Scheduled Commit

vllm_0001000.log

@@ -25,3 +25,113 @@
 (EngineCore_DP0 pid=2004825) 
 (EngineCore_DP0 pid=2004825) 
 (EngineCore_DP0 pid=2004825) 
+(EngineCore_DP0 pid=2004825) 
+(EngineCore_DP0 pid=2004825) 
+(EngineCore_DP0 pid=2004825) 
+(EngineCore_DP0 pid=2004825) 
+(EngineCore_DP0 pid=2004825) 
+(EngineCore_DP0 pid=2004825) INFO 02-02 06:56:11 [gpu_model_runner.py:3659] Model loading took 19.4031 GiB memory and 49.236307 seconds
+(EngineCore_DP0 pid=2004825) INFO 02-02 06:56:23 [backends.py:643] Using cache directory: /home/mingzhed/.cache/vllm/torch_compile_cache/fc80a2ef9e/rank_0_0/backbone for vLLM's torch.compile
+(EngineCore_DP0 pid=2004825) INFO 02-02 06:56:23 [backends.py:703] Dynamo bytecode transform time: 11.90 s
+(EngineCore_DP0 pid=2004825) INFO 02-02 06:56:33 [backends.py:226] Directly load the compiled graph(s) for compile range (1, 8192) from the cache, took 1.532 s
+(EngineCore_DP0 pid=2004825) INFO 02-02 06:56:33 [monitor.py:34] torch.compile takes 13.43 s in total
+(EngineCore_DP0 pid=2004825) INFO 02-02 06:56:34 [gpu_worker.py:375] Available KV cache memory: 8.35 GiB
+(EngineCore_DP0 pid=2004825) ERROR 02-02 06:56:35 [core.py:866] EngineCore failed to start.
+(EngineCore_DP0 pid=2004825) ERROR 02-02 06:56:35 [core.py:866] Traceback (most recent call last):
+(EngineCore_DP0 pid=2004825) ERROR 02-02 06:56:35 [core.py:866]   File "/home/mingzhed/Projects/lm-evaluation-harness/.venv/lib/python3.12/site-packages/vllm/v1/engine/core.py", line 857, in run_engine_core
+(EngineCore_DP0 pid=2004825) ERROR 02-02 06:56:35 [core.py:866]     engine_core = EngineCoreProc(*args, **kwargs)
+(EngineCore_DP0 pid=2004825) ERROR 02-02 06:56:35 [core.py:866]                   ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+(EngineCore_DP0 pid=2004825) ERROR 02-02 06:56:35 [core.py:866]   File "/home/mingzhed/Projects/lm-evaluation-harness/.venv/lib/python3.12/site-packages/vllm/v1/engine/core.py", line 637, in __init__
+(EngineCore_DP0 pid=2004825) ERROR 02-02 06:56:35 [core.py:866]     super().__init__(
+(EngineCore_DP0 pid=2004825) ERROR 02-02 06:56:35 [core.py:866]   File "/home/mingzhed/Projects/lm-evaluation-harness/.venv/lib/python3.12/site-packages/vllm/v1/engine/core.py", line 109, in __init__
+(EngineCore_DP0 pid=2004825) ERROR 02-02 06:56:35 [core.py:866]     num_gpu_blocks, num_cpu_blocks, kv_cache_config = self._initialize_kv_caches(
+(EngineCore_DP0 pid=2004825) ERROR 02-02 06:56:35 [core.py:866]                                                       ^^^^^^^^^^^^^^^^^^^^^^^^^^^
+(EngineCore_DP0 pid=2004825) ERROR 02-02 06:56:35 [core.py:866]   File "/home/mingzhed/Projects/lm-evaluation-harness/.venv/lib/python3.12/site-packages/vllm/v1/engine/core.py", line 248, in _initialize_kv_caches
+(EngineCore_DP0 pid=2004825) ERROR 02-02 06:56:35 [core.py:866]     kv_cache_configs = get_kv_cache_configs(
+(EngineCore_DP0 pid=2004825) ERROR 02-02 06:56:35 [core.py:866]                        ^^^^^^^^^^^^^^^^^^^^^
+(EngineCore_DP0 pid=2004825) ERROR 02-02 06:56:35 [core.py:866]   File "/home/mingzhed/Projects/lm-evaluation-harness/.venv/lib/python3.12/site-packages/vllm/v1/core/kv_cache_utils.py", line 1340, in get_kv_cache_configs
+(EngineCore_DP0 pid=2004825) ERROR 02-02 06:56:35 [core.py:866]     check_enough_kv_cache_memory(
+(EngineCore_DP0 pid=2004825) ERROR 02-02 06:56:35 [core.py:866]   File "/home/mingzhed/Projects/lm-evaluation-harness/.venv/lib/python3.12/site-packages/vllm/v1/core/kv_cache_utils.py", line 710, in check_enough_kv_cache_memory
+(EngineCore_DP0 pid=2004825) ERROR 02-02 06:56:35 [core.py:866]     raise ValueError(
+(EngineCore_DP0 pid=2004825) ERROR 02-02 06:56:35 [core.py:866] ValueError: To serve at least one request with the models's max seq len (40960), (10.00 GiB KV cache is needed, which is larger than the available KV cache memory (8.35 GiB). Based on the available memory, the estimated maximum model length is 34208. Try increasing `gpu_memory_utilization` or decreasing `max_model_len` when initializing the engine. See https://docs.vllm.ai/en/latest/configuration/conserving_memory/ for more details.
+(EngineCore_DP0 pid=2004825) Process EngineCore_DP0:
+(EngineCore_DP0 pid=2004825) Traceback (most recent call last):
+(EngineCore_DP0 pid=2004825)   File "/home/mingzhed/.local/share/uv/python/cpython-3.12.12-linux-x86_64-gnu/lib/python3.12/multiprocessing/process.py", line 314, in _bootstrap
+(EngineCore_DP0 pid=2004825)     self.run()
+(EngineCore_DP0 pid=2004825)   File "/home/mingzhed/.local/share/uv/python/cpython-3.12.12-linux-x86_64-gnu/lib/python3.12/multiprocessing/process.py", line 108, in run
+(EngineCore_DP0 pid=2004825)     self._target(*self._args, **self._kwargs)
+(EngineCore_DP0 pid=2004825)   File "/home/mingzhed/Projects/lm-evaluation-harness/.venv/lib/python3.12/site-packages/vllm/v1/engine/core.py", line 870, in run_engine_core
+(EngineCore_DP0 pid=2004825)     raise e
+(EngineCore_DP0 pid=2004825)   File "/home/mingzhed/Projects/lm-evaluation-harness/.venv/lib/python3.12/site-packages/vllm/v1/engine/core.py", line 857, in run_engine_core
+(EngineCore_DP0 pid=2004825)     engine_core = EngineCoreProc(*args, **kwargs)
+(EngineCore_DP0 pid=2004825)                   ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+(EngineCore_DP0 pid=2004825)   File "/home/mingzhed/Projects/lm-evaluation-harness/.venv/lib/python3.12/site-packages/vllm/v1/engine/core.py", line 637, in __init__
+(EngineCore_DP0 pid=2004825)     super().__init__(
+(EngineCore_DP0 pid=2004825)   File "/home/mingzhed/Projects/lm-evaluation-harness/.venv/lib/python3.12/site-packages/vllm/v1/engine/core.py", line 109, in __init__
+(EngineCore_DP0 pid=2004825)     num_gpu_blocks, num_cpu_blocks, kv_cache_config = self._initialize_kv_caches(
+(EngineCore_DP0 pid=2004825)                                                       ^^^^^^^^^^^^^^^^^^^^^^^^^^^
+(EngineCore_DP0 pid=2004825)   File "/home/mingzhed/Projects/lm-evaluation-harness/.venv/lib/python3.12/site-packages/vllm/v1/engine/core.py", line 248, in _initialize_kv_caches
+(EngineCore_DP0 pid=2004825)     kv_cache_configs = get_kv_cache_configs(
+(EngineCore_DP0 pid=2004825)                        ^^^^^^^^^^^^^^^^^^^^^
+(EngineCore_DP0 pid=2004825)   File "/home/mingzhed/Projects/lm-evaluation-harness/.venv/lib/python3.12/site-packages/vllm/v1/core/kv_cache_utils.py", line 1340, in get_kv_cache_configs
+(EngineCore_DP0 pid=2004825)     check_enough_kv_cache_memory(
+(EngineCore_DP0 pid=2004825)   File "/home/mingzhed/Projects/lm-evaluation-harness/.venv/lib/python3.12/site-packages/vllm/v1/core/kv_cache_utils.py", line 710, in check_enough_kv_cache_memory
+(EngineCore_DP0 pid=2004825)     raise ValueError(
+(EngineCore_DP0 pid=2004825) ValueError: To serve at least one request with the models's max seq len (40960), (10.00 GiB KV cache is needed, which is larger than the available KV cache memory (8.35 GiB). Based on the available memory, the estimated maximum model length is 34208. Try increasing `gpu_memory_utilization` or decreasing `max_model_len` when initializing the engine. See https://docs.vllm.ai/en/latest/configuration/conserving_memory/ for more details.
+[rank0]:[W202 06:56:36.498774856 ProcessGroupNCCL.cpp:1524] Warning: WARNING: destroy_process_group() was not called before program exit, which can leak resources. For more info, please see https://pytorch.org/docs/stable/distributed.html#shutdown (function operator())
+(APIServer pid=2004339) Traceback (most recent call last):
+(APIServer pid=2004339)   File "<frozen runpy>", line 198, in _run_module_as_main
+(APIServer pid=2004339)   File "<frozen runpy>", line 88, in _run_code
+(APIServer pid=2004339)   File "/home/mingzhed/Projects/lm-evaluation-harness/.venv/lib/python3.12/site-packages/vllm/entrypoints/openai/api_server.py", line 1469, in <module>
+(APIServer pid=2004339)     uvloop.run(run_server(args))
+(APIServer pid=2004339)   File "/home/mingzhed/Projects/lm-evaluation-harness/.venv/lib/python3.12/site-packages/uvloop/__init__.py", line 96, in run
+(APIServer pid=2004339)     return __asyncio.run(
+(APIServer pid=2004339)            ^^^^^^^^^^^^^^
+(APIServer pid=2004339)   File "/home/mingzhed/.local/share/uv/python/cpython-3.12.12-linux-x86_64-gnu/lib/python3.12/asyncio/runners.py", line 195, in run
+(APIServer pid=2004339)     return runner.run(main)
+(APIServer pid=2004339)            ^^^^^^^^^^^^^^^^
+(APIServer pid=2004339)   File "/home/mingzhed/.local/share/uv/python/cpython-3.12.12-linux-x86_64-gnu/lib/python3.12/asyncio/runners.py", line 118, in run
+(APIServer pid=2004339)     return self._loop.run_until_complete(task)
+(APIServer pid=2004339)            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+(APIServer pid=2004339)   File "uvloop/loop.pyx", line 1518, in uvloop.loop.Loop.run_until_complete
+(APIServer pid=2004339)   File "/home/mingzhed/Projects/lm-evaluation-harness/.venv/lib/python3.12/site-packages/uvloop/__init__.py", line 48, in wrapper
+(APIServer pid=2004339)     return await main
+(APIServer pid=2004339)            ^^^^^^^^^^
+(APIServer pid=2004339)   File "/home/mingzhed/Projects/lm-evaluation-harness/.venv/lib/python3.12/site-packages/vllm/entrypoints/openai/api_server.py", line 1398, in run_server
+(APIServer pid=2004339)     await run_server_worker(listen_address, sock, args, **uvicorn_kwargs)
+(APIServer pid=2004339)   File "/home/mingzhed/Projects/lm-evaluation-harness/.venv/lib/python3.12/site-packages/vllm/entrypoints/openai/api_server.py", line 1417, in run_server_worker
+(APIServer pid=2004339)     async with build_async_engine_client(
+(APIServer pid=2004339)                ^^^^^^^^^^^^^^^^^^^^^^^^^^
+(APIServer pid=2004339)   File "/home/mingzhed/.local/share/uv/python/cpython-3.12.12-linux-x86_64-gnu/lib/python3.12/contextlib.py", line 210, in __aenter__
+(APIServer pid=2004339)     return await anext(self.gen)
+(APIServer pid=2004339)            ^^^^^^^^^^^^^^^^^^^^^
+(APIServer pid=2004339)   File "/home/mingzhed/Projects/lm-evaluation-harness/.venv/lib/python3.12/site-packages/vllm/entrypoints/openai/api_server.py", line 172, in build_async_engine_client
+(APIServer pid=2004339)     async with build_async_engine_client_from_engine_args(
+(APIServer pid=2004339)                ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+(APIServer pid=2004339)   File "/home/mingzhed/.local/share/uv/python/cpython-3.12.12-linux-x86_64-gnu/lib/python3.12/contextlib.py", line 210, in __aenter__
+(APIServer pid=2004339)     return await anext(self.gen)
+(APIServer pid=2004339)            ^^^^^^^^^^^^^^^^^^^^^
+(APIServer pid=2004339)   File "/home/mingzhed/Projects/lm-evaluation-harness/.venv/lib/python3.12/site-packages/vllm/entrypoints/openai/api_server.py", line 213, in build_async_engine_client_from_engine_args
+(APIServer pid=2004339)     async_llm = AsyncLLM.from_vllm_config(
+(APIServer pid=2004339)                 ^^^^^^^^^^^^^^^^^^^^^^^^^^
+(APIServer pid=2004339)   File "/home/mingzhed/Projects/lm-evaluation-harness/.venv/lib/python3.12/site-packages/vllm/v1/engine/async_llm.py", line 215, in from_vllm_config
+(APIServer pid=2004339)     return cls(
+(APIServer pid=2004339)            ^^^^
+(APIServer pid=2004339)   File "/home/mingzhed/Projects/lm-evaluation-harness/.venv/lib/python3.12/site-packages/vllm/v1/engine/async_llm.py", line 134, in __init__
+(APIServer pid=2004339)     self.engine_core = EngineCoreClient.make_async_mp_client(
+(APIServer pid=2004339)                        ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+(APIServer pid=2004339)   File "/home/mingzhed/Projects/lm-evaluation-harness/.venv/lib/python3.12/site-packages/vllm/v1/engine/core_client.py", line 121, in make_async_mp_client
+(APIServer pid=2004339)     return AsyncMPClient(*client_args)
+(APIServer pid=2004339)            ^^^^^^^^^^^^^^^^^^^^^^^^^^^
+(APIServer pid=2004339)   File "/home/mingzhed/Projects/lm-evaluation-harness/.venv/lib/python3.12/site-packages/vllm/v1/engine/core_client.py", line 820, in __init__
+(APIServer pid=2004339)     super().__init__(
+(APIServer pid=2004339)   File "/home/mingzhed/Projects/lm-evaluation-harness/.venv/lib/python3.12/site-packages/vllm/v1/engine/core_client.py", line 477, in __init__
+(APIServer pid=2004339)     with launch_core_engines(vllm_config, executor_class, log_stats) as (
+(APIServer pid=2004339)          ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+(APIServer pid=2004339)   File "/home/mingzhed/.local/share/uv/python/cpython-3.12.12-linux-x86_64-gnu/lib/python3.12/contextlib.py", line 144, in __exit__
+(APIServer pid=2004339)     next(self.gen)
+(APIServer pid=2004339)   File "/home/mingzhed/Projects/lm-evaluation-harness/.venv/lib/python3.12/site-packages/vllm/v1/engine/utils.py", line 903, in launch_core_engines
+(APIServer pid=2004339)     wait_for_engine_startup(
+(APIServer pid=2004339)   File "/home/mingzhed/Projects/lm-evaluation-harness/.venv/lib/python3.12/site-packages/vllm/v1/engine/utils.py", line 960, in wait_for_engine_startup
+(APIServer pid=2004339)     raise RuntimeError(
+(APIServer pid=2004339) RuntimeError: Engine core initialization failed. See root cause above. Failed core proc(s): {}

