Scheduled Commit

vllm_0004000.log

@@ -1,20 +1,143 @@
-(APIServer pid=3232323) INFO 02-03 01:26:39 [utils.py:325] 
-(APIServer pid=3232323) INFO 02-03 01:26:39 [utils.py:325]        █     █     █▄   ▄█
-(APIServer pid=3232323) INFO 02-03 01:26:39 [utils.py:325]  ▄▄ ▄█ █     █     █ ▀▄▀ █  version 0.15.0
-(APIServer pid=3232323) INFO 02-03 01:26:39 [utils.py:325]   █▄█▀ █     █     █     █  model   Elfsong/VLM_stage_2_iter_0004000
-(APIServer pid=3232323) INFO 02-03 01:26:39 [utils.py:325]    ▀▀  ▀▀▀▀▀ ▀▀▀▀▀ ▀     ▀
-(APIServer pid=3232323) INFO 02-03 01:26:39 [utils.py:325] 
-(APIServer pid=3232323) INFO 02-03 01:26:39 [utils.py:261] non-default args: {'port': 9000, 'model': 'Elfsong/VLM_stage_2_iter_0004000', 'trust_remote_code': True, 'quantization': 'bitsandbytes', 'gpu_memory_utilization': 0.3}
-(APIServer pid=3232323) The argument `trust_remote_code` is to be used with Auto classes. It has no effect here and is ignored.
-(APIServer pid=3232323) The argument `trust_remote_code` is to be used with Auto classes. It has no effect here and is ignored.
-(APIServer pid=3232323) INFO 02-03 01:26:41 [model.py:541] Resolved architecture: Qwen3ForCausalLM
-(APIServer pid=3232323) INFO 02-03 01:26:41 [model.py:1561] Using max model len 40960
-(APIServer pid=3232323) INFO 02-03 01:26:41 [scheduler.py:226] Chunked prefill is enabled with max_num_batched_tokens=8192.
-(APIServer pid=3232323) INFO 02-03 01:26:44 [vllm.py:624] Asynchronous scheduling is enabled.
-(EngineCore_DP0 pid=3233898) INFO 02-03 01:26:58 [core.py:96] Initializing a V1 LLM engine (v0.15.0) with config: model='Elfsong/VLM_stage_2_iter_0004000', speculative_config=None, tokenizer='Elfsong/VLM_stage_2_iter_0004000', 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, enable_return_routed_experts=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, enable_mfu_metrics=False, enable_mm_processor_stats=False, enable_logging_iteration_details=False), seed=0, served_model_name=Elfsong/VLM_stage_2_iter_0004000, 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', 'vllm::rocm_aiter_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, 'assume_32_bit_indexing': True}, 'local_cache_dir': None}
-(EngineCore_DP0 pid=3233898) INFO 02-03 01:26:59 [parallel_state.py:1212] world_size=1 rank=0 local_rank=0 distributed_init_method=tcp://172.21.25.98:52225 backend=nccl
-(EngineCore_DP0 pid=3233898) INFO 02-03 01:26:59 [parallel_state.py:1423] rank 0 in world size 1 is assigned as DP rank 0, PP rank 0, PCP rank 0, TP rank 0, EP rank N/A
-(EngineCore_DP0 pid=3233898) INFO 02-03 01:27:00 [gpu_model_runner.py:4021] Starting to load model Elfsong/VLM_stage_2_iter_0004000...
-(EngineCore_DP0 pid=3233898) INFO 02-03 01:27:02 [cuda.py:364] Using FLASH_ATTN attention backend out of potential backends: ('FLASH_ATTN', 'FLASHINFER', 'TRITON_ATTN', 'FLEX_ATTENTION')
-(EngineCore_DP0 pid=3233898) INFO 02-03 01:27:03 [bitsandbytes_loader.py:786] Loading weights with BitsAndBytes quantization. May take a while ...
-Cancellation requested; stopping current tasks.
+(APIServer pid=3309550) INFO 02-03 01:40:54 [utils.py:325] 
+(APIServer pid=3309550) INFO 02-03 01:40:54 [utils.py:325]        █     █     █▄   ▄█
+(APIServer pid=3309550) INFO 02-03 01:40:54 [utils.py:325]  ▄▄ ▄█ █     █     █ ▀▄▀ █  version 0.15.0
+(APIServer pid=3309550) INFO 02-03 01:40:54 [utils.py:325]   █▄█▀ █     █     █     █  model   Elfsong/VLM_stage_2_iter_0004000
+(APIServer pid=3309550) INFO 02-03 01:40:54 [utils.py:325]    ▀▀  ▀▀▀▀▀ ▀▀▀▀▀ ▀     ▀
+(APIServer pid=3309550) INFO 02-03 01:40:54 [utils.py:325] 
+(APIServer pid=3309550) INFO 02-03 01:40:54 [utils.py:261] non-default args: {'port': 9000, 'model': 'Elfsong/VLM_stage_2_iter_0004000', 'trust_remote_code': True, 'gpu_memory_utilization': 0.4}
+(APIServer pid=3309550) The argument `trust_remote_code` is to be used with Auto classes. It has no effect here and is ignored.
+(APIServer pid=3309550) The argument `trust_remote_code` is to be used with Auto classes. It has no effect here and is ignored.
+(APIServer pid=3309550) INFO 02-03 01:40:56 [model.py:541] Resolved architecture: Qwen3ForCausalLM
+(APIServer pid=3309550) INFO 02-03 01:40:56 [model.py:1561] Using max model len 40960
+(APIServer pid=3309550) INFO 02-03 01:40:56 [scheduler.py:226] Chunked prefill is enabled with max_num_batched_tokens=8192.
+(APIServer pid=3309550) INFO 02-03 01:40:56 [vllm.py:624] Asynchronous scheduling is enabled.
+(EngineCore_DP0 pid=3310619) INFO 02-03 01:41:06 [core.py:96] Initializing a V1 LLM engine (v0.15.0) with config: model='Elfsong/VLM_stage_2_iter_0004000', speculative_config=None, tokenizer='Elfsong/VLM_stage_2_iter_0004000', 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=auto, tensor_parallel_size=1, pipeline_parallel_size=1, data_parallel_size=1, disable_custom_all_reduce=False, quantization=None, enforce_eager=False, enable_return_routed_experts=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, enable_mfu_metrics=False, enable_mm_processor_stats=False, enable_logging_iteration_details=False), seed=0, served_model_name=Elfsong/VLM_stage_2_iter_0004000, 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', 'vllm::rocm_aiter_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, 'assume_32_bit_indexing': True}, 'local_cache_dir': None}
+(EngineCore_DP0 pid=3310619) INFO 02-03 01:41:11 [parallel_state.py:1212] world_size=1 rank=0 local_rank=0 distributed_init_method=tcp://172.21.25.98:59653 backend=nccl
+(EngineCore_DP0 pid=3310619) INFO 02-03 01:41:11 [parallel_state.py:1423] rank 0 in world size 1 is assigned as DP rank 0, PP rank 0, PCP rank 0, TP rank 0, EP rank N/A
+(EngineCore_DP0 pid=3310619) INFO 02-03 01:41:12 [gpu_model_runner.py:4021] Starting to load model Elfsong/VLM_stage_2_iter_0004000...
+(EngineCore_DP0 pid=3310619) INFO 02-03 01:41:13 [cuda.py:364] Using FLASH_ATTN attention backend out of potential backends: ('FLASH_ATTN', 'FLASHINFER', 'TRITON_ATTN', 'FLEX_ATTENTION')
+(EngineCore_DP0 pid=3310619) INFO 02-03 01:42:48 [weight_utils.py:527] Time spent downloading weights for Elfsong/VLM_stage_2_iter_0004000: 92.466791 seconds
+(EngineCore_DP0 pid=3310619) 
+(EngineCore_DP0 pid=3310619) 
+(EngineCore_DP0 pid=3310619) 
+(EngineCore_DP0 pid=3310619) 
+(EngineCore_DP0 pid=3310619) 
+(EngineCore_DP0 pid=3310619) 
+(EngineCore_DP0 pid=3310619) 
+(EngineCore_DP0 pid=3310619) 
+(EngineCore_DP0 pid=3310619) 
+(EngineCore_DP0 pid=3310619) 
+(EngineCore_DP0 pid=3310619) 
+(EngineCore_DP0 pid=3310619) 
+(EngineCore_DP0 pid=3310619) 
+(EngineCore_DP0 pid=3310619) 
+(EngineCore_DP0 pid=3310619) 
+(EngineCore_DP0 pid=3310619) 
+(EngineCore_DP0 pid=3310619) 
+(EngineCore_DP0 pid=3310619) 
+(EngineCore_DP0 pid=3310619) 
+(EngineCore_DP0 pid=3310619) 
+(EngineCore_DP0 pid=3310619) INFO 02-03 01:43:08 [default_loader.py:291] Loading weights took 20.23 seconds
+(EngineCore_DP0 pid=3310619) INFO 02-03 01:43:09 [gpu_model_runner.py:4118] Model loading took 61.03 GiB memory and 115.770729 seconds
+(EngineCore_DP0 pid=3310619) INFO 02-03 01:43:25 [backends.py:805] Using cache directory: /home/mingzhe/.cache/vllm/torch_compile_cache/226ddebc06/rank_0_0/backbone for vLLM's torch.compile
+(EngineCore_DP0 pid=3310619) INFO 02-03 01:43:25 [backends.py:865] Dynamo bytecode transform time: 15.42 s
+(EngineCore_DP0 pid=3310619) INFO 02-03 01:43:38 [backends.py:302] Cache the graph of compile range (1, 8192) for later use
+(EngineCore_DP0 pid=3310619) INFO 02-03 01:44:12 [backends.py:319] Compiling a graph for compile range (1, 8192) takes 34.69 s
+(EngineCore_DP0 pid=3310619) INFO 02-03 01:44:12 [monitor.py:34] torch.compile takes 50.11 s in total
+(EngineCore_DP0 pid=3310619) INFO 02-03 01:44:15 [gpu_worker.py:356] Available KV cache memory: -77.05 GiB
+(EngineCore_DP0 pid=3310619) ERROR 02-03 01:44:15 [core.py:946] EngineCore failed to start.
+(EngineCore_DP0 pid=3310619) ERROR 02-03 01:44:15 [core.py:946] Traceback (most recent call last):
+(EngineCore_DP0 pid=3310619) ERROR 02-03 01:44:15 [core.py:946]   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/vllm/v1/engine/core.py", line 937, in run_engine_core
+(EngineCore_DP0 pid=3310619) ERROR 02-03 01:44:15 [core.py:946]     engine_core = EngineCoreProc(*args, engine_index=dp_rank, **kwargs)
+(EngineCore_DP0 pid=3310619) ERROR 02-03 01:44:15 [core.py:946]                   ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+(EngineCore_DP0 pid=3310619) ERROR 02-03 01:44:15 [core.py:946]   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/vllm/v1/engine/core.py", line 691, in __init__
+(EngineCore_DP0 pid=3310619) ERROR 02-03 01:44:15 [core.py:946]     super().__init__(
+(EngineCore_DP0 pid=3310619) ERROR 02-03 01:44:15 [core.py:946]   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/vllm/v1/engine/core.py", line 112, in __init__
+(EngineCore_DP0 pid=3310619) ERROR 02-03 01:44:15 [core.py:946]     num_gpu_blocks, num_cpu_blocks, kv_cache_config = self._initialize_kv_caches(
+(EngineCore_DP0 pid=3310619) ERROR 02-03 01:44:15 [core.py:946]                                                       ^^^^^^^^^^^^^^^^^^^^^^^^^^^
+(EngineCore_DP0 pid=3310619) ERROR 02-03 01:44:15 [core.py:946]   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/vllm/v1/engine/core.py", line 253, in _initialize_kv_caches
+(EngineCore_DP0 pid=3310619) ERROR 02-03 01:44:15 [core.py:946]     kv_cache_configs = get_kv_cache_configs(
+(EngineCore_DP0 pid=3310619) ERROR 02-03 01:44:15 [core.py:946]                        ^^^^^^^^^^^^^^^^^^^^^
+(EngineCore_DP0 pid=3310619) ERROR 02-03 01:44:15 [core.py:946]   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/vllm/v1/core/kv_cache_utils.py", line 1516, in get_kv_cache_configs
+(EngineCore_DP0 pid=3310619) ERROR 02-03 01:44:15 [core.py:946]     _check_enough_kv_cache_memory(
+(EngineCore_DP0 pid=3310619) ERROR 02-03 01:44:15 [core.py:946]   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/vllm/v1/core/kv_cache_utils.py", line 616, in _check_enough_kv_cache_memory
+(EngineCore_DP0 pid=3310619) ERROR 02-03 01:44:15 [core.py:946]     raise ValueError(
+(EngineCore_DP0 pid=3310619) ERROR 02-03 01:44:15 [core.py:946] ValueError: No available memory for the cache blocks. Try increasing `gpu_memory_utilization` when initializing the engine. See https://docs.vllm.ai/en/latest/configuration/conserving_memory/ for more details.
