cosmos_dtype": "bfloat16", "policy_device": "cpu", "runtime_mode": "real_checkpoint_loaded", "policy_load_s": 75.62602760002483, "action_shape": [1, 50, 32], "image_shape": [1, 2, 3, 224, 224], "state_shape": [1, 32], "action_mode": "delta", "action_units": "unverified", "action_joint_order": "unverified", "action_step_s": null, "control_ready": false, "external_load": null}
+127.0.0.1 - - [24/Sep/2026 01:48:01] "GET /health HTTP/1.1" 200 -
+127.0.0.1 - - [24/Sep/2026 01:48:01] "GET /props HTTP/1.1" 200 -
+127.0.0.1 - - [24/Sep/2026 01:48:07] "POST /v1/actions HTTP/1.1" 200 -
@@ -0,0 +1,67 @@
+{
+  "entrypoint": "AsyncOmni.generate",
+  "placement": "cpu",
+  "pipeline": "internvla_a1_external_whole",
+  "host_available_bytes_before": 23669141504,
+  "budget": {
+    "capacity_gib": 16,
+    "reserve_gib": 16
+  },
+  "startup_s": 134.60538950003684,
+  "request_wall_s": 4.97882199997548,
+  "outputs": [
+    {
+      "request_id": "public-internvla-1",
+      "action_shape": [
+        1,
+        50,
+        32
+      ],
+      "action_sha256": "82dd03dc31833fda60bff6620030315bad4034089b0c6c34c9de74e1c97e8d36",
+      "finite": true,
+      "metadata": {
+        "observation_timestamp_ns": 1790186560879215800,
+        "generation_timestamp_ns": 1790186565838592700,
+        "observation_age_s": 4.9593769,
+        "action_mode": "delta",
+        "action_shape": [
+          1,
+          50,
+          32
+        ],
+        "action_units": "unverified",
+        "joint_order": "unverified",
+        "action_step_s": null,
+        "control_ready": false
+      },
+      "stage_event": {
+        "request_id": "public-internvla-1-b5d14a73",
+        "stage_id": 0,
+        "epoch": 1,
+        "seq": 1,
+        "kind": "action",
+        "worker_generation": "5218d3d47e684d47af6238fd95a65f3d",
+        "buffers": [
+          {
+            "object_id": "actions",
+            "owner": "0",
+            "generation": "5218d3d47e684d47af6238fd95a65f3d",
+            "dtype": "float32",
+            "shape": [
+              1,
+              50,
+              32
+            ],
+            "nbytes": 6400,
+            "memory_domain": "host-copy",
+            "layout": "C",
+            "ready": true
+          }
+        ],
+        "terminal": true,
+        "error": null
+      }
+    }
+  ],
+  "status": "passed"
+}
@@ -0,0 +1,85 @@
+INFO 09-24 01:59:46 [patch.py:259] NVFP4 W4A4 weight_scale NaN-clamp: installed.
+INFO 09-24 01:59:46 [patch.py:510] inductor factorable-divisibility patch: installed.
+INFO 09-24 02:00:26 [omni_base.py:216] [AsyncOmni] Initializing with model \\wsl.localhost\Ubuntu\home\zhout\project\edge_infer\models\InternVLA-A1-3B-FT-Place_Markpen
+INFO 09-24 02:00:26 [async_omni_engine.py:172] [AsyncOmniEngine] Initializing with model \\wsl.localhost\Ubuntu\home\zhout\project\edge_infer\models\InternVLA-A1-3B-FT-Place_Markpen
+ERROR 09-24 02:00:26 [repo_utils.py:119] Error parsing config for \\wsl.localhost\Ubuntu\home\zhout\project\edge_infer\models\InternVLA-A1-3B-FT-Place_Markpen: Unrecognized model in \\wsl.localhost\Ubuntu\home\zhout\project\edge_infer\models\InternVLA-A1-3B-FT-Place_Markpen. Should have a `model_type` key in its config.json., retrying 1 of 2
+ERROR 09-24 02:00:28 [repo_utils.py:117] Error parsing config for \\wsl.localhost\Ubuntu\home\zhout\project\edge_infer\models\InternVLA-A1-3B-FT-Place_Markpen: Unrecognized model in \\wsl.localhost\Ubuntu\home\zhout\project\edge_infer\models\InternVLA-A1-3B-FT-Place_Markpen. Should have a `model_type` key in its config.json.
