INFO 07-07 00:08:19 [__init__.py:239] Automatically detected platform cuda. INFO 07-07 00:08:21 [config.py:209] Replacing legacy 'type' key with 'rope_type' INFO 07-07 00:08:21 [config.py:2968] Downcasting torch.float32 to torch.float16. INFO 07-07 00:08:28 [config.py:717] This model supports multiple tasks: {'generate', 'classify', 'score', 'reward', 'embed'}. Defaulting to 'generate'. INFO 07-07 00:08:28 [config.py:1770] Defaulting to use mp for distributed inference INFO 07-07 00:08:28 [config.py:2003] Chunked prefill is enabled with max_num_batched_tokens=16384. INFO 07-07 00:08:29 [core.py:58] Initializing a V1 LLM engine (v0.8.5.post1) with config: model='./merged1/phi_linear_9', speculative_config=None, tokenizer='./merged1/phi_linear_9', skip_tokenizer_init=False, tokenizer_mode=auto, revision=None, override_neuron_config=None, tokenizer_revision=None, trust_remote_code=False, dtype=torch.float16, max_seq_len=2048, download_dir=None, load_format=auto, tensor_parallel_size=4, pipeline_parallel_size=1, disable_custom_all_reduce=False, quantization=None, enforce_eager=False, kv_cache_dtype=auto, device_config=cuda, decoding_config=DecodingConfig(guided_decoding_backend='auto', 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=./merged1/phi_linear_9, num_scheduler_steps=1, multi_step_stream_outputs=True, enable_prefix_caching=True, chunked_prefill_enabled=True, use_async_output_proc=True, disable_mm_preprocessor_cache=False, mm_processor_kwargs=None, pooler_config=None, compilation_config={"level":3,"custom_ops":["none"],"splitting_ops":["vllm.unified_attention","vllm.unified_attention_with_output"],"use_inductor":true,"compile_sizes":[],"use_cudagraph":true,"cudagraph_num_of_warmups":1,"cudagraph_capture_sizes":[512,504,496,488,480,472,464,456,448,440,432,424,416,408,400,392,384,376,368,360,352,344,336,328,320,312,304,296,288,280,272,264,256,248,240,232,224,216,208,200,192,184,176,168,160,152,144,136,128,120,112,104,96,88,80,72,64,56,48,40,32,24,16,8,4,2,1],"max_capture_size":512} WARNING 07-07 00:08:29 [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. INFO 07-07 00:08:29 [shm_broadcast.py:266] vLLM message queue communication handle: Handle(local_reader_ranks=[0, 1, 2, 3], buffer_handle=(4, 10485760, 10, 'psm_b47effa6'), local_subscribe_addr='ipc:///tmp/947b9baa-eef0-4c27-aed5-34270c083d92', remote_subscribe_addr=None, remote_addr_ipv6=False) WARNING 07-07 00:08:29 [utils.py:2522] Methods determine_num_available_blocks,device_config,get_cache_block_size_bytes,initialize_cache not implemented in (VllmWorker rank=1 pid=4042133) INFO 07-07 00:08:29 [shm_broadcast.py:266] vLLM message queue communication handle: Handle(local_reader_ranks=[0], buffer_handle=(1, 10485760, 10, 'psm_01f6dea0'), local_subscribe_addr='ipc:///tmp/0d092c78-a16c-4455-8186-5ab3acb13cf3', remote_subscribe_addr=None, remote_addr_ipv6=False) WARNING 07-07 00:08:29 [utils.py:2522] Methods determine_num_available_blocks,device_config,get_cache_block_size_bytes,initialize_cache not implemented in WARNING 07-07 00:08:29 [utils.py:2522] Methods determine_num_available_blocks,device_config,get_cache_block_size_bytes,initialize_cache not implemented in WARNING 07-07 00:08:29 [utils.py:2522] Methods determine_num_available_blocks,device_config,get_cache_block_size_bytes,initialize_cache not implemented in (VllmWorker rank=0 pid=4042132) INFO 07-07 00:08:29 [shm_broadcast.py:266] vLLM message queue communication handle: Handle(local_reader_ranks=[0], buffer_handle=(1, 10485760, 10, 