vllm_0001500.log

@@ -25,3 +25,70 @@
 (EngineCore_DP0 pid=2004821) 
 (EngineCore_DP0 pid=2004821) 
 (EngineCore_DP0 pid=2004821) 
+(EngineCore_DP0 pid=2004821) 
+(EngineCore_DP0 pid=2004821) 
+(EngineCore_DP0 pid=2004821) 
+(EngineCore_DP0 pid=2004821) 
+(EngineCore_DP0 pid=2004821) 
+(EngineCore_DP0 pid=2004821) INFO 02-02 06:56:10 [gpu_model_runner.py:3659] Model loading took 19.4031 GiB memory and 48.558429 seconds
+(EngineCore_DP0 pid=2004821) INFO 02-02 06:56:23 [backends.py:643] Using cache directory: /home/mingzhed/.cache/vllm/torch_compile_cache/36416882c8/rank_0_0/backbone for vLLM's torch.compile
+(EngineCore_DP0 pid=2004821) INFO 02-02 06:56:23 [backends.py:703] Dynamo bytecode transform time: 11.92 s
+(EngineCore_DP0 pid=2004821) INFO 02-02 06:56:31 [backends.py:261] Cache the graph of compile range (1, 8192) for later use
+(EngineCore_DP0 pid=2004821) INFO 02-02 06:56:35 [backends.py:278] Compiling a graph for compile range (1, 8192) takes 4.33 s
+(EngineCore_DP0 pid=2004821) INFO 02-02 06:56:35 [monitor.py:34] torch.compile takes 16.24 s in total
+(EngineCore_DP0 pid=2004821) INFO 02-02 06:56:37 [gpu_worker.py:375] Available KV cache memory: 10.80 GiB
+(EngineCore_DP0 pid=2004821) INFO 02-02 06:56:38 [kv_cache_utils.py:1291] GPU KV cache size: 44,224 tokens
+(EngineCore_DP0 pid=2004821) INFO 02-02 06:56:38 [kv_cache_utils.py:1296] Maximum concurrency for 40,960 tokens per request: 1.08x
+(EngineCore_DP0 pid=2004821) 
+(EngineCore_DP0 pid=2004821) 
+(EngineCore_DP0 pid=2004821) INFO 02-02 06:56:56 [gpu_model_runner.py:4587] Graph capturing finished in 19 secs, took 3.22 GiB
+(EngineCore_DP0 pid=2004821) INFO 02-02 06:56:56 [core.py:259] init engine (profile, create kv cache, warmup model) took 46.18 seconds
+(APIServer pid=2004403) INFO 02-02 06:56:59 [api_server.py:1099] Supported tasks: ['generate']
+(APIServer pid=2004403) WARNING 02-02 06:56:59 [model.py:1487] Default sampling parameters have been overridden by the model's Hugging Face generation config recommended from the model creator. If this is not intended, please relaunch vLLM instance with `--generation-config vllm`.
+(APIServer pid=2004403) INFO 02-02 06:56:59 [serving_responses.py:201] Using default chat sampling params from model: {'temperature': 0.6, 'top_k': 20, 'top_p': 0.95}
+(APIServer pid=2004403) INFO 02-02 06:57:00 [serving_chat.py:137] Using default chat sampling params from model: {'temperature': 0.6, 'top_k': 20, 'top_p': 0.95}
+(APIServer pid=2004403) INFO 02-02 06:57:00 [serving_completion.py:77] Using default completion sampling params from model: {'temperature': 0.6, 'top_k': 20, 'top_p': 0.95}
+(APIServer pid=2004403) INFO 02-02 06:57:00 [serving_chat.py:137] Using default chat sampling params from model: {'temperature': 0.6, 'top_k': 20, 'top_p': 0.95}
+(APIServer pid=2004403) INFO 02-02 06:57:00 [api_server.py:1425] Starting vLLM API server 0 on http://0.0.0.0:9001
+(APIServer pid=2004403) INFO 02-02 06:57:00 [launcher.py:38] Available routes are:
+(APIServer pid=2004403) INFO 02-02 06:57:00 [launcher.py:46] Route: /openapi.json, Methods: HEAD, GET
+(APIServer pid=2004403) INFO 02-02 06:57:00 [launcher.py:46] Route: /docs, Methods: HEAD, GET
+(APIServer pid=2004403) INFO 02-02 06:57:00 [launcher.py:46] Route: /docs/oauth2-redirect, Methods: HEAD, GET
+(APIServer pid=2004403) INFO 02-02 06:57:00 [launcher.py:46] Route: /redoc, Methods: HEAD, GET
+(APIServer pid=2004403) INFO 02-02 06:57:00 [launcher.py:46] Route: /scale_elastic_ep, Methods: POST
+(APIServer pid=2004403) INFO 02-02 06:57:00 [launcher.py:46] Route: /is_scaling_elastic_ep, Methods: POST
+(APIServer pid=2004403) INFO 02-02 06:57:00 [launcher.py:46] Route: /tokenize, Methods: POST
+(APIServer pid=2004403) INFO 02-02 06:57:00 [launcher.py:46] Route: /detokenize, Methods: POST
+(APIServer pid=2004403) INFO 02-02 06:57:00 [launcher.py:46] Route: /inference/v1/generate, Methods: POST
+(APIServer pid=2004403) INFO 02-02 06:57:00 [launcher.py:46] Route: /pause, Methods: POST
+(APIServer pid=2004403) INFO 02-02 06:57:00 [launcher.py:46] Route: /resume, Methods: POST
+(APIServer pid=2004403) INFO 02-02 06:57:00 [launcher.py:46] Route: /is_paused, Methods: GET
+(APIServer pid=2004403) INFO 02-02 06:57:00 [launcher.py:46] Route: /metrics, Methods: GET
+(APIServer pid=2004403) INFO 02-02 06:57:00 [launcher.py:46] Route: /health, Methods: GET
+(APIServer pid=2004403) INFO 02-02 06:57:00 [launcher.py:46] Route: /load, Methods: GET
+(APIServer pid=2004403) INFO 02-02 06:57:00 [launcher.py:46] Route: /v1/models, Methods: GET
+(APIServer pid=2004403) INFO 02-02 06:57:00 [launcher.py:46] Route: /version, Methods: GET
+(APIServer pid=2004403) INFO 02-02 06:57:00 [launcher.py:46] Route: /v1/responses, Methods: POST
+(APIServer pid=2004403) INFO 02-02 06:57:00 [launcher.py:46] Route: /v1/responses/{response_id}, Methods: GET
+(APIServer pid=2004403) INFO 02-02 06:57:00 [launcher.py:46] Route: /v1/responses/{response_id}/cancel, Methods: POST
+(APIServer pid=2004403) INFO 02-02 06:57:00 [launcher.py:46] Route: /v1/messages, Methods: POST
+(APIServer pid=2004403) INFO 02-02 06:57:00 [launcher.py:46] Route: /v1/chat/completions, Methods: POST
+(APIServer pid=2004403) INFO 02-02 06:57:00 [launcher.py:46] Route: /v1/completions, Methods: POST
+(APIServer pid=2004403) INFO 02-02 06:57:00 [launcher.py:46] Route: /v1/audio/transcriptions, Methods: POST
+(APIServer pid=2004403) INFO 02-02 06:57:00 [launcher.py:46] Route: /v1/audio/translations, Methods: POST
+(APIServer pid=2004403) INFO 02-02 06:57:00 [launcher.py:46] Route: /ping, Methods: GET
+(APIServer pid=2004403) INFO 02-02 06:57:00 [launcher.py:46] Route: /ping, Methods: POST
+(APIServer pid=2004403) INFO 02-02 06:57:00 [launcher.py:46] Route: /invocations, Methods: POST
+(APIServer pid=2004403) INFO 02-02 06:57:00 [launcher.py:46] Route: /classify, Methods: POST
+(APIServer pid=2004403) INFO 02-02 06:57:00 [launcher.py:46] Route: /v1/embeddings, Methods: POST
+(APIServer pid=2004403) INFO 02-02 06:57:00 [launcher.py:46] Route: /score, Methods: POST
+(APIServer pid=2004403) INFO 02-02 06:57:00 [launcher.py:46] Route: /v1/score, Methods: POST
+(APIServer pid=2004403) INFO 02-02 06:57:00 [launcher.py:46] Route: /rerank, Methods: POST
+(APIServer pid=2004403) INFO 02-02 06:57:00 [launcher.py:46] Route: /v1/rerank, Methods: POST
+(APIServer pid=2004403) INFO 02-02 06:57:00 [launcher.py:46] Route: /v2/rerank, Methods: POST
+(APIServer pid=2004403) INFO 02-02 06:57:00 [launcher.py:46] Route: /pooling, Methods: POST
+(APIServer pid=2004403) INFO:     Started server process [2004403]
+(APIServer pid=2004403) INFO:     Waiting for application startup.
+(APIServer pid=2004403) INFO:     Application startup complete.
+(APIServer pid=2004403) INFO 02-02 06:57:13 [launcher.py:110] Shutting down FastAPI HTTP server.
+[rank0]:[W202 06:57:13.008751275 ProcessGroupNCCL.cpp:1524] Warning: WARNING: destroy_process_group() was not called before program exit, which can leak resources. For more info, please see https://pytorch.org/docs/stable/distributed.html#shutdown (function operator())

vllm_0002000.log

@@ -25,3 +25,69 @@
 (EngineCore_DP0 pid=2005095) 
 (EngineCore_DP0 pid=2005095) 
 (EngineCore_DP0 pid=2005095) 
+(EngineCore_DP0 pid=2005095) 
+(EngineCore_DP0 pid=2005095) 
+(EngineCore_DP0 pid=2005095) 
+(EngineCore_DP0 pid=2005095) 
+(EngineCore_DP0 pid=2005095) 
+(EngineCore_DP0 pid=2005095) INFO 02-02 06:56:08 [gpu_model_runner.py:3659] Model loading took 19.4031 GiB memory and 45.470238 seconds
+(EngineCore_DP0 pid=2005095) INFO 02-02 06:56:21 [backends.py:643] Using cache directory: /home/mingzhed/.cache/vllm/torch_compile_cache/de6285fc4b/rank_0_0/backbone for vLLM's torch.compile
+(EngineCore_DP0 pid=2005095) INFO 02-02 06:56:21 [backends.py:703] Dynamo bytecode transform time: 12.97 s
+(EngineCore_DP0 pid=2005095) INFO 02-02 06:56:31 [backends.py:226] Directly load the compiled graph(s) for compile range (1, 8192) from the cache, took 1.501 s
+(EngineCore_DP0 pid=2005095) INFO 02-02 06:56:31 [monitor.py:34] torch.compile takes 14.48 s in total
+(EngineCore_DP0 pid=2005095) INFO 02-02 06:56:32 [gpu_worker.py:375] Available KV cache memory: 10.30 GiB
+(EngineCore_DP0 pid=2005095) INFO 02-02 06:56:33 [kv_cache_utils.py:1291] GPU KV cache size: 42,176 tokens
+(EngineCore_DP0 pid=2005095) INFO 02-02 06:56:33 [kv_cache_utils.py:1296] Maximum concurrency for 40,960 tokens per request: 1.03x
+(EngineCore_DP0 pid=2005095) 
+(EngineCore_DP0 pid=2005095) 
+(EngineCore_DP0 pid=2005095) INFO 02-02 06:56:54 [gpu_model_runner.py:4587] Graph capturing finished in 21 secs, took 22.88 GiB
+(EngineCore_DP0 pid=2005095) INFO 02-02 06:56:54 [core.py:259] init engine (profile, create kv cache, warmup model) took 46.13 seconds
+(APIServer pid=2004481) INFO 02-02 06:56:57 [api_server.py:1099] Supported tasks: ['generate']
+(APIServer pid=2004481) WARNING 02-02 06:56:58 [model.py:1487] Default sampling parameters have been overridden by the model's Hugging Face generation config recommended from the model creator. If this is not intended, please relaunch vLLM instance with `--generation-config vllm`.
+(APIServer pid=2004481) INFO 02-02 06:56:58 [serving_responses.py:201] Using default chat sampling params from model: {'temperature': 0.6, 'top_k': 20, 'top_p': 0.95}
+(APIServer pid=2004481) INFO 02-02 06:56:58 [serving_chat.py:137] Using default chat sampling params from model: {'temperature': 0.6, 'top_k': 20, 'top_p': 0.95}
+(APIServer pid=2004481) INFO 02-02 06:56:58 [serving_completion.py:77] Using default completion sampling params from model: {'temperature': 0.6, 'top_k': 20, 'top_p': 0.95}
+(APIServer pid=2004481) INFO 02-02 06:56:58 [serving_chat.py:137] Using default chat sampling params from model: {'temperature': 0.6, 'top_k': 20, 'top_p': 0.95}
+(APIServer pid=2004481) INFO 02-02 06:56:58 [api_server.py:1425] Starting vLLM API server 0 on http://0.0.0.0:9002
+(APIServer pid=2004481) INFO 02-02 06:56:58 [launcher.py:38] Available routes are:
+(APIServer pid=2004481) INFO 02-02 06:56:58 [launcher.py:46] Route: /openapi.json, Methods: HEAD, GET
+(APIServer pid=2004481) INFO 02-02 06:56:58 [launcher.py:46] Route: /docs, Methods: HEAD, GET
+(APIServer pid=2004481) INFO 02-02 06:56:58 [launcher.py:46] Route: /docs/oauth2-redirect, Methods: HEAD, GET
+(APIServer pid=2004481) INFO 02-02 06:56:58 [launcher.py:46] Route: /redoc, Methods: HEAD, GET
+(APIServer pid=2004481) INFO 02-02 06:56:58 [launcher.py:46] Route: /scale_elastic_ep, Methods: POST
+(APIServer pid=2004481) INFO 02-02 06:56:58 [launcher.py:46] Route: /is_scaling_elastic_ep, Methods: POST
+(APIServer pid=2004481) INFO 02-02 06:56:58 [launcher.py:46] Route: /tokenize, Methods: POST
+(APIServer pid=2004481) INFO 02-02 06:56:58 [launcher.py:46] Route: /detokenize, Methods: POST
+(APIServer pid=2004481) INFO 02-02 06:56:58 [launcher.py:46] Route: /inference/v1/generate, Methods: POST
+(APIServer pid=2004481) INFO 02-02 06:56:58 [launcher.py:46] Route: /pause, Methods: POST
+(APIServer pid=2004481) INFO 02-02 06:56:58 [launcher.py:46] Route: /resume, Methods: POST
+(APIServer pid=2004481) INFO 02-02 06:56:58 [launcher.py:46] Route: /is_paused, Methods: GET
+(APIServer pid=2004481) INFO 02-02 06:56:58 [launcher.py:46] Route: /metrics, Methods: GET
+(APIServer pid=2004481) INFO 02-02 06:56:58 [launcher.py:46] Route: /health, Methods: GET
+(APIServer pid=2004481) INFO 02-02 06:56:58 [launcher.py:46] Route: /load, Methods: GET
+(APIServer pid=2004481) INFO 02-02 06:56:58 [launcher.py:46] Route: /v1/models, Methods: GET
+(APIServer pid=2004481) INFO 02-02 06:56:58 [launcher.py:46] Route: /version, Methods: GET
+(APIServer pid=2004481) INFO 02-02 06:56:58 [launcher.py:46] Route: /v1/responses, Methods: POST
+(APIServer pid=2004481) INFO 02-02 06:56:58 [launcher.py:46] Route: /v1/responses/{response_id}, Methods: GET
+(APIServer pid=2004481) INFO 02-02 06:56:58 [launcher.py:46] Route: /v1/responses/{response_id}/cancel, Methods: POST
+(APIServer pid=2004481) INFO 02-02 06:56:58 [launcher.py:46] Route: /v1/messages, Methods: POST
+(APIServer pid=2004481) INFO 02-02 06:56:58 [launcher.py:46] Route: /v1/chat/completions, Methods: POST
+(APIServer pid=2004481) INFO 02-02 06:56:58 [launcher.py:46] Route: /v1/completions, Methods: POST
+(APIServer pid=2004481) INFO 02-02 06:56:58 [launcher.py:46] Route: /v1/audio/transcriptions, Methods: POST
+(APIServer pid=2004481) INFO 02-02 06:56:58 [launcher.py:46] Route: /v1/audio/translations, Methods: POST
+(APIServer pid=2004481) INFO 02-02 06:56:58 [launcher.py:46] Route: /ping, Methods: GET
+(APIServer pid=2004481) INFO 02-02 06:56:58 [launcher.py:46] Route: /ping, Methods: POST
+(APIServer pid=2004481) INFO 02-02 06:56:58 [launcher.py:46] Route: /invocations, Methods: POST
+(APIServer pid=2004481) INFO 02-02 06:56:58 [launcher.py:46] Route: /classify, Methods: POST
+(APIServer pid=2004481) INFO 02-02 06:56:58 [launcher.py:46] Route: /v1/embeddings, Methods: POST
+(APIServer pid=2004481) INFO 02-02 06:56:58 [launcher.py:46] Route: /score, Methods: POST
+(APIServer pid=2004481) INFO 02-02 06:56:58 [launcher.py:46] Route: /v1/score, Methods: POST
+(APIServer pid=2004481) INFO 02-02 06:56:58 [launcher.py:46] Route: /rerank, Methods: POST
+(APIServer pid=2004481) INFO 02-02 06:56:58 [launcher.py:46] Route: /v1/rerank, Methods: POST
+(APIServer pid=2004481) INFO 02-02 06:56:58 [launcher.py:46] Route: /v2/rerank, Methods: POST
+(APIServer pid=2004481) INFO 02-02 06:56:58 [launcher.py:46] Route: /pooling, Methods: POST
+(APIServer pid=2004481) INFO:     Started server process [2004481]
+(APIServer pid=2004481) INFO:     Waiting for application startup.
+(APIServer pid=2004481) INFO:     Application startup complete.
+(APIServer pid=2004481) INFO 02-02 06:57:13 [launcher.py:110] Shutting down FastAPI HTTP server.
+[rank0]:[W202 06:57:13.013215051 ProcessGroupNCCL.cpp:1524] Warning: WARNING: destroy_process_group() was not called before program exit, which can leak resources. For more info, please see https://pytorch.org/docs/stable/distributed.html#shutdown (function operator())