+INFO 09-24 02:00:28 [async_omni_engine.py:290] [AsyncOmniEngine] Launching Orchestrator thread with 1 stages
+INFO 09-24 02:00:28 [stage_init_utils.py:768] [stage_init] Set VLLM_WORKER_MULTIPROC_METHOD=spawn
+INFO 09-24 02:00:28 [initialization.py:356] Loaded OmniTransferConfig with 0 connector configurations
+INFO 09-24 02:02:40 [orchestrator.py:604] [Orchestrator] Starting event loop
+INFO 09-24 02:02:40 [async_omni_engine.py:323] [AsyncOmniEngine] Orchestrator ready with 1 stages
+INFO 09-24 02:02:40 [omni_base.py:235] [AsyncOmni] AsyncOmniEngine initialized in 134.60 seconds
+INFO 09-24 02:02:40 [omni_base.py:255] [AsyncOmni] Initialized with 1 stages for model \\wsl.localhost\Ubuntu\home\zhout\project\edge_infer\models\InternVLA-A1-3B-FT-Place_Markpen
+INFO 09-24 02:02:45 [omni_base.py:781] [AsyncOmni] Shutting down
+INFO 09-24 02:02:45 [async_omni_engine.py:1807] [AsyncOmniEngine] Shutting down Orchestrator
+INFO 09-24 02:02:45 [orchestrator.py:705] [Orchestrator] Received shutdown signal
+INFO 09-24 02:02:45 [orchestrator.py:3092] [Orchestrator] Shutting down all 1 client(s)
+INFO 09-24 02:02:46 [stage_pool.py:1354] [StagePool] Stage 0 replica 0 shut down
+{
+  "entrypoint": "AsyncOmni.generate",
+  "placement": "cpu",
+  "pipeline": "internvla_a1_external_whole",
+  "host_available_bytes_before": 23669141504,
+  "budget": {
+    "capacity_gib": 16,
+    "reserve_gib": 16
+  },
+  "startup_s": 134.60538950003684,
+  "request_wall_s": 4.97882199997548,
+  "outputs": [
+    {
+      "request_id": "public-internvla-1",
+      "action_shape": [
+        1,
+        50,
+        32
+      ],
+      "action_sha256": "82dd03dc31833fda60bff6620030315bad4034089b0c6c34c9de74e1c97e8d36",
+      "finite": true,
+      "metadata": {
+        "observation_timestamp_ns": 1790186560879215800,
+        "generation_timestamp_ns": 1790186565838592700,
+        "observation_age_s": 4.9593769,
+        "action_mode": "delta",
+        "action_shape": [
+          1,
+          50,
+          32
+        ],
+        "action_units": "unverified",
+        "joint_order": "unverified",
+        "action_step_s": null,
+        "control_ready": false
+      },
+      "stage_event": {
+        "request_id": "public-internvla-1-b5d14a73",
+        "stage_id": 0,
+        "epoch": 1,
+        "seq": 1,
+        "kind": "action",
+        "worker_generation": "5218d3d47e684d47af6238fd95a65f3d",
+        "buffers": [
+          {
+            "object_id": "actions",
+            "owner": "0",
+            "generation": "5218d3d47e684d47af6238fd95a65f3d",
+            "dtype": "float32",
+            "shape": [
+              1,
+              50,
+              32
+            ],
+            "nbytes": 6400,
+            "memory_domain": "host-copy",
+            "layout": "C",
+            "ready": true
+          }
+        ],
+        "terminal": true,
+        "error": null
+      }
+    }
+  ],
+  "status": "passed"
+}
@@ -0,0 +1,62 @@
+{
+  "entrypoint": "AsyncOmni.generate",
+  "placement": "cpu",
+  "pipeline": "internvla_a1_external_whole",
+  "startup_s": 134.27176439994946,
+  "request_wall_s": 4.887951400014572,
+  "outputs": [
+    {
+      "request_id": "public-internvla-1",
+      "action_shape": [
+        1,
+        50,
+        32
+      ],
+      "action_sha256": "82dd03dc31833fda60bff6620030315bad4034089b0c6c34c9de74e1c97e8d36",
+      "finite": true,
+      "metadata": {
+        "observation_timestamp_ns": 1790186329520927600,
+        "generation_timestamp_ns": 1790186334391448900,
+        "observation_age_s": 4.8705213,
+        "action_mode": "delta",
+        "action_shape": [
+          1,
+          50,
+          32
+        ],
+        "action_units": "unverified",
+        "joint_order": "unverified",
+        "action_step_s": null,
+        "control_ready": false
+      },
+      "stage_event": {
+        "request_id": "public-internvla-1-98389184",
+        "stage_id": 0,
+        "epoch": 1,
+        "seq": 1,
+        "kind": "action",
+        "worker_generation": "ef0afc8c9290443981d3cfb7ed0915d4",
+        "buffers": [
+          {
+            "object_id": "actions",
+            "owner": "0",
+            "generation": "ef0afc8c9290443981d3cfb7ed0915d4",
+            "dtype": "float32",
+            "shape": [
+              1,
+              50,
+              32
+            ],
+            "nbytes": 6400,
+            "memory_domain": "host-copy",
+            "layout": "C",
+            "ready": true
+          }
+        ],
+        "terminal": true,
+        "error": null
+      }
+    }
+  ],
+  "status": "passed"
+}
@@ -0,0 +1,80 @@
+INFO 09-24 01:55:57 [patch.py:259] NVFP4 W4A4 weight_scale NaN-clamp: installed.
+INFO 09-24 01:55:57 [patch.py:510] inductor factorable-divisibility patch: installed.