+(EngineCore_DP0 pid=3310619) Process EngineCore_DP0:
+(EngineCore_DP0 pid=3310619) Traceback (most recent call last):
+(EngineCore_DP0 pid=3310619)   File "/usr/lib/python3.12/multiprocessing/process.py", line 314, in _bootstrap
+(EngineCore_DP0 pid=3310619)     self.run()
+(EngineCore_DP0 pid=3310619)   File "/usr/lib/python3.12/multiprocessing/process.py", line 108, in run
+(EngineCore_DP0 pid=3310619)     self._target(*self._args, **self._kwargs)
+(EngineCore_DP0 pid=3310619)   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/vllm/v1/engine/core.py", line 950, in run_engine_core
+(EngineCore_DP0 pid=3310619)     raise e
+(EngineCore_DP0 pid=3310619)   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/vllm/v1/engine/core.py", line 937, in run_engine_core
+(EngineCore_DP0 pid=3310619)     engine_core = EngineCoreProc(*args, engine_index=dp_rank, **kwargs)
+(EngineCore_DP0 pid=3310619)                   ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+(EngineCore_DP0 pid=3310619)   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/vllm/v1/engine/core.py", line 691, in __init__
+(EngineCore_DP0 pid=3310619)     super().__init__(
+(EngineCore_DP0 pid=3310619)   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/vllm/v1/engine/core.py", line 112, in __init__
+(EngineCore_DP0 pid=3310619)     num_gpu_blocks, num_cpu_blocks, kv_cache_config = self._initialize_kv_caches(
+(EngineCore_DP0 pid=3310619)                                                       ^^^^^^^^^^^^^^^^^^^^^^^^^^^
+(EngineCore_DP0 pid=3310619)   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/vllm/v1/engine/core.py", line 253, in _initialize_kv_caches
+(EngineCore_DP0 pid=3310619)     kv_cache_configs = get_kv_cache_configs(
+(EngineCore_DP0 pid=3310619)                        ^^^^^^^^^^^^^^^^^^^^^
+(EngineCore_DP0 pid=3310619)   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/vllm/v1/core/kv_cache_utils.py", line 1516, in get_kv_cache_configs
+(EngineCore_DP0 pid=3310619)     _check_enough_kv_cache_memory(
+(EngineCore_DP0 pid=3310619)   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/vllm/v1/core/kv_cache_utils.py", line 616, in _check_enough_kv_cache_memory
+(EngineCore_DP0 pid=3310619)     raise ValueError(
+(EngineCore_DP0 pid=3310619) ValueError: No available memory for the cache blocks. Try increasing `gpu_memory_utilization` when initializing the engine. See https://docs.vllm.ai/en/latest/configuration/conserving_memory/ for more details.
+[rank0]:[W203 01:44:17.466159054 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=3309550) Traceback (most recent call last):
+(APIServer pid=3309550)   File "<frozen runpy>", line 198, in _run_module_as_main
+(APIServer pid=3309550)   File "<frozen runpy>", line 88, in _run_code
+(APIServer pid=3309550)   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/vllm/entrypoints/openai/api_server.py", line 991, in <module>
+(APIServer pid=3309550)     uvloop.run(run_server(args))
+(APIServer pid=3309550)   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/uvloop/__init__.py", line 96, in run
+(APIServer pid=3309550)     return __asyncio.run(
+(APIServer pid=3309550)            ^^^^^^^^^^^^^^
+(APIServer pid=3309550)   File "/usr/lib/python3.12/asyncio/runners.py", line 194, in run
+(APIServer pid=3309550)     return runner.run(main)
+(APIServer pid=3309550)            ^^^^^^^^^^^^^^^^
+(APIServer pid=3309550)   File "/usr/lib/python3.12/asyncio/runners.py", line 118, in run
+(APIServer pid=3309550)     return self._loop.run_until_complete(task)
+(APIServer pid=3309550)            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+(APIServer pid=3309550)   File "uvloop/loop.pyx", line 1518, in uvloop.loop.Loop.run_until_complete
+(APIServer pid=3309550)   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/uvloop/__init__.py", line 48, in wrapper
+(APIServer pid=3309550)     return await main
+(APIServer pid=3309550)            ^^^^^^^^^^
+(APIServer pid=3309550)   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/vllm/entrypoints/openai/api_server.py", line 919, in run_server
+(APIServer pid=3309550)     await run_server_worker(listen_address, sock, args, **uvicorn_kwargs)
+(APIServer pid=3309550)   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/vllm/entrypoints/openai/api_server.py", line 938, in run_server_worker
+(APIServer pid=3309550)     async with build_async_engine_client(
+(APIServer pid=3309550)   File "/usr/lib/python3.12/contextlib.py", line 210, in __aenter__
+(APIServer pid=3309550)     return await anext(self.gen)
+(APIServer pid=3309550)            ^^^^^^^^^^^^^^^^^^^^^
+(APIServer pid=3309550)   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/vllm/entrypoints/openai/api_server.py", line 147, in build_async_engine_client
+(APIServer pid=3309550)     async with build_async_engine_client_from_engine_args(
+(APIServer pid=3309550)   File "/usr/lib/python3.12/contextlib.py", line 210, in __aenter__
+(APIServer pid=3309550)     return await anext(self.gen)
+(APIServer pid=3309550)            ^^^^^^^^^^^^^^^^^^^^^
+(APIServer pid=3309550)   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/vllm/entrypoints/openai/api_server.py", line 188, in build_async_engine_client_from_engine_args
+(APIServer pid=3309550)     async_llm = AsyncLLM.from_vllm_config(
+(APIServer pid=3309550)                 ^^^^^^^^^^^^^^^^^^^^^^^^^^
+(APIServer pid=3309550)   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/vllm/v1/engine/async_llm.py", line 228, in from_vllm_config
+(APIServer pid=3309550)     return cls(
+(APIServer pid=3309550)            ^^^^
+(APIServer pid=3309550)   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/vllm/v1/engine/async_llm.py", line 155, in __init__
+(APIServer pid=3309550)     self.engine_core = EngineCoreClient.make_async_mp_client(
+(APIServer pid=3309550)                        ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+(APIServer pid=3309550)   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/vllm/v1/engine/core_client.py", line 122, in make_async_mp_client
+(APIServer pid=3309550)     return AsyncMPClient(*client_args)
+(APIServer pid=3309550)            ^^^^^^^^^^^^^^^^^^^^^^^^^^^
+(APIServer pid=3309550)   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/vllm/v1/engine/core_client.py", line 819, in __init__
+(APIServer pid=3309550)     super().__init__(
+(APIServer pid=3309550)   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/vllm/v1/engine/core_client.py", line 479, in __init__
+(APIServer pid=3309550)     with launch_core_engines(vllm_config, executor_class, log_stats) as (
+(APIServer pid=3309550)   File "/usr/lib/python3.12/contextlib.py", line 144, in __exit__
+(APIServer pid=3309550)     next(self.gen)
+(APIServer pid=3309550)   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/vllm/v1/engine/utils.py", line 933, in launch_core_engines
+(APIServer pid=3309550)     wait_for_engine_startup(
+(APIServer pid=3309550)   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/vllm/v1/engine/utils.py", line 992, in wait_for_engine_startup
+(APIServer pid=3309550)     raise RuntimeError(
+(APIServer pid=3309550) RuntimeError: Engine core initialization failed. See root cause above. Failed core proc(s): {}

vllm_0005000.log

@@ -1,32 +1,241 @@
-(APIServer pid=3232791) INFO 02-03 01:26:45 [utils.py:325] 
-(APIServer pid=3232791) INFO 02-03 01:26:45 [utils.py:325]        █     █     █▄   ▄█
-(APIServer pid=3232791) INFO 02-03 01:26:45 [utils.py:325]  ▄▄ ▄█ █     █     █ ▀▄▀ █  version 0.15.0
-(APIServer pid=3232791) INFO 02-03 01:26:45 [utils.py:325]   █▄█▀ █     █     █     █  model   Elfsong/VLM_stage_2_iter_0005000
-(APIServer pid=3232791) INFO 02-03 01:26:45 [utils.py:325]    ▀▀  ▀▀▀▀▀ ▀▀▀▀▀ ▀     ▀
-(APIServer pid=3232791) INFO 02-03 01:26:45 [utils.py:325] 
-(APIServer pid=3232791) INFO 02-03 01:26:45 [utils.py:261] non-default args: {'port': 9001, 'model': 'Elfsong/VLM_stage_2_iter_0005000', 'trust_remote_code': True, 'quantization': 'bitsandbytes', 'gpu_memory_utilization': 0.3}
-(APIServer pid=3232791) The argument `trust_remote_code` is to be used with Auto classes. It has no effect here and is ignored.
-(APIServer pid=3232791) The argument `trust_remote_code` is to be used with Auto classes. It has no effect here and is ignored.
-(APIServer pid=3232791) INFO 02-03 01:26:46 [model.py:541] Resolved architecture: Qwen3ForCausalLM
-(APIServer pid=3232791) INFO 02-03 01:26:46 [model.py:1561] Using max model len 40960
-(APIServer pid=3232791) INFO 02-03 01:26:46 [scheduler.py:226] Chunked prefill is enabled with max_num_batched_tokens=8192.
-(APIServer pid=3232791) INFO 02-03 01:26:50 [vllm.py:624] Asynchronous scheduling is enabled.
-(EngineCore_DP0 pid=3234476) INFO 02-03 01:27:03 [core.py:96] Initializing a V1 LLM engine (v0.15.0) 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, enable_return_routed_experts=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, enable_mfu_metrics=False, enable_mm_processor_stats=False, enable_logging_iteration_details=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', 'vllm::rocm_aiter_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, 'assume_32_bit_indexing': True}, 'local_cache_dir': None}
-(EngineCore_DP0 pid=3234476) INFO 02-03 01:27:04 [parallel_state.py:1212] world_size=1 rank=0 local_rank=0 distributed_init_method=tcp://172.21.25.98:54097 backend=nccl
-(EngineCore_DP0 pid=3234476) INFO 02-03 01:27:04 [parallel_state.py:1423] rank 0 in world size 1 is assigned as DP rank 0, PP rank 0, PCP rank 0, TP rank 0, EP rank N/A
-(EngineCore_DP0 pid=3234476) INFO 02-03 01:27:06 [gpu_model_runner.py:4021] Starting to load model Elfsong/VLM_stage_2_iter_0005000...
-(EngineCore_DP0 pid=3234476) INFO 02-03 01:27:08 [cuda.py:364] Using FLASH_ATTN attention backend out of potential backends: ('FLASH_ATTN', 'FLASHINFER', 'TRITON_ATTN', 'FLEX_ATTENTION')
-(EngineCore_DP0 pid=3234476) INFO 02-03 01:27:08 [bitsandbytes_loader.py:786] Loading weights with BitsAndBytes quantization. May take a while ...
-(EngineCore_DP0 pid=3234476) INFO 02-03 01:39:24 [weight_utils.py:527] Time spent downloading weights for Elfsong/VLM_stage_2_iter_0005000: 734.294342 seconds
-(EngineCore_DP0 pid=3234476) 
-(EngineCore_DP0 pid=3234476) 
-(EngineCore_DP0 pid=3234476) 
-(EngineCore_DP0 pid=3234476) 
-(EngineCore_DP0 pid=3234476) 
-(EngineCore_DP0 pid=3234476) 
-(EngineCore_DP0 pid=3234476) 
-(EngineCore_DP0 pid=3234476) 
-(EngineCore_DP0 pid=3234476) 
-(EngineCore_DP0 pid=3234476) 
-(EngineCore_DP0 pid=3234476) 
-(EngineCore_DP0 pid=3234476) 
+(APIServer pid=3309898) INFO 02-03 01:40:59 [utils.py:325] 
+(APIServer pid=3309898) INFO 02-03 01:40:59 [utils.py:325]        █     █     █▄   ▄█
+(APIServer pid=3309898) INFO 02-03 01:40:59 [utils.py:325]  ▄▄ ▄█ █     █     █ ▀▄▀ █  version 0.15.0
+(APIServer pid=3309898) INFO 02-03 01:40:59 [utils.py:325]   █▄█▀ █     █     █     █  model   Elfsong/VLM_stage_2_iter_0005000
+(APIServer pid=3309898) INFO 02-03 01:40:59 [utils.py:325]    ▀▀  ▀▀▀▀▀ ▀▀▀▀▀ ▀     ▀
+(APIServer pid=3309898) INFO 02-03 01:40:59 [utils.py:325] 
+(APIServer pid=3309898) INFO 02-03 01:40:59 [utils.py:261] non-default args: {'port': 9001, 'model': 'Elfsong/VLM_stage_2_iter_0005000', 'trust_remote_code': True, 'gpu_memory_utilization': 0.4}
+(APIServer pid=3309898) The argument `trust_remote_code` is to be used with Auto classes. It has no effect here and is ignored.
+(APIServer pid=3309898) The argument `trust_remote_code` is to be used with Auto classes. It has no effect here and is ignored.
+(APIServer pid=3309898) INFO 02-03 01:41:01 [model.py:541] Resolved architecture: Qwen3ForCausalLM
+(APIServer pid=3309898) INFO 02-03 01:41:01 [model.py:1561] Using max model len 40960
+(APIServer pid=3309898) INFO 02-03 01:41:01 [scheduler.py:226] Chunked prefill is enabled with max_num_batched_tokens=8192.
