ThinkingCap-Qwen3.8-27B-NInfer / bench /mtp0-gpu1-145w-r3.json
Schestex's picture
RC1: D48+O44+I53+G64 + native Q8 MTP + Sharp
b78a22f verified
Raw History Blame Contribute Delete
15.8 kB
{
"schema_version": 15,
"artifact_type": "ninfer_bench_report",
"tool": "ninfer_bench",
"command": "/usr/local/bin/ninfer_bench --weights /models/thinkingcap-d48-o44-i53-g64-mtp-nativeq8-2026-09-27/thinkingcap-d48-o44-i53-g64-mtp-nativeq8.ninfer --corpus /workspace/ninfer/bench/fixtures/bench_corpus.ids --device 0 -n 256,1024 -pg 2048,512;8192,512 -r 3 --warmup 2 --max-ctx 16384 --prefill-chunk 1024 --kv-dtype fp8 --output json --output-file /out/mtp0-gpu1-r3.json",
"environment": {"gpu_name": "NVIDIA RTX PRO 4000 Blackwell", "cuda_runtime_version": "13.1", "cuda_driver_version": "13.2", "device_id": 0},
"artifact": {"path": "/models/thinkingcap-d48-o44-i53-g64-mtp-nativeq8-2026-09-27/thinkingcap-d48-o44-i53-g64-mtp-nativeq8.ninfer", "file_size_bytes": 17501292292},
"load": {
"architecture": "Qwen3_5ForCausalLM",
"name": "thinkingcap-qwen3.8-27b-d48-o44-i53-g64-mtp-nativeq8",
"formats": ["bf16","fp32","nvfp4","q4_g64_fp16","q5_g64_fp16","q8_g32_fp16"],
"prefill_signature": "2047cd3b0a44f1986d5ae5cca67c16f458371663e9769b590d96582902ce65bc",
"load_seconds": 3.215501918,
"upload_seconds": 2.351215222,
"artifact_bytes_read": 16692550665,
"host_to_device_bytes": 16679625540,
"peak_staging_bytes": 268435456,
"device_object_count": 718,
"host_object_count": 4
},
"memory": {
"device": 0,
"max_context": 16384,
"kv_capacity_mode": "explicit",
"kv_capacity": 16384,
"kv_capacity_page_groups": 256,
"kv_capacity_max_page_groups": 256,
"kv_cache": "fp8-e4m3-row256",
"weights": {"capacity_bytes": 16679684352, "used_bytes": 16679684352, "peak_used_bytes": 16679684352},
"sequence": {"capacity_bytes": 707504896, "used_bytes": 707504896, "peak_used_bytes": 707504896},
"workspace": {"capacity_bytes": 159981568, "used_bytes": 0, "peak_used_bytes": 0},
"vision_workspace": null,
"minimum_runtime_reservation_bytes": 880069376,
"kv_capacity_increment_bytes": 0,
"runtime_reservation_bytes": 880069376,
"available_after_weights_bytes": 8227717120,
"available_after_startup_bytes": 7355301888,
"kv_capacity_headroom_bytes": 0,
"planned_slack_bytes": 7347647744,
"cuda_graph_allowance_bytes": 12582912,
"kv_payload_bytes": 541065216
},
"config": {
"max_context": 16384,
"prefill_chunk": 1024,
"kv_cache": "fp8-e4m3-row256",
"speculative_backend": "none",
"draft_tokens": 0,
"proposal_head": "full",
"use_cuda_graph": true,
"decode_path": "cuda_graph",
"decode_graph_prime": {"primed": true, "output_tokens": 3},
"repetitions": 3,
"warmup": 2,
"corpus_path": "/workspace/ninfer/bench/fixtures/bench_corpus.ids",
"corpus_tokens": 65536
},
"tests": [
{
"label": "tg256",
"kind": "tg",
"n_prompt": 0,
"n_gen": 256,
"requested_output_tokens": 257,
"prefill_tok_s_mean": null,
"prefill_tok_s_stddev": null,
"decode_output_tok_s_mean": 33.684761,
"decode_output_tok_s_stddev": 0.1895257772,
"decode_engine_tok_s_mean": 33.684761,
"decode_engine_tok_s_stddev": 0.1895257772,
"prepare_seconds_mean": 1.643333333e-06,