+INFO 09-24 01:56:37 [omni_base.py:216] [AsyncOmni] Initializing with model \\wsl.localhost\Ubuntu\home\zhout\project\edge_infer\models\InternVLA-A1-3B-FT-Place_Markpen
+INFO 09-24 01:56:37 [async_omni_engine.py:172] [AsyncOmniEngine] Initializing with model \\wsl.localhost\Ubuntu\home\zhout\project\edge_infer\models\InternVLA-A1-3B-FT-Place_Markpen
+ERROR 09-24 01:56:37 [repo_utils.py:119] Error parsing config for \\wsl.localhost\Ubuntu\home\zhout\project\edge_infer\models\InternVLA-A1-3B-FT-Place_Markpen: Unrecognized model in \\wsl.localhost\Ubuntu\home\zhout\project\edge_infer\models\InternVLA-A1-3B-FT-Place_Markpen. Should have a `model_type` key in its config.json., retrying 1 of 2
+ERROR 09-24 01:56:39 [repo_utils.py:117] Error parsing config for \\wsl.localhost\Ubuntu\home\zhout\project\edge_infer\models\InternVLA-A1-3B-FT-Place_Markpen: Unrecognized model in \\wsl.localhost\Ubuntu\home\zhout\project\edge_infer\models\InternVLA-A1-3B-FT-Place_Markpen. Should have a `model_type` key in its config.json.
+INFO 09-24 01:56:39 [async_omni_engine.py:290] [AsyncOmniEngine] Launching Orchestrator thread with 1 stages
+INFO 09-24 01:56:39 [stage_init_utils.py:768] [stage_init] Set VLLM_WORKER_MULTIPROC_METHOD=spawn
+INFO 09-24 01:56:39 [initialization.py:356] Loaded OmniTransferConfig with 0 connector configurations
+INFO 09-24 01:58:49 [orchestrator.py:604] [Orchestrator] Starting event loop
+INFO 09-24 01:58:49 [async_omni_engine.py:323] [AsyncOmniEngine] Orchestrator ready with 1 stages
+INFO 09-24 01:58:49 [omni_base.py:235] [AsyncOmni] AsyncOmniEngine initialized in 132.47 seconds
+INFO 09-24 01:58:49 [omni_base.py:255] [AsyncOmni] Initialized with 1 stages for model \\wsl.localhost\Ubuntu\home\zhout\project\edge_infer\models\InternVLA-A1-3B-FT-Place_Markpen
+INFO 09-24 01:58:54 [omni_base.py:781] [AsyncOmni] Shutting down
+INFO 09-24 01:58:54 [async_omni_engine.py:1807] [AsyncOmniEngine] Shutting down Orchestrator
+INFO 09-24 01:58:54 [orchestrator.py:705] [Orchestrator] Received shutdown signal
+INFO 09-24 01:58:54 [orchestrator.py:3092] [Orchestrator] Shutting down all 1 client(s)
+INFO 09-24 01:58:54 [stage_pool.py:1354] [StagePool] Stage 0 replica 0 shut down
+{
+  "entrypoint": "AsyncOmni.generate",
+  "placement": "cpu",
+  "pipeline": "internvla_a1_external_whole",
+  "startup_s": 134.27176439994946,
+  "request_wall_s": 4.887951400014572,
+  "outputs": [
+    {
+      "request_id": "public-internvla-1",
+      "action_shape": [
+        1,
+        50,
+        32
+      ],
+      "action_sha256": "82dd03dc31833fda60bff6620030315bad4034089b0c6c34c9de74e1c97e8d36",
+      "finite": true,
+      "metadata": {
+        "observation_timestamp_ns": 1790186329520927600,
+        "generation_timestamp_ns": 1790186334391448900,
+        "observation_age_s": 4.8705213,
+        "action_mode": "delta",
+        "action_shape": [
+          1,
+          50,
+          32
+        ],
+        "action_units": "unverified",
+        "joint_order": "unverified",
+        "action_step_s": null,
+        "control_ready": false
+      },
+      "stage_event": {
+        "request_id": "public-internvla-1-98389184",
+        "stage_id": 0,
+        "epoch": 1,
+        "seq": 1,
+        "kind": "action",
+        "worker_generation": "ef0afc8c9290443981d3cfb7ed0915d4",
+        "buffers": [
+          {
+            "object_id": "actions",
+            "owner": "0",
+            "generation": "ef0afc8c9290443981d3cfb7ed0915d4",
+            "dtype": "float32",
+            "shape": [
+              1,
+              50,
+              32
+            ],
+            "nbytes": 6400,
+            "memory_domain": "host-copy",
+            "layout": "C",
+            "ready": true
+          }
+        ],
+        "terminal": true,
+        "error": null
+      }
+    }
+  ],
+  "status": "passed"
+}
@@ -0,0 +1,8 @@
+W0924 01:57:28.663000 39924 Lib\site-packages\torch\utils\cpp_extension.py:178] No CUDA runtime is found, using CUDA_HOME='C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.4'
+INFO 09-24 01:57:31 [patch.py:259] NVFP4 W4A4 weight_scale NaN-clamp: installed.
+INFO 09-24 01:57:31 [patch.py:510] inductor factorable-divisibility patch: installed.
