mudassiruddin21 commited on
Commit
a820f15
·
verified ·
1 Parent(s): deb2732

Run r3: logs and results

Browse files
runs/r3/engine_logs/20261119-045726-bench-batch4.log ADDED
@@ -0,0 +1,373 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ vllm serve /workspace/emc/models/target --served-model-name target --host 127.0.0.1 --port 21331 --dtype bfloat16 --max-model-len 16384 --gpu-memory-utilization 0.9 --max-num-seqs 4 --seed 0 --no-enable-prefix-caching --generation-config vllm --language-model-only --speculative-config {"method": "mtp", "model": "/workspace/emc/runs/main3/sub_fp8", "num_speculative_tokens": 3}
2
+
3
+ (APIServer pid=46348) INFO 11-19 04:57:46 [api_utils.py:347]
4
+ (APIServer pid=46348) INFO 11-19 04:57:46 [api_utils.py:347] █ █ █▄ ▄█
5
+ (APIServer pid=46348) INFO 11-19 04:57:46 [api_utils.py:347] ▄▄ ▄█ █ █ █ ▀▄▀ █ version 0.31.0
6
+ (APIServer pid=46348) INFO 11-19 04:57:46 [api_utils.py:347] █▄█▀ █ █ █ █ model /workspace/emc/models/target
7
+ (APIServer pid=46348) INFO 11-19 04:57:46 [api_utils.py:347] ▀▀ ▀▀▀▀▀ ▀▀▀▀▀ ▀ ▀
8
+ (APIServer pid=46348) INFO 11-19 04:57:46 [api_utils.py:347]
9
+ (APIServer pid=46348) INFO 11-19 04:57:46 [api_utils.py:286] non-default args: {'model_tag': '/workspace/emc/models/target', 'host': '127.0.0.1', 'port': 21331, 'model': '/workspace/emc/models/target', 'dtype': 'bfloat16', 'max_model_len': 16384, 'served_model_name': ['target'], 'generation_config': 'vllm', 'gpu_memory_utilization': 0.9, 'enable_prefix_caching': False, 'language_model_only': True, 'max_num_seqs': 4, 'speculative_config': {'method': 'mtp', 'model': '/workspace/emc/runs/main3/sub_fp8', 'num_speculative_tokens': 3}}
10
+ (APIServer pid=46348) INFO 11-19 04:57:46 [model.py:699] Resolved architecture: Qwen3_5ForConditionalGeneration
11
+ (APIServer pid=46348) INFO 11-19 04:57:46 [model.py:2109] Using max model len 16384
12
+ (APIServer pid=46348) INFO 11-19 04:57:48 [model.py:917] All limits of multimodal modalities supported by the model are set to 0, running in text-only mode.
13
+ (APIServer pid=46348) INFO 11-19 04:57:48 [model.py:974] Disabled mm_prefix attention mode because multimodal inputs are configuration-disabled. Attention backends without mm_prefix support may now be selected.
14
+ (APIServer pid=46348) INFO 11-19 04:57:48 [model.py:699] Resolved architecture: Qwen3_5MTP
15
+ (APIServer pid=46348) INFO 11-19 04:57:48 [model.py:2109] Using max model len 262144
16
+ (APIServer pid=46348) WARNING 11-19 04:57:48 [speculative.py:1375] Enabling num_speculative_tokens > 1 will run multiple times of forward on same MTP layer,which may result in lower acceptance rate
17
+ (APIServer pid=46348) INFO 11-19 04:57:48 [speculative.py:1663] Overriding draft model max model len from 262144 to 16384
18
+ (APIServer pid=46348) INFO 11-19 04:57:48 [scheduler.py:308] Chunked prefill is enabled with max_num_batched_tokens=8192.
19
+ (APIServer pid=46348) INFO 11-19 04:57:48 [kernel.py:408] Final IR op priority after setting platform defaults: IrOpPriorityConfig(rms_norm=['native'], fused_add_rms_norm=['native'], gelu_and_mul_sparse=['triton', 'native'])
20
+ (EngineCore pid=46447) INFO 11-19 04:58:09 [core.py:129] Initializing a V1 LLM engine (v0.31.0) with config: model='/workspace/emc/models/target', speculative_config=SpeculativeConfig(method='mtp', model='/workspace/emc/runs/main3/sub_fp8', num_spec_tokens=3), tokenizer='/workspace/emc/models/target', skip_tokenizer_init=False, tokenizer_mode=auto, revision=None, tokenizer_revision=None, trust_remote_code=False, dtype=torch.bfloat16, max_seq_len=16384, 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, 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=0, served_model_name=target, enable_prefix_caching=False, 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::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, 'compile_sizes': [], 'compile_ranges_endpoints': [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], '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, 'enable_sp': False, 'fuse_gemm_comms': 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': 32, '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_cutedsl_warmup=True, enable_jit_warmup=True, moe_backend='auto', sparse_indexer_topk_backend='auto', linear_backend='auto', linear_backend_per_quant=None)
21
+ (EngineCore pid=46447) INFO 11-19 04:58:11 [parallel_state.py:1825] world_size=1 rank=0 local_rank=0 distributed_init_method=file:///workspace/emc/tmp/vllm_dist_c3499fe3b8624bb0a60996753ab54fe8 backend=nccl
22
+ (EngineCore pid=46447) INFO 11-19 04:58:11 [parallel_state.py:2269] 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
23
+ (EngineCore pid=46447) INFO 11-19 04:58:11 [gpu_worker.py:479] Using V2 Model Runner
24
+ (EngineCore pid=46447) INFO 11-19 04:58:12 [model_runner.py:375] Loading model from scratch...
25
+ (EngineCore pid=46447) INFO 11-19 04:58:13 [cuda.py:585] Using backend AttentionBackendEnum.FLASH_ATTN for vit attention
26
+ (EngineCore pid=46447) INFO 11-19 04:58:13 [mm_encoder_attention.py:372] Using AttentionBackendEnum.FLASH_ATTN for MMEncoderAttention.
27
+ (EngineCore pid=46447) INFO 11-19 04:58:13 [qwen_gdn_linear_attn.py:177] Using FlashInfer GDN prefill kernel (requested=auto, head_k_dim=128).
28
+ (EngineCore pid=46447) INFO 11-19 04:58:13 [qwen_gdn_linear_attn.py:533] GDN decode kernel: cuda
29
+ (EngineCore pid=46447) INFO 11-19 04:58:16 [cuda.py:526] Using FLASH_ATTN attention backend out of potential backends: ['FLASH_ATTN', 'FLASHINFER', 'TRITON_ATTN', 'FLEX_ATTENTION'].
30
+ (EngineCore pid=46447) INFO 11-19 04:58:16 [flash_attn.py:1167] Using FlashAttention version 2
31
+ (EngineCore pid=46447) INFO 11-19 04:58:16 [weight_utils.py:1063] Filesystem type for checkpoints: OVERLAY. Checkpoint size: 51.75 GiB. Available RAM: 156.57 GiB.
32
+ (EngineCore pid=46447) INFO 11-19 04:58:16 [weight_utils.py:1086] Auto-prefetch is disabled because the filesystem (OVERLAY) is not a recognized network FS (NFS/Lustre). If you want to force prefetching, start vLLM with --safetensors-load-strategy=prefetch.
33
+
34
+ Loading safetensors checkpoint shards: 0% Completed | 0/18 [00:00<?, ?it/s]
35
+
36
+ Loading safetensors checkpoint shards: 6% Completed | 1/18 [00:00<00:09, 1.85it/s]
37
+
38
+ Loading safetensors checkpoint shards: 11% Completed | 2/18 [00:01<00:08, 1.90it/s]
39
+
40
+ Loading safetensors checkpoint shards: 17% Completed | 3/18 [00:01<00:07, 2.08it/s]
41
+
42
+ Loading safetensors checkpoint shards: 22% Completed | 4/18 [00:02<00:07, 1.83it/s]
43
+
44
+ Loading safetensors checkpoint shards: 28% Completed | 5/18 [00:02<00:06, 2.12it/s]
45
+
46
+ Loading safetensors checkpoint shards: 33% Completed | 6/18 [00:03<00:06, 1.83it/s]
47
+
48
+ Loading safetensors checkpoint shards: 39% Completed | 7/18 [00:03<00:05, 2.08it/s]
49
+
50
+ Loading safetensors checkpoint shards: 44% Completed | 8/18 [00:04<00:05, 1.89it/s]
51
+
52
+ Loading safetensors checkpoint shards: 50% Completed | 9/18 [00:04<00:04, 2.14it/s]
53
+
54
+ Loading safetensors checkpoint shards: 56% Completed | 10/18 [00:05<00:04, 1.92it/s]
55
+
56
+ Loading safetensors checkpoint shards: 61% Completed | 11/18 [00:05<00:03, 2.15it/s]
57
+
58
+ Loading safetensors checkpoint shards: 67% Completed | 12/18 [00:06<00:03, 1.92it/s]
59
+
60
+ Loading safetensors checkpoint shards: 72% Completed | 13/18 [00:06<00:02, 2.15it/s]
61
+
62
+ Loading safetensors checkpoint shards: 78% Completed | 14/18 [00:07<00:02, 1.89it/s]
63
+
64
+ Loading safetensors checkpoint shards: 83% Completed | 15/18 [00:07<00:01, 2.11it/s]
65
+
66
+ Loading safetensors checkpoint shards: 89% Completed | 16/18 [00:08<00:01, 1.91it/s]
67
+
68
+ Loading safetensors checkpoint shards: 94% Completed | 17/18 [00:08<00:00, 2.14it/s]
69
+
70
+ Loading safetensors checkpoint shards: 100% Completed | 18/18 [00:08<00:00, 2.22it/s]
71
+
72
+ Loading safetensors checkpoint shards: 100% Completed | 18/18 [00:08<00:00, 2.03it/s]
73
+
74
+ (EngineCore pid=46447) INFO 11-19 04:58:25 [default_loader.py:484] Loading weights took 8.88 seconds
75
+ (EngineCore pid=46447) INFO 11-19 04:58:25 [weight_utils.py:1063] Filesystem type for checkpoints: OVERLAY. Checkpoint size: 0.50 GiB. Available RAM: 157.32 GiB.
