2025-07-18T09:19:57+08:00 - gpustack.worker.backends.base - INFO - Preparing model files...
2025-07-18T09:19:57+08:00 - gpustack.worker.backends.base - INFO - Model files are ready.
2025-07-18T09:19:58+08:00 - gpustack.worker.backends.vllm - INFO - Starting vllm server
INFO 07-18 09:20:07 [__init__.py:240] Automatically detected platform cuda.
INFO 07-18 09:20:09 [api_server.py:1034] vLLM API server version 0.8.3
INFO 07-18 09:20:09 [api_server.py:1035] args: Namespace(subparser='serve', model_tag='/data/models/Qwen3-14B', config='', host='0.0.0.0', port=40045, uvicorn_log_level='info', disable_uvicorn_access_log=False, allow_credentials=False, allowed_origins=['*'], allowed_methods=['*'], allowed_headers=['*'], api_key=None, lora_modules=None, prompt_adapters=None, chat_template=None, chat_template_content_format='auto', response_role='assistant', ssl_keyfile=None, ssl_certfile=None, ssl_ca_certs=None, enable_ssl_refresh=False, ssl_cert_reqs=0, root_path=None, middleware=[], return_tokens_as_token_ids=False, disable_frontend_multiprocessing=False, enable_request_id_headers=False, enable_auto_tool_choice=False, tool_call_parser=None, tool_parser_plugin='', model='/data/models/Qwen3-14B', task='auto', tokenizer=None, hf_config_path=None, skip_tokenizer_init=False, revision=None, code_revision=None, tokenizer_revision=None, tokenizer_mode='auto', trust_remote_code=False, allowed_local_media_path=None, download_dir=None, load_format='auto', config_format=<ConfigFormat.AUTO: 'auto'>, dtype='auto', kv_cache_dtype='auto', max_model_len=8192, guided_decoding_backend='xgrammar', logits_processor_pattern=None, model_impl='auto', distributed_executor_backend=None, pipeline_parallel_size=1, tensor_parallel_size=2, data_parallel_size=1, enable_expert_parallel=False, max_parallel_loading_workers=None, ray_workers_use_nsight=False, block_size=None, enable_prefix_caching=None, prefix_caching_hash_algo='builtin', disable_sliding_window=False, use_v2_block_manager=True, num_lookahead_slots=0, seed=None, swap_space=4, cpu_offload_gb=0, gpu_memory_utilization=0.9, num_gpu_blocks_override=None, max_num_batched_tokens=None, max_num_partial_prefills=1, max_long_partial_prefills=1, long_prefill_token_threshold=0, max_num_seqs=None, max_logprobs=20, disable_log_stats=False, quantization=None, rope_scaling=None, rope_theta=None, hf_overrides=None, enforce_eager=False, max_seq_len_to_capture=8192, disable_custom_all_reduce=False, tokenizer_pool_size=0, tokenizer_pool_type='ray', tokenizer_pool_extra_config=None, limit_mm_per_prompt=None, mm_processor_kwargs=None, disable_mm_preprocessor_cache=False, enable_lora=False, enable_lora_bias=False, max_loras=1, max_lora_rank=16, lora_extra_vocab_size=256, lora_dtype='auto', long_lora_scaling_factors=None, max_cpu_loras=None, fully_sharded_loras=False, enable_prompt_adapter=False, max_prompt_adapters=1, max_prompt_adapter_token=0, device='auto', num_scheduler_steps=1, use_tqdm_on_load=True, multi_step_stream_outputs=True, scheduler_delay_factor=0.0, enable_chunked_prefill=None, speculative_config=None, model_loader_extra_config=None, ignore_patterns=[], preemption_mode=None, served_model_name=['qwen3-14b'], qlora_adapter_name_or_path=None, show_hidden_metrics_for_version=None, otlp_traces_endpoint=None, collect_detailed_traces=None, disable_async_output_proc=False, scheduling_policy='fcfs', scheduler_cls='vllm.core.scheduler.Scheduler', override_neuron_config=None, override_pooler_config=None, compilation_config=None, kv_transfer_config=None, worker_cls='auto', worker_extension_cls='', generation_config='auto', override_generation_config=None, enable_sleep_mode=False, calculate_kv_scales=False, additional_config=None, enable_reasoning=False, reasoning_parser=None, disable_cascade_attn=False, disable_log_requests=False, max_log_len=None, disable_fastapi_docs=False, enable_prompt_tokens_details=False, enable_server_load_tracking=False, dispatch_function=<function ServeSubcommand.cmd at 0xfffbb9ef9360>)