+INFO 09-24 01:57:36 [pipeline_internvla_a1.py:164] Loading InternVLA-A1 weights from \\wsl.localhost\Ubuntu\home\zhout\project\edge_infer\models\InternVLA-A1-3B-FT-Place_Markpen
+internvla-worker ready port=59514 props={"model_dir": "\\\\wsl.localhost\\Ubuntu\\home\\zhout\\project\\edge_infer\\models\\InternVLA-A1-3B-FT-Place_Markpen", "processor_dir": "\\\\wsl.localhost\\Ubuntu\\home\\zhout\\project\\edge_infer\\models\\Qwen3-VL-2B-Instruct-processor", "cosmos_dir": "\\\\wsl.localhost\\Ubuntu\\home\\zhout\\project\\edge_infer\\models\\Cosmos-Tokenizer-CI8x8", "placement": "cpu", "torch": "2.13.0+cu130", "policy_dtype": "bfloat16", "cosmos_dtype": "bfloat16", "policy_device": "cpu", "runtime_mode": "real_checkpoint_loaded", "policy_load_s": 74.91161379998084, "action_shape": [1, 50, 32], "image_shape": [1, 2, 3, 224, 224], "state_shape": [1, 32], "action_mode": "delta", "action_units": "unverified", "action_joint_order": "unverified", "action_step_s": null, "control_ready": false, "external_load": null}
+127.0.0.1 - - [24/Sep/2026 01:58:49] "GET /health HTTP/1.1" 200 -
+127.0.0.1 - - [24/Sep/2026 01:58:49] "GET /props HTTP/1.1" 200 -
+127.0.0.1 - - [24/Sep/2026 01:58:54] "POST /v1/actions HTTP/1.1" 200 -
@@ -0,0 +1,8 @@
+W0924 02:01:18.827000 6416 Lib\site-packages\torch\utils\cpp_extension.py:178] No CUDA runtime is found, using CUDA_HOME='C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.4'
+INFO 09-24 02:01:21 [patch.py:259] NVFP4 W4A4 weight_scale NaN-clamp: installed.
+INFO 09-24 02:01:21 [patch.py:510] inductor factorable-divisibility patch: installed.
+INFO 09-24 02:01:26 [pipeline_internvla_a1.py:164] Loading InternVLA-A1 weights from \\wsl.localhost\Ubuntu\home\zhout\project\edge_infer\models\InternVLA-A1-3B-FT-Place_Markpen
+internvla-worker ready port=51125 props={"model_dir": "\\\\wsl.localhost\\Ubuntu\\home\\zhout\\project\\edge_infer\\models\\InternVLA-A1-3B-FT-Place_Markpen", "processor_dir": "\\\\wsl.localhost\\Ubuntu\\home\\zhout\\project\\edge_infer\\models\\Qwen3-VL-2B-Instruct-processor", "cosmos_dir": "\\\\wsl.localhost\\Ubuntu\\home\\zhout\\project\\edge_infer\\models\\Cosmos-Tokenizer-CI8x8", "placement": "cpu", "torch": "2.13.0+cu130", "policy_dtype": "bfloat16", "cosmos_dtype": "bfloat16", "policy_device": "cpu", "runtime_mode": "real_checkpoint_loaded", "policy_load_s": 73.95817280001938, "action_shape": [1, 50, 32], "image_shape": [1, 2, 3, 224, 224], "state_shape": [1, 32], "action_mode": "delta", "action_units": "unverified", "action_joint_order": "unverified", "action_step_s": null, "control_ready": false, "external_load": null}
+127.0.0.1 - - [24/Sep/2026 02:02:40] "GET /health HTTP/1.1" 200 -
+127.0.0.1 - - [24/Sep/2026 02:02:40] "GET /props HTTP/1.1" 200 -
+127.0.0.1 - - [24/Sep/2026 02:02:45] "POST /v1/actions HTTP/1.1" 200 -
@@ -0,0 +1,37 @@
+{
+  "scope": "real InternVLA Place_Markpen checkpoint via bounded Omni whole-policy graph stage; synthetic patterned observations/noise; no robot-task quality claim",
+  "placement": "cpu",
+  "artifact_sha256": {
+    "python": "0b471133e110cfb53a061cad528ce8e517d7b9ac41a0a396c39ad795a487fc14",
+    "model": "586799539888c69c35f84d083e559cc5f0ba887801d000e70e65f58d3f85c313",
+    "model_config": "1272ca8fb93cb42f48b039bb298688b2bd50a66f4faa0afcd9e538e9762f48aa",
+    "train_config": "e50ec42e40ba1925983e0e8db36558d3738bea64e1698045ff7cc08a130e1053",
+    "stats": "354710650bb91fc0e84a461987b3554b30bef9d5de47567480624484cdfbf3a2",
+    "cosmos_encoder": "520a3b878e72ee872faccb1211b5d622d194c0fdd177f2f6df097586b8fa98e5",
+    "cosmos_decoder": "3b7791e6c8f6f92caacddf48ebb314f044bcb8302d0c7f47501d384a7a80f738",
+    "processor_tokenizer": "a5d85b6dcc535e6b93115a9ef287e6132fdbf30270da6218194ba742261173c7",
+    "processor_config": "27225450ac9c6529872ee1924fcb0962ff5634834f817040f444118116f4e516"
+  },
+  "budget": {
+    "capacities": {
+      "host_ram": 17179869184
+    },
+    "demands": {
+      "host_ram": 8589934592
+    }
+  },
+  "warmups": 0,
+  "repeats": 1,
+  "admission_refusal": "ResourceUnavailable: InternVLA weights, graph and state/workspace/headroom exceed reservation",
+  "status": "passed",
+  "ledger_after_shutdown": {
+    "capacities": {
+      "host_ram": 17179869184
+    },
+    "reserved": {
+      "host_ram": 0
+    },
+    "owners": [],
+    "quarantined": []
+  }
+}
@@ -0,0 +1,78 @@
+INFO 09-24 01:54:06 [patch.py:259] NVFP4 W4A4 weight_scale NaN-clamp: installed.
