Buckets:
| (MainProcess pid=57495) INFO 10-31 09:45:28 [model.py:733] Resolved architecture: Qwen3_5ForCausalLM | |
| (MainProcess pid=57495) INFO 10-31 09:45:28 [model.py:2152] Using max model len 32768 | |
| (MainProcess pid=57495) INFO 10-31 09:45:28 [scheduler.py:307] Chunked prefill is enabled with max_num_batched_tokens=2048. | |
| (MainProcess pid=57495) INFO 10-31 09:45:28 [config.py:688] Mamba cache mode is set to 'align' for Qwen3_5ForCausalLM by default when prefix caching is enabled | |
| (MainProcess pid=57495) INFO 10-31 09:45:28 [kernel.py:467] Final IR op priority after setting platform defaults: IrOpPriorityConfig(rms_norm=['native'], fused_add_rms_norm=['native'], gelu_and_mul_sparse=['triton', 'native']) | |
| (EngineCore pid=57770) INFO 10-31 09:45:34 [core.py:133] Initializing a V1 LLM engine (v0.32.0) with config: model='checkpoints/gold_only_merged', speculative_config=None, tokenizer='checkpoints/gold_only_merged', skip_tokenizer_init=False, tokenizer_mode=auto, revision=None, tokenizer_revision=None, trust_remote_code=False, dtype=torch.bfloat16, max_seq_len=32768, download_dir=None, load_format=auto, tensor_parallel_size=1, pipeline_parallel_size=1, data_parallel_size=1, decode_context_parallel_size=1, dcp_comm_backend=ag_rs, disable_custom_all_reduce=False, quantization=None, quantization_config=None, enforce_eager=False, aux_output_config=AuxOutputConfig(enable_return_routed_experts=False, max_bytes=None), kv_cache_dtype=auto, device_config=cuda, structured_outputs_config=StructuredOutputsConfig(backend='auto', 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, per_request_spec_decode_metrics='none', kv_cache_metrics=False, kv_cache_metrics_sample=0.01, custom_histogram_buckets=None, cudagraph_metrics=False, enable_layerwise_nvtx_tracing=False, enable_mfu_metrics=False, enable_mm_processor_stats=False, enable_logging_iteration_details=False, jit_monitor_mode='warn', jit_monitor_verbose=False), seed=17, served_model_name=checkpoints/gold_only_merged, enable_prefix_caching=True, enable_chunked_prefill=True, pooler_config=None, compilation_config={'mode': <CompilationMode.VLLM_COMPILE: 3>, 'debug_dump_path': None, 'cache_dir': '', 'compile_cache_save_format': 'binary', 'backend': 'inductor', 'custom_ops': ['none'], 'ir_enable_torch_wrap': True, 'splitting_ops': ['vllm::unified_attention_with_output', 'vllm::unified_mla_attention_with_output', 'vllm::mamba_mixer2', 'vllm::mamba_mixer', 'vllm::short_conv', 'vllm::qwen4_exp_ple_short_conv', 'vllm::qwen4_exp_qsa_with_output', 'vllm::qwen4_exp_qsa_prepare_with_output', 'vllm::linear_attention', 'vllm::qwen_gdn_attention_core', 'vllm::qwen_gdn_attention_core_fused_norm_packed', 'vllm::gdn_attention_core_xpu', 'vllm::olmo_hybrid_gdn_full_forward', 'vllm::sparse_attn_indexer', 'vllm::rocm_aiter_sparse_attn_indexer', 'vllm::deepseek_v4_attention', 'vllm::hpc_rope_norm_forward', 'vllm::unified_kv_cache_update', 'vllm::unified_mla_kv_cache_update'], 