vllm_0002500.log

@@ -22,3 +22,73 @@
 (EngineCore_DP0 pid=2005431) 
 (EngineCore_DP0 pid=2005431) 
 (EngineCore_DP0 pid=2005431) 
+(EngineCore_DP0 pid=2005431) 
+(EngineCore_DP0 pid=2005431) 
+(EngineCore_DP0 pid=2005431) 
+(EngineCore_DP0 pid=2005431) 
+(EngineCore_DP0 pid=2005431) 
+(EngineCore_DP0 pid=2005431) 
+(EngineCore_DP0 pid=2005431) 
+(EngineCore_DP0 pid=2005431) 
+(EngineCore_DP0 pid=2005431) INFO 02-02 06:56:14 [gpu_model_runner.py:3659] Model loading took 19.4031 GiB memory and 45.777633 seconds
+(EngineCore_DP0 pid=2005431) INFO 02-02 06:56:27 [backends.py:643] Using cache directory: /home/mingzhed/.cache/vllm/torch_compile_cache/15aaa291eb/rank_0_0/backbone for vLLM's torch.compile
+(EngineCore_DP0 pid=2005431) INFO 02-02 06:56:27 [backends.py:703] Dynamo bytecode transform time: 12.25 s
+(EngineCore_DP0 pid=2005431) INFO 02-02 06:56:35 [backends.py:261] Cache the graph of compile range (1, 8192) for later use
+(EngineCore_DP0 pid=2005431) INFO 02-02 06:56:40 [backends.py:278] Compiling a graph for compile range (1, 8192) takes 4.49 s
+(EngineCore_DP0 pid=2005431) INFO 02-02 06:56:40 [monitor.py:34] torch.compile takes 16.74 s in total
+(EngineCore_DP0 pid=2005431) INFO 02-02 06:56:41 [gpu_worker.py:375] Available KV cache memory: 17.66 GiB
+(EngineCore_DP0 pid=2005431) INFO 02-02 06:56:42 [kv_cache_utils.py:1291] GPU KV cache size: 72,320 tokens
+(EngineCore_DP0 pid=2005431) INFO 02-02 06:56:42 [kv_cache_utils.py:1296] Maximum concurrency for 40,960 tokens per request: 1.77x
+(EngineCore_DP0 pid=2005431) 
+(EngineCore_DP0 pid=2005431) 
+(EngineCore_DP0 pid=2005431) INFO 02-02 06:57:04 [gpu_model_runner.py:4587] Graph capturing finished in 22 secs, took 10.85 GiB
+(EngineCore_DP0 pid=2005431) INFO 02-02 06:57:04 [core.py:259] init engine (profile, create kv cache, warmup model) took 49.38 seconds
+(APIServer pid=2004680) INFO 02-02 06:57:07 [api_server.py:1099] Supported tasks: ['generate']
+(APIServer pid=2004680) WARNING 02-02 06:57:07 [model.py:1487] Default sampling parameters have been overridden by the model's Hugging Face generation config recommended from the model creator. If this is not intended, please relaunch vLLM instance with `--generation-config vllm`.
+(APIServer pid=2004680) INFO 02-02 06:57:07 [serving_responses.py:201] Using default chat sampling params from model: {'temperature': 0.6, 'top_k': 20, 'top_p': 0.95}
+(APIServer pid=2004680) INFO 02-02 06:57:07 [serving_chat.py:137] Using default chat sampling params from model: {'temperature': 0.6, 'top_k': 20, 'top_p': 0.95}
+(APIServer pid=2004680) INFO 02-02 06:57:07 [serving_completion.py:77] Using default completion sampling params from model: {'temperature': 0.6, 'top_k': 20, 'top_p': 0.95}
+(APIServer pid=2004680) INFO 02-02 06:57:07 [serving_chat.py:137] Using default chat sampling params from model: {'temperature': 0.6, 'top_k': 20, 'top_p': 0.95}
+(APIServer pid=2004680) INFO 02-02 06:57:07 [api_server.py:1425] Starting vLLM API server 0 on http://0.0.0.0:9003
+(APIServer pid=2004680) INFO 02-02 06:57:07 [launcher.py:38] Available routes are:
+(APIServer pid=2004680) INFO 02-02 06:57:07 [launcher.py:46] Route: /openapi.json, Methods: HEAD, GET
+(APIServer pid=2004680) INFO 02-02 06:57:07 [launcher.py:46] Route: /docs, Methods: HEAD, GET
+(APIServer pid=2004680) INFO 02-02 06:57:07 [launcher.py:46] Route: /docs/oauth2-redirect, Methods: HEAD, GET
+(APIServer pid=2004680) INFO 02-02 06:57:07 [launcher.py:46] Route: /redoc, Methods: HEAD, GET
+(APIServer pid=2004680) INFO 02-02 06:57:07 [launcher.py:46] Route: /scale_elastic_ep, Methods: POST
+(APIServer pid=2004680) INFO 02-02 06:57:07 [launcher.py:46] Route: /is_scaling_elastic_ep, Methods: POST
+(APIServer pid=2004680) INFO 02-02 06:57:07 [launcher.py:46] Route: /tokenize, Methods: POST
+(APIServer pid=2004680) INFO 02-02 06:57:07 [launcher.py:46] Route: /detokenize, Methods: POST
+(APIServer pid=2004680) INFO 02-02 06:57:07 [launcher.py:46] Route: /inference/v1/generate, Methods: POST
+(APIServer pid=2004680) INFO 02-02 06:57:07 [launcher.py:46] Route: /pause, Methods: POST
+(APIServer pid=2004680) INFO 02-02 06:57:07 [launcher.py:46] Route: /resume, Methods: POST
+(APIServer pid=2004680) INFO 02-02 06:57:07 [launcher.py:46] Route: /is_paused, Methods: GET
+(APIServer pid=2004680) INFO 02-02 06:57:07 [launcher.py:46] Route: /metrics, Methods: GET
+(APIServer pid=2004680) INFO 02-02 06:57:07 [launcher.py:46] Route: /health, Methods: GET
+(APIServer pid=2004680) INFO 02-02 06:57:07 [launcher.py:46] Route: /load, Methods: GET
+(APIServer pid=2004680) INFO 02-02 06:57:07 [launcher.py:46] Route: /v1/models, Methods: GET
+(APIServer pid=2004680) INFO 02-02 06:57:07 [launcher.py:46] Route: /version, Methods: GET
+(APIServer pid=2004680) INFO 02-02 06:57:07 [launcher.py:46] Route: /v1/responses, Methods: POST
+(APIServer pid=2004680) INFO 02-02 06:57:07 [launcher.py:46] Route: /v1/responses/{response_id}, Methods: GET
+(APIServer pid=2004680) INFO 02-02 06:57:07 [launcher.py:46] Route: /v1/responses/{response_id}/cancel, Methods: POST
+(APIServer pid=2004680) INFO 02-02 06:57:07 [launcher.py:46] Route: /v1/messages, Methods: POST
+(APIServer pid=2004680) INFO 02-02 06:57:07 [launcher.py:46] Route: /v1/chat/completions, Methods: POST
+(APIServer pid=2004680) INFO 02-02 06:57:07 [launcher.py:46] Route: /v1/completions, Methods: POST
+(APIServer pid=2004680) INFO 02-02 06:57:07 [launcher.py:46] Route: /v1/audio/transcriptions, Methods: POST
+(APIServer pid=2004680) INFO 02-02 06:57:07 [launcher.py:46] Route: /v1/audio/translations, Methods: POST
+(APIServer pid=2004680) INFO 02-02 06:57:07 [launcher.py:46] Route: /ping, Methods: GET
+(APIServer pid=2004680) INFO 02-02 06:57:07 [launcher.py:46] Route: /ping, Methods: POST
+(APIServer pid=2004680) INFO 02-02 06:57:07 [launcher.py:46] Route: /invocations, Methods: POST
+(APIServer pid=2004680) INFO 02-02 06:57:07 [launcher.py:46] Route: /classify, Methods: POST
+(APIServer pid=2004680) INFO 02-02 06:57:07 [launcher.py:46] Route: /v1/embeddings, Methods: POST
+(APIServer pid=2004680) INFO 02-02 06:57:07 [launcher.py:46] Route: /score, Methods: POST
+(APIServer pid=2004680) INFO 02-02 06:57:07 [launcher.py:46] Route: /v1/score, Methods: POST
+(APIServer pid=2004680) INFO 02-02 06:57:07 [launcher.py:46] Route: /rerank, Methods: POST
+(APIServer pid=2004680) INFO 02-02 06:57:07 [launcher.py:46] Route: /v1/rerank, Methods: POST
+(APIServer pid=2004680) INFO 02-02 06:57:07 [launcher.py:46] Route: /v2/rerank, Methods: POST
+(APIServer pid=2004680) INFO 02-02 06:57:07 [launcher.py:46] Route: /pooling, Methods: POST
+(APIServer pid=2004680) INFO:     Started server process [2004680]
+(APIServer pid=2004680) INFO:     Waiting for application startup.
+(APIServer pid=2004680) INFO:     Application startup complete.
+(APIServer pid=2004680) INFO 02-02 06:57:13 [launcher.py:110] Shutting down FastAPI HTTP server.
+[rank0]:[W202 06:57:13.010696572 ProcessGroupNCCL.cpp:1524] Warning: WARNING: destroy_process_group() was not called before program exit, which can leak resources. For more info, please see https://pytorch.org/docs/stable/distributed.html#shutdown (function operator())

vllm_0003000.log

@@ -19,3 +19,26 @@
 (EngineCore_DP0 pid=2006132) 
 (EngineCore_DP0 pid=2006132) 
 (EngineCore_DP0 pid=2006132) 
+(EngineCore_DP0 pid=2006132) 
+(EngineCore_DP0 pid=2006132) 
+(EngineCore_DP0 pid=2006132) 
+(EngineCore_DP0 pid=2006132) 
+(EngineCore_DP0 pid=2006132) 
+(EngineCore_DP0 pid=2006132) 
+(EngineCore_DP0 pid=2006132) 
+(EngineCore_DP0 pid=2006132) 
+(EngineCore_DP0 pid=2006132) 
+(EngineCore_DP0 pid=2006132) 
+(EngineCore_DP0 pid=2006132) 
+(EngineCore_DP0 pid=2006132) INFO 02-02 06:56:26 [gpu_model_runner.py:3659] Model loading took 19.4031 GiB memory and 50.320658 seconds
+(EngineCore_DP0 pid=2006132) INFO 02-02 06:56:39 [backends.py:643] Using cache directory: /home/mingzhed/.cache/vllm/torch_compile_cache/c48ff69b02/rank_0_0/backbone for vLLM's torch.compile
+(EngineCore_DP0 pid=2006132) INFO 02-02 06:56:39 [backends.py:703] Dynamo bytecode transform time: 12.27 s
+(EngineCore_DP0 pid=2006132) INFO 02-02 06:56:49 [backends.py:226] Directly load the compiled graph(s) for compile range (1, 8192) from the cache, took 2.376 s
+(EngineCore_DP0 pid=2006132) INFO 02-02 06:56:49 [monitor.py:34] torch.compile takes 14.65 s in total
+(EngineCore_DP0 pid=2006132) INFO 02-02 06:56:51 [gpu_worker.py:375] Available KV cache memory: 10.30 GiB
+(EngineCore_DP0 pid=2006132) INFO 02-02 06:56:51 [kv_cache_utils.py:1291] GPU KV cache size: 42,176 tokens
+(EngineCore_DP0 pid=2006132) INFO 02-02 06:56:51 [kv_cache_utils.py:1296] Maximum concurrency for 40,960 tokens per request: 1.03x
+(EngineCore_DP0 pid=2006132) 
+(EngineCore_DP0 pid=2006132) 
+(EngineCore_DP0 pid=2006132) INFO 02-02 06:57:13 [gpu_model_runner.py:4587] Graph capturing finished in 21 secs, took 21.23 GiB
+(EngineCore_DP0 pid=2006132) INFO 02-02 06:57:13 [core.py:259] init engine (profile, create kv cache, warmup model) took 46.64 seconds

vllm_0003500.log

@@ -16,3 +16,28 @@
 (EngineCore_DP0 pid=2006487) 
 (EngineCore_DP0 pid=2006487) 
 (EngineCore_DP0 pid=2006487) 
+(EngineCore_DP0 pid=2006487) 
+(EngineCore_DP0 pid=2006487) 
+(EngineCore_DP0 pid=2006487) 
+(EngineCore_DP0 pid=2006487) 
+(EngineCore_DP0 pid=2006487) 
+(EngineCore_DP0 pid=2006487) 
+(EngineCore_DP0 pid=2006487) 
+(EngineCore_DP0 pid=2006487) 
+(EngineCore_DP0 pid=2006487) 
+(EngineCore_DP0 pid=2006487) 
+(EngineCore_DP0 pid=2006487) 
+(EngineCore_DP0 pid=2006487) 
+(EngineCore_DP0 pid=2006487) 
+(EngineCore_DP0 pid=2006487) 
+(EngineCore_DP0 pid=2006487) INFO 02-02 06:56:30 [gpu_model_runner.py:3659] Model loading took 19.4031 GiB memory and 48.680980 seconds
+(EngineCore_DP0 pid=2006487) INFO 02-02 06:56:42 [backends.py:643] Using cache directory: /home/mingzhed/.cache/vllm/torch_compile_cache/71922641d4/rank_0_0/backbone for vLLM's torch.compile
+(EngineCore_DP0 pid=2006487) INFO 02-02 06:56:42 [backends.py:703] Dynamo bytecode transform time: 11.87 s
+(EngineCore_DP0 pid=2006487) INFO 02-02 06:56:51 [backends.py:261] Cache the graph of compile range (1, 8192) for later use
+(EngineCore_DP0 pid=2006487) INFO 02-02 06:56:56 [backends.py:278] Compiling a graph for compile range (1, 8192) takes 4.64 s
+(EngineCore_DP0 pid=2006487) INFO 02-02 06:56:56 [monitor.py:34] torch.compile takes 16.51 s in total
+(EngineCore_DP0 pid=2006487) INFO 02-02 06:56:57 [gpu_worker.py:375] Available KV cache memory: 17.62 GiB
+(EngineCore_DP0 pid=2006487) INFO 02-02 06:56:58 [kv_cache_utils.py:1291] GPU KV cache size: 72,144 tokens
+(EngineCore_DP0 pid=2006487) INFO 02-02 06:56:58 [kv_cache_utils.py:1296] Maximum concurrency for 40,960 tokens per request: 1.76x
+(EngineCore_DP0 pid=2006487) 
+(EngineCore_DP0 pid=2006487) 