'psm_27c8e66c'), local_subscribe_addr='ipc:///tmp/f65cdba4-f53d-4256-8656-7d1a816f9b37', remote_subscribe_addr=None, remote_addr_ipv6=False) (VllmWorker rank=3 pid=4042135) INFO 07-07 00:08:30 [shm_broadcast.py:266] vLLM message queue communication handle: Handle(local_reader_ranks=[0], buffer_handle=(1, 10485760, 10, 'psm_4b83337a'), local_subscribe_addr='ipc:///tmp/15d5657b-cd37-4835-af56-8dd3277fe73e', remote_subscribe_addr=None, remote_addr_ipv6=False) (VllmWorker rank=2 pid=4042134) INFO 07-07 00:08:30 [shm_broadcast.py:266] vLLM message queue communication handle: Handle(local_reader_ranks=[0], buffer_handle=(1, 10485760, 10, 'psm_50058d84'), local_subscribe_addr='ipc:///tmp/f762c34f-a2f9-448a-97db-1db7cc9db8c8', remote_subscribe_addr=None, remote_addr_ipv6=False) (VllmWorker rank=0 pid=4042132) INFO 07-07 00:08:32 [utils.py:1055] Found nccl from library libnccl.so.2 (VllmWorker rank=0 pid=4042132) INFO 07-07 00:08:32 [pynccl.py:69] vLLM is using nccl==2.21.5 (VllmWorker rank=1 pid=4042133) INFO 07-07 00:08:32 [utils.py:1055] Found nccl from library libnccl.so.2 (VllmWorker rank=1 pid=4042133) INFO 07-07 00:08:32 [pynccl.py:69] vLLM is using nccl==2.21.5 (VllmWorker rank=3 pid=4042135) INFO 07-07 00:08:32 [utils.py:1055] Found nccl from library libnccl.so.2 (VllmWorker rank=2 pid=4042134) INFO 07-07 00:08:32 [utils.py:1055] Found nccl from library libnccl.so.2 (VllmWorker rank=3 pid=4042135) INFO 07-07 00:08:32 [pynccl.py:69] vLLM is using nccl==2.21.5 (VllmWorker rank=2 pid=4042134) INFO 07-07 00:08:32 [pynccl.py:69] vLLM is using nccl==2.21.5 (VllmWorker rank=0 pid=4042132) WARNING 07-07 00:08:33 [custom_all_reduce.py:136] Custom allreduce is disabled because it's not supported on more than two PCIe-only GPUs. To silence this warning, specify disable_custom_all_reduce=True explicitly. (VllmWorker rank=2 pid=4042134) WARNING 07-07 00:08:33 [custom_all_reduce.py:136] Custom allreduce is disabled because it's not supported on more than two PCIe-only GPUs. To silence this warning, specify disable_custom_all_reduce=True explicitly. (VllmWorker rank=3 pid=4042135) WARNING 07-07 00:08:33 [custom_all_reduce.py:136] Custom allreduce is disabled because it's not supported on more than two PCIe-only GPUs. To silence this warning, specify disable_custom_all_reduce=True explicitly. (VllmWorker rank=1 pid=4042133) WARNING 07-07 00:08:33 [custom_all_reduce.py:136] Custom allreduce is disabled because it's not supported on more than two PCIe-only GPUs. To silence this warning, specify disable_custom_all_reduce=True explicitly. (VllmWorker rank=0 pid=4042132) INFO 07-07 00:08:33 [shm_broadcast.py:266] vLLM message queue communication handle: Handle(local_reader_ranks=[1, 2, 3], buffer_handle=(3, 4194304, 6, 'psm_4ca06e85'), local_subscribe_addr='ipc:///tmp/9aa2a8d5-9ec6-4389-9943-55ed1afa709f', remote_subscribe_addr=None, remote_addr_ipv6=False) (VllmWorker rank=2 pid=4042134) INFO 07-07 00:08:33 [parallel_state.py:1004] rank 2 in world size 4 is assigned as DP rank 0, PP rank 0, TP rank 2 (VllmWorker rank=2 pid=4042134) INFO 07-07 00:08:33 [cuda.py:221] Using Flash Attention backend on V1 engine. (VllmWorker rank=2 pid=4042134) WARNING 07-07 00:08:33 [topk_topp_sampler.py:69] FlashInfer is not available. Falling back to the PyTorch-native implementation of top-p & top-k sampling. For the best performance, please install FlashInfer. (VllmWorker rank=1 pid=4042133) INFO 07-07 00:08:33 [parallel_state.py:1004] rank 1 in world size 4 is assigned as DP rank 0, PP rank 0, TP rank 1 (VllmWorker rank=3 pid=4042135) INFO 07-07 00:08:33 [parallel_state.py:1004] rank 3 in world size 4 is assigned as DP rank 0, PP rank 0, TP rank 3 (VllmWorker rank=0 pid=4042132) INFO 07-07 00:08:33 [parallel_state.py:1004] rank 0 in world size 4 is assigned as DP rank 0, PP rank 0, TP rank 0 (VllmWorker rank=3 pid=4042135) INFO 07-07 00:08:33 [cuda.py:221] Using Flash Attention backend on V1 engine. (VllmWorker rank=1 pid=4042133) INFO 07-07 00:08:33 [cuda.py:221] Using Flash Attention backend on V1 engine. (VllmWorker rank=0 pid=4042132) INFO 07-07 00:08:33 [cuda.py:221] Using Flash Attention backend on V1 engine. (VllmWorker rank=3 pid=4042135) WARNING 07-07 00:08:33 [topk_topp_sampler.py:69] FlashInfer is not available. Falling back to the PyTorch-native implementation of top-p & top-k sampling. For the best performance, please install FlashInfer. (VllmWorker rank=1 pid=4042133) WARNING 07-07 00:08:33 [topk_topp_sampler.py:69] FlashInfer is not available. Falling back to the PyTorch-native implementation of top-p & top-k sampling. For the best performance, please install FlashInfer. (VllmWorker rank=0 pid=4042132) WARNING 07-07 00:08:33 [topk_topp_sampler.py:69] FlashInfer is not available. Falling back to the PyTorch-native implementation of top-p & top-k sampling. For the best performance, please install FlashInfer. (VllmWorker rank=2 pid=4042134) INFO 07-07 00:08:33 [gpu_model_runner.py:1329] Starting to load model ./merged1/phi_linear_9... (VllmWorker rank=3 pid=4042135) INFO 07-07 00:08:33 [gpu_model_runner.py:1329] Starting to load model ./merged1/phi_linear_9... (VllmWorker rank=1 pid=4042133) INFO 07-07 00:08:33 [gpu_model_runner.py:1329] Starting to load model ./merged1/phi_linear_9... (VllmWorker rank=0 pid=4042132) INFO 07-07 00:08:33 [gpu_model_runner.py:1329] Starting to load model ./merged1/phi_linear_9... (VllmWorker rank=3 pid=4042135) INFO 07-07 00:08:35 [loader.py:458] Loading weights took 1.51 seconds (VllmWorker rank=2 pid=4042134) INFO 07-07 00:08:35 [loader.py:458] Loading weights took 1.51 seconds (VllmWorker rank=0 pid=4042132) INFO 07-07 00:08:35 [loader.py:458] Loading weights took 1.71 seconds (VllmWorker rank=3 pid=4042135) INFO 07-07 00:08:35 [gpu_model_runner.py:1347] Model loading took 1.8196 GiB and 1.687102 seconds (VllmWorker rank=2 pid=4042134) INFO 07-07 00:08:35 [gpu_model_runner.py:1347] Model loading took 1.8196 GiB and 1.699885 seconds (VllmWorker rank=1 pid=4042133) INFO 07-07 00:08:35 [loader.py:458] Loading weights took 1.78 seconds (VllmWorker rank=0 pid=4042132) INFO 07-07 00:08:35 [gpu_model_runner.py:1347] Model loading took 1.8196 GiB and 1.911037 seconds (VllmWorker rank=1 pid=4042133) INFO 07-07 00:08:35 [gpu_model_runner.py:1347] Model loading took 1.8196 GiB and 1.998182 seconds (VllmWorker rank=0 pid=4042132) INFO 07-07 00:08:41 [backends.py:420] Using cache directory: /home/jiangli/.cache/vllm/torch_compile_cache/fe3a6231bf/rank_0_0 for vLLM's torch.compile (VllmWorker rank=0 pid=4042132) INFO 07-07 00:08:41 [backends.py:430] Dynamo bytecode transform time: 5.78 s (VllmWorker rank=3 pid=4042135) INFO 07-07 00:08:41 [backends.py:420] Using cache directory: /home/jiangli/.cache/vllm/torch_compile_cache/fe3a6231bf/rank_3_0 for vLLM's torch.compile (VllmWorker rank=3 pid=4042135) INFO 07-07 00:08:41 [backends.py:430] Dynamo bytecode transform time: 5.78 s (VllmWorker rank=2 pid=4042134) INFO 07-07 00:08:41 [backends.py:420] Using cache directory: /home/jiangli/.cache/vllm/torch_compile_cache/fe3a6231bf/rank_2_0 