| INFO 07-07 00:49:54 [__init__.py:239] Automatically detected platform cuda. |
| INFO 07-07 00:49:56 [config.py:209] Replacing legacy 'type' key with 'rope_type' |
| INFO 07-07 00:49:56 [config.py:2968] Downcasting torch.float32 to torch.float16. |
| INFO 07-07 00:50:03 [config.py:717] This model supports multiple tasks: {'generate', 'classify', 'reward', 'embed', 'score'}. Defaulting to 'generate'. |
| INFO 07-07 00:50:03 [config.py:1770] Defaulting to use mp for distributed inference |
| INFO 07-07 00:50:03 [config.py:2003] Chunked prefill is enabled with max_num_batched_tokens=16384. |
| INFO 07-07 00:50:04 [core.py:58] Initializing a V1 LLM engine (v0.8.5.post1) with config: model='./merged1/phi_ties_1', speculative_config=None, tokenizer='./merged1/phi_ties_1', 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_ties_1, 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 00:50:04 [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 00:50:04 [shm_broadcast.py:266] vLLM message queue communication handle: Handle(local_reader_ranks=[0, 1, 2, 3], buffer_handle=(4, 10485760, 10, 'psm_5fb26b87'), local_subscribe_addr='ipc:///tmp/86a46b5f-4ec0-41fa-8781-bdaad9d85662', remote_subscribe_addr=None, remote_addr_ipv6=False) |
| WARNING 07-07 00:50:05 [utils.py:2522] Methods determine_num_available_blocks,device_config,get_cache_block_size_bytes,initialize_cache not implemented in <vllm.v1.worker.gpu_worker.Worker object at 0x147f35db4d90> |
| [1;36m(VllmWorker rank=0 pid=4071265)[0;0m INFO 07-07 00:50:05 [shm_broadcast.py:266] vLLM message queue communication handle: Handle(local_reader_ranks=[0], buffer_handle=(1, 10485760, 10, 'psm_ef8f209c'), local_subscribe_addr='ipc:///tmp/990828a6-3ac7-4369-b95c-4ce7379f9390', remote_subscribe_addr=None, remote_addr_ipv6=False) |
| WARNING 07-07 00:50:05 [utils.py:2522] Methods determine_num_available_blocks,device_config,get_cache_block_size_bytes,initialize_cache not implemented in <vllm.v1.worker.gpu_worker.Worker object at 0x147f379d3fd0> |
| [1;36m(VllmWorker rank=1 pid=4071267)[0;0m INFO 07-07 00:50:05 [shm_broadcast.py:266] vLLM message queue communication handle: Handle(local_reader_ranks=[0], buffer_handle=(1, 10485760, 10, 'psm_7ec0d566'), local_subscribe_addr='ipc:///tmp/df205a33-9aae-4d18-be54-d5e1315630eb', remote_subscribe_addr=None, remote_addr_ipv6=False) |
| WARNING 07-07 00:50:05 [utils.py:2522] Methods determine_num_available_blocks,device_config,get_cache_block_size_bytes,initialize_cache not implemented in <vllm.v1.worker.gpu_worker.Worker object at 0x147f379d3d30> |
| WARNING 07-07 00:50:05 [utils.py:2522] Methods determine_num_available_blocks,device_config,get_cache_block_size_bytes,initialize_cache not implemented in <vllm.v1.worker.gpu_worker.Worker object at 0x147f379d3f70> |
| [1;36m(VllmWorker rank=2 pid=4071268)[0;0m INFO 07-07 00:50:05 [shm_broadcast.py:266] vLLM message queue communication handle: Handle(local_reader_ranks=[0], buffer_handle=(1, 10485760, 10, 'psm_6fbb8518'), local_subscribe_addr='ipc:///tmp/a5366318-c2a5-4a9e-9934-20831a46fcf5', remote_subscribe_addr=None, remote_addr_ipv6=False) |