INFO 07-18 09:20:26 [config.py:610] This model supports multiple tasks: {'embed', 'score', 'reward', 'classify', 'generate'}. Defaulting to 'generate'.
WARNING 07-18 09:20:26 [config.py:446] Please export VLLM_ENFORCE_CUDA_GRAPH=1 to enable cuda graph. For now, cuda graph is not used and --enforce-eager is disabled ,we are trying to use cuda graph as the default mode
INFO 07-18 09:20:26 [config.py:1612] Defaulting to use mp for distributed inference
WARNING 07-18 09:20:26 [cuda.py:95] To see benefits of async output processing, enable CUDA graph. Since, enforce-eager is enabled, async output processor cannot be used
INFO 07-18 09:20:26 [api_server.py:246] Started engine process with PID 3085948
INFO 07-18 09:20:35 [__init__.py:240] Automatically detected platform cuda.
INFO 07-18 09:20:37 [llm_engine.py:242] Initializing a V0 LLM engine (v0.8.3) with config: model='/data/models/Qwen3-14B', speculative_config=None, tokenizer='/data/models/Qwen3-14B', skip_tokenizer_init=False, tokenizer_mode=auto, revision=None, override_neuron_config=None, tokenizer_revision=None, trust_remote_code=False, dtype=torch.bfloat16, max_seq_len=8192, download_dir=None, load_format=auto, tensor_parallel_size=2, pipeline_parallel_size=1, disable_custom_all_reduce=False, quantization=None, enforce_eager=True, kv_cache_dtype=auto,  device_config=cuda, decoding_config=DecodingConfig(guided_decoding_backend='xgrammar', reasoning_backend=None), observability_config=ObservabilityConfig(show_hidden_metrics=False, otlp_traces_endpoint=None, collect_model_forward_time=False, collect_model_execute_time=False), seed=None, served_model_name=qwen3-14b, num_scheduler_steps=1, multi_step_stream_outputs=True, enable_prefix_caching=None, chunked_prefill_enabled=False, use_async_output_proc=False, disable_mm_preprocessor_cache=False, mm_processor_kwargs=None, pooler_config=None, compilation_config={"splitting_ops":[],"compile_sizes":[],"cudagraph_capture_sizes":[],"max_capture_size":0}, use_cached_outputs=True, 
WARNING 07-18 09:20:38 [multiproc_worker_utils.py:306] Reducing Torch parallelism from 128 threads to 1 to avoid unnecessary CPU contention. Set OMP_NUM_THREADS in the external environment to tune this value as needed.
/usr/local/corex/lib64/python3/dist-packages/vllm/executor/mp_distributed_executor.py:110: ResourceWarning: unclosed <socket.socket fd=11, family=AddressFamily.AF_INET, type=SocketKind.SOCK_DGRAM, proto=0, laddr=('172.22.0.2', 57067), raddr=('8.8.8.8', 80)>
  get_ip(), get_open_port())
ResourceWarning: Enable tracemalloc to get the object allocation traceback
INFO 07-18 09:20:38 [cuda.py:291] Using Flash Attention backend.
INFO 07-18 09:20:46 [__init__.py:240] Automatically detected platform cuda.
(VllmWorkerProcess pid=3086665) INFO 07-18 09:20:49 [multiproc_worker_utils.py:225] Worker ready; awaiting tasks
(VllmWorkerProcess pid=3086665) INFO 07-18 09:20:49 [cuda.py:291] Using Flash Attention backend.
(VllmWorkerProcess pid=3086665) INFO 07-18 09:20:50 [custom_all_reduce.py:73] Custom allreduce is disabled because of missing custom allreduce library
INFO 07-18 09:20:50 [custom_all_reduce.py:73] Custom allreduce is disabled because of missing custom allreduce library
INFO 07-18 09:20:50 [shm_broadcast.py:264] vLLM message queue communication handle: Handle(local_reader_ranks=[1], buffer_handle=(1, 4194304, 6, 'psm_b959ae60'), local_subscribe_addr='ipc:///tmp/9f0175c5-6916-456f-8c2a-115eefb3b942', remote_subscribe_addr=None, remote_addr_ipv6=False)
INFO 07-18 09:20:50 [parallel_state.py:991] rank 0 in world size 2 is assigned as DP rank 0, PP rank 0, TP rank 0
(VllmWorkerProcess pid=3086665) INFO 07-18 09:20:50 [parallel_state.py:991] rank 1 in world size 2 is assigned as DP rank 0, PP rank 0, TP rank 1
INFO 07-18 09:20:50 [model_runner.py:1110] Starting to load model /data/models/Qwen3-14B...
(VllmWorkerProcess pid=3086665) INFO 07-18 09:20:50 [model_runner.py:1110] Starting to load model /data/models/Qwen3-14B...