+INFO 09-24 01:54:06 [patch.py:510] inductor factorable-divisibility patch: installed.
+INFO 09-24 01:54:46 [stage_init_utils.py:768] [stage_init] Set VLLM_WORKER_MULTIPROC_METHOD=spawn
+WARNING 09-24 01:54:46 [stage_init_utils.py:1887] [stage_init] Failed to load transfer config: [Errno 13] Permission denied: 'C:\\Users\\zhout'
+ERROR 09-24 01:55:22 [stage_runtime.py:265] [StageRuntime] Stage initialization failed; shutting down 0 initialized client(s)
+
+ERROR 09-24 01:55:22 [stage_runtime.py:265] Traceback (most recent call last):
+
+ERROR 09-24 01:55:22 [stage_runtime.py:265]   File "\\wsl.localhost\Ubuntu\home\zhout\project\edge_infer\vllm-omni-edge\vllm_omni\engine\stage_runtime.py", line 252, in initialize
+
+ERROR 09-24 01:55:22 [stage_runtime.py:265]     initialized_clients = self._initialize_stage_replicas(stage_plans, self._stage_init_timeout)
+
+ERROR 09-24 01:55:22 [stage_runtime.py:265]                           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+
+ERROR 09-24 01:55:22 [stage_runtime.py:265]   File "\\wsl.localhost\Ubuntu\home\zhout\project\edge_infer\vllm-omni-edge\vllm_omni\engine\stage_runtime.py", line 589, in _initialize_stage_replicas
+
+ERROR 09-24 01:55:22 [stage_runtime.py:265]     raise primary_exc
+
+ERROR 09-24 01:55:22 [stage_runtime.py:265]   File "\\wsl.localhost\Ubuntu\home\zhout\project\edge_infer\vllm-omni-edge\vllm_omni\engine\stage_runtime.py", line 550, in _init_group
+
+ERROR 09-24 01:55:22 [stage_runtime.py:265]     client = self._initialize_replica(
+
+ERROR 09-24 01:55:22 [stage_runtime.py:265]              ^^^^^^^^^^^^^^^^^^^^^^^^^
+
+ERROR 09-24 01:55:22 [stage_runtime.py:265]   File "\\wsl.localhost\Ubuntu\home\zhout\project\edge_infer\vllm-omni-edge\vllm_omni\engine\stage_runtime.py", line 817, in _initialize_replica
+
+ERROR 09-24 01:55:22 [stage_runtime.py:265]     return create_graph_client(
+
+ERROR 09-24 01:55:22 [stage_runtime.py:265]            ^^^^^^^^^^^^^^^^^^^^
+
+ERROR 09-24 01:55:22 [stage_runtime.py:265]   File "\\wsl.localhost\Ubuntu\home\zhout\project\edge_infer\vllm-omni-edge\vllm_omni\engine\backends\__init__.py", line 31, in create_graph_client
+
+ERROR 09-24 01:55:22 [stage_runtime.py:265]     return InternVLAStageClient(metadata, backend, ledger, reservation)
+
+ERROR 09-24 01:55:22 [stage_runtime.py:265]            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+
+ERROR 09-24 01:55:22 [stage_runtime.py:265]   File "\\wsl.localhost\Ubuntu\home\zhout\project\edge_infer\vllm-omni-edge\vllm_omni\engine\backends\internvla.py", line 111, in __init__
+
+ERROR 09-24 01:55:22 [stage_runtime.py:265]     raise ResourceUnavailable("InternVLA weights, graph and state/workspace/headroom exceed reservation")
+
+ERROR 09-24 01:55:22 [stage_runtime.py:265] vllm_omni.engine.resource_ledger.ResourceUnavailable: InternVLA weights, graph and state/workspace/headroom exceed reservation
+{
+  "scope": "real InternVLA Place_Markpen checkpoint via bounded Omni whole-policy graph stage; synthetic patterned observations/noise; no robot-task quality claim",
+  "placement": "cpu",
+  "artifact_sha256": {
+    "python": "0b471133e110cfb53a061cad528ce8e517d7b9ac41a0a396c39ad795a487fc14",
+    "model": "586799539888c69c35f84d083e559cc5f0ba887801d000e70e65f58d3f85c313",
+    "model_config": "1272ca8fb93cb42f48b039bb298688b2bd50a66f4faa0afcd9e538e9762f48aa",
+    "train_config": "e50ec42e40ba1925983e0e8db36558d3738bea64e1698045ff7cc08a130e1053",
+    "stats": "354710650bb91fc0e84a461987b3554b30bef9d5de47567480624484cdfbf3a2",
+    "cosmos_encoder": "520a3b878e72ee872faccb1211b5d622d194c0fdd177f2f6df097586b8fa98e5",
+    "cosmos_decoder": "3b7791e6c8f6f92caacddf48ebb314f044bcb8302d0c7f47501d384a7a80f738",
+    "processor_tokenizer": "a5d85b6dcc535e6b93115a9ef287e6132fdbf30270da6218194ba742261173c7",
+    "processor_config": "27225450ac9c6529872ee1924fcb0962ff5634834f817040f444118116f4e516"
+  },
+  "budget": {
+    "capacities": {
+      "host_ram": 17179869184
+    },
+    "demands": {
+      "host_ram": 8589934592
+    }
+  },
+  "warmups": 0,
+  "repeats": 1,
+  "admission_refusal": "ResourceUnavailable: InternVLA weights, graph and state/workspace/headroom exceed reservation",
+  "status": "passed",
+  "ledger_after_shutdown": {
+    "capacities": {
+      "host_ram": 17179869184
+    },
+    "reserved": {
+      "host_ram": 0
+    },
+    "owners": [],
+    "quarantined": []
+  }
+}
@@ -2,7 +2,7 @@

 All 60 pairings have a disposition. The timing tables below remain the 2026-09-22 measurements. The configuration matrix and affected per-cell reasons incorporate the [2026-09-23 scoped recovery runs](../../../e2e_recovery_20260923/README.md) and [WSL CPU/native Windows expansion](../../../e2e_expansion_20260923/README.md). The expansion adds separate 20-request serial MiniCPM-o text-to-speech profiles on WSL CPU and native Windows GPU, a Spark output-head component profile on the HX370 AMD NPU, complete Spark text-request profiles through standalone llama.cpp and a bounded Omni whole-session stage on native Windows CPU and Radeon 890M, native Windows CPU Qwen3-TTS text-to-WAV profiles through both standalone PyTorch and a bounded Omni whole-session stage, and a hybrid CPU+Radeon 890M Qwen3-TTS profile likewise repeated through a bounded Omni whole-session stage. A pinned GGUF MiniCPM-o route adds combined image+audio to text+speech runs on WSL CPU, native Windows CPU and native Windows Radeon 890M+CPU. All three now pass a bounded complete request through an Omni graph stage; the WSL CPU path additionally passed one warmup plus 20 serial measured requests. These are scoped synthetic cases, not full release qualification.