| [1;36m(VllmWorker rank=3 pid=4071269)[0;0m INFO 07-07 00:50:05 [shm_broadcast.py:266] vLLM message queue communication handle: Handle(local_reader_ranks=[0], buffer_handle=(1, 10485760, 10, 'psm_13e65c9f'), local_subscribe_addr='ipc:///tmp/cb55b7fb-d5e6-46e9-9f1d-552bb8974ddb', remote_subscribe_addr=None, remote_addr_ipv6=False) |
| [1;36m(VllmWorker rank=0 pid=4071265)[0;0m INFO 07-07 00:50:17 [utils.py:1055] Found nccl from library libnccl.so.2 |
| [1;36m(VllmWorker rank=2 pid=4071268)[0;0m INFO 07-07 00:50:17 [utils.py:1055] Found nccl from library libnccl.so.2 |
| [1;36m(VllmWorker rank=2 pid=4071268)[0;0m INFO 07-07 00:50:17 [pynccl.py:69] vLLM is using nccl==2.21.5 |
| [1;36m(VllmWorker rank=0 pid=4071265)[0;0m INFO 07-07 00:50:17 [pynccl.py:69] vLLM is using nccl==2.21.5 |
| [1;36m(VllmWorker rank=3 pid=4071269)[0;0m INFO 07-07 00:50:17 [utils.py:1055] Found nccl from library libnccl.so.2 |
| [1;36m(VllmWorker rank=1 pid=4071267)[0;0m INFO 07-07 00:50:17 [utils.py:1055] Found nccl from library libnccl.so.2 |
| [1;36m(VllmWorker rank=1 pid=4071267)[0;0m INFO 07-07 00:50:17 [pynccl.py:69] vLLM is using nccl==2.21.5 |
| [1;36m(VllmWorker rank=3 pid=4071269)[0;0m INFO 07-07 00:50:17 [pynccl.py:69] vLLM is using nccl==2.21.5 |
| [1;36m(VllmWorker rank=2 pid=4071268)[0;0m WARNING 07-07 00:50:18 [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. |
| [1;36m(VllmWorker rank=3 pid=4071269)[0;0m WARNING 07-07 00:50:18 [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. |
| [1;36m(VllmWorker rank=0 pid=4071265)[0;0m WARNING 07-07 00:50:18 [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. |
| [1;36m(VllmWorker rank=1 pid=4071267)[0;0m WARNING 07-07 00:50:18 [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. |
| [1;36m(VllmWorker rank=0 pid=4071265)[0;0m INFO 07-07 00:50:18 [shm_broadcast.py:266] vLLM message queue communication handle: Handle(local_reader_ranks=[1, 2, 3], buffer_handle=(3, 4194304, 6, 'psm_235eb2f4'), local_subscribe_addr='ipc:///tmp/cf032ef3-3747-4b3c-93d3-42de62f2013e', remote_subscribe_addr=None, remote_addr_ipv6=False) |
| [1;36m(VllmWorker rank=3 pid=4071269)[0;0m INFO 07-07 00:50:18 [parallel_state.py:1004] rank 3 in world size 4 is assigned as DP rank 0, PP rank 0, TP rank 3 |
| [1;36m(VllmWorker rank=2 pid=4071268)[0;0m INFO 07-07 00:50:18 [parallel_state.py:1004] rank 2 in world size 4 is assigned as DP rank 0, PP rank 0, TP rank 2 |
| [1;36m(VllmWorker rank=0 pid=4071265)[0;0m INFO 07-07 00:50:18 [parallel_state.py:1004] rank 0 in world size 4 is assigned as DP rank 0, PP rank 0, TP rank 0 |
| [1;36m(VllmWorker rank=2 pid=4071268)[0;0m INFO 07-07 00:50:18 [cuda.py:221] Using Flash Attention backend on V1 engine. |
| [1;36m(VllmWorker rank=3 pid=4071269)[0;0m INFO 07-07 00:50:18 [cuda.py:221] Using Flash Attention backend on V1 engine. |
| [1;36m(VllmWorker rank=2 pid=4071268)[0;0m WARNING 07-07 00:50:18 [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. |
| [1;36m(VllmWorker rank=3 pid=4071269)[0;0m WARNING 07-07 00:50:18 [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. |
| [1;36m(VllmWorker rank=0 pid=4071265)[0;0m INFO 07-07 00:50:18 [cuda.py:221] Using Flash Attention backend on V1 engine. |
| [1;36m(VllmWorker rank=0 pid=4071265)[0;0m WARNING 07-07 00:50:18 [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. |
| [1;36m(VllmWorker rank=3 pid=4071269)[0;0m INFO 07-07 00:50:18 [gpu_model_runner.py:1329] Starting to load model ./merged1/phi_ties_1... |
