[Dev] Remove mandatory scheduler reservations for GPU validation

.agents/skills/vllm-rlt-review-pr/SKILL.md

@@ -101,9 +101,11 @@ changes need relevant regression checks, not a speedup.
 
 Use targeted CPU tests or a small reproducer to resolve a concrete uncertainty.
 Check test dependencies and skips; a green suite with GPU cases skipped does not
-validate kernels or real serving. GPU validation requires the user's applicable
-experiment scope and verified scheduler reservation; never use an unreserved
-device or silently expand a benchmark's exhausted run budget. If unavailable,
+validate kernels or real serving. Run GPU validation within the user's authorized
+experiment scope and budget. A scheduler reservation is not a prerequisite; an
+explicit GPU testing request authorizes selecting an available device without
+another confirmation solely because a scheduler is unavailable. Do not silently
+expand a benchmark's exhausted run budget. If GPU execution is unavailable,
 finish the code/evidence review and identify the remaining validation precisely.
 
 For each finding, give priority, a short defect title, the smallest relevant

docs/async-speculative.md

@@ -1196,7 +1196,7 @@ Token IDs require exact agreement; KV tolerances are FP32 `atol=rtol=4e-5` and
 BF16 `atol=0.08, rtol=0.04`. These are targeted correctness checks, not general
 model-quality evidence.
 
-For a current rerun, use the commands above on a scheduler-reserved CUDA device.
+For a current rerun, use the commands above on an explicitly selected CUDA device.
 Compare native async, synchronous speculation and async speculation with the
 same BF16 model, prompts, token budget, backend, capacity and GPU. Warm up each
 mode, alternate measurement order and retain five unprofiled observations per

docs/runtime-history.md

@@ -243,7 +243,7 @@ BF16 `atol=0.08, rtol=0.04`. These are targeted correctness checks, not general
 model-quality evidence.
 
 For a current rerun, use the [validation commands](async-speculative.md#validation)
-on a scheduler-reserved CUDA device.
+on an explicitly selected CUDA device.
 Compare native async, synchronous speculation and async speculation with the
 same BF16 model, prompts, token budget, backend, capacity and GPU. Warm up each
 mode, alternate measurement order and retain five unprofiled observations per

docs/serving.md

@@ -7,11 +7,11 @@ to `127.0.0.1:8000`; this benchmark frontend has no authentication or TLS.
 
 Prepare the Ouro checkpoint and tokenizer separately. The official model ID
 pins revision `574fa66cb8bf5abdc979642d01cf2b79b16bfab1`; local paths are supported.
-Use a GPU scheduler reservation on shared hosts:
+Select an available GPU explicitly; a scheduler is optional:
 
 ```bash
-gpu status
-gpu run --gpu-ids <available-id> --timeout 30m --note "Ouro HTTP serving" -- \
+nvidia-smi
+CUDA_VISIBLE_DEVICES=<available-id> \
   python -m vllm_rlt.entrypoints.serve \
   --model /path/to/prepared/ouro --tokenizer /path/to/prepared/ouro \
   --device cuda --dtype bfloat16 --attention-backend triton \

pyproject.toml

@@ -43,7 +43,7 @@ license-files = ["LICENSE", "NOTICE"]
 [tool.pytest.ini_options]
 testpaths = ["tests"]
 pythonpath = ["."]
-markers = ["gpu: requires a scheduler-reserved CUDA device"]
+markers = ["gpu: requires an explicitly selected CUDA device and --run-gpu"]
 
 [tool.ruff]
 line-length = 100

tests/conftest.py

@@ -6,17 +6,17 @@
 
 def pytest_addoption(parser):
     parser.addoption(
-        "--run-gpu", action="store_true", help="Run CUDA tests inside a GPU reservation"
+        "--run-gpu", action="store_true", help="Run CUDA tests on explicitly selected devices"
     )
 
 
 def pytest_collection_modifyitems(config, items):
     enabled = config.getoption("--run-gpu")
     if enabled and not os.environ.get("CUDA_VISIBLE_DEVICES"):
-        raise pytest.UsageError("--run-gpu requires scheduler-provided CUDA_VISIBLE_DEVICES")
+        raise pytest.UsageError("--run-gpu requires explicit CUDA_VISIBLE_DEVICES")
     for item in items:
         if "gpu" in item.keywords and not enabled:
-            item.add_marker(pytest.mark.skip(reason="use --run-gpu inside a GPU reservation"))
+            item.add_marker(pytest.mark.skip(reason="set CUDA_VISIBLE_DEVICES and use --run-gpu"))
 
 
 @pytest.fixture(autouse=True, scope="session")