vllm_0004000.log

@@ -13,3 +13,30 @@
 (EngineCore_DP0 pid=2006824) 
 (EngineCore_DP0 pid=2006824) 
 (EngineCore_DP0 pid=2006824) 
+(EngineCore_DP0 pid=2006824) 
+(EngineCore_DP0 pid=2006824) 
+(EngineCore_DP0 pid=2006824) 
+(EngineCore_DP0 pid=2006824) 
+(EngineCore_DP0 pid=2006824) 
+(EngineCore_DP0 pid=2006824) 
+(EngineCore_DP0 pid=2006824) 
+(EngineCore_DP0 pid=2006824) 
+(EngineCore_DP0 pid=2006824) 
+(EngineCore_DP0 pid=2006824) 
+(EngineCore_DP0 pid=2006824) 
+(EngineCore_DP0 pid=2006824) 
+(EngineCore_DP0 pid=2006824) 
+(EngineCore_DP0 pid=2006824) 
+(EngineCore_DP0 pid=2006824) 
+(EngineCore_DP0 pid=2006824) 
+(EngineCore_DP0 pid=2006824) 
+(EngineCore_DP0 pid=2006824) INFO 02-02 06:56:36 [gpu_model_runner.py:3659] Model loading took 19.4031 GiB memory and 45.800836 seconds
+(EngineCore_DP0 pid=2006824) INFO 02-02 06:56:49 [backends.py:643] Using cache directory: /home/mingzhed/.cache/vllm/torch_compile_cache/34afe90006/rank_0_0/backbone for vLLM's torch.compile
+(EngineCore_DP0 pid=2006824) INFO 02-02 06:56:49 [backends.py:703] Dynamo bytecode transform time: 12.40 s
+(EngineCore_DP0 pid=2006824) INFO 02-02 06:56:58 [backends.py:226] Directly load the compiled graph(s) for compile range (1, 8192) from the cache, took 1.548 s
+(EngineCore_DP0 pid=2006824) INFO 02-02 06:56:58 [monitor.py:34] torch.compile takes 13.94 s in total
+(EngineCore_DP0 pid=2006824) INFO 02-02 06:57:00 [gpu_worker.py:375] Available KV cache memory: 10.30 GiB
+(EngineCore_DP0 pid=2006824) INFO 02-02 06:57:00 [kv_cache_utils.py:1291] GPU KV cache size: 42,176 tokens
+(EngineCore_DP0 pid=2006824) INFO 02-02 06:57:00 [kv_cache_utils.py:1296] Maximum concurrency for 40,960 tokens per request: 1.03x
+(EngineCore_DP0 pid=2006824) 
+(EngineCore_DP0 pid=2006824) 

vllm_0004500.log

@@ -7,3 +7,30 @@
 (EngineCore_DP0 pid=2007472) INFO 02-02 06:55:57 [core.py:93] Initializing a V1 LLM engine (v0.13.0) with config: model='Elfsong/VLM_stage_2_iter_0004500', speculative_config=None, tokenizer='Elfsong/VLM_stage_2_iter_0004500', skip_tokenizer_init=False, tokenizer_mode=auto, revision=None, tokenizer_revision=None, trust_remote_code=True, dtype=torch.bfloat16, max_seq_len=40960, download_dir=None, load_format=bitsandbytes, tensor_parallel_size=1, pipeline_parallel_size=1, data_parallel_size=1, disable_custom_all_reduce=False, quantization=bitsandbytes, enforce_eager=False, kv_cache_dtype=auto, device_config=cuda, structured_outputs_config=StructuredOutputsConfig(backend='auto', disable_fallback=False, disable_any_whitespace=False, disable_additional_properties=False, reasoning_parser='', reasoning_parser_plugin='', enable_in_reasoning=False), observability_config=ObservabilityConfig(show_hidden_metrics_for_version=None, otlp_traces_endpoint=None, collect_detailed_traces=None, kv_cache_metrics=False, kv_cache_metrics_sample=0.01, cudagraph_metrics=False, enable_layerwise_nvtx_tracing=False), seed=0, served_model_name=Elfsong/VLM_stage_2_iter_0004500, enable_prefix_caching=True, enable_chunked_prefill=True, pooler_config=None, compilation_config={'level': None, 'mode': <CompilationMode.VLLM_COMPILE: 3>, 'debug_dump_path': None, 'cache_dir': '', 'compile_cache_save_format': 'binary', 'backend': 'inductor', 'custom_ops': ['none'], 'splitting_ops': ['vllm::unified_attention', 'vllm::unified_attention_with_output', 'vllm::unified_mla_attention', 'vllm::unified_mla_attention_with_output', 'vllm::mamba_mixer2', 'vllm::mamba_mixer', 'vllm::short_conv', 'vllm::linear_attention', 'vllm::plamo2_mamba_mixer', 'vllm::gdn_attention_core', 'vllm::kda_attention', 'vllm::sparse_attn_indexer'], 'compile_mm_encoder': False, 'compile_sizes': [], 'compile_ranges_split_points': [8192], 'inductor_compile_config': {'enable_auto_functionalized_v2': False, 'combo_kernels': True, 'benchmark_combo_kernel': True}, 'inductor_passes': {}, 'cudagraph_mode': <CUDAGraphMode.FULL_AND_PIECEWISE: (2, 1)>, 'cudagraph_num_of_warmups': 1, 'cudagraph_capture_sizes': [1, 2, 4, 8, 16, 24, 32, 40, 48, 56, 64, 72, 80, 88, 96, 104, 112, 120, 128, 136, 144, 152, 160, 168, 176, 184, 192, 200, 208, 216, 224, 232, 240, 248, 256, 272, 288, 304, 320, 336, 352, 368, 384, 400, 416, 432, 448, 464, 480, 496, 512], 'cudagraph_copy_inputs': False, 'cudagraph_specialize_lora': True, 'use_inductor_graph_partition': False, 'pass_config': {'fuse_norm_quant': False, 'fuse_act_quant': False, 'fuse_attn_quant': False, 'eliminate_noops': True, 'enable_sp': False, 'fuse_gemm_comms': False, 'fuse_allreduce_rms': False}, 'max_cudagraph_capture_size': 512, 'dynamic_shapes_config': {'type': <DynamicShapesType.BACKED: 'backed'>, 'evaluate_guards': False}, 'local_cache_dir': None}
 (EngineCore_DP0 pid=2007472) INFO 02-02 06:55:59 [parallel_state.py:1203] world_size=1 rank=0 local_rank=0 distributed_init_method=tcp://100.96.20.65:60831 backend=nccl
 (EngineCore_DP0 pid=2007472) INFO 02-02 06:55:59 [parallel_state.py:1411] rank 0 in world size 1 is assigned as DP rank 0, PP rank 0, PCP rank 0, TP rank 0, EP rank 0
+(EngineCore_DP0 pid=2007472) INFO 02-02 06:56:01 [gpu_model_runner.py:3562] Starting to load model Elfsong/VLM_stage_2_iter_0004500...
+(EngineCore_DP0 pid=2007472) INFO 02-02 06:56:03 [cuda.py:351] Using FLASH_ATTN attention backend out of potential backends: ('FLASH_ATTN', 'FLASHINFER', 'TRITON_ATTN', 'FLEX_ATTENTION')
+(EngineCore_DP0 pid=2007472) INFO 02-02 06:56:04 [bitsandbytes_loader.py:791] Loading weights with BitsAndBytes quantization. May take a while ...
+(EngineCore_DP0 pid=2007472) 
+(EngineCore_DP0 pid=2007472) 
+(EngineCore_DP0 pid=2007472) 
+(EngineCore_DP0 pid=2007472) 
+(EngineCore_DP0 pid=2007472) 
+(EngineCore_DP0 pid=2007472) 
+(EngineCore_DP0 pid=2007472) 
+(EngineCore_DP0 pid=2007472) 
+(EngineCore_DP0 pid=2007472) 
+(EngineCore_DP0 pid=2007472) 
+(EngineCore_DP0 pid=2007472) 
+(EngineCore_DP0 pid=2007472) 
+(EngineCore_DP0 pid=2007472) 
+(EngineCore_DP0 pid=2007472) 
+(EngineCore_DP0 pid=2007472) 
+(EngineCore_DP0 pid=2007472) 
+(EngineCore_DP0 pid=2007472) 
+(EngineCore_DP0 pid=2007472) 
+(EngineCore_DP0 pid=2007472) 
+(EngineCore_DP0 pid=2007472) 
+(EngineCore_DP0 pid=2007472) INFO 02-02 06:56:51 [gpu_model_runner.py:3659] Model loading took 19.4031 GiB memory and 48.378088 seconds
+(EngineCore_DP0 pid=2007472) INFO 02-02 06:57:03 [backends.py:643] Using cache directory: /home/mingzhed/.cache/vllm/torch_compile_cache/84789b6c78/rank_0_0/backbone for vLLM's torch.compile
+(EngineCore_DP0 pid=2007472) INFO 02-02 06:57:03 [backends.py:703] Dynamo bytecode transform time: 11.98 s
+(EngineCore_DP0 pid=2007472) INFO 02-02 06:57:12 [backends.py:261] Cache the graph of compile range (1, 8192) for later use

vllm_0005000.log

@@ -4,3 +4,32 @@
 (APIServer pid=2006738) INFO 02-02 06:55:46 [model.py:514] Resolved architecture: Qwen3ForCausalLM
 (APIServer pid=2006738) INFO 02-02 06:55:46 [model.py:1661] Using max model len 40960
 (APIServer pid=2006738) INFO 02-02 06:55:46 [scheduler.py:230] Chunked prefill is enabled with max_num_batched_tokens=8192.
+(EngineCore_DP0 pid=2007863) INFO 02-02 06:56:04 [core.py:93] Initializing a V1 LLM engine (v0.13.0) with config: model='Elfsong/VLM_stage_2_iter_0005000', speculative_config=None, tokenizer='Elfsong/VLM_stage_2_iter_0005000', skip_tokenizer_init=False, tokenizer_mode=auto, revision=None, tokenizer_revision=None, trust_remote_code=True, dtype=torch.bfloat16, max_seq_len=40960, download_dir=None, load_format=bitsandbytes, tensor_parallel_size=1, pipeline_parallel_size=1, data_parallel_size=1, disable_custom_all_reduce=False, quantization=bitsandbytes, enforce_eager=False, kv_cache_dtype=auto, device_config=cuda, structured_outputs_config=StructuredOutputsConfig(backend='auto', disable_fallback=False, disable_any_whitespace=False, disable_additional_properties=False, reasoning_parser='', reasoning_parser_plugin='', enable_in_reasoning=False), observability_config=ObservabilityConfig(show_hidden_metrics_for_version=None, otlp_traces_endpoint=None, collect_detailed_traces=None, kv_cache_metrics=False, kv_cache_metrics_sample=0.01, cudagraph_metrics=False, enable_layerwise_nvtx_tracing=False), seed=0, served_model_name=Elfsong/VLM_stage_2_iter_0005000, enable_prefix_caching=True, enable_chunked_prefill=True, pooler_config=None, compilation_config={'level': None, 'mode': <CompilationMode.VLLM_COMPILE: 3>, 'debug_dump_path': None, 'cache_dir': '', 'compile_cache_save_format': 'binary', 'backend': 'inductor', 'custom_ops': ['none'], 'splitting_ops': ['vllm::unified_attention', 'vllm::unified_attention_with_output', 'vllm::unified_mla_attention', 'vllm::unified_mla_attention_with_output', 'vllm::mamba_mixer2', 'vllm::mamba_mixer', 'vllm::short_conv', 'vllm::linear_attention', 'vllm::plamo2_mamba_mixer', 'vllm::gdn_attention_core', 'vllm::kda_attention', 'vllm::sparse_attn_indexer'], 'compile_mm_encoder': False, 'compile_sizes': [], 'compile_ranges_split_points': [8192], 'inductor_compile_config': {'enable_auto_functionalized_v2': False, 'combo_kernels': True, 'benchmark_combo_kernel': True}, 'inductor_passes': {}, 'cudagraph_mode': <CUDAGraphMode.FULL_AND_PIECEWISE: (2, 1)>, 'cudagraph_num_of_warmups': 1, 'cudagraph_capture_sizes': [1, 2, 4, 8, 16, 24, 32, 40, 48, 56, 64, 72, 80, 88, 96, 104, 112, 120, 128, 136, 144, 152, 160, 168, 176, 184, 192, 200, 208, 216, 224, 232, 240, 248, 256, 272, 288, 304, 320, 336, 352, 368, 384, 400, 416, 432, 448, 464, 480, 496, 512], 'cudagraph_copy_inputs': False, 'cudagraph_specialize_lora': True, 'use_inductor_graph_partition': False, 'pass_config': {'fuse_norm_quant': False, 'fuse_act_quant': False, 'fuse_attn_quant': False, 'eliminate_noops': True, 'enable_sp': False, 'fuse_gemm_comms': False, 'fuse_allreduce_rms': False}, 'max_cudagraph_capture_size': 512, 'dynamic_shapes_config': {'type': <DynamicShapesType.BACKED: 'backed'>, 'evaluate_guards': False}, 'local_cache_dir': None}
+(EngineCore_DP0 pid=2007863) INFO 02-02 06:56:05 [parallel_state.py:1203] world_size=1 rank=0 local_rank=0 distributed_init_method=tcp://100.96.20.65:48783 backend=nccl
+(EngineCore_DP0 pid=2007863) INFO 02-02 06:56:05 [parallel_state.py:1411] rank 0 in world size 1 is assigned as DP rank 0, PP rank 0, PCP rank 0, TP rank 0, EP rank 0
+(EngineCore_DP0 pid=2007863) INFO 02-02 06:56:06 [gpu_model_runner.py:3562] Starting to load model Elfsong/VLM_stage_2_iter_0005000...
+(EngineCore_DP0 pid=2007863) INFO 02-02 06:56:08 [cuda.py:351] Using FLASH_ATTN attention backend out of potential backends: ('FLASH_ATTN', 'FLASHINFER', 'TRITON_ATTN', 'FLEX_ATTENTION')
+(EngineCore_DP0 pid=2007863) INFO 02-02 06:56:08 [bitsandbytes_loader.py:791] Loading weights with BitsAndBytes quantization. May take a while ...
+(EngineCore_DP0 pid=2007863) 
+(EngineCore_DP0 pid=2007863) 
+(EngineCore_DP0 pid=2007863) 
+(EngineCore_DP0 pid=2007863) 
+(EngineCore_DP0 pid=2007863) 
+(EngineCore_DP0 pid=2007863) 
+(EngineCore_DP0 pid=2007863) 
+(EngineCore_DP0 pid=2007863) 
+(EngineCore_DP0 pid=2007863) 
+(EngineCore_DP0 pid=2007863) 
+(EngineCore_DP0 pid=2007863) 
+(EngineCore_DP0 pid=2007863) 
+(EngineCore_DP0 pid=2007863) 
+(EngineCore_DP0 pid=2007863) 
+(EngineCore_DP0 pid=2007863) 
+(EngineCore_DP0 pid=2007863) 
+(EngineCore_DP0 pid=2007863) 
+(EngineCore_DP0 pid=2007863) 
+(EngineCore_DP0 pid=2007863) 
+(EngineCore_DP0 pid=2007863) 
+(EngineCore_DP0 pid=2007863) INFO 02-02 06:56:52 [gpu_model_runner.py:3659] Model loading took 19.4031 GiB memory and 45.155465 seconds
+(EngineCore_DP0 pid=2007863) INFO 02-02 06:57:03 [backends.py:643] Using cache directory: /home/mingzhed/.cache/vllm/torch_compile_cache/6bbfe35b7b/rank_0_0/backbone for vLLM's torch.compile
+(EngineCore_DP0 pid=2007863) INFO 02-02 06:57:03 [backends.py:703] Dynamo bytecode transform time: 10.97 s