+(APIServer pid=3309898) INFO 02-03 01:41:01 [vllm.py:624] Asynchronous scheduling is enabled.
+(EngineCore_DP0 pid=3312712) INFO 02-03 01:41:12 [core.py:96] Initializing a V1 LLM engine (v0.15.0) 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=auto, tensor_parallel_size=1, pipeline_parallel_size=1, data_parallel_size=1, disable_custom_all_reduce=False, quantization=None, enforce_eager=False, enable_return_routed_experts=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, enable_mfu_metrics=False, enable_mm_processor_stats=False, enable_logging_iteration_details=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', 'vllm::rocm_aiter_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, 'assume_32_bit_indexing': True}, 'local_cache_dir': None}
+(EngineCore_DP0 pid=3312712) INFO 02-03 01:41:16 [parallel_state.py:1212] world_size=1 rank=0 local_rank=0 distributed_init_method=tcp://172.21.25.98:53221 backend=nccl
+(EngineCore_DP0 pid=3312712) INFO 02-03 01:41:16 [parallel_state.py:1423] rank 0 in world size 1 is assigned as DP rank 0, PP rank 0, PCP rank 0, TP rank 0, EP rank N/A
+(EngineCore_DP0 pid=3312712) INFO 02-03 01:41:17 [gpu_model_runner.py:4021] Starting to load model Elfsong/VLM_stage_2_iter_0005000...
+(EngineCore_DP0 pid=3312712) INFO 02-03 01:41:18 [cuda.py:364] Using FLASH_ATTN attention backend out of potential backends: ('FLASH_ATTN', 'FLASHINFER', 'TRITON_ATTN', 'FLEX_ATTENTION')
+(EngineCore_DP0 pid=3312712) 
+(EngineCore_DP0 pid=3312712) 
+(EngineCore_DP0 pid=3312712) 
+(EngineCore_DP0 pid=3312712) 
+(EngineCore_DP0 pid=3312712) 
+(EngineCore_DP0 pid=3312712) 
+(EngineCore_DP0 pid=3312712) 
+(EngineCore_DP0 pid=3312712) 
+(EngineCore_DP0 pid=3312712) 
+(EngineCore_DP0 pid=3312712) 
+(EngineCore_DP0 pid=3312712) 
+(EngineCore_DP0 pid=3312712) 
+(EngineCore_DP0 pid=3312712) 
+(EngineCore_DP0 pid=3312712) 
+(EngineCore_DP0 pid=3312712) 
+(EngineCore_DP0 pid=3312712) 
+(EngineCore_DP0 pid=3312712) 
+(EngineCore_DP0 pid=3312712) 
+(EngineCore_DP0 pid=3312712) 
+(EngineCore_DP0 pid=3312712) 
+(EngineCore_DP0 pid=3312712) INFO 02-03 01:41:42 [default_loader.py:291] Loading weights took 21.25 seconds
+(EngineCore_DP0 pid=3312712) INFO 02-03 01:41:43 [gpu_model_runner.py:4118] Model loading took 61.03 GiB memory and 24.133907 seconds
+(EngineCore_DP0 pid=3312712) INFO 02-03 01:42:00 [backends.py:805] Using cache directory: /home/mingzhe/.cache/vllm/torch_compile_cache/9c9322c6b2/rank_0_0/backbone for vLLM's torch.compile
+(EngineCore_DP0 pid=3312712) INFO 02-03 01:42:00 [backends.py:865] Dynamo bytecode transform time: 16.15 s
+(EngineCore_DP0 pid=3312712) INFO 02-03 01:42:20 [backends.py:302] Cache the graph of compile range (1, 8192) for later use
+(EngineCore_DP0 pid=3312712) INFO 02-03 01:44:12 [backends.py:319] Compiling a graph for compile range (1, 8192) takes 120.43 s
+(EngineCore_DP0 pid=3312712) INFO 02-03 01:44:12 [monitor.py:34] torch.compile takes 136.58 s in total
+(EngineCore_DP0 pid=3312712) ERROR 02-03 01:44:12 [core.py:946] EngineCore failed to start.
+(EngineCore_DP0 pid=3312712) ERROR 02-03 01:44:12 [core.py:946] Traceback (most recent call last):
+(EngineCore_DP0 pid=3312712) ERROR 02-03 01:44:12 [core.py:946]   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/vllm/v1/worker/gpu_model_runner.py", line 4800, in _dummy_sampler_run
+(EngineCore_DP0 pid=3312712) ERROR 02-03 01:44:12 [core.py:946]     sampler_output = self.sampler(
+(EngineCore_DP0 pid=3312712) ERROR 02-03 01:44:12 [core.py:946]                      ^^^^^^^^^^^^^
+(EngineCore_DP0 pid=3312712) ERROR 02-03 01:44:12 [core.py:946]   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/torch/nn/modules/module.py", line 1775, in _wrapped_call_impl
+(EngineCore_DP0 pid=3312712) ERROR 02-03 01:44:12 [core.py:946]     return self._call_impl(*args, **kwargs)
+(EngineCore_DP0 pid=3312712) ERROR 02-03 01:44:12 [core.py:946]            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+(EngineCore_DP0 pid=3312712) ERROR 02-03 01:44:12 [core.py:946]   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/torch/nn/modules/module.py", line 1786, in _call_impl
+(EngineCore_DP0 pid=3312712) ERROR 02-03 01:44:12 [core.py:946]     return forward_call(*args, **kwargs)
+(EngineCore_DP0 pid=3312712) ERROR 02-03 01:44:12 [core.py:946]            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+(EngineCore_DP0 pid=3312712) ERROR 02-03 01:44:12 [core.py:946]   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/vllm/v1/sample/sampler.py", line 96, in forward
+(EngineCore_DP0 pid=3312712) ERROR 02-03 01:44:12 [core.py:946]     sampled, processed_logprobs = self.sample(logits, sampling_metadata)
+(EngineCore_DP0 pid=3312712) ERROR 02-03 01:44:12 [core.py:946]                                   ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+(EngineCore_DP0 pid=3312712) ERROR 02-03 01:44:12 [core.py:946]   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/vllm/v1/sample/sampler.py", line 187, in sample
+(EngineCore_DP0 pid=3312712) ERROR 02-03 01:44:12 [core.py:946]     random_sampled, processed_logprobs = self.topk_topp_sampler(
+(EngineCore_DP0 pid=3312712) ERROR 02-03 01:44:12 [core.py:946]                                          ^^^^^^^^^^^^^^^^^^^^^^^
+(EngineCore_DP0 pid=3312712) ERROR 02-03 01:44:12 [core.py:946]   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/torch/nn/modules/module.py", line 1775, in _wrapped_call_impl
+(EngineCore_DP0 pid=3312712) ERROR 02-03 01:44:12 [core.py:946]     return self._call_impl(*args, **kwargs)
+(EngineCore_DP0 pid=3312712) ERROR 02-03 01:44:12 [core.py:946]            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+(EngineCore_DP0 pid=3312712) ERROR 02-03 01:44:12 [core.py:946]   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/torch/nn/modules/module.py", line 1786, in _call_impl
+(EngineCore_DP0 pid=3312712) ERROR 02-03 01:44:12 [core.py:946]     return forward_call(*args, **kwargs)
+(EngineCore_DP0 pid=3312712) ERROR 02-03 01:44:12 [core.py:946]            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+(EngineCore_DP0 pid=3312712) ERROR 02-03 01:44:12 [core.py:946]   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/vllm/v1/sample/ops/topk_topp_sampler.py", line 104, in forward_native
+(EngineCore_DP0 pid=3312712) ERROR 02-03 01:44:12 [core.py:946]     logits = self.apply_top_k_top_p(logits, k, p)
+(EngineCore_DP0 pid=3312712) ERROR 02-03 01:44:12 [core.py:946]              ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+(EngineCore_DP0 pid=3312712) ERROR 02-03 01:44:12 [core.py:946]   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/vllm/v1/sample/ops/topk_topp_sampler.py", line 262, in apply_top_k_top_p
+(EngineCore_DP0 pid=3312712) ERROR 02-03 01:44:12 [core.py:946]     logits_sort, logits_idx = logits.sort(dim=-1, descending=False)
+(EngineCore_DP0 pid=3312712) ERROR 02-03 01:44:12 [core.py:946]                               ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+(EngineCore_DP0 pid=3312712) ERROR 02-03 01:44:12 [core.py:946] torch.OutOfMemoryError: CUDA out of memory. Tried to allocate 1.74 GiB. GPU 0 has a total capacity of 139.80 GiB of which 368.44 MiB is free. Process 603285 has 4.61 GiB memory in use. Process 3310619 has 68.65 GiB memory in use. Including non-PyTorch memory, this process has 66.16 GiB memory in use. Of the allocated memory 65.02 GiB is allocated by PyTorch, and 416.55 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)
+(EngineCore_DP0 pid=3312712) ERROR 02-03 01:44:12 [core.py:946] 
+(EngineCore_DP0 pid=3312712) ERROR 02-03 01:44:12 [core.py:946] The above exception was the direct cause of the following exception:
+(EngineCore_DP0 pid=3312712) ERROR 02-03 01:44:12 [core.py:946] 
+(EngineCore_DP0 pid=3312712) ERROR 02-03 01:44:12 [core.py:946] Traceback (most recent call last):
+(EngineCore_DP0 pid=3312712) ERROR 02-03 01:44:12 [core.py:946]   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/vllm/v1/engine/core.py", line 937, in run_engine_core
+(EngineCore_DP0 pid=3312712) ERROR 02-03 01:44:12 [core.py:946]     engine_core = EngineCoreProc(*args, engine_index=dp_rank, **kwargs)
+(EngineCore_DP0 pid=3312712) ERROR 02-03 01:44:12 [core.py:946]                   ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+(EngineCore_DP0 pid=3312712) ERROR 02-03 01:44:12 [core.py:946]   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/vllm/v1/engine/core.py", line 691, in __init__
+(EngineCore_DP0 pid=3312712) ERROR 02-03 01:44:12 [core.py:946]     super().__init__(
+(EngineCore_DP0 pid=3312712) ERROR 02-03 01:44:12 [core.py:946]   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/vllm/v1/engine/core.py", line 112, in __init__
+(EngineCore_DP0 pid=3312712) ERROR 02-03 01:44:12 [core.py:946]     num_gpu_blocks, num_cpu_blocks, kv_cache_config = self._initialize_kv_caches(
+(EngineCore_DP0 pid=3312712) ERROR 02-03 01:44:12 [core.py:946]                                                       ^^^^^^^^^^^^^^^^^^^^^^^^^^^
+(EngineCore_DP0 pid=3312712) ERROR 02-03 01:44:12 [core.py:946]   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/vllm/v1/engine/core.py", line 242, in _initialize_kv_caches
+(EngineCore_DP0 pid=3312712) ERROR 02-03 01:44:12 [core.py:946]     available_gpu_memory = self.model_executor.determine_available_memory()
+(EngineCore_DP0 pid=3312712) ERROR 02-03 01:44:12 [core.py:946]                            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+(EngineCore_DP0 pid=3312712) ERROR 02-03 01:44:12 [core.py:946]   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/vllm/v1/executor/abstract.py", line 126, in determine_available_memory
+(EngineCore_DP0 pid=3312712) ERROR 02-03 01:44:12 [core.py:946]     return self.collective_rpc("determine_available_memory")
+(EngineCore_DP0 pid=3312712) ERROR 02-03 01:44:12 [core.py:946]            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+(EngineCore_DP0 pid=3312712) ERROR 02-03 01:44:12 [core.py:946]   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/vllm/v1/executor/uniproc_executor.py", line 75, in collective_rpc
+(EngineCore_DP0 pid=3312712) ERROR 02-03 01:44:12 [core.py:946]     result = run_method(self.driver_worker, method, args, kwargs)
+(EngineCore_DP0 pid=3312712) ERROR 02-03 01:44:12 [core.py:946]              ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+(EngineCore_DP0 pid=3312712) ERROR 02-03 01:44:12 [core.py:946]   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/vllm/v1/serial_utils.py", line 461, in run_method
+(EngineCore_DP0 pid=3312712) ERROR 02-03 01:44:12 [core.py:946]     return func(*args, **kwargs)
+(EngineCore_DP0 pid=3312712) ERROR 02-03 01:44:12 [core.py:946]            ^^^^^^^^^^^^^^^^^^^^^
+(EngineCore_DP0 pid=3312712) ERROR 02-03 01:44:12 [core.py:946]   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/torch/utils/_contextlib.py", line 120, in decorate_context
+(EngineCore_DP0 pid=3312712) ERROR 02-03 01:44:12 [core.py:946]     return func(*args, **kwargs)
+(EngineCore_DP0 pid=3312712) ERROR 02-03 01:44:12 [core.py:946]            ^^^^^^^^^^^^^^^^^^^^^
+(EngineCore_DP0 pid=3312712) ERROR 02-03 01:44:12 [core.py:946]   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/vllm/v1/worker/gpu_worker.py", line 322, in determine_available_memory
+(EngineCore_DP0 pid=3312712) ERROR 02-03 01:44:12 [core.py:946]     self.model_runner.profile_run()
+(EngineCore_DP0 pid=3312712) ERROR 02-03 01:44:12 [core.py:946]   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/vllm/v1/worker/gpu_model_runner.py", line 4981, in profile_run
+(EngineCore_DP0 pid=3312712) ERROR 02-03 01:44:12 [core.py:946]     output = self._dummy_sampler_run(last_hidden_states)
+(EngineCore_DP0 pid=3312712) ERROR 02-03 01:44:12 [core.py:946]              ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+(EngineCore_DP0 pid=3312712) ERROR 02-03 01:44:12 [core.py:946]   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/torch/utils/_contextlib.py", line 120, in decorate_context
+(EngineCore_DP0 pid=3312712) ERROR 02-03 01:44:12 [core.py:946]     return func(*args, **kwargs)
+(EngineCore_DP0 pid=3312712) ERROR 02-03 01:44:12 [core.py:946]            ^^^^^^^^^^^^^^^^^^^^^
+(EngineCore_DP0 pid=3312712) ERROR 02-03 01:44:12 [core.py:946]   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/vllm/v1/worker/gpu_model_runner.py", line 4805, in _dummy_sampler_run
+(EngineCore_DP0 pid=3312712) ERROR 02-03 01:44:12 [core.py:946]     raise RuntimeError(
+(EngineCore_DP0 pid=3312712) ERROR 02-03 01:44:12 [core.py:946] RuntimeError: CUDA out of memory occurred when warming up sampler with 1024 dummy requests. Please try lowering `max_num_seqs` or `gpu_memory_utilization` when initializing the engine.
