{ "base_url": "http://127.0.0.1:39276/v1", "completed": 5, "concurrency": 1, "context_distribution": { "4096": 5 }, "context_seed": 0, "context_tokens": 4096, "engine": "sglang", "failed": 0, "failures": [], "finished_at": "2026-09-05T03:46:14Z", "gpu": "Primary Node / RTX PRO 6000 Blackwell Max-Q / TP=1", "manifest": null, "max_context_tokens": 262144, "max_tokens": 512, "measurement_protocol": "capacity-v3", "metrics": { "content_chunks": 67, "content_chunks_s": null, "decode_tok_s_max": 120.85844470525961, "decode_tok_s_mean": 112.02222918570547, "decode_tok_s_min": 101.28051597749446, "decode_tok_s_p25": 106.54915004467354, "decode_tok_s_p50": 110.9734043997807, "decode_tok_s_p75": 120.449630801319, "decode_tok_s_p90": 120.85844470525961, "decode_tok_s_p95": 120.85844470525961, "decode_tok_s_p99": 120.85844470525961, "decode_tok_s_samples": 5, "e2e_max_ms": 919.3574000382796, "e2e_mean_ms": 592.9406800074503, "e2e_min_ms": 495.76139997225255, "e2e_p25_ms": 498.9711999660358, "e2e_p50_ms": 525.2023000502959, "e2e_p75_ms": 525.4111000103876, "e2e_p90_ms": 919.3574000382796, "e2e_p95_ms": 919.3574000382796, "e2e_p99_ms": 919.3574000382796, "e2e_samples": 5, "effective_prefill_tok_s_max": 35461.1321188502, "effective_prefill_tok_s_mean": 29075.21766952232, "effective_prefill_tok_s_min": 7766.997804796798, "effective_prefill_tok_s_p25": 33217.91799324665, "effective_prefill_tok_s_p50": 33805.31020955347, "effective_prefill_tok_s_p75": 35124.730221164486, "effective_prefill_tok_s_p90": 35461.1321188502, "effective_prefill_tok_s_p95": 35461.1321188502, "effective_prefill_tok_s_p99": 35461.1321188502, "effective_prefill_tok_s_samples": 5, "generation_max_ms": 454.1841000318527, "generation_mean_ms": 415.8277199836448, "generation_min_ms": 388.88469990342855, "generation_p25_ms": 390.2045999420807, "generation_p50_ms": 422.3403000505641, "generation_p75_ms": 423.52489999029785, "generation_p90_ms": 454.1841000318527, "generation_p95_ms": 454.1841000318527, "generation_p99_ms": 454.1841000318527, "generation_samples": 5, "mean_inter_token_latency_ms_max": 9.873567391996799, "mean_inter_token_latency_ms_mean": 8.96928871196795, "mean_inter_token_latency_ms_min": 8.274142551136778, "mean_inter_token_latency_ms_p25": 8.302225530682568, "mean_inter_token_latency_ms_p50": 9.011168084899955, "mean_inter_token_latency_ms_p75": 9.385340001123646, "mean_inter_token_latency_ms_p90": 9.873567391996799, "mean_inter_token_latency_ms_p95": 9.873567391996799, "mean_inter_token_latency_ms_p99": 9.873567391996799, "mean_inter_token_latency_ms_samples": 5, "output_token_sources": { "usage": 5 }, "output_tokens": 237, "prefix_cache_hit_avg": null, "prompt_token_samples": 5, "prompt_tokens": 18065, "prompt_tokens_max": 3613.0, "prompt_tokens_mean": 3613.0, "prompt_tokens_min": 3613.0, "prompt_tokens_p25": 3613.0, "prompt_tokens_p50": 3613.0, "prompt_tokens_p75": 3613.0, "prompt_tokens_p90": 3613.0, "prompt_tokens_p95": 3613.0, "prompt_tokens_p99": 3613.0, "prompt_tokens_samples": 5, "reasoning_chunks": 0, "throughput_tok_s": 79.91082087435166, "time_to_first_output_max_ms": 465.17330000642687, "time_to_first_output_mean_ms": 177.1129600238055, "time_to_first_output_min_ms": 101.88620002008975, "time_to_first_output_p25_ms": 102.86199999973178, "time_to_first_output_p50_ms": 106.876700068824, "time_to_first_output_p75_ms": 108.7666000239551, "time_to_first_output_p90_ms": 465.17330000642687, "time_to_first_output_p95_ms": 465.17330000642687, "time_to_first_output_p99_ms": 465.17330000642687, "time_to_first_output_samples": 5, "tpot_max_ms": 9.873567391996799, "tpot_mean_ms": 8.96928871196795, "tpot_min_ms": 8.274142551136778, "tpot_p25_ms": 8.302225530682568, "tpot_p50_ms": 9.011168084899955, "tpot_p75_ms": 9.385340001123646, "tpot_p90_ms": 9.873567391996799, "tpot_p95_ms": 9.873567391996799, "tpot_p99_ms": 9.873567391996799, "tpot_samples": 5, "ttft_max_ms": 465.1738000102341, "ttft_mean_ms": 177.11332000326365, "ttft_min_ms": 101.88660002313554, "ttft_p25_ms": 102.86220000125468, "ttft_p50_ms": 106.87689995393157, "ttft_p75_ms": 108.76710002776235, "ttft_p90_ms": 465.1738000102341, "ttft_p95_ms": 