-The WSL CPU and CPU+Radeon InternVLA cells now also have [bounded Omni complete-policy profiles](../../../e2e_expansion_20260923/evidence/internvla_omni_policy/README.md): one warmup and 20 serial patterned synthetic requests per placement, plus admission refusal, cancellation and a public hybrid `AsyncOmni.generate` request. These advance only the synthetic action-policy path. Real observations/reference actions, physical action units/order/step time, task quality, loading peak and sustained power/thermal behavior remain unverified. The earlier [Cosmos component and direct hybrid results](../../../e2e_expansion_20260923/evidence/internvla_cosmos_radeon890m/README.md) remain separate experiments.
+The WSL CPU and CPU+Radeon InternVLA cells also have [bounded Omni complete-policy profiles](../../../e2e_expansion_20260923/evidence/internvla_omni_policy/README.md): one warmup and 20 serial patterned synthetic requests per placement, plus admission refusal, cancellation and a public hybrid `AsyncOmni.generate` request. A separate [native Windows CPU Omni profile](../../../e2e_expansion_20260923/evidence/internvla_omni_windows_cpu/README.md) adds 20 serial patterned requests, refusal, cancellation and a public action request. These advance only the synthetic action-policy path. Real observations/reference actions, physical action units/order/step time, task quality, loading peak and sustained power/thermal behavior remain unverified. The earlier [Cosmos component and direct hybrid results](../../../e2e_expansion_20260923/evidence/internvla_cosmos_radeon890m/README.md) remain separate experiments.

 ## Configuration × model

@@ -17,7 +17,7 @@ NOT E2E means no complete device/model workload was verified. “Scoped E2E” i
 | Device | Spark-X2.5 | Qwen3-TTS 0.6B CustomVoice | Qwen3.8-27B | MiniCPM-o 4.5 | InternVLA-A1 |
 |---|---|---|---|---|---|
 | PC HX370: CPU / WSL | profile completed | scoped E2E: Omni CPU text → complete WAV | NOT E2E | scoped E2E: Omni GGUF combined image+audio → text+speech | synthetic policy pass: Omni CPU |
-| PC HX370: CPU / native Windows | scoped E2E: Omni Spark text | scoped E2E: Omni CPU text → complete WAV | NOT E2E | scoped E2E: Omni GGUF image+audio → text+speech | synthetic policy pass |
+| PC HX370: CPU / native Windows | scoped E2E: Omni Spark text | scoped E2E: Omni CPU text → complete WAV | NOT E2E | scoped E2E: Omni GGUF image+audio → text+speech | synthetic policy pass: Omni CPU |
 | PC HX370 + RTX 5090 Laptop / WSL | profile completed | profile completed | scoped E2E: text + image | scoped E2E: text/image/audio → speech | synthetic policy pass |
 | PC HX370 + RTX 5090 Laptop / Windows | profile completed | profile completed | NOT E2E | scoped E2E: text/image/audio → speech | synthetic policy pass |
 | PC HX370 + Radeon 890M / Windows worker | scoped E2E: Omni Spark text | scoped E2E: Omni Radeon talker/codec + CPU predictor | NOT E2E | scoped E2E: Omni Radeon+CPU image+audio → text+speech | synthetic policy pass: Omni CPU policy + Radeon Cosmos |
@@ -42,7 +42,7 @@ NOT E2E means no complete device/model workload was verified. “Scoped E2E” i
 | PC HX370: CPU / native Windows | Qwen3-TTS 0.6B CustomVoice | SCOPED_E2E_OMNI_COMPLETE_REQUEST | The pinned 0.6B CustomVoice BF16/SDPA checkpoint ran wholly on CPU through an isolated Qwen wrapper worker owned by one bounded Omni StageRuntime stage. Two named 24 kHz WAVs matched the earlier standalone PCM SHA256 values exactly. One warmup + 20 serial complete Omni requests for 4.56 s of audio had 20/20 identical PCM and nearest-rank wall p50/p95 15.32/15.60 s (RTF 3.36/3.42); independent Whisper tiny.en ASR gave WER 0.20 on the first Omni WAV, an intelligibility proxy only. A public AsyncOmni audio request passed. The stage verified pinned weights/interpreter and CPU placement, reserved 10 GiB host RAM, rejected an insufficient 4 GiB demand before worker launch, and after in-flight cancellation emitted no stale output and returned the ledger to zero. Sampled process-tree private memory peaked at 5.39 GB, not a loading peak. Native vLLM CPU attention operators remain absent; playable streaming, post-cancel restart, broader speech quality, loading peak and sustained power/thermal behavior remain unverified. | Validate playable streaming and restart/recovery, broader voice quality, peak memory, concurrency and sustained behavior; [Omni raw evidence](../../../e2e_expansion_20260923/evidence/qwen_tts_omni_cpu/README.md), [standalone baseline](../../../e2e_expansion_20260923/evidence/qwen_tts_native_cpu/README.md). |
