{ "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 -p 512,2048,8192 -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-exact16k-gpu0-r3.json", "environment": {"gpu_name": "NVIDIA RTX PRO 4000 Blackwell SFF Edition", "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": 3.100402713, "upload_seconds": 2.246084506, "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": "pp512", "kind": "pp", "n_prompt": 512, "n_gen": 0, "requested_output_tokens": 1, "prefill_tok_s_mean": 1975.39173, "prefill_tok_s_stddev": 2.38828023, "decode_output_tok_s_mean": null, "decode_output_tok_s_stddev": null, "decode_engine_tok_s_mean": null, "decode_engine_tok_s_stddev": null, "prepare_seconds_mean": 1.446666667e-06, "prepare_seconds_stddev": 3.477547028e-07, "prefill_seconds_mean": 0.2591893503, "prefill_seconds_stddev": 0.0003135025015, "decode_seconds_mean": null, "decode_seconds_stddev": null, "total_seconds_mean": 0.2592229067, "total_seconds_stddev": 0.0003021661325, "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": 1, "decode_output_tokens": null, "decode_engine_tokens": null, "timings": {"prepare_seconds": 1.84e-06, "vision_seconds": 0, "prefill_seconds": 0.258925011, "decode_seconds": 0, "total_seconds": 0.25897345}, "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": 1, "decode_output_tokens": null, "decode_engine_tokens": null, "timings": {"prepare_seconds": 1.32e-06, "vision_seconds": 0, "prefill_seconds": 0.25910733, "decode_seconds": 0, "total_seconds": 0.25913637}, "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": 1, "decode_output_tokens": null, "decode_engine_tokens": null, "timings": {"prepare_seconds": 1.18e-06, "vision_seconds": 0, "prefill_seconds": 0.25953571, "decode_seconds": 0, "total_seconds": 0.2595589}, "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", "kind": "pp", "n_prompt": 2048, "n_gen": 0, "requested_output_tokens": 1, "prefill_tok_s_mean": 2208.341671, "prefill_tok_s_stddev": 2.695990854, "decode_output_tok_s_mean": null, "decode_output_tok_s_stddev": null, "decode_engine_tok_s_mean": null, "decode_engine_tok_s_stddev": null, "prepare_seconds_mean": 7.096666667e-06, "prepare_seconds_stddev": 1.484362939e-07, "prefill_seconds_mean": 0.9273936463, "prefill_seconds_stddev": 0.001132072088, "decode_seconds_mean": null, "decode_seconds_stddev": null, "total_seconds_mean": 0.927424723, "total_seconds_stddev": 0.001130915645, "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": 1, "decode_output_tokens": null, "decode_engine_tokens": null, "timings": {"prepare_seconds": 6.97e-06, "vision_seconds": 0, "prefill_seconds": 0.926232878, "decode_seconds": 0, "total_seconds": 0.926265318}, "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": 1, "decode_output_tokens": null, "decode_engine_tokens": null, "timings": {"prepare_seconds": 7.26e-06, "vision_seconds": 0, "prefill_seconds": 0.927453406, "decode_seconds": 0, "total_seconds": 0.927484036}, "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": 1, "decode_output_tokens": null, "decode_engine_tokens": null, "timings": {"prepare_seconds": 7.06e-06, "vision_seconds": 0, "prefill_seconds": 0.928494655, "decode_seconds": 0, "total_seconds": 0.928524815}, "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", "kind": "pp", "n_prompt": 8192, "n_gen": 0, "requested_output_tokens": 1, "prefill_tok_s_mean": 2077.149766, "prefill_tok_s_stddev": 9.2611386, "decode_output_tok_s_mean": null, "decode_output_tok_s_stddev": null, "decode_engine_tok_s_mean": null, "decode_engine_tok_s_stddev": null, "prepare_seconds_mean": 2.819666667e-05, "prepare_seconds_stddev": 1.043567599e-06, "prefill_seconds_mean": 3.943918046, "prefill_seconds_stddev": 0.01756471008, "decode_seconds_mean": null, "decode_seconds_stddev": null, "total_seconds_mean": 3.944006923, "total_seconds_stddev": 0.01756820149, "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": 1, "decode_output_tokens": null, "decode_engine_tokens": null, "timings": {"prepare_seconds": 2.702e-05, "vision_seconds": 0, "prefill_seconds": 3.925049937, "decode_seconds": 0, "total_seconds": 3.925134537}, "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": 1, "decode_output_tokens": null, "decode_engine_tokens": null, "timings": {"prepare_seconds": 2.901e-05, "vision_seconds": 0, "prefill_seconds": 3.946908897, "decode_seconds": 0, "total_seconds": 3.946999787}, "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": 1, "decode_output_tokens": null, "decode_engine_tokens": null, "timings": {"prepare_seconds": 2.856e-05, "vision_seconds": 