76
+
77
+ Loading safetensors checkpoint shards: 0% Completed | 0/1 [00:00<?, ?it/s]
78
+
79
+ Loading safetensors checkpoint shards: 100% Completed | 1/1 [00:00<00:00, 3.50it/s]
80
+
81
+ Loading safetensors checkpoint shards: 100% Completed | 1/1 [00:00<00:00, 3.50it/s]
82
+
83
+ (EngineCore pid=46447) INFO 11-19 04:58:26 [default_loader.py:484] Loading weights took 0.29 seconds
84
+ (EngineCore pid=46447) INFO 11-19 04:58:26 [model_runner.py:407] Model loading took 51.02 GiB memory and 14.375547 seconds
85
+ (EngineCore pid=46447) INFO 11-19 04:58:26 [topk_topp_sampler.py:96] FlashInfer top-p/top-k sampling disabled via VLLM_USE_FLASHINFER_SAMPLER=0.
86
+ (EngineCore pid=46447) WARNING 11-19 04:58:26 [torch_utils.py:274] OMP_NUM_THREADS=20 is set; leaving Torch threads at 20 for serving. Multi-threaded torch CPU ops during serving can degrade performance through spin-wait contention and cgroup CPU-quota throttling.
87
+ (EngineCore pid=46447) INFO 11-19 04:58:26 [interface.py:989] Setting attention block size to 800 tokens to ensure that attention page size is >= mamba page size.
88
+ (EngineCore pid=46447) INFO 11-19 04:58:26 [interface.py:1013] Padding mamba page size by 0.25% to ensure that mamba page size and attention page size are exactly equal.
89
+ (EngineCore pid=46447) INFO 11-19 04:58:26 [utils.py:343] Using LBNHC KV cache layout.
90
+ (EngineCore pid=46447) INFO 11-19 04:58:28 [caching.py:342] reconstructed serializable fn from standalone compile artifacts. num_artifacts=20 num_submods=65
91
+ (EngineCore pid=46447) INFO 11-19 04:58:28 [decorators.py:312] Directly load AOT compilation from path /workspace/emc/work/cache/vllm/torch_compile_cache/torch_aot_compile/2bf75a0f005d511d44b2e0ebedfdffd0cbe6a7f8d0aa3da736fdfd02e918092c/rank_0_0/model
92
+ (EngineCore pid=46447) INFO 11-19 04:58:28 [monitor.py:53] torch.compile took 0.95 s in total
93
+ (EngineCore pid=46447) INFO 11-19 04:58:42 [monitor.py:81] Initial profiling/warmup run took 13.99 s
94
+ (EngineCore pid=46447) INFO 11-19 04:58:42 [caching.py:342] reconstructed serializable fn from standalone compile artifacts. num_artifacts=2 num_submods=2
95
+ (EngineCore pid=46447) INFO 11-19 04:58:42 [decorators.py:312] Directly load AOT compilation from path /workspace/emc/work/cache/vllm/torch_compile_cache/torch_aot_compile/18479e3edf45fa305490e52c44dc54aa80cc7798504d276363b2fcb0c5cc2807/rank_0_0/model
96
+ (EngineCore pid=46447) INFO 11-19 04:58:42 [monitor.py:53] torch.compile took 0.07 s in total
97
+ (EngineCore pid=46447) INFO 11-19 04:58:42 [monitor.py:81] Initial profiling/warmup run took 0.10 s
98
+ (EngineCore pid=46447) WARNING 11-19 04:58:43 [kv_cache_utils.py:1574] Add 3 padding layers, may waste at most 6.25% KV cache memory
99
+
100
+ Capturing CUDA graphs (PIECEWISE): 0%| | 0/7 [00:00<?, ?it/s]
101
+ Capturing CUDA graphs (PIECEWISE): 14%|█▍ | 1/7 [00:06<00:39, 6.61s/it]
102
+ Capturing CUDA graphs (PIECEWISE): 29%|██▊ | 2/7 [00:06<00:14, 2.82s/it]
103
+ Capturing CUDA graphs (PIECEWISE): 43%|████▎ | 3/7 [00:06<00:06, 1.60s/it]
104
+ Capturing CUDA graphs (PIECEWISE): 57%|█████▋ | 4/7 [00:07<00:03, 1.03s/it]
105
+ Capturing CUDA graphs (PIECEWISE): 71%|███████▏ | 5/7 [00:07<00:01, 1.38it/s]
106
+ Capturing CUDA graphs (PIECEWISE): 100%|██████████| 7/7 [00:07<00:00, 2.43it/s]
107
+ Capturing CUDA graphs (PIECEWISE): 100%|██████████| 7/7 [00:07<00:00, 1.07s/it]
108
+
109
+ Capturing CUDA graphs (FULL): 0%| | 0/2 [00:00<?, ?it/s]
110
+ Capturing CUDA graphs (FULL): 100%|██████████| 2/2 [00:00<00:00, 13.12it/s]
111
+ Capturing CUDA graphs (FULL): 100%|██████████| 2/2 [00:00<00:00, 13.09it/s]
112
+ (EngineCore pid=46447) INFO 11-19 04:58:51 [speculator.py:144] Capturing model for speculator...
113
+
114
+ Capturing prefill CUDA graphs (PIECEWISE): 0%| | 0/7 [00:00<?, ?it/s]
115
+ Capturing prefill CUDA graphs (PIECEWISE): 100%|██████████| 7/7 [00:00<00:00, 137.64it/s]
116
+
117
+ Capturing prefill CUDA graphs (FULL): 0%| | 0/3 [00:00<?, ?it/s]
118
+ Capturing prefill CUDA graphs (FULL): 100%|██████████| 3/3 [00:00<00:00, 159.21it/s]
119
+
120
+ Capturing decode CUDA graphs (FULL): 0%| | 0/3 [00:00<?, ?it/s]
121
+ Capturing decode CUDA graphs (FULL): 100%|██████████| 3/3 [00:00<00:00, 91.78it/s]
122
+ (EngineCore pid=46447) INFO 11-19 04:58:52 [model_runner.py:1087] Graph capturing finished in 8 secs, took 0.23 GiB
123
+ (EngineCore pid=46447) INFO 11-19 04:58:52 [gpu_worker.py:692] Available KV cache memory: 32.65 GiB
124
+ (EngineCore pid=46447) INFO 11-19 04:58:52 [gpu_worker.py:707] CUDA graph memory profiling is enabled (default since v0.21.0). The current --gpu-memory-utilization=0.9000 is equivalent to --gpu-memory-utilization=0.8975 without CUDA graph memory profiling. To maintain the same effective KV cache size as before, increase --gpu-memory-utilization to 0.9025. To disable, set VLLM_MEMORY_PROFILER_ESTIMATE_CUDAGRAPHS=0.
125
+ (EngineCore pid=46447) WARNING 11-19 04:58:52 [kv_cache_utils.py:1574] Add 3 padding layers, may waste at most 6.25% KV cache memory
126
+ (EngineCore pid=46447) INFO 11-19 04:58:52 [kv_cache_utils.py:2464] GPU KV cache size: 312,288 tokens, Maximum concurrency for 16,384 tokens per request: 19.06x
127
+ (EngineCore pid=46447) INFO 11-19 04:58:52 [kernel_warmup.py:172] JIT kernel warmup starting.
128
+ (EngineCore pid=46447) INFO 11-19 04:58:52 [kernel_warmup.py:185] JIT kernel warmup finished in 0.00s.
129
+ (EngineCore pid=46447) INFO 11-19 04:58:52 [qwen_triton_warmup.py:294] Warming up Qwen GDN Triton kernels for model_type=qwen3_5_text.
130
+ (EngineCore pid=46447) INFO 11-19 04:58:52 [qwen_vl_triton_warmup.py:98] Warmed M-RoPE Triton kernels.
131
+ (EngineCore pid=46447) INFO 11-19 04:58:52 [mamba_triton_warmup.py:42] Warmed Mamba batch_memcpy_kernel.
132
+ (EngineCore pid=46447) INFO 11-19 04:58:53 [kernel_warmup.py:437] Using FlashInfer autotune cache file: /workspace/emc/work/cache/vllm/flashinfer_autotune_cache/0.7.0.post1/120f/fd30e073b649e1eb00be702764c5abb39723166a735906b2f854d86e1534bb12/autotune_configs.json
133
+ 2026-11-19 04:58:54,080 - INFO - autotuner.py:3691 - flashinfer.jit: [Autotuner]: Loaded 0 configs from /workspace/emc/work/cache/vllm/flashinfer_autotune_cache/0.7.0.post1/120f/fd30e073b649e1eb00be702764c5abb39723166a735906b2f854d86e1534bb12/autotune_configs.json
134
+ 2026-11-19 04:58:54,080 - INFO - autotuner.py:1046 - flashinfer.jit: [Autotuner]: Autotuning process starts ...
135
+ (EngineCore pid=46447) INFO 11-19 04:58:54 [kernel_warmup.py:359] Running FlashInfer autotune with 8192 tokens and token buckets (1, 2, 4, 8, 16, 32, 64, 128, 256, 512, 768, 1024, 1280, 1536, 1792, 2048, 2560, 3072, 3584, 4096, 8192).