 | PC HX370: CPU / native Windows | Qwen3.8-27B | NOT_E2E_PROFILEABLE | The installed native Windows vLLM extension lacks cpu_attn_reshape_and_cache and cpu_attn_get_scheduler_metadata. The separate ORT CPU graph worker is not a whole-model text/TTS adapter; a qualified CPU build/binding is missing. | Supply and validate the missing matching artifact/backend, then run the complete specified workload. |
 | PC HX370: CPU / native Windows | MiniCPM-o 4.5 | SCOPED_E2E_OMNI_COMPLETE_REQUEST | The pinned Q4_K_M/F16 GGUF set ran one combined synthetic JPEG+spoken-question → text+speech request through a bounded Omni whole-session graph stage on native Windows CPU. It answered “The square in the image is red.” and produced 2.24 s of 24 kHz speech in 24.34 s complete-request wall time (one cold sample, zero warmups); sampled worker RSS peaked at 13.11 GB. The stage verified ten weight hashes and the CLI hash, observed CPU placement, reserved 24 GiB shared RAM, and emitted a terminal audio event. An 8 GiB reservation was refused before worker launch; in-flight cancellation left no stale output and released the ledger. The prior standalone prompted run took 24.66 s and independently transcribed output with WER 0. This is a whole-request C++ worker under Omni control, not native vLLM execution or playable streaming; one synthetic quality case is not general qualification. | Validate real image/audio suites, playable streaming, post-cancel restart, loading peak, repeated latency and sustained thermal behavior. [Omni raw evidence](../../../e2e_expansion_20260923/evidence/minicpmo_omni_cpp/README.md), [standalone prompted evidence](../../../e2e_expansion_20260923/evidence/minicpmo_cpp_windows_cpu_prompted/README.md), [failed first run](../../../e2e_expansion_20260923/evidence/minicpmo_cpp_windows_cpu_attempt1/README.md). |
-| PC HX370: CPU / native Windows | InternVLA-A1 | SYNTHETIC_POLICY_PASS | The Place_Markpen checkpoint strict-loaded from a UNC path and the direct CPU policy produced finite `[1, 50, 32]` actions on synthetic zero observations/noise. Policy/output CPU placement was checked; one forward took 3.36 s. Relative L2 difference from WSL CPU was 1.08% (no task tolerance). This direct policy route does not require native vLLM CPU attention operators. | Validate real observations/reference actions, units/timestamps, memory and repeated latency before task qualification; [native evidence](../../../e2e_expansion_20260923/README.md). |
+| PC HX370: CPU / native Windows | InternVLA-A1 | SYNTHETIC_POLICY_PASS_OMNI; NOT_TASK_E2E | The real Place_Markpen checkpoint now runs through an admitted native Windows Omni StageRuntime whole-policy stage with the policy and output on CPU. One warmup plus 20 serial patterned synthetic camera/state/noise requests returned identical finite `[1,50,32]` action chunks; nearest-rank complete-request wall p50/p95 was 4.878/4.902 s, excluding 135.60 s startup. A separate public `AsyncOmni.generate` request returned a terminal action event with the same hash. The stage verified checkpoint/interpreter hashes, used a 16 GiB host-RAM capacity and demand within observed Windows availability, refused 8 GiB before worker launch, and after in-flight abort emitted no stale output and released the ledger. Compared with the WSL CPU Omni output for this fixture, action relative L2 difference was 2.1795% and cosine 0.999764, without a task tolerance. The earlier one-forward zero-input policy result is a different fixture. | Validate real observations/reference actions, physical units/order/step time and deadlines, loading peak, post-cancel restart, concurrency and sustained power/thermal behavior; [native Omni evidence](../../../e2e_expansion_20260923/evidence/internvla_omni_windows_cpu/README.md), [earlier direct policy evidence](../../../e2e_expansion_20260923/README.md). |
 | PC HX370 + RTX 5090 Laptop / WSL | Spark-X2.5 | PROFILE_COMPLETED | not fully qualified: reference-quality and remaining reliability/performance gates must be reviewed | Broaden only with a named workload/device/precision qualification. |
 | PC HX370 + RTX 5090 Laptop / WSL | Qwen3-TTS 0.6B CustomVoice | PROFILE_COMPLETED | not fully qualified: reference-quality and remaining reliability/performance gates must be reviewed | Resolve and measure playback stalls; run quality, interruption and thermal tests. |
 | PC HX370 + RTX 5090 Laptop / WSL | Qwen3.8-27B | SCOPED_E2E | A new one-stage Omni binding ran the real NVFP4 checkpoint: one 16-token text request and one synthetic red-image request (answer `Red`) at an explicit 512-token context with host offload. This is not the full multimodal, long-context or quality workload. | Qualify image/video suites, longer contexts, latency distributions and sustained power; [recovery evidence](../../../e2e_recovery_20260923/README.md). |
@@ -57,6 +57,14 @@ the corrected process-tree retirement and reruns ended at zero. Real
 observation/reference-action quality, physical units/order/step time,
 post-cancel restart, loading peak and sustained benefit remain open.