Loading safetensors checkpoint shards:   0% Completed | 0/8 [00:00<?, ?it/s]

Loading safetensors checkpoint shards:  12% Completed | 1/8 [00:00<00:04,  1.61it/s]

Loading safetensors checkpoint shards:  25% Completed | 2/8 [00:01<00:04,  1.48it/s]

Loading safetensors checkpoint shards:  38% Completed | 3/8 [00:02<00:03,  1.43it/s]

Loading safetensors checkpoint shards:  50% Completed | 4/8 [00:02<00:02,  1.41it/s]

Loading safetensors checkpoint shards:  62% Completed | 5/8 [00:03<00:02,  1.40it/s]
(VllmWorkerProcess pid=3086665) INFO 07-18 09:20:55 [loader.py:448] Loading weights took 4.38 seconds
(VllmWorkerProcess pid=3086665) INFO 07-18 09:20:55 [model_runner.py:1146] Model loading took 13.8818 GiB and 4.782220 seconds

Loading safetensors checkpoint shards:  75% Completed | 6/8 [00:04<00:01,  1.40it/s]

Loading safetensors checkpoint shards:  88% Completed | 7/8 [00:04<00:00,  1.42it/s]

Loading safetensors checkpoint shards: 100% Completed | 8/8 [00:05<00:00,  1.77it/s]

Loading safetensors checkpoint shards: 100% Completed | 8/8 [00:05<00:00,  1.54it/s]