136
+ 2026-11-19 04:58:54,109 - INFO - autotuner.py:1069 - flashinfer.jit: [Autotuner]: Autotuning process ends
137
+ 2026-11-19 04:58:54,110 - INFO - autotuner.py:3554 - flashinfer.jit: [Autotuner]: Saved 0 configs to /workspace/emc/work/cache/vllm/flashinfer_autotune_cache/0.7.0.post1/120f/fd30e073b649e1eb00be702764c5abb39723166a735906b2f854d86e1534bb12/autotune_configs.json (0 new, 0 from previous config)
138
+
139
+ Capturing CUDA graphs (PIECEWISE): 0%| | 0/7 [00:00<?, ?it/s]
140
+ Capturing CUDA graphs (PIECEWISE): 14%|█▍ | 1/7 [00:00<00:00, 6.50it/s]
141
+ Capturing CUDA graphs (PIECEWISE): 29%|██▊ | 2/7 [00:00<00:00, 6.38it/s]
142
+ Capturing CUDA graphs (PIECEWISE): 43%|████▎ | 3/7 [00:00<00:00, 6.41it/s]
143
+ Capturing CUDA graphs (PIECEWISE): 57%|█████▋ | 4/7 [00:00<00:00, 6.42it/s]
144
+ Capturing CUDA graphs (PIECEWISE): 71%|███████▏ | 5/7 [00:00<00:00, 7.30it/s]
145
+ Capturing CUDA graphs (PIECEWISE): 100%|██████████| 7/7 [00:00<00:00, 8.57it/s]
146
+ Capturing CUDA graphs (PIECEWISE): 100%|██████████| 7/7 [00:00<00:00, 7.60it/s]
147
+
148
+ Capturing CUDA graphs (FULL): 0%| | 0/3 [00:00<?, ?it/s]
149
+ Capturing CUDA graphs (FULL): 67%|██████▋ | 2/3 [00:00<00:00, 14.72it/s]
150
+ Capturing CUDA graphs (FULL): 100%|██████████| 3/3 [00:00<00:00, 14.49it/s]
151
+ (EngineCore pid=46447) INFO 11-19 04:58:57 [speculator.py:144] Capturing model for speculator...
152
+
153
+ Capturing prefill CUDA graphs (PIECEWISE): 0%| | 0/7 [00:00<?, ?it/s]
154
+ Capturing prefill CUDA graphs (PIECEWISE): 100%|██████████| 7/7 [00:00<00:00, 146.26it/s]
155
+
156
+ Capturing prefill CUDA graphs (FULL): 0%| | 0/3 [00:00<?, ?it/s]
157
+ Capturing prefill CUDA graphs (FULL): 100%|██████████| 3/3 [00:00<00:00, 170.80it/s]
158
+
159
+ Capturing decode CUDA graphs (FULL): 0%| | 0/3 [00:00<?, ?it/s]
160
+ Capturing decode CUDA graphs (FULL): 100%|██████████| 3/3 [00:00<00:00, 105.15it/s]
161
+ (EngineCore pid=46447) INFO 11-19 04:58:57 [model_runner.py:1087] Graph capturing finished in 2 secs, took 0.17 GiB
162
+ (EngineCore pid=46447) INFO 11-19 04:58:57 [gpu_worker.py:892] CUDA graph pool memory: 0.17 GiB (actual), 0.24 GiB (estimated), difference: 0.07 GiB (40.7%).
163
+ (EngineCore pid=46447) INFO 11-19 04:58:57 [gpu_worker.py:955] Free memory on device (94.43/94.97 GiB) on startup. Desired GPU memory utilization is (0.9, 85.48 GiB). Actual usage is 51.37 GiB for consumed memory (weights + non-torch), 1.46 GiB for peak activation, and 0.17 GiB for CUDAGraph memory. Replace gpu_memory_utilization config with `--kv-cache-memory=34723144500` (32.34 GiB) to fit into requested memory, or `--kv-cache-memory=44335427584` (41.29 GiB) to fully utilize gpu memory. Current kv cache memory in use is 32.65 GiB.
164
+ (EngineCore pid=46447) INFO 11-19 04:58:59 [jit_monitor.py:84] Kernel JIT monitor activated; monitored JIT compilations during inference will use mode=warn.
165
+ (EngineCore pid=46447) INFO 11-19 04:58:59 [core.py:373] init engine (profile, create kv cache, warmup model) took 33.04 s (compilation: 1.02 s)
166
+ (EngineCore pid=46447) INFO 11-19 04:58:59 [kv_cache_utils.py:772] kv cache group sizes [16384, 16384, 16384, 800]
167
+ (EngineCore pid=46447) INFO 11-19 04:58:59 [kv_cache_utils.py:773] kv lcm block sizes 409600
168
+ (EngineCore pid=46447) INFO 11-19 04:58:59 [kernel.py:408] Final IR op priority after setting platform defaults: IrOpPriorityConfig(rms_norm=['native'], fused_add_rms_norm=['native'], gelu_and_mul_sparse=['triton', 'native'])
169
+ (APIServer pid=46348) INFO 11-19 04:58:59 [entry.py:132] Supported tasks: ['generate']
170
+ (APIServer pid=46348) INFO 11-19 04:59:00 [factories.py:76] Scale-out endpoints are disabled. Set --enable-scale-out to enable them.
171
+ (APIServer pid=46348) INFO 11-19 04:59:00 [hf.py:646] Detected the chat template content format to be 'openai'. You can set `--chat-template-content-format` to override this.
172
+ (APIServer pid=46348) WARNING 11-19 04:59:00 [serving.py:221] The chat template requires system-first ordering, so inline system messages in /v1/messages requests (e.g. host Code's per-turn reminders) are merged into the leading system prompt. Each new one changes the prompt prefix, so the rest of the conversation misses the prefix cache. Pass a --chat-template that accepts non-leading system messages to avoid this.
173
+ (APIServer pid=46348) WARNING 11-19 04:59:00 [serving.py:221] The chat template requires system-first ordering, so inline system messages in /v1/messages requests (e.g. host Code's per-turn reminders) are merged into the leading system prompt. Each new one changes the prompt prefix, so the rest of the conversation misses the prefix cache. Pass a --chat-template that accepts non-leading system messages to avoid this.
174
+ (APIServer pid=46348) INFO 11-19 04:59:00 [entry.py:136] Starting vLLM server on http://127.0.0.1:21331
175
+ (APIServer pid=46348) INFO 11-19 04:59:00 [launcher.py:60] Available routes are:
176
+ (APIServer pid=46348) INFO 11-19 04:59:00 [launcher.py:69] Route: /openapi.json, Methods: GET, HEAD
177
+ (APIServer pid=46348) INFO 11-19 04:59:00 [launcher.py:69] Route: /docs, Methods: GET, HEAD
178
+ (APIServer pid=46348) INFO 11-19 04:59:00 [launcher.py:69] Route: /docs/oauth2-redirect, Methods: GET, HEAD
179
+ (APIServer pid=46348) INFO 11-19 04:59:00 [launcher.py:69] Route: /redoc, Methods: GET, HEAD
180
+ (APIServer pid=46348) INFO 11-19 04:59:00 [launcher.py:69] Route: /load, Methods: GET
181
+ (APIServer pid=46348) INFO 11-19 04:59:00 [launcher.py:69] Route: /version, Methods: GET
182
+ (APIServer pid=46348) INFO 11-19 04:59:00 [launcher.py:69] Route: /kv_event_sources, Methods: GET
183
+ (APIServer pid=46348) INFO 11-19 04:59:00 [launcher.py:69] Route: /health, Methods: GET
184
+ (APIServer pid=46348) INFO 11-19 04:59:00 [launcher.py:69] Route: /metrics, Methods: GET
185
+ (APIServer pid=46348) INFO 11-19 04:59:00 [launcher.py:69] Route: /tokenize, Methods: POST
186
+ (APIServer pid=46348) INFO 11-19 04:59:00 [launcher.py:69] Route: /detokenize, Methods: POST
187
+ (APIServer pid=46348) INFO 11-19 04:59:00 [launcher.py:69] Route: /v1/models, Methods: GET
188
+ (APIServer pid=46348) INFO 11-19 04:59:00 [launcher.py:69] Route: /ping, Methods: GET
189
+ (APIServer pid=46348) INFO 11-19 04:59:00 [launcher.py:69] Route: /ping, Methods: POST
190
+ (APIServer pid=46348) INFO 11-19 04:59:00 [launcher.py:69] Route: /invocations, Methods: POST
191
+ (APIServer pid=46348) INFO 11-19 04:59:00 [launcher.py:69] Route: /v1/chat/completions, Methods: POST
192
+ (APIServer pid=46348) INFO 11-19 04:59:00 [launcher.py:69] Route: /v1/chat/completions/batch, Methods: POST
193
+ (APIServer pid=46348) INFO 11-19 04:59:00 [launcher.py:69] Route: /v1/responses, Methods: POST
194
+ (APIServer pid=46348) INFO 11-19 04:59:00 [launcher.py:69] Route: /v1/responses/{response_id}, Methods: GET
195
+ (APIServer pid=46348) INFO 11-19 04:59:00 [launcher.py:69] Route: /v1/responses/{response_id}/cancel, Methods: POST
196
+ (APIServer pid=46348) INFO 11-19 04:59:00 [launcher.py:69] Route: /v1/completions, Methods: POST
197
+ (APIServer pid=46348) INFO 11-19 04:59:00 [launcher.py:69] Route: /v1/messages, Methods: POST
198
+ (APIServer pid=46348) INFO 11-19 04:59:00 [launcher.py:69] Route: /v1/messages/count_tokens, Methods: POST
199
+ (APIServer pid=46348) INFO 11-19 04:59:00 [launcher.py:69] Route: /generative_scoring, Methods: POST
200
+ (APIServer pid=46348) INFO 11-19 04:59:00 [launcher.py:69] Route: /scale_elastic_ep, Methods: POST
201
+ (APIServer pid=46348) INFO 11-19 04:59:00 [launcher.py:69] Route: /is_scaling_elastic_ep, Methods: POST
202
+ INFO: Started server process [46348]
203
+ INFO: Waiting for application startup.