+The [native Windows CPU InternVLA Omni stage](../../benchmarks/edge_harness/results/e2e_expansion_20260923/evidence/internvla_omni_windows_cpu/README.md)
+also ran one warmup plus 20 serial synthetic complete-policy requests under
+a 16 GiB shared-RAM ceiling: p50/p95 4.878/4.902 s, with 20/20 identical
+finite actions. The public action entrypoint, 8 GiB pre-load refusal and
+in-flight abort passed. Native-vs-WSL CPU action relative L2 difference on
+the same fixture was 2.1795%, without a task tolerance. This deepens the
+native Windows CPU cell, not any robot-task or mobile qualification.
+
 The WSL RTX MiniCPM-o image-to-text+speech path now passes for one 448×448
 synthetic red-square image under an explicit 0.61 thinker GPU budget. The
 previous 0.58 budget failed the 2,048-token KV admission gate; the successful
@@ -211,7 +219,7 @@ capability and lifecycle implementation; a state-handle type alone is insufficie
 | M2 | Qwen3-TTS | Resolve desktop playback and shutdown findings and CPU compatibility; then mobile talker/predictor plus GPU vocoder | Complete PCM and tail, history/chunk correctness, interruption/recovery, quality checks, no post-startup underruns and RTF below 1 in the declared workload |
 | M3 | AMD stages | Connect compatible real encoders or other useful coarse stages through current Omni workers; start from the existing 890M vision evidence | Actual node/device placement, numerical and task quality, complete downstream output, shared-memory and handoff costs; reject unhelpful splits |
 | M4a | MiniCPM-o 4.5 | Three-stage desktop text+WAV produced coherent text and nonzero WAV on RTX 5090 Laptop under WSL and native Windows, and WSL CPU, using separate constrained plans. One synthetic red-square image and one synthetic tone each passed through the three-stage WSL RTX, native Windows RTX and WSL CPU paths to text and nonzero WAV. RTX text/token IDs matched across OS environments; CPU wording/token IDs differed while remaining plausible. The WSL RTX image plan needed a thinker GPU budget increase from 0.58 to 0.61 for KV admission. WSL multimodal shutdown cleanup remains open, including forced stage-0 termination on CPU. The CPU plan needed vLLM 0.28/0.29 processor compatibility and used swap; 20 serial **text** CPU requests measured p50 21.86 s/p95 22.97 s. Native Windows needed longer startup limits; 20 serial text requests measured p50 18.38 s/p95 18.73 s with host RAM/pagefile pressure. Add video encoder and persistent-state reference before mobile artifacts | Speech intelligibility/alignment, loading-peak admission, concurrency, broad image/audio-understanding suites, video, shutdown lifecycle, then combined streaming and interruption |
-| M4b | InternVLA-A1 | Strict-load synthetic policy forwards pass on WSL and native Windows CPU/RTX. A bounded WSL Omni whole-policy stage also passes on CPU and CPU+Radeon Cosmos with explicit shared-RAM admission, terminal action metadata and in-flight cancellation; the public hybrid entrypoint passes. This is not robot-task qualification. | Reference action agreement on real observations, physical units/order/step time, observation age and deadlines; post-cancel restart, loading peak, concurrency, power and thermal behavior. |
+| M4b | InternVLA-A1 | Strict-load synthetic policy forwards pass on WSL and native Windows CPU/RTX. Bounded Omni whole-policy stages pass on WSL CPU, WSL CPU+Radeon Cosmos and native Windows CPU with explicit shared-RAM admission, terminal action metadata and in-flight cancellation; public hybrid and native CPU entrypoints pass. This is not robot-task qualification. | Reference action agreement on real observations, physical units/order/step time, observation age and deadlines; post-cancel restart, loading peak, concurrency, power and thermal behavior. |
 | Independent | Qwen3.8-27B | Public Omni pipeline binding; text, then image/video and longer contexts; same-model CPU/iGPU artifact and explicit offload evaluation | Modality-specific quality, actual loading/runtime memory and measured performance; each precision/context receives its own qualification |

 The initial M0 desktop text acceptance remains scoped to its tested checkpoint,