"prepare_seconds_stddev": 1.594783162e-07,
"prefill_seconds_mean": null,
"prefill_seconds_stddev": null,
"decode_seconds_mean": 7.600036103,
"decode_seconds_stddev": 0.04273637934,
"total_seconds_mean": 7.630126484,
"total_seconds_stddev": 0.04289000507,
"workspace_peak_bytes": 159981568,
"workspace_allocator_peak_bytes": 159981568,
"speculative": {
"backend": "none",
"enabled": false,
"draft_window": 0,
"rounds": 0,
"drafted_tokens": 0,
"accepted_tokens": 0,
"fallback_steps": 0,
"acceptance_rate": null,
"acceptance_length": null,
"accepted_per_position": []
},
"reps": [
{
"generated_output_tokens": 257,
"decode_output_tokens": 256,
"decode_engine_tokens": 256,
"timings": {"prepare_seconds": 1.72e-06, "vision_seconds": 0, "prefill_seconds": 0.029530101, "decode_seconds": 7.555931071, "total_seconds": 7.585861902},
"speculative": {
"backend": "none",
"enabled": false,
"draft_window": 0,
"rounds": 0,
"drafted_tokens": 0,
"accepted_tokens": 0,
"fallback_steps": 0,
"acceptance_rate": null,
"acceptance_length": null,
"accepted_per_position": []
}
},
{
"generated_output_tokens": 257,
"decode_output_tokens": 256,
"decode_engine_tokens": 256,
"timings": {"prepare_seconds": 1.75e-06, "vision_seconds": 0, "prefill_seconds": 0.029666951, "decode_seconds": 7.602919433, "total_seconds": 7.633022412},
"speculative": {
"backend": "none",
"enabled": false,
"draft_window": 0,
"rounds": 0,
"drafted_tokens": 0,
"accepted_tokens": 0,
"fallback_steps": 0,
"acceptance_rate": null,
"acceptance_length": null,
"accepted_per_position": []
}
},
{
"generated_output_tokens": 257,
"decode_output_tokens": 256,
"decode_engine_tokens": 256,
"timings": {"prepare_seconds": 1.46e-06, "vision_seconds": 0, "prefill_seconds": 0.029836681, "decode_seconds": 7.641257806, "total_seconds": 7.671495137},
"speculative": {
"backend": "none",
"enabled": false,
"draft_window": 0,
"rounds": 0,
"drafted_tokens": 0,
"accepted_tokens": 0,
"fallback_steps": 0,
"acceptance_rate": null,
"acceptance_length": null,
"accepted_per_position": []
}
}
]
},
{
"label": "tg1024",
"kind": "tg",
"n_prompt": 0,
"n_gen": 1024,
"requested_output_tokens": 1025,
"prefill_tok_s_mean": null,
"prefill_tok_s_stddev": null,
"decode_output_tok_s_mean": 33.56648354,
"decode_output_tok_s_stddev": 0.07796054842,
"decode_engine_tok_s_mean": 33.56648354,
"decode_engine_tok_s_stddev": 0.07796054842,
"prepare_seconds_mean": 2.236666667e-06,
"prepare_seconds_stddev": 2.804163571e-07,
"prefill_seconds_mean": null,
"prefill_seconds_stddev": null,
"decode_seconds_mean": 30.50673092,
"decode_seconds_stddev": 0.07083484686,
"total_seconds_mean": 30.53810507,
"total_seconds_stddev": 0.07092740454,
"workspace_peak_bytes": 159981568,
"workspace_allocator_peak_bytes": 159981568,
"speculative": {
"backend": "none",
"enabled": false,
"draft_window": 0,
"rounds": 0,
"drafted_tokens": 0,
"accepted_tokens": 0,
"fallback_steps": 0,
"acceptance_rate": null,
"acceptance_length": null,
"accepted_per_position": []
},
"reps": [
{
"generated_output_tokens": 1025,
"decode_output_tokens": 1024,
"decode_engine_tokens": 1024,