204
+ INFO: Application startup complete.
205
+ INFO: 127.0.0.1:29401 - "GET /health HTTP/1.1" 200 OK
206
+ INFO: 127.0.0.1:61159 - "POST /v1/completions HTTP/1.1" 200 OK
207
+ (EngineCore pid=46447) WARNING 11-19 04:59:02 [jit_monitor.py:140] Triton kernel JIT compilation during inference: _compute_local_logits_stats_kernel. This causes a latency spike; consider extending warmup to cover this shape/config.
208
+ (EngineCore pid=46447) WARNING 11-19 04:59:02 [jit_monitor.py:140] Triton kernel JIT compilation during inference: _rejection_kernel. This causes a latency spike; consider extending warmup to cover this shape/config.
209
+ (EngineCore pid=46447) WARNING 11-19 04:59:02 [jit_monitor.py:140] Triton kernel JIT compilation during inference: _resample_kernel. This causes a latency spike; consider extending warmup to cover this shape/config.
210
+ INFO: 127.0.0.1:39194 - "POST /v1/completions HTTP/1.1" 200 OK
211
+ INFO: 127.0.0.1:29260 - "POST /v1/completions HTTP/1.1" 200 OK
212
+ (APIServer pid=46348) INFO 11-19 04:59:11 [loggers.py:320] Engine 000: Avg prompt throughput: 28.6 tokens/s, Avg generation throughput: 43.3 tokens/s, Running: 1 reqs, Waiting: 0 reqs, GPU KV cache usage: 2.1%
213
+ (APIServer pid=46348) INFO 11-19 04:59:11 [metrics.py:120] SpecDecoding metrics: Mean acceptance length: 3.24, Accepted throughput: 29.94 tokens/s, Drafted throughput: 40.10 tokens/s, Accepted: 345 tokens, Drafted: 462 tokens, Per-position acceptance rate: 0.890, 0.740, 0.610, Avg Draft acceptance rate: 74.7%
214
+ INFO: 127.0.0.1:58391 - "POST /v1/completions HTTP/1.1" 200 OK
215
+ INFO: 127.0.0.1:62060 - "POST /v1/completions HTTP/1.1" 200 OK
216
+ INFO: 127.0.0.1:41210 - "POST /v1/completions HTTP/1.1" 200 OK
217
+ (APIServer pid=46348) INFO 11-19 04:59:21 [loggers.py:320] Engine 000: Avg prompt throughput: 38.4 tokens/s, Avg generation throughput: 67.9 tokens/s, Running: 1 reqs, Waiting: 0 reqs, GPU KV cache usage: 2.1%
218
+ (APIServer pid=46348) INFO 11-19 04:59:21 [metrics.py:120] SpecDecoding metrics: Mean acceptance length: 3.83, Accepted throughput: 50.29 tokens/s, Drafted throughput: 53.39 tokens/s, Accepted: 503 tokens, Drafted: 534 tokens, Per-position acceptance rate: 0.989, 0.949, 0.888, Avg Draft acceptance rate: 94.2%
219
+ INFO: 127.0.0.1:55682 - "POST /v1/completions HTTP/1.1" 200 OK
220
+ (APIServer pid=46348) INFO 11-19 04:59:31 [loggers.py:320] Engine 000: Avg prompt throughput: 3.9 tokens/s, Avg generation throughput: 50.6 tokens/s, Running: 1 reqs, Waiting: 0 reqs, GPU KV cache usage: 2.1%
221
+ (APIServer pid=46348) INFO 11-19 04:59:31 [metrics.py:120] SpecDecoding metrics: Mean acceptance length: 2.77, Accepted throughput: 32.30 tokens/s, Drafted throughput: 54.60 tokens/s, Accepted: 323 tokens, Drafted: 546 tokens, Per-position acceptance rate: 0.780, 0.588, 0.407, Avg Draft acceptance rate: 59.2%
222
+ (APIServer pid=46348) INFO 11-19 04:59:41 [loggers.py:320] Engine 000: Avg prompt throughput: 0.0 tokens/s, Avg generation throughput: 48.6 tokens/s, Running: 1 reqs, Waiting: 0 reqs, GPU KV cache usage: 2.2%
223
+ (APIServer pid=46348) INFO 11-19 04:59:41 [metrics.py:120] SpecDecoding metrics: Mean acceptance length: 2.64, Accepted throughput: 30.20 tokens/s, Drafted throughput: 55.19 tokens/s, Accepted: 302 tokens, Drafted: 552 tokens, Per-position acceptance rate: 0.804, 0.533, 0.304, Avg Draft acceptance rate: 54.7%
224
+ INFO: 127.0.0.1:56820 - "POST /v1/completions HTTP/1.1" 200 OK
225
+ (APIServer pid=46348) INFO 11-19 04:59:51 [loggers.py:320] Engine 000: Avg prompt throughput: 7.8 tokens/s, Avg generation throughput: 45.7 tokens/s, Running: 1 reqs, Waiting: 0 reqs, GPU KV cache usage: 2.1%
226
+ (APIServer pid=46348) INFO 11-19 04:59:51 [metrics.py:120] SpecDecoding metrics: Mean acceptance length: 2.50, Accepted throughput: 27.40 tokens/s, Drafted throughput: 54.89 tokens/s, Accepted: 274 tokens, Drafted: 549 tokens, Per-position acceptance rate: 0.760, 0.437, 0.301, Avg Draft acceptance rate: 49.9%
227
+ (APIServer pid=46348) INFO 11-19 05:00:01 [loggers.py:320] Engine 000: Avg prompt throughput: 0.0 tokens/s, Avg generation throughput: 43.2 tokens/s, Running: 1 reqs, Waiting: 0 reqs, GPU KV cache usage: 2.2%
228
+ (APIServer pid=46348) INFO 11-19 05:00:01 [metrics.py:120] SpecDecoding metrics: Mean acceptance length: 2.35, Accepted throughput: 24.80 tokens/s, Drafted throughput: 55.19 tokens/s, Accepted: 248 tokens, Drafted: 552 tokens, Per-position acceptance rate: 0.712, 0.391, 0.245, Avg Draft acceptance rate: 44.9%
229
+ INFO: 127.0.0.1:39020 - "GET /metrics HTTP/1.1" 200 OK
230
+ INFO: 127.0.0.1:24078 - "POST /v1/completions HTTP/1.1" 200 OK
231
+ (APIServer pid=46348) INFO 11-19 05:00:11 [loggers.py:320] Engine 000: Avg prompt throughput: 7.2 tokens/s, Avg generation throughput: 45.5 tokens/s, Running: 1 reqs, Waiting: 0 reqs, GPU KV cache usage: 2.1%
232
+ (APIServer pid=46348) INFO 11-19 05:00:11 [metrics.py:120] SpecDecoding metrics: Mean acceptance length: 2.50, Accepted throughput: 27.30 tokens/s, Drafted throughput: 54.59 tokens/s, Accepted: 273 tokens, Drafted: 546 tokens, Per-position acceptance rate: 0.736, 0.473, 0.291, Avg Draft acceptance rate: 50.0%
233
+ INFO: 127.0.0.1:20147 - "POST /v1/completions HTTP/1.1" 200 OK
234
+ (APIServer pid=46348) INFO 11-19 05:00:21 [loggers.py:320] Engine 000: Avg prompt throughput: 3.9 tokens/s, Avg generation throughput: 56.1 tokens/s, Running: 1 reqs, Waiting: 0 reqs, GPU KV cache usage: 2.1%
235
+ (APIServer pid=46348) INFO 11-19 05:00:21 [metrics.py:120] SpecDecoding metrics: Mean acceptance length: 3.07, Accepted throughput: 37.80 tokens/s, Drafted throughput: 54.89 tokens/s, Accepted: 378 tokens, Drafted: 549 tokens, Per-position acceptance rate: 0.858, 0.694, 0.514, Avg Draft acceptance rate: 68.9%
236
+ INFO: 127.0.0.1:50437 - "POST /v1/completions HTTP/1.1" 200 OK
237
+ (APIServer pid=46348) INFO 11-19 05:00:31 [loggers.py:320] Engine 000: Avg prompt throughput: 6.9 tokens/s, Avg generation throughput: 64.4 tokens/s, Running: 1 reqs, Waiting: 0 reqs, GPU KV cache usage: 2.1%
238
+ (APIServer pid=46348) INFO 11-19 05:00:31 [metrics.py:120] SpecDecoding metrics: Mean acceptance length: 3.51, Accepted throughput: 46.19 tokens/s, Drafted throughput: 55.19 tokens/s, Accepted: 462 tokens, Drafted: 552 tokens, Per-position acceptance rate: 0.940, 0.837, 0.734, Avg Draft acceptance rate: 83.7%
239
+ INFO: 127.0.0.1:48413 - "POST /v1/completions HTTP/1.1" 200 OK
240
+ (APIServer pid=46348) INFO 11-19 05:00:41 [loggers.py:320] Engine 000: Avg prompt throughput: 5.0 tokens/s, Avg generation throughput: 63.4 tokens/s, Running: 1 reqs, Waiting: 0 reqs, GPU KV cache usage: 2.1%
241
+ (APIServer pid=46348) INFO 11-19 05:00:41 [metrics.py:120] SpecDecoding metrics: Mean acceptance length: 3.45, Accepted throughput: 45.09 tokens/s, Drafted throughput: 55.19 tokens/s, Accepted: 451 tokens, Drafted: 552 tokens, Per-position acceptance rate: 0.935, 0.804, 0.712, Avg Draft acceptance rate: 81.7%