'compile_mm_encoder': False, 'cudagraph_mm_encoder': False, 'encoder_cudagraph_token_budgets': [], 'encoder_cudagraph_max_vision_items_per_batch': 0, 'encoder_cudagraph_max_frames_per_batch': None, 'cudagraph_decoder_replay': True, 'decoder_replay_cudagraph_capture_sizes': [], 'decoder_replay_trim_threshold': 768, 'compile_sizes': [], 'compile_ranges_endpoints': [2048], '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], '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, 'fuse_allreduce_rms': False, 'enable_qk_norm_rope_fusion': False, 'fuse_rope_kvcache_cat_mla': False, 'fuse_act_padding': False, 'fuse_qk_norm_rope_kvcache': False}, 'max_cudagraph_capture_size': 256, 'dynamic_shapes_config': {'type': <DynamicShapesType.BACKED: 'backed'>, 'evaluate_guards': False, 'assume_32_bit_indexing': False}, 'local_cache_dir': None, 'fast_moe_cold_start': False, 'static_all_moe_layers': []}, kernel_config=KernelConfig(ir_op_priority=IrOpPriorityConfig(rms_norm=['native'], fused_add_rms_norm=['native'], gelu_and_mul_sparse=['triton', 'native']), enable_flashinfer_autotune=True, enable_rocm_segmented_attn_autotune=False, enable_cutedsl_warmup=True, enable_jit_warmup=True, moe_backend='auto', sparse_indexer_topk_backend='auto', linear_backend='auto', linear_backend_per_quant=None) | |
| (EngineCore pid=57770) INFO 10-31 09:45:36 [parallel_state.py:1906] world_size=1 rank=0 local_rank=0 distributed_init_method=file:///tmp/vllm_dist_cded85161b374b0f9cd4fead486e56d8 backend=nccl | |
| (EngineCore pid=57770) INFO 10-31 09:45:36 [parallel_state.py:2350] rank 0 in world size 1 is assigned as DP rank 0, PP rank 0, PCP rank 0, TP rank 0, ETP rank 0, EP rank N/A, EPLB rank N/A | |
| (EngineCore pid=57770) INFO 10-31 09:45:36 [gpu_worker.py:518] Using V2 Model Runner | |
| (EngineCore pid=57770) INFO 10-31 09:45:37 [model_runner.py:392] Loading model from scratch... | |
| (EngineCore pid=57770) INFO 10-31 09:45:37 [qwen_gdn_linear_attn.py:221] Using Triton/FLA GDN prefill kernel (requested=auto, head_k_dim=128). | |
| (EngineCore pid=57770) INFO 10-31 09:45:39 [cuda.py:506] Using FLASH_ATTN attention backend out of potential backends: ['FLASH_ATTN', 'FLASHINFER', 'TRITON_ATTN', 'FLEX_ATTENTION']. | |
| (EngineCore pid=57770) INFO 10-31 09:45:39 [flash_attn.py:1117] Using FlashAttention version 2 | |
| (EngineCore pid=57770) INFO 10-31 09:45:39 [base_loader.py:82] Initializing model took 1.584827 seconds | |
| (EngineCore pid=57770) INFO 10-31 09:45:39 [weight_utils.py:1108] Filesystem type for checkpoints: XFS. Checkpoint size: 7.83 GiB. Available RAM: 162.14 GiB. | |
| (EngineCore pid=57770) INFO 10-31 09:45:39 [weight_utils.py:1131] Auto-prefetch is disabled because the filesystem (XFS) is not a recognized network FS (NFS/Lustre). If you want to force prefetching, start vLLM with --safetensors-load-strategy=prefetch. | |
| Loading safetensors checkpoint shards: 0% Completed | 0/1 [00:00<?, ?it/s] | |