vllm_0005500.log

@@ -4,3 +4,270 @@
 (APIServer pid=2007088) INFO 02-02 06:55:47 [model.py:514] Resolved architecture: Qwen3ForCausalLM
 (APIServer pid=2007088) INFO 02-02 06:55:47 [model.py:1661] Using max model len 40960
 (APIServer pid=2007088) INFO 02-02 06:55:47 [scheduler.py:230] Chunked prefill is enabled with max_num_batched_tokens=8192.
+(EngineCore_DP0 pid=2007963) INFO 02-02 06:56:04 [core.py:93] Initializing a V1 LLM engine (v0.13.0) with config: model='Elfsong/VLM_stage_2_iter_0005500', speculative_config=None, tokenizer='Elfsong/VLM_stage_2_iter_0005500', skip_tokenizer_init=False, tokenizer_mode=auto, revision=None, tokenizer_revision=None, trust_remote_code=True, dtype=torch.bfloat16, max_seq_len=40960, download_dir=None, load_format=bitsandbytes, tensor_parallel_size=1, pipeline_parallel_size=1, data_parallel_size=1, disable_custom_all_reduce=False, quantization=bitsandbytes, enforce_eager=False, kv_cache_dtype=auto, device_config=cuda, structured_outputs_config=StructuredOutputsConfig(backend='auto', disable_fallback=False, disable_any_whitespace=False, disable_additional_properties=False, reasoning_parser='', reasoning_parser_plugin='', enable_in_reasoning=False), observability_config=ObservabilityConfig(show_hidden_metrics_for_version=None, otlp_traces_endpoint=None, collect_detailed_traces=None, kv_cache_metrics=False, kv_cache_metrics_sample=0.01, cudagraph_metrics=False, enable_layerwise_nvtx_tracing=False), seed=0, served_model_name=Elfsong/VLM_stage_2_iter_0005500, enable_prefix_caching=True, enable_chunked_prefill=True, pooler_config=None, compilation_config={'level': None, 'mode': <CompilationMode.VLLM_COMPILE: 3>, 'debug_dump_path': None, 'cache_dir': '', 'compile_cache_save_format': 'binary', 'backend': 'inductor', 'custom_ops': ['none'], 'splitting_ops': ['vllm::unified_attention', 'vllm::unified_attention_with_output', 'vllm::unified_mla_attention', 'vllm::unified_mla_attention_with_output', 'vllm::mamba_mixer2', 'vllm::mamba_mixer', 'vllm::short_conv', 'vllm::linear_attention', 'vllm::plamo2_mamba_mixer', 'vllm::gdn_attention_core', 'vllm::kda_attention', 'vllm::sparse_attn_indexer'], 'compile_mm_encoder': False, 'compile_sizes': [], 'compile_ranges_split_points': [8192], 'inductor_compile_config': {'enable_auto_functionalized_v2': False, 'combo_kernels': True, 'benchmark_combo_kernel': True}, 'inductor_passes': {}, 'cudagraph_mode': <CUDAGraphMode.FULL_AND_PIECEWISE: (2, 1)>, 'cudagraph_num_of_warmups': 1, 'cudagraph_capture_sizes': [1, 2, 4, 8, 16, 24, 32, 40, 48, 56, 64, 72, 80, 88, 96, 104, 112, 120, 128, 136, 144, 152, 160, 168, 176, 184, 192, 200, 208, 216, 224, 232, 240, 248, 256, 272, 288, 304, 320, 336, 352, 368, 384, 400, 416, 432, 448, 464, 480, 496, 512], 'cudagraph_copy_inputs': False, 'cudagraph_specialize_lora': True, 'use_inductor_graph_partition': False, 'pass_config': {'fuse_norm_quant': False, 'fuse_act_quant': False, 'fuse_attn_quant': False, 'eliminate_noops': True, 'enable_sp': False, 'fuse_gemm_comms': False, 'fuse_allreduce_rms': False}, 'max_cudagraph_capture_size': 512, 'dynamic_shapes_config': {'type': <DynamicShapesType.BACKED: 'backed'>, 'evaluate_guards': False}, 'local_cache_dir': None}
+(EngineCore_DP0 pid=2007963) INFO 02-02 06:56:05 [parallel_state.py:1203] world_size=1 rank=0 local_rank=0 distributed_init_method=tcp://100.96.20.65:44781 backend=nccl
+(EngineCore_DP0 pid=2007963) INFO 02-02 06:56:05 [parallel_state.py:1411] rank 0 in world size 1 is assigned as DP rank 0, PP rank 0, PCP rank 0, TP rank 0, EP rank 0
+(EngineCore_DP0 pid=2007963) INFO 02-02 06:56:06 [gpu_model_runner.py:3562] Starting to load model Elfsong/VLM_stage_2_iter_0005500...
+(EngineCore_DP0 pid=2007963) INFO 02-02 06:56:08 [cuda.py:351] Using FLASH_ATTN attention backend out of potential backends: ('FLASH_ATTN', 'FLASHINFER', 'TRITON_ATTN', 'FLEX_ATTENTION')
+(EngineCore_DP0 pid=2007963) INFO 02-02 06:56:08 [bitsandbytes_loader.py:791] Loading weights with BitsAndBytes quantization. May take a while ...
+(EngineCore_DP0 pid=2007963) 
+(EngineCore_DP0 pid=2007963) 
+(EngineCore_DP0 pid=2007963) 
+(EngineCore_DP0 pid=2007963) 
+(EngineCore_DP0 pid=2007963) 
+(EngineCore_DP0 pid=2007963) 
+(EngineCore_DP0 pid=2007963) 
+(EngineCore_DP0 pid=2007963) 
+(EngineCore_DP0 pid=2007963) 
+(EngineCore_DP0 pid=2007963) 
+(EngineCore_DP0 pid=2007963) 
+(EngineCore_DP0 pid=2007963) 
+(EngineCore_DP0 pid=2007963) 
+(EngineCore_DP0 pid=2007963) 
+(EngineCore_DP0 pid=2007963) 
+(EngineCore_DP0 pid=2007963) 
+(EngineCore_DP0 pid=2007963) 
+(EngineCore_DP0 pid=2007963) 
+(EngineCore_DP0 pid=2007963) 
+(EngineCore_DP0 pid=2007963) 
+(EngineCore_DP0 pid=2007963) INFO 02-02 06:56:53 [gpu_model_runner.py:3659] Model loading took 19.4031 GiB memory and 46.079784 seconds
+(EngineCore_DP0 pid=2007963) INFO 02-02 06:57:05 [backends.py:643] Using cache directory: /home/mingzhed/.cache/vllm/torch_compile_cache/366f26cb47/rank_0_0/backbone for vLLM's torch.compile
+(EngineCore_DP0 pid=2007963) INFO 02-02 06:57:05 [backends.py:703] Dynamo bytecode transform time: 11.91 s
+(EngineCore_DP0 pid=2007963) ERROR 02-02 06:57:13 [core.py:866] EngineCore failed to start.
+(EngineCore_DP0 pid=2007963) ERROR 02-02 06:57:13 [core.py:866] Traceback (most recent call last):
+(EngineCore_DP0 pid=2007963) ERROR 02-02 06:57:13 [core.py:866]   File "/home/mingzhed/Projects/lm-evaluation-harness/.venv/lib/python3.12/site-packages/torch/_dynamo/convert_frame.py", line 1433, in _compile
+(EngineCore_DP0 pid=2007963) ERROR 02-02 06:57:13 [core.py:866]     guarded_code, tracer_output = compile_inner(code, one_graph, hooks)
+(EngineCore_DP0 pid=2007963) ERROR 02-02 06:57:13 [core.py:866]                                   ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+(EngineCore_DP0 pid=2007963) ERROR 02-02 06:57:13 [core.py:866]   File "/home/mingzhed/Projects/lm-evaluation-harness/.venv/lib/python3.12/site-packages/torch/_utils_internal.py", line 92, in wrapper_function
+(EngineCore_DP0 pid=2007963) ERROR 02-02 06:57:13 [core.py:866]     return function(*args, **kwargs)
+(EngineCore_DP0 pid=2007963) ERROR 02-02 06:57:13 [core.py:866]            ^^^^^^^^^^^^^^^^^^^^^^^^^
+(EngineCore_DP0 pid=2007963) ERROR 02-02 06:57:13 [core.py:866]   File "/home/mingzhed/Projects/lm-evaluation-harness/.venv/lib/python3.12/site-packages/torch/_dynamo/convert_frame.py", line 1117, in compile_inner
+(EngineCore_DP0 pid=2007963) ERROR 02-02 06:57:13 [core.py:866]     return _compile_inner(code, one_graph, hooks)
+(EngineCore_DP0 pid=2007963) ERROR 02-02 06:57:13 [core.py:866]            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+(EngineCore_DP0 pid=2007963) ERROR 02-02 06:57:13 [core.py:866]   File "/home/mingzhed/Projects/lm-evaluation-harness/.venv/lib/python3.12/site-packages/torch/_dynamo/convert_frame.py", line 1251, in _compile_inner
+(EngineCore_DP0 pid=2007963) ERROR 02-02 06:57:13 [core.py:866]     check_fn = dynamo_output.build_guards(
+(EngineCore_DP0 pid=2007963) ERROR 02-02 06:57:13 [core.py:866]                ^^^^^^^^^^^^^^^^^^^^^^^^^^^
+(EngineCore_DP0 pid=2007963) ERROR 02-02 06:57:13 [core.py:866]   File "/home/mingzhed/Projects/lm-evaluation-harness/.venv/lib/python3.12/site-packages/torch/_dynamo/convert_frame.py", line 856, in build_guards
+(EngineCore_DP0 pid=2007963) ERROR 02-02 06:57:13 [core.py:866]     return CheckFunctionManager(
+(EngineCore_DP0 pid=2007963) ERROR 02-02 06:57:13 [core.py:866]            ^^^^^^^^^^^^^^^^^^^^^
+(EngineCore_DP0 pid=2007963) ERROR 02-02 06:57:13 [core.py:866]   File "/home/mingzhed/Projects/lm-evaluation-harness/.venv/lib/python3.12/site-packages/torch/_dynamo/guards.py", line 3339, in __init__
+(EngineCore_DP0 pid=2007963) ERROR 02-02 06:57:13 [core.py:866]     builder, guard_manager = self.build_guards(
+(EngineCore_DP0 pid=2007963) ERROR 02-02 06:57:13 [core.py:866]                              ^^^^^^^^^^^^^^^^^^
+(EngineCore_DP0 pid=2007963) ERROR 02-02 06:57:13 [core.py:866]   File "/home/mingzhed/Projects/lm-evaluation-harness/.venv/lib/python3.12/site-packages/torch/_dynamo/guards.py", line 3674, in build_guards
+(EngineCore_DP0 pid=2007963) ERROR 02-02 06:57:13 [core.py:866]     guard.create(builder)
+(EngineCore_DP0 pid=2007963) ERROR 02-02 06:57:13 [core.py:866]   File "/home/mingzhed/Projects/lm-evaluation-harness/.venv/lib/python3.12/site-packages/torch/_guards.py", line 366, in create
+(EngineCore_DP0 pid=2007963) ERROR 02-02 06:57:13 [core.py:866]     return self.create_fn(builder, self)
+(EngineCore_DP0 pid=2007963) ERROR 02-02 06:57:13 [core.py:866]            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+(EngineCore_DP0 pid=2007963) ERROR 02-02 06:57:13 [core.py:866]   File "/home/mingzhed/Projects/lm-evaluation-harness/.venv/lib/python3.12/site-packages/torch/_dynamo/guards.py", line 2304, in SEQUENCE_LENGTH
+(EngineCore_DP0 pid=2007963) ERROR 02-02 06:57:13 [core.py:866]     len(value), get_verbose_code_parts(code, guard)
+(EngineCore_DP0 pid=2007963) ERROR 02-02 06:57:13 [core.py:866]                 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+(EngineCore_DP0 pid=2007963) ERROR 02-02 06:57:13 [core.py:866]   File "/home/mingzhed/Projects/lm-evaluation-harness/.venv/lib/python3.12/site-packages/torch/_dynamo/guards.py", line 795, in get_verbose_code_parts
+(EngineCore_DP0 pid=2007963) ERROR 02-02 06:57:13 [core.py:866]     get_verbose_code_part(code_part, guard) for code_part in code_parts
+(EngineCore_DP0 pid=2007963) ERROR 02-02 06:57:13 [core.py:866]     ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+(EngineCore_DP0 pid=2007963) ERROR 02-02 06:57:13 [core.py:866]   File "/home/mingzhed/Projects/lm-evaluation-harness/.venv/lib/python3.12/site-packages/torch/_dynamo/guards.py", line 768, in get_verbose_code_part
+(EngineCore_DP0 pid=2007963) ERROR 02-02 06:57:13 [core.py:866]     extra = f"  # {format_frame(fs, line=True)}"
+(EngineCore_DP0 pid=2007963) ERROR 02-02 06:57:13 [core.py:866]                    ^^^^^^^^^^^^^^^^^^^^^^^^^^^
+(EngineCore_DP0 pid=2007963) ERROR 02-02 06:57:13 [core.py:866]   File "/home/mingzhed/Projects/lm-evaluation-harness/.venv/lib/python3.12/site-packages/torch/utils/_traceback.py", line 156, in format_frame
+(EngineCore_DP0 pid=2007963) ERROR 02-02 06:57:13 [core.py:866]     return f"{extra_line}{shorten_filename(frame.filename, base=base)}:{frame.lineno} in {frame.name}"
+(EngineCore_DP0 pid=2007963) ERROR 02-02 06:57:13 [core.py:866]                           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+(EngineCore_DP0 pid=2007963) ERROR 02-02 06:57:13 [core.py:866]   File "/home/mingzhed/Projects/lm-evaluation-harness/.venv/lib/python3.12/site-packages/torch/utils/_traceback.py", line 140, in shorten_filename
+(EngineCore_DP0 pid=2007963) ERROR 02-02 06:57:13 [core.py:866]     prefix = os.path.commonpath([fn, base])
+(EngineCore_DP0 pid=2007963) ERROR 02-02 06:57:13 [core.py:866]              ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+(EngineCore_DP0 pid=2007963) ERROR 02-02 06:57:13 [core.py:866]   File "<frozen posixpath>", line 560, in commonpath
+(EngineCore_DP0 pid=2007963) ERROR 02-02 06:57:13 [core.py:866]   File "/home/mingzhed/Projects/lm-evaluation-harness/.venv/lib/python3.12/site-packages/vllm/v1/engine/core.py", line 838, in signal_handler
+(EngineCore_DP0 pid=2007963) ERROR 02-02 06:57:13 [core.py:866]     raise SystemExit()
+(EngineCore_DP0 pid=2007963) ERROR 02-02 06:57:13 [core.py:866] SystemExit
+(EngineCore_DP0 pid=2007963) ERROR 02-02 06:57:13 [core.py:866] 
+(EngineCore_DP0 pid=2007963) ERROR 02-02 06:57:13 [core.py:866] During handling of the above exception, another exception occurred:
+(EngineCore_DP0 pid=2007963) ERROR 02-02 06:57:13 [core.py:866] 
+(EngineCore_DP0 pid=2007963) ERROR 02-02 06:57:13 [core.py:866] Traceback (most recent call last):
+(EngineCore_DP0 pid=2007963) ERROR 02-02 06:57:13 [core.py:866]   File "/home/mingzhed/Projects/lm-evaluation-harness/.venv/lib/python3.12/site-packages/vllm/v1/engine/core.py", line 857, in run_engine_core
+(EngineCore_DP0 pid=2007963) ERROR 02-02 06:57:13 [core.py:866]     engine_core = EngineCoreProc(*args, **kwargs)
+(EngineCore_DP0 pid=2007963) ERROR 02-02 06:57:13 [core.py:866]                   ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+(EngineCore_DP0 pid=2007963) ERROR 02-02 06:57:13 [core.py:866]   File "/home/mingzhed/Projects/lm-evaluation-harness/.venv/lib/python3.12/site-packages/vllm/v1/engine/core.py", line 637, in __init__
+(EngineCore_DP0 pid=2007963) ERROR 02-02 06:57:13 [core.py:866]     super().__init__(
+(EngineCore_DP0 pid=2007963) ERROR 02-02 06:57:13 [core.py:866]   File "/home/mingzhed/Projects/lm-evaluation-harness/.venv/lib/python3.12/site-packages/vllm/v1/engine/core.py", line 109, in __init__
+(EngineCore_DP0 pid=2007963) ERROR 02-02 06:57:13 [core.py:866]     num_gpu_blocks, num_cpu_blocks, kv_cache_config = self._initialize_kv_caches(
+(EngineCore_DP0 pid=2007963) ERROR 02-02 06:57:13 [core.py:866]                                                       ^^^^^^^^^^^^^^^^^^^^^^^^^^^
+(EngineCore_DP0 pid=2007963) ERROR 02-02 06:57:13 [core.py:866]   File "/home/mingzhed/Projects/lm-evaluation-harness/.venv/lib/python3.12/site-packages/vllm/v1/engine/core.py", line 240, in _initialize_kv_caches
+(EngineCore_DP0 pid=2007963) ERROR 02-02 06:57:13 [core.py:866]     available_gpu_memory = self.model_executor.determine_available_memory()
+(EngineCore_DP0 pid=2007963) ERROR 02-02 06:57:13 [core.py:866]                            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+(EngineCore_DP0 pid=2007963) ERROR 02-02 06:57:13 [core.py:866]   File "/home/mingzhed/Projects/lm-evaluation-harness/.venv/lib/python3.12/site-packages/vllm/v1/executor/abstract.py", line 126, in determine_available_memory
+(EngineCore_DP0 pid=2007963) ERROR 02-02 06:57:13 [core.py:866]     return self.collective_rpc("determine_available_memory")
+(EngineCore_DP0 pid=2007963) ERROR 02-02 06:57:13 [core.py:866]            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+(EngineCore_DP0 pid=2007963) ERROR 02-02 06:57:13 [core.py:866]   File "/home/mingzhed/Projects/lm-evaluation-harness/.venv/lib/python3.12/site-packages/vllm/v1/executor/uniproc_executor.py", line 75, in collective_rpc
+(EngineCore_DP0 pid=2007963) ERROR 02-02 06:57:13 [core.py:866]     result = run_method(self.driver_worker, method, args, kwargs)
+(EngineCore_DP0 pid=2007963) ERROR 02-02 06:57:13 [core.py:866]              ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+(EngineCore_DP0 pid=2007963) ERROR 02-02 06:57:13 [core.py:866]   File "/home/mingzhed/Projects/lm-evaluation-harness/.venv/lib/python3.12/site-packages/vllm/v1/serial_utils.py", line 461, in run_method
+(EngineCore_DP0 pid=2007963) ERROR 02-02 06:57:13 [core.py:866]     return func(*args, **kwargs)
+(EngineCore_DP0 pid=2007963) ERROR 02-02 06:57:13 [core.py:866]            ^^^^^^^^^^^^^^^^^^^^^
+(EngineCore_DP0 pid=2007963) ERROR 02-02 06:57:13 [core.py:866]   File "/home/mingzhed/Projects/lm-evaluation-harness/.venv/lib/python3.12/site-packages/torch/utils/_contextlib.py", line 120, in decorate_context
+(EngineCore_DP0 pid=2007963) ERROR 02-02 06:57:13 [core.py:866]     return func(*args, **kwargs)
+(EngineCore_DP0 pid=2007963) ERROR 02-02 06:57:13 [core.py:866]            ^^^^^^^^^^^^^^^^^^^^^
+(EngineCore_DP0 pid=2007963) ERROR 02-02 06:57:13 [core.py:866]   File "/home/mingzhed/Projects/lm-evaluation-harness/.venv/lib/python3.12/site-packages/vllm/v1/worker/gpu_worker.py", line 340, in determine_available_memory
+(EngineCore_DP0 pid=2007963) ERROR 02-02 06:57:13 [core.py:866]     self.model_runner.profile_run()
+(EngineCore_DP0 pid=2007963) ERROR 02-02 06:57:13 [core.py:866]   File "/home/mingzhed/Projects/lm-evaluation-harness/.venv/lib/python3.12/site-packages/vllm/v1/worker/gpu_model_runner.py", line 4474, in profile_run
+(EngineCore_DP0 pid=2007963) ERROR 02-02 06:57:13 [core.py:866]     hidden_states, last_hidden_states = self._dummy_run(
+(EngineCore_DP0 pid=2007963) ERROR 02-02 06:57:13 [core.py:866]                                         ^^^^^^^^^^^^^^^^
+(EngineCore_DP0 pid=2007963) ERROR 02-02 06:57:13 [core.py:866]   File "/home/mingzhed/Projects/lm-evaluation-harness/.venv/lib/python3.12/site-packages/torch/utils/_contextlib.py", line 120, in decorate_context
+(EngineCore_DP0 pid=2007963) ERROR 02-02 06:57:13 [core.py:866]     return func(*args, **kwargs)
+(EngineCore_DP0 pid=2007963) ERROR 02-02 06:57:13 [core.py:866]            ^^^^^^^^^^^^^^^^^^^^^
+(EngineCore_DP0 pid=2007963) ERROR 02-02 06:57:13 [core.py:866]   File "/home/mingzhed/Projects/lm-evaluation-harness/.venv/lib/python3.12/site-packages/vllm/v1/worker/gpu_model_runner.py", line 4198, in _dummy_run
+(EngineCore_DP0 pid=2007963) ERROR 02-02 06:57:13 [core.py:866]     outputs = self.model(
+(EngineCore_DP0 pid=2007963) ERROR 02-02 06:57:13 [core.py:866]               ^^^^^^^^^^^
+(EngineCore_DP0 pid=2007963) ERROR 02-02 06:57:13 [core.py:866]   File "/home/mingzhed/Projects/lm-evaluation-harness/.venv/lib/python3.12/site-packages/vllm/compilation/cuda_graph.py", line 220, in __call__
+(EngineCore_DP0 pid=2007963) ERROR 02-02 06:57:13 [core.py:866]     return self.runnable(*args, **kwargs)
+(EngineCore_DP0 pid=2007963) ERROR 02-02 06:57:13 [core.py:866]            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+(EngineCore_DP0 pid=2007963) ERROR 02-02 06:57:13 [core.py:866]   File "/home/mingzhed/Projects/lm-evaluation-harness/.venv/lib/python3.12/site-packages/torch/nn/modules/module.py", line 1775, in _wrapped_call_impl