INFO 07-18 09:20:56 [loader.py:448] Loading weights took 5.19 seconds
INFO 07-18 09:20:56 [model_runner.py:1146] Model loading took 13.8818 GiB and 6.014946 seconds
(VllmWorkerProcess pid=3086665) INFO 07-18 09:20:59 [worker.py:268] Memory profiling takes 2.15 seconds
(VllmWorkerProcess pid=3086665) INFO 07-18 09:20:59 [worker.py:268] the current vLLM instance can use total_gpu_memory (31.95GiB) x gpu_memory_utilization (0.90) = 28.75GiB
(VllmWorkerProcess pid=3086665) INFO 07-18 09:20:59 [worker.py:268] model weights take 13.88GiB; non_torch_memory takes 1.13GiB; PyTorch activation peak memory takes 0.71GiB; the rest of the memory reserved for KV Cache is 13.03GiB.
INFO 07-18 09:20:59 [worker.py:268] Memory profiling takes 2.50 seconds
INFO 07-18 09:20:59 [worker.py:268] the current vLLM instance can use total_gpu_memory (31.95GiB) x gpu_memory_utilization (0.90) = 28.75GiB
INFO 07-18 09:20:59 [worker.py:268] model weights take 13.88GiB; non_torch_memory takes 1.13GiB; PyTorch activation peak memory takes 1.46GiB; the rest of the memory reserved for KV Cache is 12.28GiB.
INFO 07-18 09:20:59 [executor_base.py:112] # cuda blocks: 10061, # CPU blocks: 3276
INFO 07-18 09:20:59 [executor_base.py:117] Maximum concurrency for 8192 tokens per request: 19.65x
INFO 07-18 09:21:00 [llm_engine.py:455] init engine (profile, create kv cache, warmup model) took 3.46 seconds
WARNING 07-18 09:21:00 [config.py:1100] 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`.
INFO 07-18 09:21:00 [serving_chat.py:117] Using default chat sampling params from model: {'temperature': 0.6, 'top_k': 20, 'top_p': 0.95}
INFO 07-18 09:21:00 [serving_completion.py:61] Using default completion sampling params from model: {'temperature': 0.6, 'top_k': 20, 'top_p': 0.95}
INFO 07-18 09:21:00 [api_server.py:1081] Starting vLLM API server on http://0.0.0.0:40045
INFO 07-18 09:21:00 [launcher.py:26] Available routes are:
INFO 07-18 09:21:00 [launcher.py:34] Route: /openapi.json, Methods: HEAD, GET
INFO 07-18 09:21:00 [launcher.py:34] Route: /docs, Methods: HEAD, GET
INFO 07-18 09:21:00 [launcher.py:34] Route: /docs/oauth2-redirect, Methods: HEAD, GET
INFO 07-18 09:21:00 [launcher.py:34] Route: /redoc, Methods: HEAD, GET
INFO 07-18 09:21:00 [launcher.py:34] Route: /health, Methods: GET
INFO 07-18 09:21:00 [launcher.py:34] Route: /load, Methods: GET
INFO 07-18 09:21:00 [launcher.py:34] Route: /ping, Methods: GET, POST
INFO 07-18 09:21:00 [launcher.py:34] Route: /tokenize, Methods: POST
INFO 07-18 09:21:00 [launcher.py:34] Route: /detokenize, Methods: POST
INFO 07-18 09:21:00 [launcher.py:34] Route: /v1/models, Methods: GET
INFO 07-18 09:21:00 [launcher.py:34] Route: /version, Methods: GET
INFO 07-18 09:21:00 [launcher.py:34] Route: /v1/chat/completions, Methods: POST
INFO 07-18 09:21:00 [launcher.py:34] Route: /v1/completions, Methods: POST
INFO 07-18 09:21:00 [launcher.py:34] Route: /v1/embeddings, Methods: POST
INFO 07-18 09:21:00 [launcher.py:34] Route: /pooling, Methods: POST
INFO 07-18 09:21:00 [launcher.py:34] Route: /score, Methods: POST
INFO 07-18 09:21:00 [launcher.py:34] Route: /v1/score, Methods: POST
INFO 07-18 09:21:00 [launcher.py:34] Route: /v1/audio/transcriptions, Methods: POST
INFO 07-18 09:21:00 [launcher.py:34] Route: /rerank, Methods: POST
INFO 07-18 09:21:00 [launcher.py:34] Route: /v1/rerank, Methods: POST
INFO 07-18 09:21:00 [launcher.py:34] Route: /v2/rerank, Methods: POST
INFO 07-18 09:21:00 [launcher.py:34] Route: /invocations, Methods: POST
INFO 07-18 09:21:00 [launcher.py:34] Route: /metrics, Methods: GET
/usr/local/corex/lib64/python3/dist-packages/websockets/legacy/__init__.py:6: DeprecationWarning: websockets.legacy is deprecated; see https://websockets.readthedocs.io/en/stable/howto/upgrade.html for upgrade instructions
  warnings.warn(  # deprecated in 14.0 - 2024-11-09
/usr/local/lib/python3.10/site-packages/uvicorn/protocols/websockets/websockets_impl.py:16: DeprecationWarning: websockets.server.WebSocketServerProtocol is deprecated
  from websockets.server import WebSocketServerProtocol
INFO:     Started server process [3084405]
INFO:     Waiting for application startup.
INFO:     Application startup complete.
INFO:     127.0.0.1:45260 - "GET /v1/models HTTP/1.1" 200 OK
INFO:     127.0.0.1:45272 - "GET /v1/models HTTP/1.1" 200 OK
INFO 07-18 10:20:37 [multiproc_worker_utils.py:137] Terminating local vLLM worker processes
INFO 07-18 10:20:37 [launcher.py:74] Shutting down FastAPI HTTP server.
INFO:     Shutting down
INFO:     Waiting for application shutdown.
INFO:     Application shutdown complete.
/usr/local/lib/python3.10/tempfile.py:860: ResourceWarning: Implicitly cleaning up <TemporaryDirectory '/tmp/tmpcr3gfs55'>
  _warnings.warn(warn_message, ResourceWarning)