"timings": {"prepare_seconds": 1.93e-06, "vision_seconds": 0, "prefill_seconds": 0.029781792, "decode_seconds": 30.43330908, "total_seconds": 30.46457702},
"speculative": {
"backend": "none",
"enabled": false,
"draft_window": 0,
"rounds": 0,
"drafted_tokens": 0,
"accepted_tokens": 0,
"fallback_steps": 0,
"acceptance_rate": null,
"acceptance_length": null,
"accepted_per_position": []
}
},
{
"generated_output_tokens": 1025,
"decode_output_tokens": 1024,
"decode_engine_tokens": 1024,
"timings": {"prepare_seconds": 2.48e-06, "vision_seconds": 0, "prefill_seconds": 0.029931181, "decode_seconds": 30.51222486, "total_seconds": 30.54362944},
"speculative": {
"backend": "none",
"enabled": false,
"draft_window": 0,
"rounds": 0,
"drafted_tokens": 0,
"accepted_tokens": 0,
"fallback_steps": 0,
"acceptance_rate": null,
"acceptance_length": null,
"accepted_per_position": []
}
},
{
"generated_output_tokens": 1025,
"decode_output_tokens": 1024,
"decode_engine_tokens": 1024,
"timings": {"prepare_seconds": 2.3e-06, "vision_seconds": 0, "prefill_seconds": 0.029967602, "decode_seconds": 30.57465883, "total_seconds": 30.60610875},
"speculative": {
"backend": "none",
"enabled": false,
"draft_window": 0,
"rounds": 0,
"drafted_tokens": 0,
"accepted_tokens": 0,
"fallback_steps": 0,
"acceptance_rate": null,
"acceptance_length": null,
"accepted_per_position": []
}
}
]
},
{
"label": "pp2048+tg512",
"kind": "pp+tg",
"n_prompt": 2048,
"n_gen": 512,
"requested_output_tokens": 513,
"prefill_tok_s_mean": 3658.542902,
"prefill_tok_s_stddev": 3.445873284,
"decode_output_tok_s_mean": 33.28000528,
"decode_output_tok_s_stddev": 0.006441816527,
"decode_engine_tok_s_mean": 33.28000528,
"decode_engine_tok_s_stddev": 0.006441816527,
"prepare_seconds_mean": 1.159666667e-05,
"prepare_seconds_stddev": 5.299371032e-07,
"prefill_seconds_mean": 0.5597860313,
"prefill_seconds_stddev": 0.0005275218527,
"decode_seconds_mean": 15.38461333,
"decode_seconds_stddev": 0.002977803412,
"total_seconds_mean": 15.9452091,
"total_seconds_stddev": 0.003136026081,
"workspace_peak_bytes": 159981568,
"workspace_allocator_peak_bytes": 159981568,
"speculative": {
"backend": "none",
"enabled": false,
"draft_window": 0,
"rounds": 0,
"drafted_tokens": 0,
"accepted_tokens": 0,
"fallback_steps": 0,
"acceptance_rate": null,
"acceptance_length": null,
"accepted_per_position": []
},
"reps": [
{
"generated_output_tokens": 513,
"decode_output_tokens": 512,
"decode_engine_tokens": 512,
"timings": {"prepare_seconds": 1.103e-05, "vision_seconds": 0, "prefill_seconds": 0.559531288, "decode_seconds": 15.38738746, "total_seconds": 15.94772397},
"speculative": {
"backend": "none",
"enabled": false,
"draft_window": 0,
"rounds": 0,
"drafted_tokens": 0,
"accepted_tokens": 0,
"fallback_steps": 0,
"acceptance_rate": null,
"acceptance_length": null,
"accepted_per_position": []
}
},
{
"generated_output_tokens": 513,
"decode_output_tokens": 512,
"decode_engine_tokens": 512,
"timings": {"prepare_seconds": 1.168e-05, "vision_seconds": 0, "prefill_seconds": 0.560392578, "decode_seconds": 15.38498566, "total_seconds": 15.94620803},
"speculative": {
"backend": "none",
"enabled": false,