+(EngineCore_DP0 pid=2007963) ERROR 02-02 06:57:13 [core.py:866]     return self._call_impl(*args, **kwargs)
+(EngineCore_DP0 pid=2007963) ERROR 02-02 06:57:13 [core.py:866]            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+(EngineCore_DP0 pid=2007963) ERROR 02-02 06:57:13 [core.py:866]   File "/home/mingzhed/Projects/lm-evaluation-harness/.venv/lib/python3.12/site-packages/torch/nn/modules/module.py", line 1786, in _call_impl
+(EngineCore_DP0 pid=2007963) ERROR 02-02 06:57:13 [core.py:866]     return forward_call(*args, **kwargs)
+(EngineCore_DP0 pid=2007963) ERROR 02-02 06:57:13 [core.py:866]            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+(EngineCore_DP0 pid=2007963) ERROR 02-02 06:57:13 [core.py:866]   File "/home/mingzhed/Projects/lm-evaluation-harness/.venv/lib/python3.12/site-packages/vllm/model_executor/models/qwen3.py", line 314, in forward
+(EngineCore_DP0 pid=2007963) ERROR 02-02 06:57:13 [core.py:866]     hidden_states = self.model(
+(EngineCore_DP0 pid=2007963) ERROR 02-02 06:57:13 [core.py:866]                     ^^^^^^^^^^^
+(EngineCore_DP0 pid=2007963) ERROR 02-02 06:57:13 [core.py:866]   File "/home/mingzhed/Projects/lm-evaluation-harness/.venv/lib/python3.12/site-packages/vllm/compilation/decorators.py", line 526, in __call__
+(EngineCore_DP0 pid=2007963) ERROR 02-02 06:57:13 [core.py:866]     output = TorchCompileWithNoGuardsWrapper.__call__(self, *args, **kwargs)
+(EngineCore_DP0 pid=2007963) ERROR 02-02 06:57:13 [core.py:866]              ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+(EngineCore_DP0 pid=2007963) ERROR 02-02 06:57:13 [core.py:866]   File "/home/mingzhed/Projects/lm-evaluation-harness/.venv/lib/python3.12/site-packages/vllm/compilation/wrapper.py", line 218, in __call__
+(EngineCore_DP0 pid=2007963) ERROR 02-02 06:57:13 [core.py:866]     return self._call_with_optional_nvtx_range(
+(EngineCore_DP0 pid=2007963) ERROR 02-02 06:57:13 [core.py:866]            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+(EngineCore_DP0 pid=2007963) ERROR 02-02 06:57:13 [core.py:866]   File "/home/mingzhed/Projects/lm-evaluation-harness/.venv/lib/python3.12/site-packages/vllm/compilation/wrapper.py", line 109, in _call_with_optional_nvtx_range
+(EngineCore_DP0 pid=2007963) ERROR 02-02 06:57:13 [core.py:866]     return callable_fn(*args, **kwargs)
+(EngineCore_DP0 pid=2007963) ERROR 02-02 06:57:13 [core.py:866]            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+(EngineCore_DP0 pid=2007963) ERROR 02-02 06:57:13 [core.py:866]   File "/home/mingzhed/Projects/lm-evaluation-harness/.venv/lib/python3.12/site-packages/torch/_dynamo/eval_frame.py", line 832, in compile_wrapper
+(EngineCore_DP0 pid=2007963) ERROR 02-02 06:57:13 [core.py:866]     return fn(*args, **kwargs)
+(EngineCore_DP0 pid=2007963) ERROR 02-02 06:57:13 [core.py:866]            ^^^^^^^^^^^^^^^^^^^
+(EngineCore_DP0 pid=2007963) ERROR 02-02 06:57:13 [core.py:866]   File "/home/mingzhed/Projects/lm-evaluation-harness/.venv/lib/python3.12/site-packages/torch/_dynamo/convert_frame.py", line 1874, in __call__
+(EngineCore_DP0 pid=2007963) ERROR 02-02 06:57:13 [core.py:866]     result = self._torchdynamo_orig_backend(
+(EngineCore_DP0 pid=2007963) ERROR 02-02 06:57:13 [core.py:866]              ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+(EngineCore_DP0 pid=2007963) ERROR 02-02 06:57:13 [core.py:866]   File "/home/mingzhed/Projects/lm-evaluation-harness/.venv/lib/python3.12/site-packages/torch/_dynamo/convert_frame.py", line 688, in __call__
+(EngineCore_DP0 pid=2007963) ERROR 02-02 06:57:13 [core.py:866]     result = _compile(
+(EngineCore_DP0 pid=2007963) ERROR 02-02 06:57:13 [core.py:866]              ^^^^^^^^^
+(EngineCore_DP0 pid=2007963) ERROR 02-02 06:57:13 [core.py:866]   File "/home/mingzhed/Projects/lm-evaluation-harness/.venv/lib/python3.12/site-packages/torch/_dynamo/convert_frame.py", line 1509, in _compile
+(EngineCore_DP0 pid=2007963) ERROR 02-02 06:57:13 [core.py:866]     if tracer_output:
+(EngineCore_DP0 pid=2007963) ERROR 02-02 06:57:13 [core.py:866]        ^^^^^^^^^^^^^
+(EngineCore_DP0 pid=2007963) ERROR 02-02 06:57:13 [core.py:866] UnboundLocalError: cannot access local variable 'tracer_output' where it is not associated with a value
+(EngineCore_DP0 pid=2007963) Process EngineCore_DP0:
+(EngineCore_DP0 pid=2007963) Traceback (most recent call last):
+(EngineCore_DP0 pid=2007963)   File "/home/mingzhed/Projects/lm-evaluation-harness/.venv/lib/python3.12/site-packages/torch/_dynamo/convert_frame.py", line 1433, in _compile
+(EngineCore_DP0 pid=2007963)     guarded_code, tracer_output = compile_inner(code, one_graph, hooks)
+(EngineCore_DP0 pid=2007963)                                   ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+(EngineCore_DP0 pid=2007963)   File "/home/mingzhed/Projects/lm-evaluation-harness/.venv/lib/python3.12/site-packages/torch/_utils_internal.py", line 92, in wrapper_function
+(EngineCore_DP0 pid=2007963)     return function(*args, **kwargs)
+(EngineCore_DP0 pid=2007963)            ^^^^^^^^^^^^^^^^^^^^^^^^^
+(EngineCore_DP0 pid=2007963)   File "/home/mingzhed/Projects/lm-evaluation-harness/.venv/lib/python3.12/site-packages/torch/_dynamo/convert_frame.py", line 1117, in compile_inner
+(EngineCore_DP0 pid=2007963)     return _compile_inner(code, one_graph, hooks)
+(EngineCore_DP0 pid=2007963)            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+(EngineCore_DP0 pid=2007963)   File "/home/mingzhed/Projects/lm-evaluation-harness/.venv/lib/python3.12/site-packages/torch/_dynamo/convert_frame.py", line 1251, in _compile_inner
+(EngineCore_DP0 pid=2007963)     check_fn = dynamo_output.build_guards(
+(EngineCore_DP0 pid=2007963)                ^^^^^^^^^^^^^^^^^^^^^^^^^^^
+(EngineCore_DP0 pid=2007963)   File "/home/mingzhed/Projects/lm-evaluation-harness/.venv/lib/python3.12/site-packages/torch/_dynamo/convert_frame.py", line 856, in build_guards
+(EngineCore_DP0 pid=2007963)     return CheckFunctionManager(
+(EngineCore_DP0 pid=2007963)            ^^^^^^^^^^^^^^^^^^^^^
+(EngineCore_DP0 pid=2007963)   File "/home/mingzhed/Projects/lm-evaluation-harness/.venv/lib/python3.12/site-packages/torch/_dynamo/guards.py", line 3339, in __init__
+(EngineCore_DP0 pid=2007963)     builder, guard_manager = self.build_guards(
+(EngineCore_DP0 pid=2007963)                              ^^^^^^^^^^^^^^^^^^
+(EngineCore_DP0 pid=2007963)   File "/home/mingzhed/Projects/lm-evaluation-harness/.venv/lib/python3.12/site-packages/torch/_dynamo/guards.py", line 3674, in build_guards
+(EngineCore_DP0 pid=2007963)     guard.create(builder)
+(EngineCore_DP0 pid=2007963)   File "/home/mingzhed/Projects/lm-evaluation-harness/.venv/lib/python3.12/site-packages/torch/_guards.py", line 366, in create
+(EngineCore_DP0 pid=2007963)     return self.create_fn(builder, self)
+(EngineCore_DP0 pid=2007963)            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+(EngineCore_DP0 pid=2007963)   File "/home/mingzhed/Projects/lm-evaluation-harness/.venv/lib/python3.12/site-packages/torch/_dynamo/guards.py", line 2304, in SEQUENCE_LENGTH
+(EngineCore_DP0 pid=2007963)     len(value), get_verbose_code_parts(code, guard)
+(EngineCore_DP0 pid=2007963)                 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+(EngineCore_DP0 pid=2007963)   File "/home/mingzhed/Projects/lm-evaluation-harness/.venv/lib/python3.12/site-packages/torch/_dynamo/guards.py", line 795, in get_verbose_code_parts
+(EngineCore_DP0 pid=2007963)     get_verbose_code_part(code_part, guard) for code_part in code_parts
+(EngineCore_DP0 pid=2007963)     ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+(EngineCore_DP0 pid=2007963)   File "/home/mingzhed/Projects/lm-evaluation-harness/.venv/lib/python3.12/site-packages/torch/_dynamo/guards.py", line 768, in get_verbose_code_part
+(EngineCore_DP0 pid=2007963)     extra = f"  # {format_frame(fs, line=True)}"
+(EngineCore_DP0 pid=2007963)                    ^^^^^^^^^^^^^^^^^^^^^^^^^^^
+(EngineCore_DP0 pid=2007963)   File "/home/mingzhed/Projects/lm-evaluation-harness/.venv/lib/python3.12/site-packages/torch/utils/_traceback.py", line 156, in format_frame
+(EngineCore_DP0 pid=2007963)     return f"{extra_line}{shorten_filename(frame.filename, base=base)}:{frame.lineno} in {frame.name}"
+(EngineCore_DP0 pid=2007963)                           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+(EngineCore_DP0 pid=2007963)   File "/home/mingzhed/Projects/lm-evaluation-harness/.venv/lib/python3.12/site-packages/torch/utils/_traceback.py", line 140, in shorten_filename
+(EngineCore_DP0 pid=2007963)     prefix = os.path.commonpath([fn, base])
+(EngineCore_DP0 pid=2007963)              ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+(EngineCore_DP0 pid=2007963)   File "<frozen posixpath>", line 560, in commonpath
+(EngineCore_DP0 pid=2007963)   File "/home/mingzhed/Projects/lm-evaluation-harness/.venv/lib/python3.12/site-packages/vllm/v1/engine/core.py", line 838, in signal_handler
+(EngineCore_DP0 pid=2007963)     raise SystemExit()
+(EngineCore_DP0 pid=2007963) SystemExit
+(EngineCore_DP0 pid=2007963) 
+(EngineCore_DP0 pid=2007963) During handling of the above exception, another exception occurred:
+(EngineCore_DP0 pid=2007963) 
+(EngineCore_DP0 pid=2007963) Traceback (most recent call last):
+(EngineCore_DP0 pid=2007963)   File "/home/mingzhed/.local/share/uv/python/cpython-3.12.12-linux-x86_64-gnu/lib/python3.12/multiprocessing/process.py", line 314, in _bootstrap
+(EngineCore_DP0 pid=2007963)     self.run()
+(EngineCore_DP0 pid=2007963)   File "/home/mingzhed/.local/share/uv/python/cpython-3.12.12-linux-x86_64-gnu/lib/python3.12/multiprocessing/process.py", line 108, in run
+(EngineCore_DP0 pid=2007963)     self._target(*self._args, **self._kwargs)
+(EngineCore_DP0 pid=2007963)   File "/home/mingzhed/Projects/lm-evaluation-harness/.venv/lib/python3.12/site-packages/vllm/v1/engine/core.py", line 870, in run_engine_core
+(EngineCore_DP0 pid=2007963)     raise e
+(EngineCore_DP0 pid=2007963)   File "/home/mingzhed/Projects/lm-evaluation-harness/.venv/lib/python3.12/site-packages/vllm/v1/engine/core.py", line 857, in run_engine_core
+(EngineCore_DP0 pid=2007963)     engine_core = EngineCoreProc(*args, **kwargs)
+(EngineCore_DP0 pid=2007963)                   ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+(EngineCore_DP0 pid=2007963)   File "/home/mingzhed/Projects/lm-evaluation-harness/.venv/lib/python3.12/site-packages/vllm/v1/engine/core.py", line 637, in __init__
+(EngineCore_DP0 pid=2007963)     super().__init__(
+(EngineCore_DP0 pid=2007963)   File "/home/mingzhed/Projects/lm-evaluation-harness/.venv/lib/python3.12/site-packages/vllm/v1/engine/core.py", line 109, in __init__
+(EngineCore_DP0 pid=2007963)     num_gpu_blocks, num_cpu_blocks, kv_cache_config = self._initialize_kv_caches(
+(EngineCore_DP0 pid=2007963)                                                       ^^^^^^^^^^^^^^^^^^^^^^^^^^^
+(EngineCore_DP0 pid=2007963)   File "/home/mingzhed/Projects/lm-evaluation-harness/.venv/lib/python3.12/site-packages/vllm/v1/engine/core.py", line 240, in _initialize_kv_caches
+(EngineCore_DP0 pid=2007963)     available_gpu_memory = self.model_executor.determine_available_memory()
+(EngineCore_DP0 pid=2007963)                            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+(EngineCore_DP0 pid=2007963)   File "/home/mingzhed/Projects/lm-evaluation-harness/.venv/lib/python3.12/site-packages/vllm/v1/executor/abstract.py", line 126, in determine_available_memory
+(EngineCore_DP0 pid=2007963)     return self.collective_rpc("determine_available_memory")
+(EngineCore_DP0 pid=2007963)            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+(EngineCore_DP0 pid=2007963)   File "/home/mingzhed/Projects/lm-evaluation-harness/.venv/lib/python3.12/site-packages/vllm/v1/executor/uniproc_executor.py", line 75, in collective_rpc
+(EngineCore_DP0 pid=2007963)     result = run_method(self.driver_worker, method, args, kwargs)
+(EngineCore_DP0 pid=2007963)              ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+(EngineCore_DP0 pid=2007963)   File "/home/mingzhed/Projects/lm-evaluation-harness/.venv/lib/python3.12/site-packages/vllm/v1/serial_utils.py", line 461, in run_method
+(EngineCore_DP0 pid=2007963)     return func(*args, **kwargs)
+(EngineCore_DP0 pid=2007963)            ^^^^^^^^^^^^^^^^^^^^^
+(EngineCore_DP0 pid=2007963)   File "/home/mingzhed/Projects/lm-evaluation-harness/.venv/lib/python3.12/site-packages/torch/utils/_contextlib.py", line 120, in decorate_context
+(EngineCore_DP0 pid=2007963)     return func(*args, **kwargs)
+(EngineCore_DP0 pid=2007963)            ^^^^^^^^^^^^^^^^^^^^^
+(EngineCore_DP0 pid=2007963)   File "/home/mingzhed/Projects/lm-evaluation-harness/.venv/lib/python3.12/site-packages/vllm/v1/worker/gpu_worker.py", line 340, in determine_available_memory
+(EngineCore_DP0 pid=2007963)     self.model_runner.profile_run()
+(EngineCore_DP0 pid=2007963)   File "/home/mingzhed/Projects/lm-evaluation-harness/.venv/lib/python3.12/site-packages/vllm/v1/worker/gpu_model_runner.py", line 4474, in profile_run
+(EngineCore_DP0 pid=2007963)     hidden_states, last_hidden_states = self._dummy_run(
+(EngineCore_DP0 pid=2007963)                                         ^^^^^^^^^^^^^^^^
+(EngineCore_DP0 pid=2007963)   File "/home/mingzhed/Projects/lm-evaluation-harness/.venv/lib/python3.12/site-packages/torch/utils/_contextlib.py", line 120, in decorate_context
+(EngineCore_DP0 pid=2007963)     return func(*args, **kwargs)
+(EngineCore_DP0 pid=2007963)            ^^^^^^^^^^^^^^^^^^^^^
+(EngineCore_DP0 pid=2007963)   File "/home/mingzhed/Projects/lm-evaluation-harness/.venv/lib/python3.12/site-packages/vllm/v1/worker/gpu_model_runner.py", line 4198, in _dummy_run
+(EngineCore_DP0 pid=2007963)     outputs = self.model(
+(EngineCore_DP0 pid=2007963)               ^^^^^^^^^^^
+(EngineCore_DP0 pid=2007963)   File "/home/mingzhed/Projects/lm-evaluation-harness/.venv/lib/python3.12/site-packages/vllm/compilation/cuda_graph.py", line 220, in __call__
+(EngineCore_DP0 pid=2007963)     return self.runnable(*args, **kwargs)
+(EngineCore_DP0 pid=2007963)            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+(EngineCore_DP0 pid=2007963)   File "/home/mingzhed/Projects/lm-evaluation-harness/.venv/lib/python3.12/site-packages/torch/nn/modules/module.py", line 1775, in _wrapped_call_impl
+(EngineCore_DP0 pid=2007963)     return self._call_impl(*args, **kwargs)
+(EngineCore_DP0 pid=2007963)            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+(EngineCore_DP0 pid=2007963)   File "/home/mingzhed/Projects/lm-evaluation-harness/.venv/lib/python3.12/site-packages/torch/nn/modules/module.py", line 1786, in _call_impl
+(EngineCore_DP0 pid=2007963)     return forward_call(*args, **kwargs)
+(EngineCore_DP0 pid=2007963)            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+(EngineCore_DP0 pid=2007963)   File "/home/mingzhed/Projects/lm-evaluation-harness/.venv/lib/python3.12/site-packages/vllm/model_executor/models/qwen3.py", line 314, in forward
+(EngineCore_DP0 pid=2007963)     hidden_states = self.model(
+(EngineCore_DP0 pid=2007963)                     ^^^^^^^^^^^
+(EngineCore_DP0 pid=2007963)   File "/home/mingzhed/Projects/lm-evaluation-harness/.venv/lib/python3.12/site-packages/vllm/compilation/decorators.py", line 526, in __call__
+(EngineCore_DP0 pid=2007963)     output = TorchCompileWithNoGuardsWrapper.__call__(self, *args, **kwargs)
+(EngineCore_DP0 pid=2007963)              ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+(EngineCore_DP0 pid=2007963)   File "/home/mingzhed/Projects/lm-evaluation-harness/.venv/lib/python3.12/site-packages/vllm/compilation/wrapper.py", line 218, in __call__
+(EngineCore_DP0 pid=2007963)     return self._call_with_optional_nvtx_range(
+(EngineCore_DP0 pid=2007963)            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+(EngineCore_DP0 pid=2007963)   File "/home/mingzhed/Projects/lm-evaluation-harness/.venv/lib/python3.12/site-packages/vllm/compilation/wrapper.py", line 109, in _call_with_optional_nvtx_range
+(EngineCore_DP0 pid=2007963)     return callable_fn(*args, **kwargs)
+(EngineCore_DP0 pid=2007963)            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+(EngineCore_DP0 pid=2007963)   File "/home/mingzhed/Projects/lm-evaluation-harness/.venv/lib/python3.12/site-packages/torch/_dynamo/eval_frame.py", line 832, in compile_wrapper
+(EngineCore_DP0 pid=2007963)     return fn(*args, **kwargs)
+(EngineCore_DP0 pid=2007963)            ^^^^^^^^^^^^^^^^^^^
+(EngineCore_DP0 pid=2007963)   File "/home/mingzhed/Projects/lm-evaluation-harness/.venv/lib/python3.12/site-packages/torch/_dynamo/convert_frame.py", line 1874, in __call__
+(EngineCore_DP0 pid=2007963)     result = self._torchdynamo_orig_backend(
+(EngineCore_DP0 pid=2007963)              ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+(EngineCore_DP0 pid=2007963)   File "/home/mingzhed/Projects/lm-evaluation-harness/.venv/lib/python3.12/site-packages/torch/_dynamo/convert_frame.py", line 688, in __call__
+(EngineCore_DP0 pid=2007963)     result = _compile(
+(EngineCore_DP0 pid=2007963)              ^^^^^^^^^
+(EngineCore_DP0 pid=2007963)   File "/home/mingzhed/Projects/lm-evaluation-harness/.venv/lib/python3.12/site-packages/torch/_dynamo/convert_frame.py", line 1509, in _compile
+(EngineCore_DP0 pid=2007963)     if tracer_output:
+(EngineCore_DP0 pid=2007963)        ^^^^^^^^^^^^^
+(EngineCore_DP0 pid=2007963) UnboundLocalError: cannot access local variable 'tracer_output' where it is not associated with a value