242
+ INFO: 127.0.0.1:41415 - "POST /v1/completions HTTP/1.1" 200 OK
243
+ (APIServer pid=46348) INFO 11-19 05:00:51 [loggers.py:320] Engine 000: Avg prompt throughput: 10.3 tokens/s, Avg generation throughput: 51.2 tokens/s, Running: 1 reqs, Waiting: 0 reqs, GPU KV cache usage: 2.1%
244
+ (APIServer pid=46348) INFO 11-19 05:00:51 [metrics.py:120] SpecDecoding metrics: Mean acceptance length: 2.80, Accepted throughput: 32.90 tokens/s, Drafted throughput: 54.90 tokens/s, Accepted: 329 tokens, Drafted: 549 tokens, Per-position acceptance rate: 0.792, 0.590, 0.415, Avg Draft acceptance rate: 59.9%
245
+ INFO: 127.0.0.1:47746 - "POST /v1/completions HTTP/1.1" 200 OK
246
+ INFO: 127.0.0.1:35794 - "POST /v1/completions HTTP/1.1" 200 OK
247
+ (APIServer pid=46348) INFO 11-19 05:01:01 [loggers.py:320] Engine 000: Avg prompt throughput: 9.0 tokens/s, Avg generation throughput: 57.0 tokens/s, Running: 1 reqs, Waiting: 0 reqs, GPU KV cache usage: 2.1%
248
+ (APIServer pid=46348) INFO 11-19 05:01:01 [metrics.py:120] SpecDecoding metrics: Mean acceptance length: 3.15, Accepted throughput: 39.00 tokens/s, Drafted throughput: 54.29 tokens/s, Accepted: 390 tokens, Drafted: 543 tokens, Per-position acceptance rate: 0.884, 0.702, 0.569, Avg Draft acceptance rate: 71.8%
249
+ INFO: 127.0.0.1:20320 - "POST /v1/completions HTTP/1.1" 200 OK
250
+ (APIServer pid=46348) INFO 11-19 05:01:11 [loggers.py:320] Engine 000: Avg prompt throughput: 22.7 tokens/s, Avg generation throughput: 51.2 tokens/s, Running: 1 reqs, Waiting: 0 reqs, GPU KV cache usage: 2.1%
251
+ (APIServer pid=46348) INFO 11-19 05:01:11 [metrics.py:120] SpecDecoding metrics: Mean acceptance length: 2.81, Accepted throughput: 32.99 tokens/s, Drafted throughput: 54.59 tokens/s, Accepted: 330 tokens, Drafted: 546 tokens, Per-position acceptance rate: 0.852, 0.582, 0.379, Avg Draft acceptance rate: 60.4%
252
+ INFO: 127.0.0.1:50125 - "POST /v1/completions HTTP/1.1" 200 OK
253
+ INFO: 127.0.0.1:52697 - "POST /v1/completions HTTP/1.1" 200 OK
254
+ (APIServer pid=46348) INFO 11-19 05:01:21 [loggers.py:320] Engine 000: Avg prompt throughput: 32.5 tokens/s, Avg generation throughput: 68.5 tokens/s, Running: 1 reqs, Waiting: 0 reqs, GPU KV cache usage: 2.1%
255
+ (APIServer pid=46348) INFO 11-19 05:01:21 [metrics.py:120] SpecDecoding metrics: Mean acceptance length: 3.79, Accepted throughput: 50.49 tokens/s, Drafted throughput: 54.29 tokens/s, Accepted: 505 tokens, Drafted: 543 tokens, Per-position acceptance rate: 0.983, 0.928, 0.878, Avg Draft acceptance rate: 93.0%
256
+ INFO: 127.0.0.1:37869 - "POST /v1/completions HTTP/1.1" 200 OK
257
+ INFO: 127.0.0.1:44139 - "POST /v1/completions HTTP/1.1" 200 OK
258
+ INFO: 127.0.0.1:17087 - "POST /v1/completions HTTP/1.1" 200 OK
259
+ (APIServer pid=46348) INFO 11-19 05:01:31 [loggers.py:320] Engine 000: Avg prompt throughput: 69.3 tokens/s, Avg generation throughput: 69.9 tokens/s, Running: 1 reqs, Waiting: 0 reqs, GPU KV cache usage: 2.1%
260
+ (APIServer pid=46348) INFO 11-19 05:01:31 [metrics.py:120] SpecDecoding metrics: Mean acceptance length: 3.94, Accepted throughput: 52.29 tokens/s, Drafted throughput: 53.39 tokens/s, Accepted: 523 tokens, Drafted: 534 tokens, Per-position acceptance rate: 1.000, 0.983, 0.955, Avg Draft acceptance rate: 97.9%
261
+ INFO: 127.0.0.1:37636 - "POST /v1/completions HTTP/1.1" 200 OK
262
+ INFO: 127.0.0.1:19162 - "POST /v1/completions HTTP/1.1" 200 OK
263
+ (APIServer pid=46348) INFO 11-19 05:01:41 [loggers.py:320] Engine 000: Avg prompt throughput: 33.3 tokens/s, Avg generation throughput: 69.5 tokens/s, Running: 1 reqs, Waiting: 0 reqs, GPU KV cache usage: 2.1%
264
+ (APIServer pid=46348) INFO 11-19 05:01:41 [metrics.py:120] SpecDecoding metrics: Mean acceptance length: 3.87, Accepted throughput: 51.69 tokens/s, Drafted throughput: 53.99 tokens/s, Accepted: 517 tokens, Drafted: 540 tokens, Per-position acceptance rate: 0.994, 0.950, 0.928, Avg Draft acceptance rate: 95.7%
265
+ INFO: 127.0.0.1:28343 - "POST /v1/completions HTTP/1.1" 200 OK
266
+ INFO: 127.0.0.1:46297 - "POST /v1/completions HTTP/1.1" 200 OK
267
+ (APIServer pid=46348) INFO 11-19 05:01:51 [loggers.py:320] Engine 000: Avg prompt throughput: 23.0 tokens/s, Avg generation throughput: 67.0 tokens/s, Running: 1 reqs, Waiting: 0 reqs, GPU KV cache usage: 2.1%
268
+ (APIServer pid=46348) INFO 11-19 05:01:51 [metrics.py:120] SpecDecoding metrics: Mean acceptance length: 3.70, Accepted throughput: 48.89 tokens/s, Drafted throughput: 54.29 tokens/s, Accepted: 489 tokens, Drafted: 543 tokens, Per-position acceptance rate: 0.945, 0.895, 0.862, Avg Draft acceptance rate: 90.1%
269
+ INFO: 127.0.0.1:48437 - "POST /v1/completions HTTP/1.1" 200 OK
270
+ INFO: 127.0.0.1:57658 - "POST /v1/completions HTTP/1.1" 200 OK
271
+ (APIServer pid=46348) INFO 11-19 05:02:01 [loggers.py:320] Engine 000: Avg prompt throughput: 18.9 tokens/s, Avg generation throughput: 67.4 tokens/s, Running: 1 reqs, Waiting: 0 reqs, GPU KV cache usage: 2.1%
272
+ (APIServer pid=46348) INFO 11-19 05:02:01 [metrics.py:120] SpecDecoding metrics: Mean acceptance length: 3.73, Accepted throughput: 49.49 tokens/s, Drafted throughput: 54.29 tokens/s, Accepted: 495 tokens, Drafted: 543 tokens, Per-position acceptance rate: 0.978, 0.912, 0.845, Avg Draft acceptance rate: 91.2%
273
+ INFO: 127.0.0.1:29822 - "POST /v1/completions HTTP/1.1" 200 OK
274
+ INFO: 127.0.0.1:19486 - "POST /v1/completions HTTP/1.1" 200 OK
275
+ INFO: 127.0.0.1:41903 - "POST /v1/completions HTTP/1.1" 200 OK
276
+ (APIServer pid=46348) INFO 11-19 05:02:11 [loggers.py:320] Engine 000: Avg prompt throughput: 36.0 tokens/s, Avg generation throughput: 62.6 tokens/s, Running: 1 reqs, Waiting: 0 reqs, GPU KV cache usage: 2.1%
277
+ (APIServer pid=46348) INFO 11-19 05:02:11 [metrics.py:120] SpecDecoding metrics: Mean acceptance length: 3.54, Accepted throughput: 45.20 tokens/s, Drafted throughput: 53.40 tokens/s, Accepted: 452 tokens, Drafted: 534 tokens, Per-position acceptance rate: 0.955, 0.843, 0.742, Avg Draft acceptance rate: 84.6%
278
+ INFO: 127.0.0.1:35748 - "POST /v1/completions HTTP/1.1" 200 OK
279
+ INFO: 127.0.0.1:61777 - "POST /v1/completions HTTP/1.1" 200 OK
280
+ (APIServer pid=46348) INFO 11-19 05:02:21 [loggers.py:320] Engine 000: Avg prompt throughput: 21.9 tokens/s, Avg generation throughput: 66.2 tokens/s, Running: 1 reqs, Waiting: 0 reqs, GPU KV cache usage: 2.1%
281
+ (APIServer pid=46348) INFO 11-19 05:02:21 [metrics.py:120] SpecDecoding metrics: Mean acceptance length: 3.65, Accepted throughput: 47.99 tokens/s, Drafted throughput: 54.29 tokens/s, Accepted: 480 tokens, Drafted: 543 tokens, Per-position acceptance rate: 0.945, 0.890, 0.818, Avg Draft acceptance rate: 88.4%
282
+ INFO: 127.0.0.1:21996 - "POST /v1/completions HTTP/1.1" 200 OK
283
+ (APIServer pid=46348) INFO 11-19 05:02:31 [loggers.py:320] Engine 000: Avg prompt throughput: 4.2 tokens/s, Avg generation throughput: 57.6 tokens/s, Running: 1 reqs, Waiting: 0 reqs, GPU KV cache usage: 2.1%