| Loading safetensors checkpoint shards: 100% Completed | 1/1 [00:00<00:00, 1.22it/s] | |
| Loading safetensors checkpoint shards: 100% Completed | 1/1 [00:00<00:00, 1.22it/s] | |
| (EngineCore pid=57770) INFO 10-31 09:45:40 [default_loader.py:501] Loading weights took 0.84 seconds | |
| (EngineCore pid=57770) INFO 10-31 09:45:42 [model_runner.py:424] Model loading took 7.94 GiB memory and 4.588046 seconds | |
| (EngineCore pid=57770) INFO 10-31 09:45:42 [topk_topp_sampler.py:124] Using FlashInfer for top-p & top-k sampling. | |
| (EngineCore pid=57770) INFO 10-31 09:45:42 [torch_utils.py:295] Reducing Torch threads from 128 to 1 for serving. Set OMP_NUM_THREADS in the external environment to override. | |
| (EngineCore pid=57770) INFO 10-31 09:45:42 [interface.py:1096] Setting attention block size to 528 tokens to ensure that attention page size is >= mamba page size. | |
| (EngineCore pid=57770) INFO 10-31 09:45:42 [interface.py:1120] Padding mamba page size by 0.76% to ensure that mamba page size and attention page size are exactly equal. | |
| (EngineCore pid=57770) INFO 10-31 09:45:42 [utils.py:347] Using LBNHC KV cache layout. | |
| (EngineCore pid=57770) INFO 10-31 09:45:43 [monitor.py:40] [backbone] Starting torch.compile | |
| (EngineCore pid=57770) INFO 10-31 09:45:48 [backends.py:1104] [backbone] Using cache directory: /root/.cache/vllm/torch_compile_cache/8a9d0bdb28/rank_0_0/backbone for vLLM's torch.compile | |
| (EngineCore pid=57770) INFO 10-31 09:45:48 [backends.py:1166] [backbone] Dynamo bytecode transform time: 5.48 s | |
| (EngineCore pid=57770) INFO 10-31 09:46:06 [backends.py:398] [backbone] Compiling a graph for compile range (1, 2048) takes 17.35 s | |
| (EngineCore pid=57770) INFO 10-31 09:46:08 [backends.py:927] collected artifacts: 33 entries, 12 artifacts, 20433929 bytes total | |
| (EngineCore pid=57770) INFO 10-31 09:46:08 [decorators.py:743] [backbone] saved AOT compiled function to /root/.cache/vllm/torch_compile_cache/torch_aot_compile/a412c7d5e23763e33eecbcc5f7b227476f13259ff5446f4d73fb6b16a8691e68/rank_0_0/model | |
| (EngineCore pid=57770) INFO 10-31 09:46:08 [monitor.py:62] [backbone] torch.compile took 25.85 s in total | |
| (EngineCore pid=57770) INFO 10-31 09:47:37 [monitor.py:91] [backbone] Initial profiling/warmup run took 88.35 s | |
| Capturing CUDA graphs (PIECEWISE): 0%| | 0/35 [00:00<?, ?it/s] Capturing CUDA graphs (PIECEWISE): 3%|▎ | 1/35 [00:00<00:32, 1.03it/s] Capturing CUDA graphs (PIECEWISE): 6%|▌ | 2/35 [00:01<00:25, 1.27it/s] Capturing CUDA graphs (PIECEWISE): 9%|▊ | 3/35 [00:02<00:20, 1.55it/s] Capturing CUDA graphs (PIECEWISE): 11%|█▏ | 4/35 [00:02<00:15, 2.05it/s] Capturing CUDA graphs (PIECEWISE): 14%|█▍ | 5/35 [00:02<00:14, 2.05it/s] Capturing CUDA graphs (PIECEWISE): 17%|█▋ | 6/35 [00:03<00:11, 2.51it/s] Capturing CUDA graphs (PIECEWISE): 20%|██ | 7/35 [00:03<00:11, 2.41it/s] Capturing CUDA graphs (PIECEWISE): 23%|██▎ | 8/35 [00:03<00:09, 2.81it/s] Capturing CUDA graphs (PIECEWISE): 