| [1;36m(VllmWorker rank=2 pid=4071268)[0;0m INFO 07-07 00:50:18 [gpu_model_runner.py:1329] Starting to load model ./merged1/phi_ties_1... |
| [1;36m(VllmWorker rank=1 pid=4071267)[0;0m INFO 07-07 00:50:18 [parallel_state.py:1004] rank 1 in world size 4 is assigned as DP rank 0, PP rank 0, TP rank 1 |
| [1;36m(VllmWorker rank=0 pid=4071265)[0;0m INFO 07-07 00:50:18 [gpu_model_runner.py:1329] Starting to load model ./merged1/phi_ties_1... |
| [1;36m(VllmWorker rank=1 pid=4071267)[0;0m INFO 07-07 00:50:18 [cuda.py:221] Using Flash Attention backend on V1 engine. |
| [1;36m(VllmWorker rank=1 pid=4071267)[0;0m WARNING 07-07 00:50:18 [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. |
| [1;36m(VllmWorker rank=1 pid=4071267)[0;0m INFO 07-07 00:50:18 [gpu_model_runner.py:1329] Starting to load model ./merged1/phi_ties_1... |
| [1;36m(VllmWorker rank=2 pid=4071268)[0;0m INFO 07-07 00:50:30 [loader.py:458] Loading weights took 12.48 seconds |
| [1;36m(VllmWorker rank=0 pid=4071265)[0;0m INFO 07-07 00:50:30 [loader.py:458] Loading weights took 12.45 seconds |
| [1;36m(VllmWorker rank=3 pid=4071269)[0;0m INFO 07-07 00:50:30 [loader.py:458] Loading weights took 12.49 seconds |
| [1;36m(VllmWorker rank=1 pid=4071267)[0;0m INFO 07-07 00:50:30 [loader.py:458] Loading weights took 12.45 seconds |
| [1;36m(VllmWorker rank=3 pid=4071269)[0;0m INFO 07-07 00:50:30 [gpu_model_runner.py:1347] Model loading took 1.8196 GiB and 12.689570 seconds |
| [1;36m(VllmWorker rank=2 pid=4071268)[0;0m INFO 07-07 00:50:30 [gpu_model_runner.py:1347] Model loading took 1.8196 GiB and 12.676910 seconds |
| [1;36m(VllmWorker rank=0 pid=4071265)[0;0m INFO 07-07 00:50:30 [gpu_model_runner.py:1347] Model loading took 1.8196 GiB and 12.683249 seconds |
| [1;36m(VllmWorker rank=1 pid=4071267)[0;0m INFO 07-07 00:50:31 [gpu_model_runner.py:1347] Model loading took 1.8196 GiB and 12.677000 seconds |
| [1;36m(VllmWorker rank=1 pid=4071267)[0;0m INFO 07-07 00:50:36 [backends.py:420] Using cache directory: /home/jiangli/.cache/vllm/torch_compile_cache/89b1d99067/rank_1_0 for vLLM's torch.compile |
| [1;36m(VllmWorker rank=1 pid=4071267)[0;0m INFO 07-07 00:50:36 [backends.py:430] Dynamo bytecode transform time: 5.84 s |
| [1;36m(VllmWorker rank=2 pid=4071268)[0;0m INFO 07-07 00:50:36 [backends.py:420] Using cache directory: /home/jiangli/.cache/vllm/torch_compile_cache/89b1d99067/rank_2_0 for vLLM's torch.compile |
| [1;36m(VllmWorker rank=2 pid=4071268)[0;0m INFO 07-07 00:50:36 [backends.py:430] Dynamo bytecode transform time: 5.94 s |
| [1;36m(VllmWorker rank=3 pid=4071269)[0;0m INFO 07-07 00:50:37 [backends.py:420] Using cache directory: /home/jiangli/.cache/vllm/torch_compile_cache/89b1d99067/rank_3_0 for vLLM's torch.compile |
| [1;36m(VllmWorker rank=3 pid=4071269)[0;0m INFO 07-07 00:50:37 [backends.py:430] Dynamo bytecode transform time: 6.02 s |
| [1;36m(VllmWorker rank=0 pid=4071265)[0;0m INFO 07-07 00:50:37 [backends.py:420] Using cache directory: /home/jiangli/.cache/vllm/torch_compile_cache/89b1d99067/rank_0_0 for vLLM's torch.compile |
| [1;36m(VllmWorker rank=0 pid=4071265)[0;0m INFO 07-07 00:50:37 [backends.py:430] Dynamo bytecode transform time: 6.07 s |
| [1;36m(VllmWorker rank=1 pid=4071267)[0;0m INFO 07-07 00:50:42 [backends.py:118] Directly load the compiled graph(s) for shape None from the cache, took 4.671 s |