284
+ (APIServer pid=46348) INFO 11-19 05:02:31 [metrics.py:120] SpecDecoding metrics: Mean acceptance length: 3.15, Accepted throughput: 39.29 tokens/s, Drafted throughput: 54.89 tokens/s, Accepted: 393 tokens, Drafted: 549 tokens, Per-position acceptance rate: 0.842, 0.710, 0.596, Avg Draft acceptance rate: 71.6%
285
+ (APIServer pid=46348) INFO 11-19 05:02:41 [loggers.py:320] Engine 000: Avg prompt throughput: 0.0 tokens/s, Avg generation throughput: 48.7 tokens/s, Running: 1 reqs, Waiting: 0 reqs, GPU KV cache usage: 2.1%
286
+ (APIServer pid=46348) INFO 11-19 05:02:41 [metrics.py:120] SpecDecoding metrics: Mean acceptance length: 2.63, Accepted throughput: 30.19 tokens/s, Drafted throughput: 55.49 tokens/s, Accepted: 302 tokens, Drafted: 555 tokens, Per-position acceptance rate: 0.789, 0.508, 0.335, Avg Draft acceptance rate: 54.4%
287
+ INFO: 127.0.0.1:48968 - "POST /v1/completions HTTP/1.1" 200 OK
288
+ (APIServer pid=46348) INFO 11-19 05:02:51 [loggers.py:320] Engine 000: Avg prompt throughput: 3.8 tokens/s, Avg generation throughput: 48.2 tokens/s, Running: 1 reqs, Waiting: 0 reqs, GPU KV cache usage: 2.1%
289
+ (APIServer pid=46348) INFO 11-19 05:02:51 [metrics.py:120] SpecDecoding metrics: Mean acceptance length: 2.65, Accepted throughput: 30.00 tokens/s, Drafted throughput: 54.59 tokens/s, Accepted: 300 tokens, Drafted: 546 tokens, Per-position acceptance rate: 0.758, 0.566, 0.324, Avg Draft acceptance rate: 54.9%
290
+ (APIServer pid=46348) INFO 11-19 05:03:01 [loggers.py:320] Engine 000: Avg prompt throughput: 0.0 tokens/s, Avg generation throughput: 46.8 tokens/s, Running: 1 reqs, Waiting: 0 reqs, GPU KV cache usage: 2.1%
291
+ (APIServer pid=46348) INFO 11-19 05:03:01 [metrics.py:120] SpecDecoding metrics: Mean acceptance length: 2.53, Accepted throughput: 28.30 tokens/s, Drafted throughput: 55.49 tokens/s, Accepted: 283 tokens, Drafted: 555 tokens, Per-position acceptance rate: 0.784, 0.497, 0.249, Avg Draft acceptance rate: 51.0%
292
+ INFO: 127.0.0.1:27789 - "POST /v1/completions HTTP/1.1" 200 OK
293
+ (APIServer pid=46348) INFO 11-19 05:03:11 [loggers.py:320] Engine 000: Avg prompt throughput: 4.2 tokens/s, Avg generation throughput: 43.7 tokens/s, Running: 1 reqs, Waiting: 0 reqs, GPU KV cache usage: 2.1%
294
+ (APIServer pid=46348) INFO 11-19 05:03:11 [metrics.py:120] SpecDecoding metrics: Mean acceptance length: 2.38, Accepted throughput: 25.30 tokens/s, Drafted throughput: 54.90 tokens/s, Accepted: 253 tokens, Drafted: 549 tokens, Per-position acceptance rate: 0.710, 0.426, 0.246, Avg Draft acceptance rate: 46.1%
295
+ (APIServer pid=46348) INFO 11-19 05:03:21 [loggers.py:320] Engine 000: Avg prompt throughput: 0.0 tokens/s, Avg generation throughput: 46.4 tokens/s, Running: 1 reqs, Waiting: 0 reqs, GPU KV cache usage: 2.2%
296
+ (APIServer pid=46348) INFO 11-19 05:03:21 [metrics.py:120] SpecDecoding metrics: Mean acceptance length: 2.51, Accepted throughput: 27.90 tokens/s, Drafted throughput: 55.50 tokens/s, Accepted: 279 tokens, Drafted: 555 tokens, Per-position acceptance rate: 0.735, 0.486, 0.286, Avg Draft acceptance rate: 50.3%
297
+ INFO: 127.0.0.1:24816 - "POST /v1/completions HTTP/1.1" 200 OK
298
+ (APIServer pid=46348) INFO 11-19 05:03:31 [loggers.py:320] Engine 000: Avg prompt throughput: 7.0 tokens/s, Avg generation throughput: 50.4 tokens/s, Running: 1 reqs, Waiting: 0 reqs, GPU KV cache usage: 2.1%
299
+ (APIServer pid=46348) INFO 11-19 05:03:31 [metrics.py:120] SpecDecoding metrics: Mean acceptance length: 2.76, Accepted throughput: 32.10 tokens/s, Drafted throughput: 54.59 tokens/s, Accepted: 321 tokens, Drafted: 546 tokens, Per-position acceptance rate: 0.797, 0.571, 0.396, Avg Draft acceptance rate: 58.8%
300
+ (APIServer pid=46348) INFO 11-19 05:03:41 [loggers.py:320] Engine 000: Avg prompt throughput: 0.0 tokens/s, Avg generation throughput: 47.2 tokens/s, Running: 1 reqs, Waiting: 0 reqs, GPU KV cache usage: 2.2%
301
+ (APIServer pid=46348) INFO 11-19 05:03:41 [metrics.py:120] SpecDecoding metrics: Mean acceptance length: 2.55, Accepted throughput: 28.70 tokens/s, Drafted throughput: 55.49 tokens/s, Accepted: 287 tokens, Drafted: 555 tokens, Per-position acceptance rate: 0.768, 0.481, 0.303, Avg Draft acceptance rate: 51.7%
302
+ INFO: 127.0.0.1:44535 - "POST /v1/completions HTTP/1.1" 200 OK
303
+ (APIServer pid=46348) INFO 11-19 05:03:51 [loggers.py:320] Engine 000: Avg prompt throughput: 6.7 tokens/s, Avg generation throughput: 43.1 tokens/s, Running: 1 reqs, Waiting: 0 reqs, GPU KV cache usage: 2.1%
304
+ (APIServer pid=46348) INFO 11-19 05:03:51 [metrics.py:120] SpecDecoding metrics: Mean acceptance length: 2.35, Accepted throughput: 24.70 tokens/s, Drafted throughput: 54.89 tokens/s, Accepted: 247 tokens, Drafted: 549 tokens, Per-position acceptance rate: 0.667, 0.432, 0.251, Avg Draft acceptance rate: 45.0%
305
+ (APIServer pid=46348) INFO 11-19 05:04:01 [loggers.py:320] Engine 000: Avg prompt throughput: 0.0 tokens/s, Avg generation throughput: 44.0 tokens/s, Running: 1 reqs, Waiting: 0 reqs, GPU KV cache usage: 2.1%
306
+ (APIServer pid=46348) INFO 11-19 05:04:01 [metrics.py:120] SpecDecoding metrics: Mean acceptance length: 2.38, Accepted throughput: 25.50 tokens/s, Drafted throughput: 55.49 tokens/s, Accepted: 255 tokens, Drafted: 555 tokens, Per-position acceptance rate: 0.686, 0.411, 0.281, Avg Draft acceptance rate: 45.9%
307
+ INFO: 127.0.0.1:60361 - "POST /v1/completions HTTP/1.1" 200 OK
308
+ (APIServer pid=46348) INFO 11-19 05:04:11 [loggers.py:320] Engine 000: Avg prompt throughput: 4.6 tokens/s, Avg generation throughput: 45.6 tokens/s, Running: 1 reqs, Waiting: 0 reqs, GPU KV cache usage: 2.1%
309
+ (APIServer pid=46348) INFO 11-19 05:04:11 [metrics.py:120] SpecDecoding metrics: Mean acceptance length: 2.50, Accepted throughput: 27.30 tokens/s, Drafted throughput: 54.59 tokens/s, Accepted: 273 tokens, Drafted: 546 tokens, Per-position acceptance rate: 0.769, 0.451, 0.280, Avg Draft acceptance rate: 50.0%
310
+ (APIServer pid=46348) INFO 11-19 05:04:21 [loggers.py:320] Engine 000: Avg prompt throughput: 0.0 tokens/s, Avg generation throughput: 46.6 tokens/s, Running: 1 reqs, Waiting: 0 reqs, GPU KV cache usage: 2.1%
311
+ (APIServer pid=46348) INFO 11-19 05:04:21 [metrics.py:120] SpecDecoding metrics: Mean acceptance length: 2.52, Accepted throughput: 28.10 tokens/s, Drafted throughput: 55.49 tokens/s, Accepted: 281 tokens, Drafted: 555 tokens, Per-position acceptance rate: 0.757, 0.481, 0.281, Avg Draft acceptance rate: 50.6%
312
+ INFO: 127.0.0.1:31438 - "POST /v1/completions HTTP/1.1" 200 OK
313
+ (APIServer pid=46348) INFO 11-19 05:04:31 [loggers.py:320] Engine 000: Avg prompt throughput: 4.3 tokens/s, Avg generation throughput: 43.9 tokens/s, Running: 1 reqs, Waiting: 0 reqs, GPU KV cache usage: 2.1%
314