26%|██▌ | 9/35 [00:04<00:10, 2.54it/s] Capturing CUDA graphs (PIECEWISE): 29%|██▊ | 10/35 [00:04<00:10, 2.43it/s] Capturing CUDA graphs (PIECEWISE): 31%|███▏ | 11/35 [00:04<00:08, 2.78it/s] Capturing CUDA graphs (PIECEWISE): 34%|███▍ | 12/35 [00:05<00:08, 2.64it/s] Capturing CUDA graphs (PIECEWISE): 37%|███▋ | 13/35 [00:05<00:07, 3.14it/s] Capturing CUDA graphs (PIECEWISE): 40%|████ | 14/35 [00:05<00:05, 3.61it/s] Capturing CUDA graphs (PIECEWISE): 43%|████▎ | 15/35 [00:06<00:05, 3.41it/s] Capturing CUDA graphs (PIECEWISE): 46%|████▌ | 16/35 [00:06<00:04, 4.13it/s] Capturing CUDA graphs (PIECEWISE): 49%|████▊ | 17/35 [00:07<00:09, 1.96it/s] Capturing CUDA graphs (PIECEWISE): 51%|█████▏ | 18/35 [00:07<00:07, 2.33it/s] Capturing CUDA graphs (PIECEWISE): 57%|█████▋ | 20/35 [00:07<00:03, 3.92it/s] Capturing CUDA graphs (PIECEWISE): 60%|██████ | 21/35 [00:07<00:03, 3.94it/s] Capturing CUDA graphs (PIECEWISE): 66%|██████▌ | 23/35 [00:07<00:02, 5.82it/s] Capturing CUDA graphs (PIECEWISE): 74%|███████▍ | 26/35 [00:08<00:01, 8.83it/s] Capturing CUDA graphs (PIECEWISE): 80%|████████ | 28/35 [00:08<00:00, 10.20it/s] Capturing CUDA graphs (PIECEWISE): 86%|████████▌ | 30/35 [00:11<00:02, 2.04it/s] Capturing CUDA graphs (PIECEWISE): 91%|█████████▏| 32/35 [00:11<00:01, 2.76it/s] Capturing CUDA graphs (PIECEWISE): 97%|█████████▋| 34/35 [00:11<00:00, 3.71it/s] Capturing CUDA graphs (PIECEWISE): 100%|██████████| 35/35 [00:11<00:00, 3.01it/s] | |
| Capturing CUDA graphs (FULL): 0%| | 0/2 [00:00<?, ?it/s] Capturing CUDA graphs (FULL): 100%|██████████| 2/2 [00:00<00:00, 23.80it/s] | |
| (EngineCore pid=57770) INFO 10-31 09:47:52 [model_runner.py:1127] Graph capturing finished in 13 secs, took 0.30 GiB | |
| (EngineCore pid=57770) INFO 10-31 09:47:53 [gpu_worker.py:759] Available KV cache memory: 59.38 GiB | |
| (EngineCore pid=57770) INFO 10-31 09:47:53 [gpu_worker.py:774] CUDA graph memory profiling is enabled (default since v0.21.0). The current --gpu-memory-utilization=0.9000 is equivalent to --gpu-memory-utilization=0.8960 without CUDA graph memory profiling. To maintain the same effective KV cache size as before, increase --gpu-memory-utilization to 0.9040. To disable, set VLLM_MEMORY_PROFILER_ESTIMATE_CUDAGRAPHS=0. | |
| (EngineCore pid=57770) INFO 10-31 09:47:53 [kv_cache_utils.py:2327] GPU KV cache size: 1,749,526 tokens, Maximum concurrency for 32,768 tokens per request: 53.39x | |
| (EngineCore pid=57770) INFO 10-31 09:47:53 [kv_cache_utils.py:853] kv cache group sizes (528, 528, 528, 528) | |
| (EngineCore pid=57770) INFO 10-31 09:47:53 [kv_cache_utils.py:854] kv lcm block sizes 528 | |
| (EngineCore pid=57770) INFO 10-31 09:47:53 [kernel_warmup.py:176] JIT kernel warmup starting. | |
| (EngineCore pid=57770) INFO 10-31 09:47:53 [kernel_warmup.py:189] JIT kernel warmup finished in 0.00s. | |
| (EngineCore pid=57770) INFO 10-31 09:47:53 [qwen_triton_warmup.py:296] Warming up Qwen GDN Triton kernels for model_type=qwen3_5_text. | |