vllm_0006500.log

@@ -0,0 +1,29 @@
+(APIServer pid=2007746) INFO 02-02 06:56:02 [api_server.py:1351] vLLM API server version 0.13.0
+(APIServer pid=2007746) INFO 02-02 06:56:02 [utils.py:253] non-default args: {'port': 9011, 'model': 'Elfsong/VLM_stage_2_iter_0006500', 'trust_remote_code': True, 'quantization': 'bitsandbytes', 'gpu_memory_utilization': 0.4}
+(APIServer pid=2007746) The argument `trust_remote_code` is to be used with Auto classes. It has no effect here and is ignored.
+(APIServer pid=2007746) INFO 02-02 06:56:04 [model.py:514] Resolved architecture: Qwen3ForCausalLM
+(APIServer pid=2007746) INFO 02-02 06:56:04 [model.py:1661] Using max model len 40960
+(APIServer pid=2007746) INFO 02-02 06:56:04 [scheduler.py:230] Chunked prefill is enabled with max_num_batched_tokens=8192.
+(EngineCore_DP0 pid=2009129) INFO 02-02 06:56:24 [core.py:93] Initializing a V1 LLM engine (v0.13.0) with config: model='Elfsong/VLM_stage_2_iter_0006500', speculative_config=None, tokenizer='Elfsong/VLM_stage_2_iter_0006500', skip_tokenizer_init=False, tokenizer_mode=auto, revision=None, tokenizer_revision=None, trust_remote_code=True, dtype=torch.bfloat16, max_seq_len=40960, download_dir=None, load_format=bitsandbytes, tensor_parallel_size=1, pipeline_parallel_size=1, data_parallel_size=1, disable_custom_all_reduce=False, quantization=bitsandbytes, enforce_eager=False, kv_cache_dtype=auto, device_config=cuda, structured_outputs_config=StructuredOutputsConfig(backend='auto', disable_fallback=False, disable_any_whitespace=False, disable_additional_properties=False, reasoning_parser='', reasoning_parser_plugin='', enable_in_reasoning=False), observability_config=ObservabilityConfig(show_hidden_metrics_for_version=None, otlp_traces_endpoint=None, collect_detailed_traces=None, kv_cache_metrics=False, kv_cache_metrics_sample=0.01, cudagraph_metrics=False, enable_layerwise_nvtx_tracing=False), seed=0, served_model_name=Elfsong/VLM_stage_2_iter_0006500, enable_prefix_caching=True, enable_chunked_prefill=True, pooler_config=None, compilation_config={'level': None, 'mode': <CompilationMode.VLLM_COMPILE: 3>, 'debug_dump_path': None, 'cache_dir': '', 'compile_cache_save_format': 'binary', 'backend': 'inductor', 'custom_ops': ['none'], 'splitting_ops': ['vllm::unified_attention', 'vllm::unified_attention_with_output', 'vllm::unified_mla_attention', 'vllm::unified_mla_attention_with_output', 'vllm::mamba_mixer2', 'vllm::mamba_mixer', 'vllm::short_conv', 'vllm::linear_attention', 'vllm::plamo2_mamba_mixer', 'vllm::gdn_attention_core', 'vllm::kda_attention', 'vllm::sparse_attn_indexer'], 'compile_mm_encoder': False, 'compile_sizes': [], 'compile_ranges_split_points': [8192], 'inductor_compile_config': {'enable_auto_functionalized_v2': False, 'combo_kernels': True, 'benchmark_combo_kernel': True}, 'inductor_passes': {}, 'cudagraph_mode': <CUDAGraphMode.FULL_AND_PIECEWISE: (2, 1)>, 'cudagraph_num_of_warmups': 1, 'cudagraph_capture_sizes': [1, 2, 4, 8, 16, 24, 32, 40, 48, 56, 64, 72, 80, 88, 96, 104, 112, 120, 128, 136, 144, 152, 160, 168, 176, 184, 192, 200, 208, 216, 224, 232, 240, 248, 256, 272, 288, 304, 320, 336, 352, 368, 384, 400, 416, 432, 448, 464, 480, 496, 512], 'cudagraph_copy_inputs': False, 'cudagraph_specialize_lora': True, 'use_inductor_graph_partition': False, 'pass_config': {'fuse_norm_quant': False, 'fuse_act_quant': False, 'fuse_attn_quant': False, 'eliminate_noops': True, 'enable_sp': False, 'fuse_gemm_comms': False, 'fuse_allreduce_rms': False}, 'max_cudagraph_capture_size': 512, 'dynamic_shapes_config': {'type': <DynamicShapesType.BACKED: 'backed'>, 'evaluate_guards': False}, 'local_cache_dir': None}
+(EngineCore_DP0 pid=2009129) INFO 02-02 06:56:26 [parallel_state.py:1203] world_size=1 rank=0 local_rank=0 distributed_init_method=tcp://100.96.20.65:53439 backend=nccl
+(EngineCore_DP0 pid=2009129) INFO 02-02 06:56:26 [parallel_state.py:1411] rank 0 in world size 1 is assigned as DP rank 0, PP rank 0, PCP rank 0, TP rank 0, EP rank 0
+(EngineCore_DP0 pid=2009129) INFO 02-02 06:56:29 [gpu_model_runner.py:3562] Starting to load model Elfsong/VLM_stage_2_iter_0006500...
+(EngineCore_DP0 pid=2009129) INFO 02-02 06:56:31 [cuda.py:351] Using FLASH_ATTN attention backend out of potential backends: ('FLASH_ATTN', 'FLASHINFER', 'TRITON_ATTN', 'FLEX_ATTENTION')
+(EngineCore_DP0 pid=2009129) INFO 02-02 06:56:31 [bitsandbytes_loader.py:791] Loading weights with BitsAndBytes quantization. May take a while ...
+(EngineCore_DP0 pid=2009129) 
+(EngineCore_DP0 pid=2009129) 
+(EngineCore_DP0 pid=2009129) 
+(EngineCore_DP0 pid=2009129) 
+(EngineCore_DP0 pid=2009129) 
+(EngineCore_DP0 pid=2009129) 
+(EngineCore_DP0 pid=2009129) 
+(EngineCore_DP0 pid=2009129) 
+(EngineCore_DP0 pid=2009129) 
+(EngineCore_DP0 pid=2009129) 
+(EngineCore_DP0 pid=2009129) 
+(EngineCore_DP0 pid=2009129) 
+(EngineCore_DP0 pid=2009129) 
+(EngineCore_DP0 pid=2009129) 
+(EngineCore_DP0 pid=2009129) 
+(EngineCore_DP0 pid=2009129) 
+(EngineCore_DP0 pid=2009129) 