+ (APIServer pid=46348) INFO 11-19 05:04:31 [metrics.py:120] SpecDecoding metrics: Mean acceptance length: 2.41, Accepted throughput: 25.80 tokens/s, Drafted throughput: 54.89 tokens/s, Accepted: 258 tokens, Drafted: 549 tokens, Per-position acceptance rate: 0.765, 0.415, 0.230, Avg Draft acceptance rate: 47.0%
315
+ (APIServer pid=46348) INFO 11-19 05:04:41 [loggers.py:320] Engine 000: Avg prompt throughput: 0.0 tokens/s, Avg generation throughput: 41.8 tokens/s, Running: 1 reqs, Waiting: 0 reqs, GPU KV cache usage: 2.1%
316
+ (APIServer pid=46348) INFO 11-19 05:04:41 [metrics.py:120] SpecDecoding metrics: Mean acceptance length: 2.26, Accepted throughput: 23.30 tokens/s, Drafted throughput: 55.49 tokens/s, Accepted: 233 tokens, Drafted: 555 tokens, Per-position acceptance rate: 0.681, 0.373, 0.205, Avg Draft acceptance rate: 42.0%
317
+ INFO: 127.0.0.1:22693 - "POST /v1/completions HTTP/1.1" 200 OK
318
+ (APIServer pid=46348) INFO 11-19 05:04:51 [loggers.py:320] Engine 000: Avg prompt throughput: 7.0 tokens/s, Avg generation throughput: 46.9 tokens/s, Running: 1 reqs, Waiting: 0 reqs, GPU KV cache usage: 2.1%
319
+ (APIServer pid=46348) INFO 11-19 05:04:51 [metrics.py:120] SpecDecoding metrics: Mean acceptance length: 2.59, Accepted throughput: 28.90 tokens/s, Drafted throughput: 54.59 tokens/s, Accepted: 289 tokens, Drafted: 546 tokens, Per-position acceptance rate: 0.753, 0.495, 0.341, Avg Draft acceptance rate: 52.9%
320
+ (APIServer pid=46348) INFO 11-19 05:05:01 [loggers.py:320] Engine 000: Avg prompt throughput: 0.0 tokens/s, Avg generation throughput: 44.0 tokens/s, Running: 1 reqs, Waiting: 0 reqs, GPU KV cache usage: 2.1%
321
+ (APIServer pid=46348) INFO 11-19 05:05:01 [metrics.py:120] SpecDecoding metrics: Mean acceptance length: 2.38, Accepted throughput: 25.50 tokens/s, Drafted throughput: 55.49 tokens/s, Accepted: 255 tokens, Drafted: 555 tokens, Per-position acceptance rate: 0.746, 0.405, 0.227, Avg Draft acceptance rate: 45.9%
322
+ (APIServer pid=46348) INFO 11-19 05:05:11 [loggers.py:320] Engine 000: Avg prompt throughput: 0.0 tokens/s, Avg generation throughput: 44.0 tokens/s, Running: 1 reqs, Waiting: 0 reqs, GPU KV cache usage: 2.2%
323
+ (APIServer pid=46348) INFO 11-19 05:05:11 [metrics.py:120] SpecDecoding metrics: Mean acceptance length: 2.39, Accepted throughput: 25.60 tokens/s, Drafted throughput: 55.20 tokens/s, Accepted: 256 tokens, Drafted: 552 tokens, Per-position acceptance rate: 0.696, 0.424, 0.272, Avg Draft acceptance rate: 46.4%
324
+ INFO: 127.0.0.1:54056 - "GET /metrics HTTP/1.1" 200 OK
325
+ INFO: 127.0.0.1:27419 - "GET /metrics HTTP/1.1" 200 OK
326
+ INFO: 127.0.0.1:22449 - "POST /v1/completions HTTP/1.1" 200 OK
327
+ INFO: 127.0.0.1:32176 - "POST /v1/completions HTTP/1.1" 200 OK
328
+ INFO: 127.0.0.1:31936 - "POST /v1/completions HTTP/1.1" 200 OK
329
+ INFO: 127.0.0.1:35639 - "POST /v1/completions HTTP/1.1" 200 OK
330
+ (APIServer pid=46348) INFO 11-19 05:05:21 [loggers.py:320] Engine 000: Avg prompt throughput: 18.0 tokens/s, Avg generation throughput: 114.5 tokens/s, Running: 1 reqs, Waiting: 0 reqs, GPU KV cache usage: 2.1%
331
+ (APIServer pid=46348) INFO 11-19 05:05:21 [metrics.py:120] SpecDecoding metrics: Mean acceptance length: 3.15, Accepted throughput: 78.19 tokens/s, Drafted throughput: 108.88 tokens/s, Accepted: 782 tokens, Drafted: 1089 tokens, Per-position acceptance rate: 0.887, 0.694, 0.573, Avg Draft acceptance rate: 71.8%
332
+ INFO: 127.0.0.1:34955 - "POST /v1/completions HTTP/1.1" 200 OK
333
+ INFO: 127.0.0.1:43333 - "POST /v1/completions HTTP/1.1" 200 OK
334
+ (APIServer pid=46348) INFO 11-19 05:05:31 [loggers.py:320] Engine 000: Avg prompt throughput: 21.7 tokens/s, Avg generation throughput: 119.1 tokens/s, Running: 2 reqs, Waiting: 0 reqs, GPU KV cache usage: 4.1%
335
+ (APIServer pid=46348) INFO 11-19 05:05:31 [metrics.py:120] SpecDecoding metrics: Mean acceptance length: 3.28, Accepted throughput: 82.80 tokens/s, Drafted throughput: 108.90 tokens/s, Accepted: 828 tokens, Drafted: 1089 tokens, Per-position acceptance rate: 0.884, 0.760, 0.636, Avg Draft acceptance rate: 76.0%
336
+ INFO: 127.0.0.1:25501 - "POST /v1/completions HTTP/1.1" 200 OK
337
+ INFO: 127.0.0.1:24531 - "POST /v1/completions HTTP/1.1" 200 OK
338
+ INFO: 127.0.0.1:48986 - "POST /v1/completions HTTP/1.1" 200 OK
339
+ (APIServer pid=46348) INFO 11-19 05:05:41 [loggers.py:320] Engine 000: Avg prompt throughput: 37.2 tokens/s, Avg generation throughput: 110.5 tokens/s, Running: 2 reqs, Waiting: 0 reqs, GPU KV cache usage: 4.1%
340
+ (APIServer pid=46348) INFO 11-19 05:05:41 [metrics.py:120] SpecDecoding metrics: Mean acceptance length: 3.05, Accepted throughput: 74.49 tokens/s, Drafted throughput: 108.88 tokens/s, Accepted: 745 tokens, Drafted: 1089 tokens, Per-position acceptance rate: 0.873, 0.667, 0.512, Avg Draft acceptance rate: 68.4%
341
+ INFO: 127.0.0.1:58918 - "POST /v1/completions HTTP/1.1" 200 OK
342
+ INFO: 127.0.0.1:23917 - "POST /v1/completions HTTP/1.1" 200 OK
343
+ INFO: 127.0.0.1:54139 - "POST /v1/completions HTTP/1.1" 200 OK
344
+ INFO: 127.0.0.1:16442 - "POST /v1/completions HTTP/1.1" 200 OK
345
+ (APIServer pid=46348) INFO 11-19 05:05:51 [loggers.py:320] Engine 000: Avg prompt throughput: 89.9 tokens/s, Avg generation throughput: 125.5 tokens/s, Running: 2 reqs, Waiting: 0 reqs, GPU KV cache usage: 4.1%
346
+ (APIServer pid=46348) INFO 11-19 05:05:51 [metrics.py:120] SpecDecoding metrics: Mean acceptance length: 3.52, Accepted throughput: 90.09 tokens/s, Drafted throughput: 107.39 tokens/s, Accepted: 901 tokens, Drafted: 1074 tokens, Per-position acceptance rate: 0.930, 0.844, 0.743, Avg Draft acceptance rate: 83.9%
347
+ INFO: 127.0.0.1:28723 - "POST /v1/completions HTTP/1.1" 200 OK
348
+ INFO: 127.0.0.1:25219 - "POST /v1/completions HTTP/1.1" 200 OK
349
+ INFO: 127.0.0.1:52811 - "POST /v1/completions HTTP/1.1" 200 OK
350
+ INFO: 127.0.0.1:32938 - "POST /v1/completions HTTP/1.1" 200 OK
351
+ (APIServer pid=46348) INFO 11-19 05:06:01 [loggers.py:320] Engine 000: Avg prompt throughput: 49.6 tokens/s, Avg generation throughput: 138.0 tokens/s, Running: 1 reqs, Waiting: 0 reqs, GPU KV cache usage: 2.1%
352
+ (APIServer pid=46348) INFO 11-19 05:06:01 [metrics.py:120] SpecDecoding metrics: Mean acceptance length: 3.82, Accepted throughput: 102.19 tokens/s, Drafted throughput: 108.59 tokens/s, Accepted: 1022 tokens, Drafted: 1086 tokens, Per-position acceptance rate: 0.975, 0.936, 0.912, Avg Draft acceptance rate: 94.1%
353
+ INFO: 127.0.0.1:58106 - "POST /v1/completions HTTP/1.1" 200 OK
354
+ INFO: 127.0.0.1:31663 - "POST /v1/completions HTTP/1.1" 200 OK
355
+ INFO: 127.0.0.1:60389 - "POST /v1/completions HTTP/1.1" 200 OK
356
+ INFO: 127.0.0.1:24904 - "POST /v1/completions HTTP/1.1" 200 OK
357
+ (APIServer pid=46348) INFO 11-19 05:06:11 [loggers.py:320] Engine 000: Avg prompt throughput: 47.7 tokens/s, Avg generation throughput: 131.5 tokens/s, Running: 2 reqs, Waiting: 0 reqs, GPU KV cache usage: 4.1%
358