| (EngineCore pid=57770) INFO 10-31 09:47:54 [mamba_triton_warmup.py:42] Warmed Mamba batch_memcpy_kernel. | |
| Capturing CUDA graphs (PIECEWISE): 0%| | 0/35 [00:00<?, ?it/s] Capturing CUDA graphs (PIECEWISE): 3%|▎ | 1/35 [00:00<00:12, 2.77it/s] Capturing CUDA graphs (PIECEWISE): 6%|▌ | 2/35 [00:00<00:14, 2.30it/s] Capturing CUDA graphs (PIECEWISE): 9%|▊ | 3/35 [00:01<00:12, 2.66it/s] Capturing CUDA graphs (PIECEWISE): 11%|█▏ | 4/35 [00:01<00:12, 2.51it/s] Capturing CUDA graphs (PIECEWISE): 14%|█▍ | 5/35 [00:01<00:11, 2.68it/s] Capturing CUDA graphs (PIECEWISE): 17%|█▋ | 6/35 [00:02<00:10, 2.82it/s] Capturing CUDA graphs (PIECEWISE): 20%|██ | 7/35 [00:02<00:09, 2.98it/s] Capturing CUDA graphs (PIECEWISE): 23%|██▎ | 8/35 [00:02<00:09, 2.99it/s] Capturing CUDA graphs (PIECEWISE): 26%|██▌ | 9/35 [00:03<00:10, 2.59it/s] Capturing CUDA graphs (PIECEWISE): 29%|██▊ | 10/35 [00:03<00:08, 2.88it/s] Capturing CUDA graphs (PIECEWISE): 31%|███▏ | 11/35 [00:04<00:09, 2.66it/s] Capturing CUDA graphs (PIECEWISE): 34%|███▍ | 12/35 [00:04<00:07, 3.11it/s] Capturing CUDA graphs (PIECEWISE): 37%|███▋ | 13/35 [00:04<00:07, 2.86it/s] Capturing CUDA graphs (PIECEWISE): 40%|████ | 14/35 [00:04<00:06, 3.42it/s] Capturing CUDA graphs (PIECEWISE): 43%|████▎ | 15/35 [00:04<00:04, 4.11it/s] Capturing CUDA graphs (PIECEWISE): 46%|████▌ | 16/35 [00:05<00:05, 3.60it/s] Capturing CUDA graphs (PIECEWISE): 54%|█████▍ | 19/35 [00:05<00:02, 6.79it/s] Capturing CUDA graphs (PIECEWISE): 63%|██████▎ | 22/35 [00:05<00:01, 9.85it/s] Capturing CUDA graphs (PIECEWISE): 69%|██████▊ | 24/35 [00:05<00:01, 10.45it/s] Capturing CUDA graphs (PIECEWISE): 74%|███████▍ | 26/35 [00:06<00:00, 9.58it/s] Capturing CUDA graphs (PIECEWISE): 83%|████████▎ | 29/35 [00:06<00:00, 12.37it/s] Capturing CUDA graphs (PIECEWISE): 91%|█████████▏| 32/35 [00:06<00:00, 14.68it/s] Capturing CUDA graphs (PIECEWISE): 97%|█████████▋| 34/35 [00:06<00:00, 12.17it/s] Capturing CUDA graphs (PIECEWISE): 100%|██████████| 35/35 [00:06<00:00, 5.29it/s] | |
| Capturing CUDA graphs (FULL): 0%| | 0/19 [00:00<?, ?it/s] Capturing CUDA graphs (FULL): 16%|█▌ | 3/19 [00:00<00:00, 28.04it/s] Capturing CUDA graphs (FULL): 32%|███▏ | 6/19 [00:00<00:00, 28.70it/s] Capturing CUDA graphs (FULL): 47%|████▋ | 9/19 [00:00<00:00, 29.04it/s] Capturing CUDA graphs (FULL): 63%|██████▎ | 12/19 [00:00<00:00, 18.63it/s] Capturing CUDA graphs (FULL): 79%|███████▉ | 15/19 [00:00<00:00, 19.84it/s] Capturing CUDA graphs (FULL): 95%|█████████▍| 18/19 [00:00<00:00, 21.96it/s] Capturing CUDA graphs (FULL): 100%|██████████| 19/19 [00:00<00:00, 22.65it/s] | |
| (EngineCore pid=57770) INFO 10-31 09:48:21 [model_runner.py:1127] Graph capturing finished in 9 secs, took 0.16 GiB | |
| (EngineCore pid=57770) INFO 10-31 09:48:21 [gpu_worker.py:971] CUDA graph pool memory: 0.16 GiB (actual), 0.32 GiB (estimated), difference: 0.16 GiB (103.8%). | |