for vLLM's torch.compile (VllmWorker rank=2 pid=4042134) INFO 07-07 00:08:41 [backends.py:430] Dynamo bytecode transform time: 5.79 s (VllmWorker rank=1 pid=4042133) INFO 07-07 00:08:41 [backends.py:420] Using cache directory: /home/jiangli/.cache/vllm/torch_compile_cache/fe3a6231bf/rank_1_0 for vLLM's torch.compile (VllmWorker rank=1 pid=4042133) INFO 07-07 00:08:41 [backends.py:430] Dynamo bytecode transform time: 5.90 s (VllmWorker rank=2 pid=4042134) INFO 07-07 00:08:46 [backends.py:118] Directly load the compiled graph(s) for shape None from the cache, took 4.410 s (VllmWorker rank=0 pid=4042132) INFO 07-07 00:08:46 [backends.py:118] Directly load the compiled graph(s) for shape None from the cache, took 4.430 s (VllmWorker rank=3 pid=4042135) INFO 07-07 00:08:46 [backends.py:118] Directly load the compiled graph(s) for shape None from the cache, took 4.417 s (VllmWorker rank=1 pid=4042133) INFO 07-07 00:08:46 [backends.py:118] Directly load the compiled graph(s) for shape None from the cache, took 4.486 s (VllmWorker rank=1 pid=4042133) INFO 07-07 00:08:52 [monitor.py:33] torch.compile takes 5.90 s in total (VllmWorker rank=3 pid=4042135) INFO 07-07 00:08:52 [monitor.py:33] torch.compile takes 5.78 s in total (VllmWorker rank=2 pid=4042134) INFO 07-07 00:08:52 [monitor.py:33] torch.compile takes 5.79 s in total (VllmWorker rank=0 pid=4042132) INFO 07-07 00:08:52 [monitor.py:33] torch.compile takes 5.78 s in total INFO 07-07 00:08:53 [kv_cache_utils.py:634] GPU KV cache size: 2,007,088 tokens INFO 07-07 00:08:53 [kv_cache_utils.py:637] Maximum concurrency for 2,048 tokens per request: 980.02x INFO 07-07 00:08:53 [kv_cache_utils.py:634] GPU KV cache size: 2,006,832 tokens INFO 07-07 00:08:53 [kv_cache_utils.py:637] Maximum concurrency for 2,048 tokens per request: 979.90x INFO 07-07 00:08:53 [kv_cache_utils.py:634] GPU KV cache size: 2,006,832 tokens INFO 07-07 00:08:53 [kv_cache_utils.py:637] Maximum concurrency for 2,048 tokens per request: 979.90x INFO 07-07 00:08:53 [kv_cache_utils.py:634] GPU KV cache size: 2,008,112 tokens INFO 07-07 00:08:53 [kv_cache_utils.py:637] Maximum concurrency for 2,048 tokens per request: 980.52x (VllmWorker rank=2 pid=4042134) INFO 07-07 00:09:19 [gpu_model_runner.py:1686] Graph capturing finished in 26 secs, took 3.00 GiB (VllmWorker rank=3 pid=4042135) INFO 07-07 00:09:19 [gpu_model_runner.py:1686] Graph capturing finished in 26 secs, took 3.00 GiB (VllmWorker rank=0 pid=4042132) INFO 07-07 00:09:19 [gpu_model_runner.py:1686] Graph capturing finished in 26 secs, took 3.00 GiB (VllmWorker rank=1 pid=4042133) INFO 07-07 00:09:19 [gpu_model_runner.py:1686] Graph capturing finished in 26 secs, took 3.00 GiB INFO 07-07 00:09:19 [core.py:159] init engine (profile, create kv cache, warmup model) took 43.82 seconds INFO 07-07 00:09:19 [core_client.py:439] Core engine process 0 ready. INFO 07-07 00:13:43 [importing.py:53] Triton module has been replaced with a placeholder. INFO 07-07 00:13:43 [__init__.py:239] Automatically detected platform cuda. | Task |Version| Metric |Value | |Stderr| |------------------|------:|---------------------|-----:|---|-----:| |all | |math_pass@1:1_samples|0.8659|± |0.0348| | | |sem |0.4659|± |0.0286| |mm\|arc_challenge\|0| 0|sem |0.4724|± |0.0256| |mm\|arc_easy\|0 | 0|sem |0.5512|± |0.0162| |mm\|commonsenseqa\|0| 0|sem |0.3937|± |0.0274| |mm\|gsm8k\|0 | 0|math_pass@1:1_samples|0.8568|± |0.0166| |mm\|math_500\|0 | 3|math_pass@1:1_samples|0.8750|± |0.0530| |mm\|truthfulqa\|0 | 0|sem |0.4463|± |0.0454|