0, "prefill_seconds": 3.959795305, "decode_seconds": 0, "total_seconds": 3.959886445}, "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": "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": 20.41920607, "decode_output_tok_s_stddev": 0.09921701512, "decode_engine_tok_s_mean": 20.41920607, "decode_engine_tok_s_stddev": 0.09921701512, "prepare_seconds_mean": 1.693333333e-06, "prepare_seconds_stddev": 2.450170062e-07, "prefill_seconds_mean": null, "prefill_seconds_stddev": null, "decode_seconds_mean": 12.5374134, "decode_seconds_stddev": 0.06089365302, "total_seconds_mean": 12.58602146, "total_seconds_stddev": 0.06112850327, "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": 1.94e-06, "vision_seconds": 0, "prefill_seconds": 0.047897554, "decode_seconds": 12.47487263, "total_seconds": 12.52324359}, "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": 1.69e-06, "vision_seconds": 0, "prefill_seconds": 0.048181833, "decode_seconds": 12.54085348, "total_seconds": 12.58946582}, "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": 1.45e-06, "vision_seconds": 0, "prefill_seconds": 0.048394783, "decode_seconds": 12.59651409, "total_seconds": 12.64535495}, "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": 19.89475339, "decode_output_tok_s_stddev": 0.1495446743, "decode_engine_tok_s_mean": 19.89475339, "decode_engine_tok_s_stddev": 0.1495446743, "prepare_seconds_mean": 1.723333333e-06, "prepare_seconds_stddev": 3.592121008e-07, "prefill_seconds_mean": null, "prefill_seconds_stddev": null, "decode_seconds_mean": 51.47280398, "decode_seconds_stddev": 0.3885958408, "total_seconds_mean": 51.52387485, "total_seconds_stddev": 0.3886190529, "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": 1.31e-06, "vision_seconds": 0, "prefill_seconds": 0.049595022, "decode_seconds": 51.92149158, "total_seconds": 51.97259201}, "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": 1.96e-06, "vision_seconds": 0, "prefill_seconds": 0.049710342, "decode_seconds": 51.24442193, "total_seconds": 51.2957152}, "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": 1.9e-06, "vision_seconds": 0, "prefill_seconds": 0.049256332, "decode_seconds": 51.25249843, "total_seconds": 51.30331733}, "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": 2067.40482, "prefill_tok_s_stddev": 2.735276698, "decode_output_tok_s_mean": 19.89388005, "decode_output_tok_s_stddev": 0.01015019764, "decode_engine_tok_s_mean": 19.89388005, "decode_engine_tok_s_stddev": 0.01015019764, "prepare_seconds_mean": 1.122e-05, "prepare_seconds_stddev": 1.216552506e-07, "prefill_seconds_mean": 0.990615079, "prefill_seconds_stddev": 0.001310202558, "decode_seconds_mean": 25.73656258, "decode_seconds_stddev": 0.01313507293, "total_seconds_mean": 26.727997, "total_seconds_stddev": 0.01421923245, "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": 1.13e-05, "vision_seconds": 0, "prefill_seconds": 0.989207997, "decode_seconds": 25.72843637, "total_seconds": 26.71847092}, "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": 1.128e-05, "vision_seconds": 0, "prefill_seconds": 0.990837246, "decode_seconds": 25.72953499, "total_seconds": 26.72117872}, "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": 1.108e-05, "vision_seconds": 0, "prefill_seconds": 0.991799994, "decode_seconds": 25.75171639, "total_seconds": 26.74434136}, "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": 1978.625365, "prefill_tok_s_stddev": 0.8762491362, "decode_output_tok_s_mean": 19.67408967, "decode_output_tok_s_stddev": 0.006378937548, "decode_engine_tok_s_mean": 19.67408967, "decode_engine_tok_s_stddev": 0.006378937548, "prepare_seconds_mean": 2.876666667e-05, "prepare_seconds_stddev": 3.701801363e-07, "prefill_seconds_mean": 4.140248688, "prefill_seconds_stddev": 0.001833116877, "decode_seconds_mean": 26.02407758, "decode_seconds_stddev": 0.008438874743, "total_seconds_mean": 30.16520794, "total_seconds_stddev": 0.009293615679, "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": 2.839e-05, "vision_seconds": 0, "prefill_seconds": 4.141565419, "decode_seconds": 26.03346076, "total_seconds": 30.17589377}, "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": 2.878e-05, "vision_seconds": 0, "prefill_seconds": 4.141025587, "decode_seconds": 26.0171097, "total_seconds": 30.15901005}, "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": 2.913e-05, "vision_seconds": 0, "prefill_seconds": 4.138155059, "decode_seconds": 26.02166228, "total_seconds": 30.16071998}, "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": [] } } ] } ] }