+(EngineCore_DP0 pid=3312712) Process EngineCore_DP0:
+(EngineCore_DP0 pid=3312712) Traceback (most recent call last):
+(EngineCore_DP0 pid=3312712)   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/vllm/v1/worker/gpu_model_runner.py", line 4800, in _dummy_sampler_run
+(EngineCore_DP0 pid=3312712)     sampler_output = self.sampler(
+(EngineCore_DP0 pid=3312712)                      ^^^^^^^^^^^^^
+(EngineCore_DP0 pid=3312712)   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/torch/nn/modules/module.py", line 1775, in _wrapped_call_impl
+(EngineCore_DP0 pid=3312712)     return self._call_impl(*args, **kwargs)
+(EngineCore_DP0 pid=3312712)            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+(EngineCore_DP0 pid=3312712)   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/torch/nn/modules/module.py", line 1786, in _call_impl
+(EngineCore_DP0 pid=3312712)     return forward_call(*args, **kwargs)
+(EngineCore_DP0 pid=3312712)            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+(EngineCore_DP0 pid=3312712)   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/vllm/v1/sample/sampler.py", line 96, in forward
+(EngineCore_DP0 pid=3312712)     sampled, processed_logprobs = self.sample(logits, sampling_metadata)
+(EngineCore_DP0 pid=3312712)                                   ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+(EngineCore_DP0 pid=3312712)   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/vllm/v1/sample/sampler.py", line 187, in sample
+(EngineCore_DP0 pid=3312712)     random_sampled, processed_logprobs = self.topk_topp_sampler(
+(EngineCore_DP0 pid=3312712)                                          ^^^^^^^^^^^^^^^^^^^^^^^
+(EngineCore_DP0 pid=3312712)   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/torch/nn/modules/module.py", line 1775, in _wrapped_call_impl
+(EngineCore_DP0 pid=3312712)     return self._call_impl(*args, **kwargs)
+(EngineCore_DP0 pid=3312712)            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+(EngineCore_DP0 pid=3312712)   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/torch/nn/modules/module.py", line 1786, in _call_impl
+(EngineCore_DP0 pid=3312712)     return forward_call(*args, **kwargs)
+(EngineCore_DP0 pid=3312712)            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+(EngineCore_DP0 pid=3312712)   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/vllm/v1/sample/ops/topk_topp_sampler.py", line 104, in forward_native
+(EngineCore_DP0 pid=3312712)     logits = self.apply_top_k_top_p(logits, k, p)
+(EngineCore_DP0 pid=3312712)              ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+(EngineCore_DP0 pid=3312712)   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/vllm/v1/sample/ops/topk_topp_sampler.py", line 262, in apply_top_k_top_p
+(EngineCore_DP0 pid=3312712)     logits_sort, logits_idx = logits.sort(dim=-1, descending=False)
+(EngineCore_DP0 pid=3312712)                               ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+(EngineCore_DP0 pid=3312712) torch.OutOfMemoryError: CUDA out of memory. Tried to allocate 1.74 GiB. GPU 0 has a total capacity of 139.80 GiB of which 368.44 MiB is free. Process 603285 has 4.61 GiB memory in use. Process 3310619 has 68.65 GiB memory in use. Including non-PyTorch memory, this process has 66.16 GiB memory in use. Of the allocated memory 65.02 GiB is allocated by PyTorch, and 416.55 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)
+(EngineCore_DP0 pid=3312712) 
+(EngineCore_DP0 pid=3312712) The above exception was the direct cause of the following exception:
+(EngineCore_DP0 pid=3312712) 
+(EngineCore_DP0 pid=3312712) Traceback (most recent call last):
+(EngineCore_DP0 pid=3312712)   File "/usr/lib/python3.12/multiprocessing/process.py", line 314, in _bootstrap
+(EngineCore_DP0 pid=3312712)     self.run()
+(EngineCore_DP0 pid=3312712)   File "/usr/lib/python3.12/multiprocessing/process.py", line 108, in run
+(EngineCore_DP0 pid=3312712)     self._target(*self._args, **self._kwargs)
+(EngineCore_DP0 pid=3312712)   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/vllm/v1/engine/core.py", line 950, in run_engine_core
+(EngineCore_DP0 pid=3312712)     raise e
+(EngineCore_DP0 pid=3312712)   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/vllm/v1/engine/core.py", line 937, in run_engine_core
+(EngineCore_DP0 pid=3312712)     engine_core = EngineCoreProc(*args, engine_index=dp_rank, **kwargs)
+(EngineCore_DP0 pid=3312712)                   ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+(EngineCore_DP0 pid=3312712)   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/vllm/v1/engine/core.py", line 691, in __init__
+(EngineCore_DP0 pid=3312712)     super().__init__(
+(EngineCore_DP0 pid=3312712)   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/vllm/v1/engine/core.py", line 112, in __init__
+(EngineCore_DP0 pid=3312712)     num_gpu_blocks, num_cpu_blocks, kv_cache_config = self._initialize_kv_caches(
+(EngineCore_DP0 pid=3312712)                                                       ^^^^^^^^^^^^^^^^^^^^^^^^^^^
+(EngineCore_DP0 pid=3312712)   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/vllm/v1/engine/core.py", line 242, in _initialize_kv_caches
+(EngineCore_DP0 pid=3312712)     available_gpu_memory = self.model_executor.determine_available_memory()
+(EngineCore_DP0 pid=3312712)                            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+(EngineCore_DP0 pid=3312712)   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/vllm/v1/executor/abstract.py", line 126, in determine_available_memory
+(EngineCore_DP0 pid=3312712)     return self.collective_rpc("determine_available_memory")
+(EngineCore_DP0 pid=3312712)            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+(EngineCore_DP0 pid=3312712)   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/vllm/v1/executor/uniproc_executor.py", line 75, in collective_rpc
+(EngineCore_DP0 pid=3312712)     result = run_method(self.driver_worker, method, args, kwargs)
+(EngineCore_DP0 pid=3312712)              ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+(EngineCore_DP0 pid=3312712)   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/vllm/v1/serial_utils.py", line 461, in run_method
+(EngineCore_DP0 pid=3312712)     return func(*args, **kwargs)
+(EngineCore_DP0 pid=3312712)            ^^^^^^^^^^^^^^^^^^^^^
+(EngineCore_DP0 pid=3312712)   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/torch/utils/_contextlib.py", line 120, in decorate_context
+(EngineCore_DP0 pid=3312712)     return func(*args, **kwargs)
+(EngineCore_DP0 pid=3312712)            ^^^^^^^^^^^^^^^^^^^^^
+(EngineCore_DP0 pid=3312712)   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/vllm/v1/worker/gpu_worker.py", line 322, in determine_available_memory
+(EngineCore_DP0 pid=3312712)     self.model_runner.profile_run()
+(EngineCore_DP0 pid=3312712)   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/vllm/v1/worker/gpu_model_runner.py", line 4981, in profile_run
+(EngineCore_DP0 pid=3312712)     output = self._dummy_sampler_run(last_hidden_states)
+(EngineCore_DP0 pid=3312712)              ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+(EngineCore_DP0 pid=3312712)   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/torch/utils/_contextlib.py", line 120, in decorate_context
+(EngineCore_DP0 pid=3312712)     return func(*args, **kwargs)
+(EngineCore_DP0 pid=3312712)            ^^^^^^^^^^^^^^^^^^^^^
+(EngineCore_DP0 pid=3312712)   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/vllm/v1/worker/gpu_model_runner.py", line 4805, in _dummy_sampler_run
+(EngineCore_DP0 pid=3312712)     raise RuntimeError(
+(EngineCore_DP0 pid=3312712) RuntimeError: CUDA out of memory occurred when warming up sampler with 1024 dummy requests. Please try lowering `max_num_seqs` or `gpu_memory_utilization` when initializing the engine.
+[rank0]:[W203 01:44:15.196915434 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=3309898) Traceback (most recent call last):
+(APIServer pid=3309898)   File "<frozen runpy>", line 198, in _run_module_as_main
+(APIServer pid=3309898)   File "<frozen runpy>", line 88, in _run_code
+(APIServer pid=3309898)   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/vllm/entrypoints/openai/api_server.py", line 991, in <module>
+(APIServer pid=3309898)     uvloop.run(run_server(args))
+(APIServer pid=3309898)   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/uvloop/__init__.py", line 96, in run
+(APIServer pid=3309898)     return __asyncio.run(
+(APIServer pid=3309898)            ^^^^^^^^^^^^^^
+(APIServer pid=3309898)   File "/usr/lib/python3.12/asyncio/runners.py", line 194, in run
+(APIServer pid=3309898)     return runner.run(main)
+(APIServer pid=3309898)            ^^^^^^^^^^^^^^^^
+(APIServer pid=3309898)   File "/usr/lib/python3.12/asyncio/runners.py", line 118, in run
+(APIServer pid=3309898)     return self._loop.run_until_complete(task)
+(APIServer pid=3309898)            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+(APIServer pid=3309898)   File "uvloop/loop.pyx", line 1518, in uvloop.loop.Loop.run_until_complete
+(APIServer pid=3309898)   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/uvloop/__init__.py", line 48, in wrapper
+(APIServer pid=3309898)     return await main
+(APIServer pid=3309898)            ^^^^^^^^^^
+(APIServer pid=3309898)   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/vllm/entrypoints/openai/api_server.py", line 919, in run_server
+(APIServer pid=3309898)     await run_server_worker(listen_address, sock, args, **uvicorn_kwargs)
+(APIServer pid=3309898)   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/vllm/entrypoints/openai/api_server.py", line 938, in run_server_worker
+(APIServer pid=3309898)     async with build_async_engine_client(
+(APIServer pid=3309898)   File "/usr/lib/python3.12/contextlib.py", line 210, in __aenter__
+(APIServer pid=3309898)     return await anext(self.gen)
+(APIServer pid=3309898)            ^^^^^^^^^^^^^^^^^^^^^
+(APIServer pid=3309898)   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/vllm/entrypoints/openai/api_server.py", line 147, in build_async_engine_client
+(APIServer pid=3309898)     async with build_async_engine_client_from_engine_args(
+(APIServer pid=3309898)   File "/usr/lib/python3.12/contextlib.py", line 210, in __aenter__
+(APIServer pid=3309898)     return await anext(self.gen)
+(APIServer pid=3309898)            ^^^^^^^^^^^^^^^^^^^^^
+(APIServer pid=3309898)   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/vllm/entrypoints/openai/api_server.py", line 188, in build_async_engine_client_from_engine_args
+(APIServer pid=3309898)     async_llm = AsyncLLM.from_vllm_config(
+(APIServer pid=3309898)                 ^^^^^^^^^^^^^^^^^^^^^^^^^^
+(APIServer pid=3309898)   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/vllm/v1/engine/async_llm.py", line 228, in from_vllm_config
+(APIServer pid=3309898)     return cls(
+(APIServer pid=3309898)            ^^^^
+(APIServer pid=3309898)   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/vllm/v1/engine/async_llm.py", line 155, in __init__
+(APIServer pid=3309898)     self.engine_core = EngineCoreClient.make_async_mp_client(
+(APIServer pid=3309898)                        ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+(APIServer pid=3309898)   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/vllm/v1/engine/core_client.py", line 122, in make_async_mp_client
+(APIServer pid=3309898)     return AsyncMPClient(*client_args)
+(APIServer pid=3309898)            ^^^^^^^^^^^^^^^^^^^^^^^^^^^
+(APIServer pid=3309898)   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/vllm/v1/engine/core_client.py", line 819, in __init__
+(APIServer pid=3309898)     super().__init__(
+(APIServer pid=3309898)   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/vllm/v1/engine/core_client.py", line 479, in __init__
+(APIServer pid=3309898)     with launch_core_engines(vllm_config, executor_class, log_stats) as (
+(APIServer pid=3309898)   File "/usr/lib/python3.12/contextlib.py", line 144, in __exit__
+(APIServer pid=3309898)     next(self.gen)
+(APIServer pid=3309898)   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/vllm/v1/engine/utils.py", line 933, in launch_core_engines
+(APIServer pid=3309898)     wait_for_engine_startup(
+(APIServer pid=3309898)   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/vllm/v1/engine/utils.py", line 992, in wait_for_engine_startup
+(APIServer pid=3309898)     raise RuntimeError(
+(APIServer pid=3309898) RuntimeError: Engine core initialization failed. See root cause above. Failed core proc(s): {}

vllm_0006000.log

@@ -1,20 +1,18 @@
-(APIServer pid=3233259) INFO 02-03 01:26:50 [utils.py:325] 
-(APIServer pid=3233259) INFO 02-03 01:26:50 [utils.py:325]        █     █     █▄   ▄█
-(APIServer pid=3233259) INFO 02-03 01:26:50 [utils.py:325]  ▄▄ ▄█ █     █     █ ▀▄▀ █  version 0.15.0
-(APIServer pid=3233259) INFO 02-03 01:26:50 [utils.py:325]   █▄█▀ █     █     █     █  model   Elfsong/VLM_stage_2_iter_0006000
-(APIServer pid=3233259) INFO 02-03 01:26:50 [utils.py:325]    ▀▀  ▀▀▀▀▀ ▀▀▀▀▀ ▀     ▀
-(APIServer pid=3233259) INFO 02-03 01:26:50 [utils.py:325] 
-(APIServer pid=3233259) INFO 02-03 01:26:50 [utils.py:261] non-default args: {'port': 9002, 'model': 'Elfsong/VLM_stage_2_iter_0006000', 'trust_remote_code': True, 'quantization': 'bitsandbytes', 'gpu_memory_utilization': 0.3}
-(APIServer pid=3233259) The argument `trust_remote_code` is to be used with Auto classes. It has no effect here and is ignored.