465.1738000102341, "ttft_p99_ms": 465.1738000102341, "ttft_samples": 5 }, "model": "qwen38-27b-inferact-nvfp4-pro6000-dflash2-k12-chunk1k-mamba40", "request_timings": [ { "content_chunks": 15, "decode_tok_s": 101.28051597749446, "e2e_ms": 919.3574000382796, "effective_prefill_tok_s": 7766.997804796798, "generation_ms": 454.1841000318527, "mean_inter_token_latency_ms": 9.873567391996799, "output_token_source": "usage", "output_tokens": 47, "planned_context_tokens": 4096, "prompt_tokens": 3613, "reasoning_chunks": 0, "request_index": 0, "time_to_first_output_ms": 465.17330000642687, "tpot_ms": 9.873567391996799, "ttft_ms": 465.1738000102341, "visible_generation_ms": 454.1836000280455 }, { "content_chunks": 13, "decode_tok_s": 120.449630801319, "e2e_ms": 498.9711999660358, "effective_prefill_tok_s": 33217.91799324665, "generation_ms": 390.2045999420807, "mean_inter_token_latency_ms": 8.302225530682568, "output_token_source": "usage", "output_tokens": 48, "planned_context_tokens": 4096, "prompt_tokens": 3613, "reasoning_chunks": 0, "request_index": 1, "time_to_first_output_ms": 108.7666000239551, "tpot_ms": 8.302225530682568, "ttft_ms": 108.76710002776235, "visible_generation_ms": 390.20409993827343 }, { "content_chunks": 13, "decode_tok_s": 106.54915004467354, "e2e_ms": 525.2023000502959, "effective_prefill_tok_s": 35124.730221164486, "generation_ms": 422.3403000505641, "mean_inter_token_latency_ms": 9.385340001123646, "output_token_source": "usage", "output_tokens": 46, "planned_context_tokens": 4096, "prompt_tokens": 3613, "reasoning_chunks": 0, "request_index": 2, "time_to_first_output_ms": 102.86199999973178, "tpot_ms": 9.385340001123646, "ttft_ms": 102.86220000125468, "visible_generation_ms": 422.3401000490412 }, { "content_chunks": 13, "decode_tok_s": 110.9734043997807, "e2e_ms": 525.4111000103876, "effective_prefill_tok_s": 35461.1321188502, "generation_ms": 423.52489999029785, "mean_inter_token_latency_ms": 9.011168084899955, "output_token_source": "usage", "output_tokens": 48, "planned_context_tokens": 4096, "prompt_tokens": 3613, "reasoning_chunks": 0, "request_index": 3, "time_to_first_output_ms": 101.88620002008975, "tpot_ms": 9.011168084899955, "ttft_ms": 101.88660002313554, "visible_generation_ms": 423.52449998725206 }, { "content_chunks": 13, "decode_tok_s": 120.85844470525961, "e2e_ms": 495.76139997225255, "effective_prefill_tok_s": 33805.31020955347, "generation_ms": 388.88469990342855, "mean_inter_token_latency_ms": 8.274142551136778, "output_token_source": "usage", "output_tokens": 48, "planned_context_tokens": 4096, "prompt_tokens": 3613, "reasoning_chunks": 0, "request_index": 4, "time_to_first_output_ms": 106.876700068824, "tpot_ms": 8.274142551136778, "ttft_ms": 106.87689995393157, "visible_generation_ms": 388.884500018321 } ], "requests": 5, "run_id": "benchmark-20260905T034611Z", "schema": "anvil-serving.benchmark/v1", "serve_flags": { "no_thinking": false, "reasoning_effort": null, "shared_prefix_burst": false, "thinking_mode": "disabled" }, "source_recipe": null, "started_at": "2026-09-05T03:46:11Z", "tier": null, "timing_methodology": { "clock": "client time.perf_counter", "decode": "usage completion tokens after the first token divided by client-observed generation time; completion tokens may include reasoning tokens", "distributions": "request-level mean, min, p25, p50, p75, p90, p95, p99, and max using the nearest-rank percentile method", "effective_prefill": "usage.prompt_tokens divided by client-observed time to first output; includes queueing, scheduling, prefill, and first-token work", "generation": "client-observed E2E minus time to first output", "mean_inter_token_latency": "client-observed generation time divided by completion tokens after the first token; not a raw per-token timestamp trace", "time_to_first_output": "request start through the first non-empty streamed reasoning or content delta", "tpot": "alias of mean_inter_token_latency for capacity-v3: client-observed generation time divided by completion tokens after the first token", "ttft": "request start through the first non-empty streamed content delta" }, "wall_clock_ms": 2965.8060999354348 }