+ (APIServer pid=46348) INFO 11-19 05:06:11 [metrics.py:120] SpecDecoding metrics: Mean acceptance length: 3.68, Accepted throughput: 95.98 tokens/s, Drafted throughput: 107.38 tokens/s, Accepted: 960 tokens, Drafted: 1074 tokens, Per-position acceptance rate: 0.969, 0.897, 0.816, Avg Draft acceptance rate: 89.4%
359
+ INFO: 127.0.0.1:48380 - "POST /v1/completions HTTP/1.1" 200 OK
360
+ INFO: 127.0.0.1:22565 - "POST /v1/completions HTTP/1.1" 200 OK
361
+ INFO: 127.0.0.1:49259 - "POST /v1/completions HTTP/1.1" 200 OK
362
+ INFO: 127.0.0.1:63885 - "POST /v1/completions HTTP/1.1" 200 OK
363
+ (APIServer pid=46348) INFO 11-19 05:06:21 [loggers.py:320] Engine 000: Avg prompt throughput: 40.0 tokens/s, Avg generation throughput: 128.7 tokens/s, Running: 2 reqs, Waiting: 0 reqs, GPU KV cache usage: 4.1%
364
+ (APIServer pid=46348) INFO 11-19 05:06:21 [metrics.py:120] SpecDecoding metrics: Mean acceptance length: 3.58, Accepted throughput: 92.98 tokens/s, Drafted throughput: 107.98 tokens/s, Accepted: 930 tokens, Drafted: 1080 tokens, Per-position acceptance rate: 0.944, 0.864, 0.775, Avg Draft acceptance rate: 86.1%
365
+ INFO: 127.0.0.1:42278 - "POST /v1/completions HTTP/1.1" 200 OK
366
+ (APIServer pid=46348) INFO 11-19 05:06:31 [loggers.py:320] Engine 000: Avg prompt throughput: 3.8 tokens/s, Avg generation throughput: 94.5 tokens/s, Running: 2 reqs, Waiting: 0 reqs, GPU KV cache usage: 4.1%
367
+ (APIServer pid=46348) INFO 11-19 05:06:31 [metrics.py:120] SpecDecoding metrics: Mean acceptance length: 2.55, Accepted throughput: 57.59 tokens/s, Drafted throughput: 111.28 tokens/s, Accepted: 576 tokens, Drafted: 1113 tokens, Per-position acceptance rate: 0.739, 0.488, 0.326, Avg Draft acceptance rate: 51.8%
368
+ INFO: 127.0.0.1:55630 - "POST /v1/completions HTTP/1.1" 200 OK
369
+ INFO: 127.0.0.1:30446 - "POST /v1/completions HTTP/1.1" 200 OK
370
+ (APIServer pid=46348) INFO 11-19 05:06:41 [loggers.py:320] Engine 000: Avg prompt throughput: 11.2 tokens/s, Avg generation throughput: 95.2 tokens/s, Running: 2 reqs, Waiting: 0 reqs, GPU KV cache usage: 4.1%
371
+ (APIServer pid=46348) INFO 11-19 05:06:41 [metrics.py:120] SpecDecoding metrics: Mean acceptance length: 2.60, Accepted throughput: 58.39 tokens/s, Drafted throughput: 109.79 tokens/s, Accepted: 584 tokens, Drafted: 1098 tokens, Per-position acceptance rate: 0.757, 0.516, 0.322, Avg Draft acceptance rate: 53.2%
372
+ (APIServer pid=46348) INFO 11-19 05:06:51 [loggers.py:320] Engine 000: Avg prompt throughput: 0.0 tokens/s, Avg generation throughput: 94.1 tokens/s, Running: 2 reqs, Waiting: 0 reqs, GPU KV cache usage: 4.1%
373
+ (APIServer pid=46348) INFO 11-19 05:06:51 [metrics.py:120] SpecDecoding metrics: Mean acceptance length: 2.53, Accepted throughput: 56.90 tokens/s, Drafted throughput: 111.60 tokens/s, Accepted: 569 tokens, Drafted: 1116 tokens, Per-position acceptance rate: 0.758, 0.500, 0.272, Avg Draft acceptance rate: 51.0%
runs/r3/logs/bench_draft.log ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ [04:57:26] starting vLLM (log: /workspace/emc/work/logs/20261119-045726-bench-batch4.log)
2
+ [04:59:02] vLLM ready after 96s
3
+ [05:05:11] batch 1: {'requests': 32, 'concurrency': 1, 'output_tokens': 16195, 'wall_s': 303.54, 'throughput_tok_s': 53.4, 'tpot_median_ms': 15.39, 'tpot_p90_ms': 22.04, 'ttft_median_ms': 113.6, 'ttft_max_ms': 136.61, 'acceptance_length': 2.9237, 'engine_max_num_seqs': 4}
runs/r3/logs/pipeline.log CHANGED
@@ -37,3 +37,4 @@
37
  [2026-11-19 04:57:23] done choose (0s)
38
  [2026-11-19 04:57:24] start report
39
  [2026-11-19 04:57:24] done report (0s)
 
 
37
  [2026-11-19 04:57:23] done choose (0s)
38
  [2026-11-19 04:57:24] start report
39
  [2026-11-19 04:57:24] done report (0s)
40
+ [2026-11-19 04:57:26] start bench_draft
runs/r3/logs/pipeline.out CHANGED
@@ -37,3 +37,4 @@
37
  [2026-11-19 04:57:23] done choose (0s)
38
  [2026-11-19 04:57:24] start report
39
  [2026-11-19 04:57:24] done report (0s)
 
 
37
  [2026-11-19 04:57:23] done choose (0s)
38
  [2026-11-19 04:57:24] start report
39
  [2026-11-19 04:57:24] done report (0s)
40
+ [2026-11-19 04:57:26] start bench_draft
runs/r3/logs/stats.log CHANGED
@@ -65,3 +65,5 @@
65
  [2026-11-19 04:57:23] ram 0.2/172 GB used (cache 0.0, shmem 0.0) | cgroup 0.2/172 GB | disk 144 GB used | gpu 0.0/95.6 GiB, 0%, 84 W | load 0.0 | largest python 0.1 GB, python 0.1 GB, python 0.0 GB
66
  [2026-11-19 04:57:24] ram 0.2/172 GB used (cache 0.0, shmem 0.0) | cgroup 0.2/172 GB | disk 144 GB used | gpu 0.0/95.6 GiB, 0%, 62 W | load 0.0 | largest python 0.1 GB, python 0.1 GB, python 0.0 GB
67
  [2026-11-19 04:57:24] ram 0.2/172 GB used (cache 0.0, shmem 0.0) | cgroup 0.2/172 GB | disk 144 GB used | gpu 0.0/95.6 GiB, 0%, 62 W | load 0.0 | largest python 0.1 GB, python 0.1 GB, python 0.0 GB
 
 
 
65
  [2026-11-19 04:57:23] ram 0.2/172 GB used (cache 0.0, shmem 0.0) | cgroup 0.2/172 GB | disk 144 GB used | gpu 0.0/95.6 GiB, 0%, 84 W | load 0.0 | largest python 0.1 GB, python 0.1 GB, python 0.0 GB
66
  [2026-11-19 04:57:24] ram 0.2/172 GB used (cache 0.0, shmem 0.0) | cgroup 0.2/172 GB | disk 144 GB used | gpu 0.0/95.6 GiB, 0%, 62 W | load 0.0 | largest python 0.1 GB, python 0.1 GB, python 0.0 GB
67
  [2026-11-19 04:57:24] ram 0.2/172 GB used (cache 0.0, shmem 0.0) | cgroup 0.2/172 GB | disk 144 GB used | gpu 0.0/95.6 GiB, 0%, 62 W | load 0.0 | largest python 0.1 GB, python 0.1 GB, python 0.0 GB
68
+ [2026-11-19 04:57:26] ram 0.2/172 GB used (cache 0.0, shmem 0.0) | cgroup 0.2/172 GB | disk 144 GB used | gpu 0.0/95.6 GiB, 0%, 62 W | load 0.0 | largest python 0.1 GB, python 0.1 GB, python 0.0 GB
69
+ [2026-11-19 05:06:52] ram 4.1/172 GB used (cache 0.3, shmem 0.0) | cgroup 4.1/172 GB | disk 144 GB used | gpu 84.9/95.6 GiB, 100%, 298 W | load 0.0 | largest VLLM::EngineCor 4.3 GB, vllm 2.5 GB, python 0.1 GB
runs/r3/logs/sync.log CHANGED
@@ -52,3 +52,4 @@ uploaded logs/results to mudassiruddin21/EMC:runs/r3
52
  uploaded logs/results to mudassiruddin21/EMC:runs/r3
53
  uploaded logs/results to mudassiruddin21/EMC:runs/r3
54
  uploaded logs/results to mudassiruddin21/EMC:runs/r3
 
 
52
  uploaded logs/results to mudassiruddin21/EMC:runs/r3
53
  uploaded logs/results to mudassiruddin21/EMC:runs/r3
54
  uploaded logs/results to mudassiruddin21/EMC:runs/r3
55
+ uploaded logs/results to mudassiruddin21/EMC:runs/r3
runs/r3/results/bench_draft.json ADDED
@@ -0,0 +1,24 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "tag": "draft",
3
+ "draft": "/workspace/emc/runs/main3/sub_fp8",
4
+ "prompts": 32,
5
+ "official_max_num_seqs": 4,
6
+ "batch": [
7
+ {
8
+ "requests": 32,
9
+ "concurrency": 1,
10
+ "output_tokens": 16195,
11
+ "wall_s": 303.54,
12
+ "throughput_tok_s": 53.4,
13
+ "tpot_median_ms": 15.39,
14
+ "tpot_p90_ms": 22.04,
15
+ "ttft_median_ms": 113.6,
16
+ "ttft_max_ms": 136.61,
17
+ "acceptance_length": 2.9237,
18
+ "engine_max_num_seqs": 4
19
+ }
20
+ ],
21
+ "long": [],
22
+ "k": [],
23
+ "errors": []
24
+ }