| (EngineCore pid=57770) INFO 10-31 09:48:21 [gpu_worker.py:1034] Free memory on device (78.84/79.25 GiB) on startup. Desired GPU memory utilization is (0.9, 71.33 GiB). Actual usage is 10.39 GiB for consumed memory (weights + non-torch), 1.55 GiB for peak activation, and 0.16 GiB for CUDAGraph memory. Replace gpu_memory_utilization config with `--kv-cache-memory=63430039962` (59.07 GiB) to fit into requested memory, or `--kv-cache-memory=71496105984` (66.59 GiB) to fully utilize gpu memory. Current kv cache memory in use is 59.38 GiB. | |
| (EngineCore pid=57770) INFO 10-31 09:48:22 [jit_monitor.py:84] Kernel JIT monitor activated; monitored JIT compilations during inference will use mode=warn. | |
| (EngineCore pid=57770) INFO 10-31 09:48:23 [core.py:395] init engine (profile, create kv cache, warmup model) took 160.87 s (compilation: 25.85 s) | |
| (EngineCore pid=57770) INFO 10-31 09:48:23 [kernel.py:467] Final IR op priority after setting platform defaults: IrOpPriorityConfig(rms_norm=['native'], fused_add_rms_norm=['native'], gelu_and_mul_sparse=['triton', 'native']) | |
| (MainProcess pid=57495) WARNING 10-31 09:48:24 [input_processor.py:408] Passing raw prompts to InputProcessor is deprecated and will be removed in the future. You should instead pass the outputs of Renderer.render_cmpl() or Renderer.render_chat(). | |
| (EngineCore pid=57770) WARNING 10-31 09:48:24 [jit_monitor.py:140] Triton kernel JIT compilation during inference: _topk_log_softmax_kernel. This causes a latency spike; consider extending warmup to cover this shape/config. | |
| readout gold_only: 5957 items in 114s | |
| (MainProcess pid=57495) INFO 10-31 09:50:16 [core_client.py:770] [shutdown] MPClient: start timeout=default | |
| (MainProcess pid=57495) INFO 10-31 09:50:16 [core_client.py:772] [shutdown] MPClient: stopping engine manager | |
| (MainProcess pid=57495) INFO 10-31 09:50:16 [utils.py:628] [shutdown] Process manager: send sigterm to process EngineCore | |
| (EngineCore pid=57770) INFO 10-31 09:50:16 [core.py:1454] [shutdown] EngineCore: trigger received signal=SIGTERM | |
| (EngineCore pid=57770) INFO 10-31 09:50:16 [core.py:1635] [shutdown] EngineCore: start mode=abort timeout=0s | |
| (EngineCore pid=57770) INFO 10-31 09:50:16 [core.py:1666] [shutdown] EngineCore: request processing complete; starting resource teardown | |
| (EngineCore pid=57770) INFO 10-31 09:50:16 [core.py:1499] [shutdown] EngineCore: exiting busy loop | |
| (MainProcess pid=57495) INFO 10-31 09:50:19 [core_client.py:774] [shutdown] MPClient: engine manager stopped | |
| (MainProcess pid=57495) INFO 10-31 09:50:19 [core_client.py:775] [shutdown] MPClient: cleaning up background resources | |
| (MainProcess pid=57495) INFO 10-31 09:50:19 [core_client.py:777] [shutdown] MPClient: complete | |
Xet Storage Details
- Size:
- 18.9 kB
- Xet hash:
- 346c8d4bbcd305b6acf450c7b8721f30674c7f9208ac7729268950ece1e2ac1c
·
Xet efficiently stores files, intelligently splitting them into unique chunks and accelerating uploads and downloads. More info.