| [1;36m(VllmWorker rank=2 pid=4071268)[0;0m INFO 07-07 00:50:42 [backends.py:118] Directly load the compiled graph(s) for shape None from the cache, took 5.105 s |
| [1;36m(VllmWorker rank=0 pid=4071265)[0;0m INFO 07-07 00:50:42 [backends.py:118] Directly load the compiled graph(s) for shape None from the cache, took 5.042 s |
| [1;36m(VllmWorker rank=3 pid=4071269)[0;0m INFO 07-07 00:50:42 [backends.py:118] Directly load the compiled graph(s) for shape None from the cache, took 5.148 s |
| [1;36m(VllmWorker rank=0 pid=4071265)[0;0m INFO 07-07 00:50:48 [monitor.py:33] torch.compile takes 6.07 s in total |
| [1;36m(VllmWorker rank=2 pid=4071268)[0;0m INFO 07-07 00:50:48 [monitor.py:33] torch.compile takes 5.94 s in total |
| [1;36m(VllmWorker rank=3 pid=4071269)[0;0m INFO 07-07 00:50:48 [monitor.py:33] torch.compile takes 6.02 s in total |
| [1;36m(VllmWorker rank=1 pid=4071267)[0;0m INFO 07-07 00:50:48 [monitor.py:33] torch.compile takes 5.84 s in total |
| INFO 07-07 00:50:49 [kv_cache_utils.py:634] GPU KV cache size: 2,007,088 tokens |
| INFO 07-07 00:50:49 [kv_cache_utils.py:637] Maximum concurrency for 2,048 tokens per request: 980.02x |
| INFO 07-07 00:50:49 [kv_cache_utils.py:634] GPU KV cache size: 2,006,832 tokens |
| INFO 07-07 00:50:49 [kv_cache_utils.py:637] Maximum concurrency for 2,048 tokens per request: 979.90x |
| INFO 07-07 00:50:49 [kv_cache_utils.py:634] GPU KV cache size: 2,006,832 tokens |
| INFO 07-07 00:50:49 [kv_cache_utils.py:637] Maximum concurrency for 2,048 tokens per request: 979.90x |
| INFO 07-07 00:50:49 [kv_cache_utils.py:634] GPU KV cache size: 2,008,112 tokens |
| INFO 07-07 00:50:49 [kv_cache_utils.py:637] Maximum concurrency for 2,048 tokens per request: 980.52x |
| [1;36m(VllmWorker rank=2 pid=4071268)[0;0m INFO 07-07 00:51:16 [gpu_model_runner.py:1686] Graph capturing finished in 26 secs, took 3.00 GiB |
| [1;36m(VllmWorker rank=0 pid=4071265)[0;0m INFO 07-07 00:51:16 [gpu_model_runner.py:1686] Graph capturing finished in 26 secs, took 3.00 GiB |
| [1;36m(VllmWorker rank=1 pid=4071267)[0;0m INFO 07-07 00:51:16 [gpu_model_runner.py:1686] Graph capturing finished in 27 secs, took 3.00 GiB |
| [1;36m(VllmWorker rank=3 pid=4071269)[0;0m INFO 07-07 00:51:16 [gpu_model_runner.py:1686] Graph capturing finished in 27 secs, took 3.00 GiB |
| INFO 07-07 00:51:16 [core.py:159] init engine (profile, create kv cache, warmup model) took 45.43 seconds |
| INFO 07-07 00:51:16 [core_client.py:439] Core engine process 0 ready. |
| INFO 07-07 00:52:32 [importing.py:53] Triton module has been replaced with a placeholder. |
| INFO 07-07 00:52:32 [__init__.py:239] Automatically detected platform cuda. |
| | Task |Version| Metric |Value | |Stderr| |
| |------------------|------:|---------------------|-----:|---|-----:| |
| |all | |sem |0.8810|± |0.0192| |
| | | |math_pass@1:1_samples|0.9573|± |0.0181| |
| |mm\|arc_challenge\|0| 0|sem |0.9318|± |0.0129| |
| |mm\|arc_easy\|0 | 0|sem |0.9736|± |0.0052| |
| |mm\|commonsenseqa\|0| 0|sem |0.8500|± |0.0200| |
| |mm\|gsm8k\|0 | 0|math_pass@1:1_samples|0.9396|± |0.0113| |
| |mm\|math_500\|0 | 3|math_pass@1:1_samples|0.9750|± |0.0250| |
| |mm\|truthfulqa\|0 | 0|sem |0.7686|± |0.0385| |
|
|
|
|