vllm_0007000.log

@@ -0,0 +1,29 @@
+(APIServer pid=2007973) INFO 02-02 06:56:03 [api_server.py:1351] vLLM API server version 0.13.0
+(APIServer pid=2007973) INFO 02-02 06:56:03 [utils.py:253] non-default args: {'port': 9012, 'model': 'Elfsong/VLM_stage_2_iter_0007000', 'trust_remote_code': True, 'quantization': 'bitsandbytes', 'gpu_memory_utilization': 0.4}
+(APIServer pid=2007973) The argument `trust_remote_code` is to be used with Auto classes. It has no effect here and is ignored.
+(APIServer pid=2007973) INFO 02-02 06:56:04 [model.py:514] Resolved architecture: Qwen3ForCausalLM
+(APIServer pid=2007973) INFO 02-02 06:56:04 [model.py:1661] Using max model len 40960
+(APIServer pid=2007973) INFO 02-02 06:56:04 [scheduler.py:230] Chunked prefill is enabled with max_num_batched_tokens=8192.
+(EngineCore_DP0 pid=2009143) INFO 02-02 06:56:24 [core.py:93] Initializing a V1 LLM engine (v0.13.0) with config: model='Elfsong/VLM_stage_2_iter_0007000', speculative_config=None, tokenizer='Elfsong/VLM_stage_2_iter_0007000', skip_tokenizer_init=False, tokenizer_mode=auto, revision=None, tokenizer_revision=None, trust_remote_code=True, dtype=torch.bfloat16, max_seq_len=40960, download_dir=None, load_format=bitsandbytes, tensor_parallel_size=1, pipeline_parallel_size=1, data_parallel_size=1, disable_custom_all_reduce=False, quantization=bitsandbytes, enforce_eager=False, kv_cache_dtype=auto, device_config=cuda, structured_outputs_config=StructuredOutputsConfig(backend='auto', disable_fallback=False, disable_any_whitespace=False, disable_additional_properties=False, reasoning_parser='', reasoning_parser_plugin='', enable_in_reasoning=False), observability_config=ObservabilityConfig(show_hidden_metrics_for_version=None, otlp_traces_endpoint=None, collect_detailed_traces=None, kv_cache_metrics=False, kv_cache_metrics_sample=0.01, cudagraph_metrics=False, enable_layerwise_nvtx_tracing=False), seed=0, served_model_name=Elfsong/VLM_stage_2_iter_0007000, enable_prefix_caching=True, enable_chunked_prefill=True, pooler_config=None, compilation_config={'level': None, 'mode': <CompilationMode.VLLM_COMPILE: 3>, 'debug_dump_path': None, 'cache_dir': '', 'compile_cache_save_format': 'binary', 'backend': 'inductor', 'custom_ops': ['none'], 'splitting_ops': ['vllm::unified_attention', 'vllm::unified_attention_with_output', 'vllm::unified_mla_attention', 'vllm::unified_mla_attention_with_output', 'vllm::mamba_mixer2', 'vllm::mamba_mixer', 'vllm::short_conv', 'vllm::linear_attention', 'vllm::plamo2_mamba_mixer', 'vllm::gdn_attention_core', 'vllm::kda_attention', 'vllm::sparse_attn_indexer'], 'compile_mm_encoder': False, 'compile_sizes': [], 'compile_ranges_split_points': [8192], 'inductor_compile_config': {'enable_auto_functionalized_v2': False, 'combo_kernels': True, 'benchmark_combo_kernel': True}, 'inductor_passes': {}, 'cudagraph_mode': <CUDAGraphMode.FULL_AND_PIECEWISE: (2, 1)>, 'cudagraph_num_of_warmups': 1, 'cudagraph_capture_sizes': [1, 2, 4, 8, 16, 24, 32, 40, 48, 56, 64, 72, 80, 88, 96, 104, 112, 120, 128, 136, 144, 152, 160, 168, 176, 184, 192, 200, 208, 216, 224, 232, 240, 248, 256, 272, 288, 304, 320, 336, 352, 368, 384, 400, 416, 432, 448, 464, 480, 496, 512], 'cudagraph_copy_inputs': False, 'cudagraph_specialize_lora': True, 'use_inductor_graph_partition': False, 'pass_config': {'fuse_norm_quant': False, 'fuse_act_quant': False, 'fuse_attn_quant': False, 'eliminate_noops': True, 'enable_sp': False, 'fuse_gemm_comms': False, 'fuse_allreduce_rms': False}, 'max_cudagraph_capture_size': 512, 'dynamic_shapes_config': {'type': <DynamicShapesType.BACKED: 'backed'>, 'evaluate_guards': False}, 'local_cache_dir': None}
+(EngineCore_DP0 pid=2009143) INFO 02-02 06:56:26 [parallel_state.py:1203] world_size=1 rank=0 local_rank=0 distributed_init_method=tcp://100.96.20.65:32907 backend=nccl
+(EngineCore_DP0 pid=2009143) INFO 02-02 06:56:26 [parallel_state.py:1411] rank 0 in world size 1 is assigned as DP rank 0, PP rank 0, PCP rank 0, TP rank 0, EP rank 0
+(EngineCore_DP0 pid=2009143) INFO 02-02 06:56:29 [gpu_model_runner.py:3562] Starting to load model Elfsong/VLM_stage_2_iter_0007000...
+(EngineCore_DP0 pid=2009143) INFO 02-02 06:56:31 [cuda.py:351] Using FLASH_ATTN attention backend out of potential backends: ('FLASH_ATTN', 'FLASHINFER', 'TRITON_ATTN', 'FLEX_ATTENTION')
+(EngineCore_DP0 pid=2009143) INFO 02-02 06:56:31 [bitsandbytes_loader.py:791] Loading weights with BitsAndBytes quantization. May take a while ...
+(EngineCore_DP0 pid=2009143) 
+(EngineCore_DP0 pid=2009143) 
+(EngineCore_DP0 pid=2009143) 
+(EngineCore_DP0 pid=2009143) 
+(EngineCore_DP0 pid=2009143) 
+(EngineCore_DP0 pid=2009143) 
+(EngineCore_DP0 pid=2009143) 
+(EngineCore_DP0 pid=2009143) 
+(EngineCore_DP0 pid=2009143) 
+(EngineCore_DP0 pid=2009143) 
+(EngineCore_DP0 pid=2009143) 
+(EngineCore_DP0 pid=2009143) 
+(EngineCore_DP0 pid=2009143) 
+(EngineCore_DP0 pid=2009143) 
+(EngineCore_DP0 pid=2009143) 
+(EngineCore_DP0 pid=2009143) 
+(EngineCore_DP0 pid=2009143) 

vllm_0007500.log

@@ -0,0 +1,26 @@
+(APIServer pid=2008308) INFO 02-02 06:56:15 [api_server.py:1351] vLLM API server version 0.13.0
+(APIServer pid=2008308) INFO 02-02 06:56:15 [utils.py:253] non-default args: {'port': 9013, 'model': 'Elfsong/VLM_stage_2_iter_0007500', 'trust_remote_code': True, 'quantization': 'bitsandbytes', 'gpu_memory_utilization': 0.4}
+(APIServer pid=2008308) The argument `trust_remote_code` is to be used with Auto classes. It has no effect here and is ignored.
+(APIServer pid=2008308) INFO 02-02 06:56:17 [model.py:514] Resolved architecture: Qwen3ForCausalLM
+(APIServer pid=2008308) INFO 02-02 06:56:17 [model.py:1661] Using max model len 40960
+(APIServer pid=2008308) INFO 02-02 06:56:17 [scheduler.py:230] Chunked prefill is enabled with max_num_batched_tokens=8192.
+(EngineCore_DP0 pid=2009629) INFO 02-02 06:56:39 [core.py:93] Initializing a V1 LLM engine (v0.13.0) with config: model='Elfsong/VLM_stage_2_iter_0007500', speculative_config=None, tokenizer='Elfsong/VLM_stage_2_iter_0007500', skip_tokenizer_init=False, tokenizer_mode=auto, revision=None, tokenizer_revision=None, trust_remote_code=True, dtype=torch.bfloat16, max_seq_len=40960, download_dir=None, load_format=bitsandbytes, tensor_parallel_size=1, pipeline_parallel_size=1, data_parallel_size=1, disable_custom_all_reduce=False, quantization=bitsandbytes, enforce_eager=False, kv_cache_dtype=auto, device_config=cuda, structured_outputs_config=StructuredOutputsConfig(backend='auto', disable_fallback=False, disable_any_whitespace=False, disable_additional_properties=False, reasoning_parser='', reasoning_parser_plugin='', enable_in_reasoning=False), observability_config=ObservabilityConfig(show_hidden_metrics_for_version=None, otlp_traces_endpoint=None, collect_detailed_traces=None, kv_cache_metrics=False, kv_cache_metrics_sample=0.01, cudagraph_metrics=False, enable_layerwise_nvtx_tracing=False), seed=0, served_model_name=Elfsong/VLM_stage_2_iter_0007500, enable_prefix_caching=True, enable_chunked_prefill=True, pooler_config=None, compilation_config={'level': None, 'mode': <CompilationMode.VLLM_COMPILE: 3>, 'debug_dump_path': None, 'cache_dir': '', 'compile_cache_save_format': 'binary', 'backend': 'inductor', 'custom_ops': ['none'], 'splitting_ops': ['vllm::unified_attention', 'vllm::unified_attention_with_output', 'vllm::unified_mla_attention', 'vllm::unified_mla_attention_with_output', 'vllm::mamba_mixer2', 'vllm::mamba_mixer', 'vllm::short_conv', 'vllm::linear_attention', 'vllm::plamo2_mamba_mixer', 'vllm::gdn_attention_core', 'vllm::kda_attention', 'vllm::sparse_attn_indexer'], 'compile_mm_encoder': False, 'compile_sizes': [], 'compile_ranges_split_points': [8192], 'inductor_compile_config': {'enable_auto_functionalized_v2': False, 'combo_kernels': True, 'benchmark_combo_kernel': True}, 'inductor_passes': {}, 'cudagraph_mode': <CUDAGraphMode.FULL_AND_PIECEWISE: (2, 1)>, 'cudagraph_num_of_warmups': 1, 'cudagraph_capture_sizes': [1, 2, 4, 8, 16, 24, 32, 40, 48, 56, 64, 72, 80, 88, 96, 104, 112, 120, 128, 136, 144, 152, 160, 168, 176, 184, 192, 200, 208, 216, 224, 232, 240, 248, 256, 272, 288, 304, 320, 336, 352, 368, 384, 400, 416, 432, 448, 464, 480, 496, 512], 'cudagraph_copy_inputs': False, 'cudagraph_specialize_lora': True, 'use_inductor_graph_partition': False, 'pass_config': {'fuse_norm_quant': False, 'fuse_act_quant': False, 'fuse_attn_quant': False, 'eliminate_noops': True, 'enable_sp': False, 'fuse_gemm_comms': False, 'fuse_allreduce_rms': False}, 'max_cudagraph_capture_size': 512, 'dynamic_shapes_config': {'type': <DynamicShapesType.BACKED: 'backed'>, 'evaluate_guards': False}, 'local_cache_dir': None}
+(EngineCore_DP0 pid=2009629) INFO 02-02 06:56:40 [parallel_state.py:1203] world_size=1 rank=0 local_rank=0 distributed_init_method=tcp://100.96.20.65:39393 backend=nccl
+(EngineCore_DP0 pid=2009629) INFO 02-02 06:56:40 [parallel_state.py:1411] rank 0 in world size 1 is assigned as DP rank 0, PP rank 0, PCP rank 0, TP rank 0, EP rank 0
+(EngineCore_DP0 pid=2009629) INFO 02-02 06:56:41 [gpu_model_runner.py:3562] Starting to load model Elfsong/VLM_stage_2_iter_0007500...
+(EngineCore_DP0 pid=2009629) INFO 02-02 06:56:43 [cuda.py:351] Using FLASH_ATTN attention backend out of potential backends: ('FLASH_ATTN', 'FLASHINFER', 'TRITON_ATTN', 'FLEX_ATTENTION')
+(EngineCore_DP0 pid=2009629) INFO 02-02 06:56:43 [bitsandbytes_loader.py:791] Loading weights with BitsAndBytes quantization. May take a while ...
+(EngineCore_DP0 pid=2009629) 
+(EngineCore_DP0 pid=2009629) 
+(EngineCore_DP0 pid=2009629) 
+(EngineCore_DP0 pid=2009629) 
+(EngineCore_DP0 pid=2009629) 
+(EngineCore_DP0 pid=2009629) 
+(EngineCore_DP0 pid=2009629) 
+(EngineCore_DP0 pid=2009629) 
+(EngineCore_DP0 pid=2009629) 
+(EngineCore_DP0 pid=2009629) 
+(EngineCore_DP0 pid=2009629) 
+(EngineCore_DP0 pid=2009629) 
+(EngineCore_DP0 pid=2009629) 
+(EngineCore_DP0 pid=2009629)