"draft_window": 0,
"rounds": 0,
"drafted_tokens": 0,
"accepted_tokens": 0,
"fallback_steps": 0,
"acceptance_rate": null,
"acceptance_length": null,
"accepted_per_position": []
}
},
{
"generated_output_tokens": 513,
"decode_output_tokens": 512,
"decode_engine_tokens": 512,
"timings": {"prepare_seconds": 1.208e-05, "vision_seconds": 0, "prefill_seconds": 0.559434228, "decode_seconds": 15.38146687, "total_seconds": 15.94169529},
"speculative": {
"backend": "none",
"enabled": false,
"draft_window": 0,
"rounds": 0,
"drafted_tokens": 0,
"accepted_tokens": 0,
"fallback_steps": 0,
"acceptance_rate": null,
"acceptance_length": null,
"accepted_per_position": []
}
}
]
},
{
"label": "pp8192+tg512",
"kind": "pp+tg",
"n_prompt": 8192,
"n_gen": 512,
"requested_output_tokens": 513,
"prefill_tok_s_mean": 3474.769135,
"prefill_tok_s_stddev": 1.539211318,
"decode_output_tok_s_mean": 32.77703692,
"decode_output_tok_s_stddev": 0.009201321588,
"decode_engine_tok_s_mean": 32.77703692,
"decode_engine_tok_s_stddev": 0.009201321588,
"prepare_seconds_mean": 2.889e-05,
"prepare_seconds_stddev": 2.343074903e-07,
"prefill_seconds_mean": 2.357567008,
"prefill_seconds_stddev": 0.001044130902,
"decode_seconds_mean": 15.62069287,
"decode_seconds_stddev": 0.004384481473,
"total_seconds_mean": 17.97908915,
"total_seconds_stddev": 0.005416429425,
"workspace_peak_bytes": 159981568,
"workspace_allocator_peak_bytes": 159981568,
"speculative": {
"backend": "none",
"enabled": false,
"draft_window": 0,
"rounds": 0,
"drafted_tokens": 0,
"accepted_tokens": 0,
"fallback_steps": 0,
"acceptance_rate": null,
"acceptance_length": null,
"accepted_per_position": []
},
"reps": [
{
"generated_output_tokens": 513,
"decode_output_tokens": 512,
"decode_engine_tokens": 512,
"timings": {"prepare_seconds": 2.877e-05, "vision_seconds": 0, "prefill_seconds": 2.356398688, "decode_seconds": 15.61569361, "total_seconds": 17.9729261},
"speculative": {
"backend": "none",
"enabled": false,
"draft_window": 0,
"rounds": 0,
"drafted_tokens": 0,
"accepted_tokens": 0,
"fallback_steps": 0,
"acceptance_rate": null,
"acceptance_length": null,
"accepted_per_position": []
}
},
{
"generated_output_tokens": 513,
"decode_output_tokens": 512,
"decode_engine_tokens": 512,
"timings": {"prepare_seconds": 2.874e-05, "vision_seconds": 0, "prefill_seconds": 2.357893328, "decode_seconds": 15.6225002, "total_seconds": 17.98124853},
"speculative": {
"backend": "none",
"enabled": false,
"draft_window": 0,
"rounds": 0,
"drafted_tokens": 0,
"accepted_tokens": 0,
"fallback_steps": 0,
"acceptance_rate": null,
"acceptance_length": null,
"accepted_per_position": []
}
},
{
"generated_output_tokens": 513,
"decode_output_tokens": 512,
"decode_engine_tokens": 512,
"timings": {"prepare_seconds": 2.916e-05, "vision_seconds": 0, "prefill_seconds": 2.358409007, "decode_seconds": 15.62388479, "total_seconds": 17.98309282},
"speculative": {
"backend": "none",
"enabled": false,
"draft_window": 0,
"rounds": 0,
"drafted_tokens": 0,
"accepted_tokens": 0,
"fallback_steps": 0,
"acceptance_rate": null,
"acceptance_length": null,
"accepted_per_position": []
}
}
]
}
]
}