-(APIServer pid=3233259) The argument `trust_remote_code` is to be used with Auto classes. It has no effect here and is ignored.
-(APIServer pid=3233259) INFO 02-03 01:26:51 [model.py:541] Resolved architecture: Qwen3ForCausalLM
-(APIServer pid=3233259) INFO 02-03 01:26:51 [model.py:1561] Using max model len 40960
-(APIServer pid=3233259) INFO 02-03 01:26:52 [scheduler.py:226] Chunked prefill is enabled with max_num_batched_tokens=8192.
-(APIServer pid=3233259) INFO 02-03 01:26:55 [vllm.py:624] Asynchronous scheduling is enabled.
-(EngineCore_DP0 pid=3235235) INFO 02-03 01:27:09 [core.py:96] Initializing a V1 LLM engine (v0.15.0) with config: model='Elfsong/VLM_stage_2_iter_0006000', speculative_config=None, tokenizer='Elfsong/VLM_stage_2_iter_0006000', 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, enable_return_routed_experts=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, enable_mfu_metrics=False, enable_mm_processor_stats=False, enable_logging_iteration_details=False), seed=0, served_model_name=Elfsong/VLM_stage_2_iter_0006000, 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', 'vllm::rocm_aiter_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, 'assume_32_bit_indexing': True}, 'local_cache_dir': None}
-(EngineCore_DP0 pid=3235235) INFO 02-03 01:27:10 [parallel_state.py:1212] world_size=1 rank=0 local_rank=0 distributed_init_method=tcp://172.21.25.98:60911 backend=nccl
-(EngineCore_DP0 pid=3235235) INFO 02-03 01:27:10 [parallel_state.py:1423] rank 0 in world size 1 is assigned as DP rank 0, PP rank 0, PCP rank 0, TP rank 0, EP rank N/A
-(EngineCore_DP0 pid=3235235) INFO 02-03 01:27:11 [gpu_model_runner.py:4021] Starting to load model Elfsong/VLM_stage_2_iter_0006000...
-(EngineCore_DP0 pid=3235235) INFO 02-03 01:27:13 [cuda.py:364] Using FLASH_ATTN attention backend out of potential backends: ('FLASH_ATTN', 'FLASHINFER', 'TRITON_ATTN', 'FLEX_ATTENTION')
-(EngineCore_DP0 pid=3235235) INFO 02-03 01:27:14 [bitsandbytes_loader.py:786] Loading weights with BitsAndBytes quantization. May take a while ...
-Cancellation requested; stopping current tasks.
+(APIServer pid=3310473) INFO 02-03 01:41:05 [utils.py:325] 
+(APIServer pid=3310473) INFO 02-03 01:41:05 [utils.py:325]        █     █     █▄   ▄█
+(APIServer pid=3310473) INFO 02-03 01:41:05 [utils.py:325]  ▄▄ ▄█ █     █     █ ▀▄▀ █  version 0.15.0
+(APIServer pid=3310473) INFO 02-03 01:41:05 [utils.py:325]   █▄█▀ █     █     █     █  model   Elfsong/VLM_stage_2_iter_0006000
+(APIServer pid=3310473) INFO 02-03 01:41:05 [utils.py:325]    ▀▀  ▀▀▀▀▀ ▀▀▀▀▀ ▀     ▀
+(APIServer pid=3310473) INFO 02-03 01:41:05 [utils.py:325] 
+(APIServer pid=3310473) INFO 02-03 01:41:05 [utils.py:261] non-default args: {'port': 9002, 'model': 'Elfsong/VLM_stage_2_iter_0006000', 'trust_remote_code': True, 'gpu_memory_utilization': 0.4}
+(APIServer pid=3310473) The argument `trust_remote_code` is to be used with Auto classes. It has no effect here and is ignored.
+(APIServer pid=3310473) The argument `trust_remote_code` is to be used with Auto classes. It has no effect here and is ignored.
+(APIServer pid=3310473) INFO 02-03 01:41:06 [model.py:541] Resolved architecture: Qwen3ForCausalLM
+(APIServer pid=3310473) INFO 02-03 01:41:06 [model.py:1561] Using max model len 40960
+(APIServer pid=3310473) INFO 02-03 01:41:06 [scheduler.py:226] Chunked prefill is enabled with max_num_batched_tokens=8192.
+(APIServer pid=3310473) INFO 02-03 01:41:06 [vllm.py:624] Asynchronous scheduling is enabled.
+(EngineCore_DP0 pid=3313628) INFO 02-03 01:41:18 [core.py:96] Initializing a V1 LLM engine (v0.15.0) with config: model='Elfsong/VLM_stage_2_iter_0006000', speculative_config=None, tokenizer='Elfsong/VLM_stage_2_iter_0006000', 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=auto, tensor_parallel_size=1, pipeline_parallel_size=1, data_parallel_size=1, disable_custom_all_reduce=False, quantization=None, enforce_eager=False, enable_return_routed_experts=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, enable_mfu_metrics=False, enable_mm_processor_stats=False, enable_logging_iteration_details=False), seed=0, served_model_name=Elfsong/VLM_stage_2_iter_0006000, 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', 'vllm::rocm_aiter_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, 'assume_32_bit_indexing': True}, 'local_cache_dir': None}
+(EngineCore_DP0 pid=3313628) INFO 02-03 01:41:21 [parallel_state.py:1212] world_size=1 rank=0 local_rank=0 distributed_init_method=tcp://172.21.25.98:47213 backend=nccl
+(EngineCore_DP0 pid=3313628) INFO 02-03 01:41:21 [parallel_state.py:1423] rank 0 in world size 1 is assigned as DP rank 0, PP rank 0, PCP rank 0, TP rank 0, EP rank N/A
+(EngineCore_DP0 pid=3313628) INFO 02-03 01:41:23 [gpu_model_runner.py:4021] Starting to load model Elfsong/VLM_stage_2_iter_0006000...
+(EngineCore_DP0 pid=3313628) INFO 02-03 01:41:24 [cuda.py:364] Using FLASH_ATTN attention backend out of potential backends: ('FLASH_ATTN', 'FLASHINFER', 'TRITON_ATTN', 'FLEX_ATTENTION')

vllm_0006500.log

@@ -1,97 +1,203 @@
-(APIServer pid=2014082) INFO 02-02 06:58:38 [api_server.py:1351] vLLM API server version 0.13.0
-(APIServer pid=2014082) INFO 02-02 06:58:38 [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=2014082) The argument `trust_remote_code` is to be used with Auto classes. It has no effect here and is ignored.
-(APIServer pid=2014082) INFO 02-02 06:58:39 [model.py:514] Resolved architecture: Qwen3ForCausalLM
-(APIServer pid=2014082) INFO 02-02 06:58:39 [model.py:1661] Using max model len 40960
-(APIServer pid=2014082) INFO 02-02 06:58:40 [scheduler.py:230] Chunked prefill is enabled with max_num_batched_tokens=8192.
-(EngineCore_DP0 pid=2015333) INFO 02-02 06:59:00 [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=2015333) INFO 02-02 06:59:01 [parallel_state.py:1203] world_size=1 rank=0 local_rank=0 distributed_init_method=tcp://100.96.20.65:51155 backend=nccl
-(EngineCore_DP0 pid=2015333) INFO 02-02 06:59:01 [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=2015333) INFO 02-02 06:59:02 [gpu_model_runner.py:3562] Starting to load model Elfsong/VLM_stage_2_iter_0006500...
-(EngineCore_DP0 pid=2015333) INFO 02-02 06:59:04 [cuda.py:351] Using FLASH_ATTN attention backend out of potential backends: ('FLASH_ATTN', 'FLASHINFER', 'TRITON_ATTN', 'FLEX_ATTENTION')
-(EngineCore_DP0 pid=2015333) INFO 02-02 06:59:04 [bitsandbytes_loader.py:791] Loading weights with BitsAndBytes quantization. May take a while ...
-(EngineCore_DP0 pid=2015333) 
-(EngineCore_DP0 pid=2015333) 
-(EngineCore_DP0 pid=2015333) 
-(EngineCore_DP0 pid=2015333) 
-(EngineCore_DP0 pid=2015333) 
-(EngineCore_DP0 pid=2015333) 
-(EngineCore_DP0 pid=2015333) 
-(EngineCore_DP0 pid=2015333) 
-(EngineCore_DP0 pid=2015333) 
-(EngineCore_DP0 pid=2015333) 
-(EngineCore_DP0 pid=2015333) 
-(EngineCore_DP0 pid=2015333) 
-(EngineCore_DP0 pid=2015333) 
-(EngineCore_DP0 pid=2015333) 
-(EngineCore_DP0 pid=2015333) 
-(EngineCore_DP0 pid=2015333) 
-(EngineCore_DP0 pid=2015333) 
-(EngineCore_DP0 pid=2015333) 
-(EngineCore_DP0 pid=2015333) 
-(EngineCore_DP0 pid=2015333) 
-(EngineCore_DP0 pid=2015333) INFO 02-02 06:59:51 [gpu_model_runner.py:3659] Model loading took 19.4031 GiB memory and 48.128302 seconds
-(EngineCore_DP0 pid=2015333) INFO 02-02 07:00:02 [backends.py:643] Using cache directory: /home/mingzhed/.cache/vllm/torch_compile_cache/7553580953/rank_0_0/backbone for vLLM's torch.compile
-(EngineCore_DP0 pid=2015333) INFO 02-02 07:00:02 [backends.py:703] Dynamo bytecode transform time: 10.49 s
-(EngineCore_DP0 pid=2015333) INFO 02-02 07:00:10 [backends.py:261] Cache the graph of compile range (1, 8192) for later use
-(EngineCore_DP0 pid=2015333) INFO 02-02 07:00:13 [backends.py:278] Compiling a graph for compile range (1, 8192) takes 3.70 s
-(EngineCore_DP0 pid=2015333) INFO 02-02 07:00:13 [monitor.py:34] torch.compile takes 14.19 s in total
-(EngineCore_DP0 pid=2015333) INFO 02-02 07:00:14 [gpu_worker.py:375] Available KV cache memory: 49.18 GiB
-(EngineCore_DP0 pid=2015333) INFO 02-02 07:00:15 [kv_cache_utils.py:1291] GPU KV cache size: 201,440 tokens
-(EngineCore_DP0 pid=2015333) INFO 02-02 07:00:15 [kv_cache_utils.py:1296] Maximum concurrency for 40,960 tokens per request: 4.92x
-(EngineCore_DP0 pid=2015333) 
-(EngineCore_DP0 pid=2015333) 
-(EngineCore_DP0 pid=2015333) INFO 02-02 07:00:31 [gpu_model_runner.py:4587] Graph capturing finished in 16 secs, took 3.22 GiB
-(EngineCore_DP0 pid=2015333) INFO 02-02 07:00:31 [core.py:259] init engine (profile, create kv cache, warmup model) took 40.24 seconds
-(APIServer pid=2014082) INFO 02-02 07:00:33 [api_server.py:1099] Supported tasks: ['generate']
-(APIServer pid=2014082) WARNING 02-02 07:00:34 [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=2014082) INFO 02-02 07:00:34 [serving_responses.py:201] Using default chat sampling params from model: {'temperature': 0.6, 'top_k': 20, 'top_p': 0.95}
-(APIServer pid=2014082) INFO 02-02 07:00:34 [serving_chat.py:137] Using default chat sampling params from model: {'temperature': 0.6, 'top_k': 20, 'top_p': 0.95}
-(APIServer pid=2014082) INFO 02-02 07:00:34 [serving_completion.py:77] Using default completion sampling params from model: {'temperature': 0.6, 'top_k': 20, 'top_p': 0.95}
-(APIServer pid=2014082) INFO 02-02 07:00:34 [serving_chat.py:137] Using default chat sampling params from model: {'temperature': 0.6, 'top_k': 20, 'top_p': 0.95}
-(APIServer pid=2014082) INFO 02-02 07:00:34 [api_server.py:1425] Starting vLLM API server 0 on http://0.0.0.0:9011
-(APIServer pid=2014082) INFO 02-02 07:00:34 [launcher.py:38] Available routes are:
-(APIServer pid=2014082) INFO 02-02 07:00:34 [launcher.py:46] Route: /openapi.json, Methods: GET, HEAD
-(APIServer pid=2014082) INFO 02-02 07:00:34 [launcher.py:46] Route: /docs, Methods: GET, HEAD
-(APIServer pid=2014082) INFO 02-02 07:00:34 [launcher.py:46] Route: /docs/oauth2-redirect, Methods: GET, HEAD
-(APIServer pid=2014082) INFO 02-02 07:00:34 [launcher.py:46] Route: /redoc, Methods: GET, HEAD
-(APIServer pid=2014082) INFO 02-02 07:00:34 [launcher.py:46] Route: /scale_elastic_ep, Methods: POST
-(APIServer pid=2014082) INFO 02-02 07:00:34 [launcher.py:46] Route: /is_scaling_elastic_ep, Methods: POST
-(APIServer pid=2014082) INFO 02-02 07:00:34 [launcher.py:46] Route: /tokenize, Methods: POST
-(APIServer pid=2014082) INFO 02-02 07:00:34 [launcher.py:46] Route: /detokenize, Methods: POST
-(APIServer pid=2014082) INFO 02-02 07:00:34 [launcher.py:46] Route: /inference/v1/generate, Methods: POST
-(APIServer pid=2014082) INFO 02-02 07:00:34 [launcher.py:46] Route: /pause, Methods: POST
-(APIServer pid=2014082) INFO 02-02 07:00:34 [launcher.py:46] Route: /resume, Methods: POST
-(APIServer pid=2014082) INFO 02-02 07:00:34 [launcher.py:46] Route: /is_paused, Methods: GET
-(APIServer pid=2014082) INFO 02-02 07:00:34 [launcher.py:46] Route: /metrics, Methods: GET
-(APIServer pid=2014082) INFO 02-02 07:00:34 [launcher.py:46] Route: /health, Methods: GET
-(APIServer pid=2014082) INFO 02-02 07:00:34 [launcher.py:46] Route: /load, Methods: GET
-(APIServer pid=2014082) INFO 02-02 07:00:34 [launcher.py:46] Route: /v1/models, Methods: GET
-(APIServer pid=2014082) INFO 02-02 07:00:34 [launcher.py:46] Route: /version, Methods: GET
-(APIServer pid=2014082) INFO 02-02 07:00:34 [launcher.py:46] Route: /v1/responses, Methods: POST
-(APIServer pid=2014082) INFO 02-02 07:00:34 [launcher.py:46] Route: /v1/responses/{response_id}, Methods: GET
-(APIServer pid=2014082) INFO 02-02 07:00:34 [launcher.py:46] Route: /v1/responses/{response_id}/cancel, Methods: POST
-(APIServer pid=2014082) INFO 02-02 07:00:34 [launcher.py:46] Route: /v1/messages, Methods: POST
-(APIServer pid=2014082) INFO 02-02 07:00:34 [launcher.py:46] Route: /v1/chat/completions, Methods: POST
-(APIServer pid=2014082) INFO 02-02 07:00:34 [launcher.py:46] Route: /v1/completions, Methods: POST
-(APIServer pid=2014082) INFO 02-02 07:00:34 [launcher.py:46] Route: /v1/audio/transcriptions, Methods: POST
-(APIServer pid=2014082) INFO 02-02 07:00:34 [launcher.py:46] Route: /v1/audio/translations, Methods: POST
-(APIServer pid=2014082) INFO 02-02 07:00:34 [launcher.py:46] Route: /ping, Methods: GET
-(APIServer pid=2014082) INFO 02-02 07:00:34 [launcher.py:46] Route: /ping, Methods: POST
-(APIServer pid=2014082) INFO 02-02 07:00:34 [launcher.py:46] Route: /invocations, Methods: POST
-(APIServer pid=2014082) INFO 02-02 07:00:34 [launcher.py:46] Route: /classify, Methods: POST
-(APIServer pid=2014082) INFO 02-02 07:00:34 [launcher.py:46] Route: /v1/embeddings, Methods: POST
-(APIServer pid=2014082) INFO 02-02 07:00:34 [launcher.py:46] Route: /score, Methods: POST
-(APIServer pid=2014082) INFO 02-02 07:00:34 [launcher.py:46] Route: /v1/score, Methods: POST
-(APIServer pid=2014082) INFO 02-02 07:00:34 [launcher.py:46] Route: /rerank, Methods: POST
-(APIServer pid=2014082) INFO 02-02 07:00:34 [launcher.py:46] Route: /v1/rerank, Methods: POST
-(APIServer pid=2014082) INFO 02-02 07:00:34 [launcher.py:46] Route: /v2/rerank, Methods: POST
-(APIServer pid=2014082) INFO 02-02 07:00:34 [launcher.py:46] Route: /pooling, Methods: POST
-(APIServer pid=2014082) INFO:     Started server process [2014082]
-(APIServer pid=2014082) INFO:     Waiting for application startup.
-(APIServer pid=2014082) INFO:     Application startup complete.
-(APIServer pid=2014082) INFO 02-02 09:26:22 [launcher.py:110] Shutting down FastAPI HTTP server.
-[rank0]:[W202 09:26:22.195432165 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=2014082) INFO:     Shutting down
-(APIServer pid=2014082) INFO:     Waiting for application shutdown.
-(APIServer pid=2014082) INFO:     Application shutdown complete.
+(APIServer pid=3311584) INFO 02-03 01:41:10 [utils.py:325] 
+(APIServer pid=3311584) INFO 02-03 01:41:10 [utils.py:325]        █     █     █▄   ▄█
+(APIServer pid=3311584) INFO 02-03 01:41:10 [utils.py:325]  ▄▄ ▄█ █     █     █ ▀▄▀ █  version 0.15.0
+(APIServer pid=3311584) INFO 02-03 01:41:10 [utils.py:325]   █▄█▀ █     █     █     █  model   Elfsong/VLM_stage_2_iter_0006500
+(APIServer pid=3311584) INFO 02-03 01:41:10 [utils.py:325]    ▀▀  ▀▀▀▀▀ ▀▀▀▀▀ ▀     ▀
+(APIServer pid=3311584) INFO 02-03 01:41:10 [utils.py:325] 
+(APIServer pid=3311584) INFO 02-03 01:41:10 [utils.py:261] non-default args: {'port': 9003, 'model': 'Elfsong/VLM_stage_2_iter_0006500', 'trust_remote_code': True, 'gpu_memory_utilization': 0.4}
+(APIServer pid=3311584) The argument `trust_remote_code` is to be used with Auto classes. It has no effect here and is ignored.
+(APIServer pid=3311584) The argument `trust_remote_code` is to be used with Auto classes. It has no effect here and is ignored.
+(APIServer pid=3311584) INFO 02-03 01:41:12 [model.py:541] Resolved architecture: Qwen3ForCausalLM
+(APIServer pid=3311584) INFO 02-03 01:41:12 [model.py:1561] Using max model len 40960
+(APIServer pid=3311584) INFO 02-03 01:41:12 [scheduler.py:226] Chunked prefill is enabled with max_num_batched_tokens=8192.
+(APIServer pid=3311584) INFO 02-03 01:41:12 [vllm.py:624] Asynchronous scheduling is enabled.
+(EngineCore_DP0 pid=3315747) INFO 02-03 01:41:31 [core.py:96] Initializing a V1 LLM engine (v0.15.0) 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=auto, tensor_parallel_size=1, pipeline_parallel_size=1, data_parallel_size=1, disable_custom_all_reduce=False, quantization=None, enforce_eager=False, enable_return_routed_experts=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, enable_mfu_metrics=False, enable_mm_processor_stats=False, enable_logging_iteration_details=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', 'vllm::rocm_aiter_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, 'assume_32_bit_indexing': True}, 'local_cache_dir': None}
+(EngineCore_DP0 pid=3315747) INFO 02-03 01:41:35 [parallel_state.py:1212] world_size=1 rank=0 local_rank=0 distributed_init_method=tcp://172.21.25.98:59489 backend=nccl
+(EngineCore_DP0 pid=3315747) INFO 02-03 01:41:35 [parallel_state.py:1423] rank 0 in world size 1 is assigned as DP rank 0, PP rank 0, PCP rank 0, TP rank 0, EP rank N/A
+(EngineCore_DP0 pid=3315747) INFO 02-03 01:41:37 [gpu_model_runner.py:4021] Starting to load model Elfsong/VLM_stage_2_iter_0006500...
+(EngineCore_DP0 pid=3315747) INFO 02-03 01:41:38 [cuda.py:364] Using FLASH_ATTN attention backend out of potential backends: ('FLASH_ATTN', 'FLASHINFER', 'TRITON_ATTN', 'FLEX_ATTENTION')
+(EngineCore_DP0 pid=3315747) ERROR 02-03 01:41:41 [gpu_model_runner.py:4116] Failed to load model - not enough GPU memory. Try lowering --gpu-memory-utilization to free memory for weights, increasing --tensor-parallel-size, or using --quantization. See https://docs.vllm.ai/en/latest/configuration/conserving_memory/ for more tips. (original error: CUDA out of memory. Tried to allocate 500.00 MiB. GPU 0 has a total capacity of 139.80 GiB of which 312.25 MiB is free. Process 3284142 has 5.32 GiB memory in use. Process 3243122 has 2.91 GiB memory in use. Process 3243120 has 2.80 GiB memory in use. Process 3243121 has 3.62 GiB memory in use. Process 3243119 has 3.37 GiB memory in use. Process 3315388 has 61.82 GiB memory in use. Including non-PyTorch memory, this process has 59.60 GiB memory in use. Of the allocated memory 58.94 GiB is allocated by PyTorch, and 278.00 KiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables))
+(EngineCore_DP0 pid=3315747) ERROR 02-03 01:41:41 [core.py:946] EngineCore failed to start.
+(EngineCore_DP0 pid=3315747) ERROR 02-03 01:41:41 [core.py:946] Traceback (most recent call last):
+(EngineCore_DP0 pid=3315747) ERROR 02-03 01:41:41 [core.py:946]   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/vllm/v1/engine/core.py", line 937, in run_engine_core
+(EngineCore_DP0 pid=3315747) ERROR 02-03 01:41:41 [core.py:946]     engine_core = EngineCoreProc(*args, engine_index=dp_rank, **kwargs)
+(EngineCore_DP0 pid=3315747) ERROR 02-03 01:41:41 [core.py:946]                   ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+(EngineCore_DP0 pid=3315747) ERROR 02-03 01:41:41 [core.py:946]   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/vllm/v1/engine/core.py", line 691, in __init__
+(EngineCore_DP0 pid=3315747) ERROR 02-03 01:41:41 [core.py:946]     super().__init__(
+(EngineCore_DP0 pid=3315747) ERROR 02-03 01:41:41 [core.py:946]   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/vllm/v1/engine/core.py", line 105, in __init__
+(EngineCore_DP0 pid=3315747) ERROR 02-03 01:41:41 [core.py:946]     self.model_executor = executor_class(vllm_config)
+(EngineCore_DP0 pid=3315747) ERROR 02-03 01:41:41 [core.py:946]                           ^^^^^^^^^^^^^^^^^^^^^^^^^^^
+(EngineCore_DP0 pid=3315747) ERROR 02-03 01:41:41 [core.py:946]   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/vllm/v1/executor/abstract.py", line 101, in __init__
+(EngineCore_DP0 pid=3315747) ERROR 02-03 01:41:41 [core.py:946]     self._init_executor()
+(EngineCore_DP0 pid=3315747) ERROR 02-03 01:41:41 [core.py:946]   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/vllm/v1/executor/uniproc_executor.py", line 48, in _init_executor
+(EngineCore_DP0 pid=3315747) ERROR 02-03 01:41:41 [core.py:946]     self.driver_worker.load_model()
+(EngineCore_DP0 pid=3315747) ERROR 02-03 01:41:41 [core.py:946]   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/vllm/v1/worker/gpu_worker.py", line 275, in load_model
+(EngineCore_DP0 pid=3315747) ERROR 02-03 01:41:41 [core.py:946]     self.model_runner.load_model(eep_scale_up=eep_scale_up)
+(EngineCore_DP0 pid=3315747) ERROR 02-03 01:41:41 [core.py:946]   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/vllm/v1/worker/gpu_model_runner.py", line 4117, in load_model
+(EngineCore_DP0 pid=3315747) ERROR 02-03 01:41:41 [core.py:946]     raise e
+(EngineCore_DP0 pid=3315747) ERROR 02-03 01:41:41 [core.py:946]   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/vllm/v1/worker/gpu_model_runner.py", line 4040, in load_model
+(EngineCore_DP0 pid=3315747) ERROR 02-03 01:41:41 [core.py:946]     self.model = model_loader.load_model(
+(EngineCore_DP0 pid=3315747) ERROR 02-03 01:41:41 [core.py:946]                  ^^^^^^^^^^^^^^^^^^^^^^^^
+(EngineCore_DP0 pid=3315747) ERROR 02-03 01:41:41 [core.py:946]   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/vllm/model_executor/model_loader/base_loader.py", line 50, in load_model
+(EngineCore_DP0 pid=3315747) ERROR 02-03 01:41:41 [core.py:946]     model = initialize_model(
+(EngineCore_DP0 pid=3315747) ERROR 02-03 01:41:41 [core.py:946]             ^^^^^^^^^^^^^^^^^
+(EngineCore_DP0 pid=3315747) ERROR 02-03 01:41:41 [core.py:946]   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/vllm/model_executor/model_loader/utils.py", line 48, in initialize_model
+(EngineCore_DP0 pid=3315747) ERROR 02-03 01:41:41 [core.py:946]     return model_class(vllm_config=vllm_config, prefix=prefix)
+(EngineCore_DP0 pid=3315747) ERROR 02-03 01:41:41 [core.py:946]            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+(EngineCore_DP0 pid=3315747) ERROR 02-03 01:41:41 [core.py:946]   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/vllm/model_executor/models/qwen3.py", line 274, in __init__
+(EngineCore_DP0 pid=3315747) ERROR 02-03 01:41:41 [core.py:946]     self.model = Qwen3Model(
+(EngineCore_DP0 pid=3315747) ERROR 02-03 01:41:41 [core.py:946]                  ^^^^^^^^^^^
+(EngineCore_DP0 pid=3315747) ERROR 02-03 01:41:41 [core.py:946]   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/vllm/compilation/decorators.py", line 306, in __init__
+(EngineCore_DP0 pid=3315747) ERROR 02-03 01:41:41 [core.py:946]     old_init(self, **kwargs)
+(EngineCore_DP0 pid=3315747) ERROR 02-03 01:41:41 [core.py:946]   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/vllm/model_executor/models/qwen3.py", line 248, in __init__
+(EngineCore_DP0 pid=3315747) ERROR 02-03 01:41:41 [core.py:946]     super().__init__(
+(EngineCore_DP0 pid=3315747) ERROR 02-03 01:41:41 [core.py:946]   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/vllm/compilation/decorators.py", line 306, in __init__
+(EngineCore_DP0 pid=3315747) ERROR 02-03 01:41:41 [core.py:946]     old_init(self, **kwargs)
+(EngineCore_DP0 pid=3315747) ERROR 02-03 01:41:41 [core.py:946]   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/vllm/model_executor/models/qwen2.py", line 394, in __init__
+(EngineCore_DP0 pid=3315747) ERROR 02-03 01:41:41 [core.py:946]     self.start_layer, self.end_layer, self.layers = make_layers(
+(EngineCore_DP0 pid=3315747) ERROR 02-03 01:41:41 [core.py:946]                                                     ^^^^^^^^^^^^
+(EngineCore_DP0 pid=3315747) ERROR 02-03 01:41:41 [core.py:946]   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/vllm/model_executor/models/utils.py", line 707, in make_layers
+(EngineCore_DP0 pid=3315747) ERROR 02-03 01:41:41 [core.py:946]     maybe_offload_to_cpu(layer_fn(prefix=f"{prefix}.{idx}"))
+(EngineCore_DP0 pid=3315747) ERROR 02-03 01:41:41 [core.py:946]                          ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+(EngineCore_DP0 pid=3315747) ERROR 02-03 01:41:41 [core.py:946]   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/vllm/model_executor/models/qwen2.py", line 396, in <lambda>
+(EngineCore_DP0 pid=3315747) ERROR 02-03 01:41:41 [core.py:946]     lambda prefix: decoder_layer_type(
+(EngineCore_DP0 pid=3315747) ERROR 02-03 01:41:41 [core.py:946]                    ^^^^^^^^^^^^^^^^^^^
+(EngineCore_DP0 pid=3315747) ERROR 02-03 01:41:41 [core.py:946]   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/vllm/model_executor/models/qwen3.py", line 196, in __init__
+(EngineCore_DP0 pid=3315747) ERROR 02-03 01:41:41 [core.py:946]     self.mlp = Qwen3MLP(
+(EngineCore_DP0 pid=3315747) ERROR 02-03 01:41:41 [core.py:946]                ^^^^^^^^^
+(EngineCore_DP0 pid=3315747) ERROR 02-03 01:41:41 [core.py:946]   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/vllm/model_executor/models/qwen2.py", line 87, in __init__
+(EngineCore_DP0 pid=3315747) ERROR 02-03 01:41:41 [core.py:946]     self.gate_up_proj = MergedColumnParallelLinear(
+(EngineCore_DP0 pid=3315747) ERROR 02-03 01:41:41 [core.py:946]                         ^^^^^^^^^^^^^^^^^^^^^^^^^^^
+(EngineCore_DP0 pid=3315747) ERROR 02-03 01:41:41 [core.py:946]   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/vllm/model_executor/layers/linear.py", line 670, in __init__
+(EngineCore_DP0 pid=3315747) ERROR 02-03 01:41:41 [core.py:946]     super().__init__(
+(EngineCore_DP0 pid=3315747) ERROR 02-03 01:41:41 [core.py:946]   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/vllm/model_executor/layers/linear.py", line 495, in __init__
+(EngineCore_DP0 pid=3315747) ERROR 02-03 01:41:41 [core.py:946]     self.quant_method.create_weights(
+(EngineCore_DP0 pid=3315747) ERROR 02-03 01:41:41 [core.py:946]   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/vllm/model_executor/layers/linear.py", line 224, in create_weights
+(EngineCore_DP0 pid=3315747) ERROR 02-03 01:41:41 [core.py:946]     data=torch.empty(
+(EngineCore_DP0 pid=3315747) ERROR 02-03 01:41:41 [core.py:946]          ^^^^^^^^^^^^
+(EngineCore_DP0 pid=3315747) ERROR 02-03 01:41:41 [core.py:946]   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/torch/utils/_device.py", line 103, in __torch_function__
+(EngineCore_DP0 pid=3315747) ERROR 02-03 01:41:41 [core.py:946]     return func(*args, **kwargs)
+(EngineCore_DP0 pid=3315747) ERROR 02-03 01:41:41 [core.py:946]            ^^^^^^^^^^^^^^^^^^^^^
+(EngineCore_DP0 pid=3315747) ERROR 02-03 01:41:41 [core.py:946] torch.OutOfMemoryError: CUDA out of memory. Tried to allocate 500.00 MiB. GPU 0 has a total capacity of 139.80 GiB of which 312.25 MiB is free. Process 3284142 has 5.32 GiB memory in use. Process 3243122 has 2.91 GiB memory in use. Process 3243120 has 2.80 GiB memory in use. Process 3243121 has 3.62 GiB memory in use. Process 3243119 has 3.37 GiB memory in use. Process 3315388 has 61.82 GiB memory in use. Including non-PyTorch memory, this process has 59.60 GiB memory in use. Of the allocated memory 58.94 GiB is allocated by PyTorch, and 278.00 KiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)
+(EngineCore_DP0 pid=3315747) Process EngineCore_DP0:
+(EngineCore_DP0 pid=3315747) Traceback (most recent call last):
+(EngineCore_DP0 pid=3315747)   File "/usr/lib/python3.12/multiprocessing/process.py", line 314, in _bootstrap
+(EngineCore_DP0 pid=3315747)     self.run()
+(EngineCore_DP0 pid=3315747)   File "/usr/lib/python3.12/multiprocessing/process.py", line 108, in run
+(EngineCore_DP0 pid=3315747)     self._target(*self._args, **self._kwargs)
+(EngineCore_DP0 pid=3315747)   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/vllm/v1/engine/core.py", line 950, in run_engine_core
+(EngineCore_DP0 pid=3315747)     raise e
+(EngineCore_DP0 pid=3315747)   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/vllm/v1/engine/core.py", line 937, in run_engine_core
+(EngineCore_DP0 pid=3315747)     engine_core = EngineCoreProc(*args, engine_index=dp_rank, **kwargs)
+(EngineCore_DP0 pid=3315747)                   ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+(EngineCore_DP0 pid=3315747)   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/vllm/v1/engine/core.py", line 691, in __init__
+(EngineCore_DP0 pid=3315747)     super().__init__(
+(EngineCore_DP0 pid=3315747)   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/vllm/v1/engine/core.py", line 105, in __init__
+(EngineCore_DP0 pid=3315747)     self.model_executor = executor_class(vllm_config)
+(EngineCore_DP0 pid=3315747)                           ^^^^^^^^^^^^^^^^^^^^^^^^^^^
+(EngineCore_DP0 pid=3315747)   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/vllm/v1/executor/abstract.py", line 101, in __init__
+(EngineCore_DP0 pid=3315747)     self._init_executor()
+(EngineCore_DP0 pid=3315747)   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/vllm/v1/executor/uniproc_executor.py", line 48, in _init_executor
+(EngineCore_DP0 pid=3315747)     self.driver_worker.load_model()
+(EngineCore_DP0 pid=3315747)   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/vllm/v1/worker/gpu_worker.py", line 275, in load_model
+(EngineCore_DP0 pid=3315747)     self.model_runner.load_model(eep_scale_up=eep_scale_up)
+(EngineCore_DP0 pid=3315747)   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/vllm/v1/worker/gpu_model_runner.py", line 4117, in load_model
+(EngineCore_DP0 pid=3315747)     raise e
+(EngineCore_DP0 pid=3315747)   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/vllm/v1/worker/gpu_model_runner.py", line 4040, in load_model
+(EngineCore_DP0 pid=3315747)     self.model = model_loader.load_model(
+(EngineCore_DP0 pid=3315747)                  ^^^^^^^^^^^^^^^^^^^^^^^^
+(EngineCore_DP0 pid=3315747)   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/vllm/model_executor/model_loader/base_loader.py", line 50, in load_model
+(EngineCore_DP0 pid=3315747)     model = initialize_model(
+(EngineCore_DP0 pid=3315747)             ^^^^^^^^^^^^^^^^^
+(EngineCore_DP0 pid=3315747)   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/vllm/model_executor/model_loader/utils.py", line 48, in initialize_model
+(EngineCore_DP0 pid=3315747)     return model_class(vllm_config=vllm_config, prefix=prefix)
+(EngineCore_DP0 pid=3315747)            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+(EngineCore_DP0 pid=3315747)   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/vllm/model_executor/models/qwen3.py", line 274, in __init__
+(EngineCore_DP0 pid=3315747)     self.model = Qwen3Model(
+(EngineCore_DP0 pid=3315747)                  ^^^^^^^^^^^
+(EngineCore_DP0 pid=3315747)   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/vllm/compilation/decorators.py", line 306, in __init__
+(EngineCore_DP0 pid=3315747)     old_init(self, **kwargs)
+(EngineCore_DP0 pid=3315747)   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/vllm/model_executor/models/qwen3.py", line 248, in __init__
+(EngineCore_DP0 pid=3315747)     super().__init__(
+(EngineCore_DP0 pid=3315747)   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/vllm/compilation/decorators.py", line 306, in __init__
+(EngineCore_DP0 pid=3315747)     old_init(self, **kwargs)
+(EngineCore_DP0 pid=3315747)   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/vllm/model_executor/models/qwen2.py", line 394, in __init__
+(EngineCore_DP0 pid=3315747)     self.start_layer, self.end_layer, self.layers = make_layers(
+(EngineCore_DP0 pid=3315747)                                                     ^^^^^^^^^^^^
+(EngineCore_DP0 pid=3315747)   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/vllm/model_executor/models/utils.py", line 707, in make_layers
+(EngineCore_DP0 pid=3315747)     maybe_offload_to_cpu(layer_fn(prefix=f"{prefix}.{idx}"))
+(EngineCore_DP0 pid=3315747)                          ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+(EngineCore_DP0 pid=3315747)   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/vllm/model_executor/models/qwen2.py", line 396, in <lambda>
+(EngineCore_DP0 pid=3315747)     lambda prefix: decoder_layer_type(
+(EngineCore_DP0 pid=3315747)                    ^^^^^^^^^^^^^^^^^^^
+(EngineCore_DP0 pid=3315747)   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/vllm/model_executor/models/qwen3.py", line 196, in __init__
+(EngineCore_DP0 pid=3315747)     self.mlp = Qwen3MLP(
+(EngineCore_DP0 pid=3315747)                ^^^^^^^^^
+(EngineCore_DP0 pid=3315747)   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/vllm/model_executor/models/qwen2.py", line 87, in __init__
+(EngineCore_DP0 pid=3315747)     self.gate_up_proj = MergedColumnParallelLinear(
+(EngineCore_DP0 pid=3315747)                         ^^^^^^^^^^^^^^^^^^^^^^^^^^^
+(EngineCore_DP0 pid=3315747)   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/vllm/model_executor/layers/linear.py", line 670, in __init__
+(EngineCore_DP0 pid=3315747)     super().__init__(
+(EngineCore_DP0 pid=3315747)   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/vllm/model_executor/layers/linear.py", line 495, in __init__
+(EngineCore_DP0 pid=3315747)     self.quant_method.create_weights(
+(EngineCore_DP0 pid=3315747)   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/vllm/model_executor/layers/linear.py", line 224, in create_weights
+(EngineCore_DP0 pid=3315747)     data=torch.empty(
+(EngineCore_DP0 pid=3315747)          ^^^^^^^^^^^^
+(EngineCore_DP0 pid=3315747)   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/torch/utils/_device.py", line 103, in __torch_function__
+(EngineCore_DP0 pid=3315747)     return func(*args, **kwargs)
+(EngineCore_DP0 pid=3315747)            ^^^^^^^^^^^^^^^^^^^^^
+(EngineCore_DP0 pid=3315747) torch.OutOfMemoryError: CUDA out of memory. Tried to allocate 500.00 MiB. GPU 0 has a total capacity of 139.80 GiB of which 312.25 MiB is free. Process 3284142 has 5.32 GiB memory in use. Process 3243122 has 2.91 GiB memory in use. Process 3243120 has 2.80 GiB memory in use. Process 3243121 has 3.62 GiB memory in use. Process 3243119 has 3.37 GiB memory in use. Process 3315388 has 61.82 GiB memory in use. Including non-PyTorch memory, this process has 59.60 GiB memory in use. Of the allocated memory 58.94 GiB is allocated by PyTorch, and 278.00 KiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)
+[rank0]:[W203 01:41:43.753415460 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=3311584) Traceback (most recent call last):
+(APIServer pid=3311584)   File "<frozen runpy>", line 198, in _run_module_as_main
+(APIServer pid=3311584)   File "<frozen runpy>", line 88, in _run_code
+(APIServer pid=3311584)   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/vllm/entrypoints/openai/api_server.py", line 991, in <module>
+(APIServer pid=3311584)     uvloop.run(run_server(args))
+(APIServer pid=3311584)   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/uvloop/__init__.py", line 96, in run
+(APIServer pid=3311584)     return __asyncio.run(
+(APIServer pid=3311584)            ^^^^^^^^^^^^^^
+(APIServer pid=3311584)   File "/usr/lib/python3.12/asyncio/runners.py", line 194, in run
+(APIServer pid=3311584)     return runner.run(main)
+(APIServer pid=3311584)            ^^^^^^^^^^^^^^^^
+(APIServer pid=3311584)   File "/usr/lib/python3.12/asyncio/runners.py", line 118, in run
+(APIServer pid=3311584)     return self._loop.run_until_complete(task)
+(APIServer pid=3311584)            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+(APIServer pid=3311584)   File "uvloop/loop.pyx", line 1518, in uvloop.loop.Loop.run_until_complete
+(APIServer pid=3311584)   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/uvloop/__init__.py", line 48, in wrapper
+(APIServer pid=3311584)     return await main
+(APIServer pid=3311584)            ^^^^^^^^^^
+(APIServer pid=3311584)   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/vllm/entrypoints/openai/api_server.py", line 919, in run_server
+(APIServer pid=3311584)     await run_server_worker(listen_address, sock, args, **uvicorn_kwargs)
+(APIServer pid=3311584)   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/vllm/entrypoints/openai/api_server.py", line 938, in run_server_worker
+(APIServer pid=3311584)     async with build_async_engine_client(
+(APIServer pid=3311584)   File "/usr/lib/python3.12/contextlib.py", line 210, in __aenter__
+(APIServer pid=3311584)     return await anext(self.gen)
+(APIServer pid=3311584)            ^^^^^^^^^^^^^^^^^^^^^
+(APIServer pid=3311584)   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/vllm/entrypoints/openai/api_server.py", line 147, in build_async_engine_client
+(APIServer pid=3311584)     async with build_async_engine_client_from_engine_args(
+(APIServer pid=3311584)   File "/usr/lib/python3.12/contextlib.py", line 210, in __aenter__
+(APIServer pid=3311584)     return await anext(self.gen)
+(APIServer pid=3311584)            ^^^^^^^^^^^^^^^^^^^^^
+(APIServer pid=3311584)   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/vllm/entrypoints/openai/api_server.py", line 188, in build_async_engine_client_from_engine_args
+(APIServer pid=3311584)     async_llm = AsyncLLM.from_vllm_config(
+(APIServer pid=3311584)                 ^^^^^^^^^^^^^^^^^^^^^^^^^^
+(APIServer pid=3311584)   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/vllm/v1/engine/async_llm.py", line 228, in from_vllm_config
+(APIServer pid=3311584)     return cls(
+(APIServer pid=3311584)            ^^^^
+(APIServer pid=3311584)   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/vllm/v1/engine/async_llm.py", line 155, in __init__
+(APIServer pid=3311584)     self.engine_core = EngineCoreClient.make_async_mp_client(
+(APIServer pid=3311584)                        ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+(APIServer pid=3311584)   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/vllm/v1/engine/core_client.py", line 122, in make_async_mp_client
+(APIServer pid=3311584)     return AsyncMPClient(*client_args)
+(APIServer pid=3311584)            ^^^^^^^^^^^^^^^^^^^^^^^^^^^
+(APIServer pid=3311584)   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/vllm/v1/engine/core_client.py", line 819, in __init__
+(APIServer pid=3311584)     super().__init__(
+(APIServer pid=3311584)   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/vllm/v1/engine/core_client.py", line 479, in __init__
+(APIServer pid=3311584)     with launch_core_engines(vllm_config, executor_class, log_stats) as (
+(APIServer pid=3311584)   File "/usr/lib/python3.12/contextlib.py", line 144, in __exit__
+(APIServer pid=3311584)     next(self.gen)
+(APIServer pid=3311584)   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/vllm/v1/engine/utils.py", line 933, in launch_core_engines
+(APIServer pid=3311584)     wait_for_engine_startup(
+(APIServer pid=3311584)   File "/home/mingzhe/Projects/Arena/.venv/lib/python3.12/site-packages/vllm/v1/engine/utils.py", line 992, in wait_for_engine_startup
+(APIServer pid=3311584)     raise RuntimeError(
+(APIServer pid=3311584) RuntimeError: Engine core initialization failed. See root cause above. Failed core proc(s): {}

vllm_0007000.log

@@ -1,20 +1,18 @@
-(APIServer pid=3233856) INFO 02-03 01:26:55 [utils.py:325] 
-(APIServer pid=3233856) INFO 02-03 01:26:55 [utils.py:325]        █     █     █▄   ▄█
-(APIServer pid=3233856) INFO 02-03 01:26:55 [utils.py:325]  ▄▄ ▄█ █     █     █ ▀▄▀ █  version 0.15.0
-(APIServer pid=3233856) INFO 02-03 01:26:55 [utils.py:325]   █▄█▀ █     █     █     █  model   Elfsong/VLM_stage_2_iter_0007000
-(APIServer pid=3233856) INFO 02-03 01:26:55 [utils.py:325]    ▀▀  ▀▀▀▀▀ ▀▀▀▀▀ ▀     ▀
-(APIServer pid=3233856) INFO 02-03 01:26:55 [utils.py:325] 
-(APIServer pid=3233856) INFO 02-03 01:26:55 [utils.py:261] non-default args: {'port': 9003, 'model': 'Elfsong/VLM_stage_2_iter_0007000', 'trust_remote_code': True, 'quantization': 'bitsandbytes', 'gpu_memory_utilization': 0.3}
-(APIServer pid=3233856) The argument `trust_remote_code` is to be used with Auto classes. It has no effect here and is ignored.
-(APIServer pid=3233856) The argument `trust_remote_code` is to be used with Auto classes. It has no effect here and is ignored.
-(APIServer pid=3233856) INFO 02-03 01:26:57 [model.py:541] Resolved architecture: Qwen3ForCausalLM
-(APIServer pid=3233856) INFO 02-03 01:26:57 [model.py:1561] Using max model len 40960
-(APIServer pid=3233856) INFO 02-03 01:26:57 [scheduler.py:226] Chunked prefill is enabled with max_num_batched_tokens=8192.
-(APIServer pid=3233856) INFO 02-03 01:27:00 [vllm.py:624] Asynchronous scheduling is enabled.
-(EngineCore_DP0 pid=3236204) INFO 02-03 01:27:14 [core.py:96] Initializing a V1 LLM engine (v0.15.0) 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, enable_return_routed_experts=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, enable_mfu_metrics=False, enable_mm_processor_stats=False, enable_logging_iteration_details=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', 'vllm::rocm_aiter_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, 'assume_32_bit_indexing': True}, 'local_cache_dir': None}
-(EngineCore_DP0 pid=3236204) INFO 02-03 01:27:15 [parallel_state.py:1212] world_size=1 rank=0 local_rank=0 distributed_init_method=tcp://172.21.25.98:33921 backend=nccl
-(EngineCore_DP0 pid=3236204) INFO 02-03 01:27:15 [parallel_state.py:1423] rank 0 in world size 1 is assigned as DP rank 0, PP rank 0, PCP rank 0, TP rank 0, EP rank N/A
-(EngineCore_DP0 pid=3236204) INFO 02-03 01:27:16 [gpu_model_runner.py:4021] Starting to load model Elfsong/VLM_stage_2_iter_0007000...
-(EngineCore_DP0 pid=3236204) INFO 02-03 01:27:18 [cuda.py:364] Using FLASH_ATTN attention backend out of potential backends: ('FLASH_ATTN', 'FLASHINFER', 'TRITON_ATTN', 'FLEX_ATTENTION')
-(EngineCore_DP0 pid=3236204) INFO 02-03 01:27:19 [bitsandbytes_loader.py:786] Loading weights with BitsAndBytes quantization. May take a while ...
-Cancellation requested; stopping current tasks.
+(APIServer pid=3313448) INFO 02-03 01:41:15 [utils.py:325] 
+(APIServer pid=3313448) INFO 02-03 01:41:15 [utils.py:325]        █     █     █▄   ▄█
+(APIServer pid=3313448) INFO 02-03 01:41:15 [utils.py:325]  ▄▄ ▄█ █     █     █ ▀▄▀ █  version 0.15.0
+(APIServer pid=3313448) INFO 02-03 01:41:15 [utils.py:325]   █▄█▀ █     █     █     █  model   Elfsong/VLM_stage_2_iter_0007000
+(APIServer pid=3313448) INFO 02-03 01:41:15 [utils.py:325]    ▀▀  ▀▀▀▀▀ ▀▀▀▀▀ ▀     ▀
+(APIServer pid=3313448) INFO 02-03 01:41:15 [utils.py:325] 
+(APIServer pid=3313448) INFO 02-03 01:41:15 [utils.py:261] non-default args: {'port': 9004, 'model': 'Elfsong/VLM_stage_2_iter_0007000', 'trust_remote_code': True, 'gpu_memory_utilization': 0.4}
+(APIServer pid=3313448) The argument `trust_remote_code` is to be used with Auto classes. It has no effect here and is ignored.
+(APIServer pid=3313448) The argument `trust_remote_code` is to be used with Auto classes. It has no effect here and is ignored.
+(APIServer pid=3313448) INFO 02-03 01:41:16 [model.py:541] Resolved architecture: Qwen3ForCausalLM
+(APIServer pid=3313448) INFO 02-03 01:41:16 [model.py:1561] Using max model len 40960
+(APIServer pid=3313448) INFO 02-03 01:41:16 [scheduler.py:226] Chunked prefill is enabled with max_num_batched_tokens=8192.
+(APIServer pid=3313448) INFO 02-03 01:41:16 [vllm.py:624] Asynchronous scheduling is enabled.
+(EngineCore_DP0 pid=3315388) INFO 02-03 01:41:28 [core.py:96] Initializing a V1 LLM engine (v0.15.0) 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=auto, tensor_parallel_size=1, pipeline_parallel_size=1, data_parallel_size=1, disable_custom_all_reduce=False, quantization=None, enforce_eager=False, enable_return_routed_experts=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, enable_mfu_metrics=False, enable_mm_processor_stats=False, enable_logging_iteration_details=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', 'vllm::rocm_aiter_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, 'assume_32_bit_indexing': True}, 'local_cache_dir': None}
+(EngineCore_DP0 pid=3315388) INFO 02-03 01:41:32 [parallel_state.py:1212] world_size=1 rank=0 local_rank=0 distributed_init_method=tcp://172.21.25.98:53245 backend=nccl
+(EngineCore_DP0 pid=3315388) INFO 02-03 01:41:32 [parallel_state.py:1423] rank 0 in world size 1 is assigned as DP rank 0, PP rank 0, PCP rank 0, TP rank 0, EP rank N/A
+(EngineCore_DP0 pid=3315388) INFO 02-03 01:41:33 [gpu_model_runner.py:4021] Starting to load model Elfsong/VLM_stage_2_iter_0007000...
+(EngineCore_DP0 pid=3315388) INFO 02-03 01:41:34 [cuda.py:364] Using FLASH_ATTN attention backend out of potential backends: ('FLASH_ATTN', 'FLASHINFER', 'TRITON_ATTN', 'FLEX_ATTENTION')