{ "base_url": "http://127.0.0.1:39274/v1", "completed": 3, "concurrency": 1, "context_distribution": { "32768": 3 }, "context_seed": 0, "context_tokens": 32768, "engine": "sglang", "failed": 0, "failures": [], "finished_at": "2026-09-05T03:37:34Z", "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": 62, "content_chunks_s": null, "decode_tok_s_max": 111.17481605200948, "decode_tok_s_mean": 94.17808326841997, "decode_tok_s_min": 84.21256014155688, "decode_tok_s_p25": 84.21256014155688, "decode_tok_s_p50": 87.14687361169354, "decode_tok_s_p75": 111.17481605200948, "decode_tok_s_p90": 111.17481605200948, "decode_tok_s_p95": 111.17481605200948, "decode_tok_s_p99": 111.17481605200948, "decode_tok_s_samples": 3, "e2e_max_ms": 4855.882200063206, "e2e_mean_ms": 4501.613899987812, "e2e_min_ms": 4304.464699933305, "e2e_p25_ms": 4304.464699933305, "e2e_p50_ms": 4344.494799966924, "e2e_p75_ms": 4855.882200063206, "e2e_p90_ms": 4855.882200063206, "e2e_p95_ms": 4855.882200063206, "e2e_p99_ms": 4855.882200063206, "e2e_samples": 3, "effective_prefill_tok_s_max": 8227.630540946167, "effective_prefill_tok_s_mean": 7726.733795290881, "effective_prefill_tok_s_min": 6845.319266950189, "effective_prefill_tok_s_p25": 6845.319266950189, "effective_prefill_tok_s_p50": 8107.251577976287, "effective_prefill_tok_s_p75": 8227.630540946167, "effective_prefill_tok_s_p90": 8227.630540946167, "effective_prefill_tok_s_p95": 8227.630540946167, "effective_prefill_tok_s_p99": 8227.630540946167, "effective_prefill_tok_s_samples": 3, "generation_max_ms": 722.9174999520183, "generation_mean_ms": 618.662833274963, "generation_min_ms": 503.7111999699846, "generation_p25_ms": 503.7111999699846, "generation_p50_ms": 629.3597999028862, "generation_p75_ms": 722.9174999520183, "generation_p90_ms": 722.9174999520183, "generation_p95_ms": 722.9174999520183, "generation_p99_ms": 722.9174999520183, "generation_samples": 3, "mean_inter_token_latency_ms_max": 11.874713205714833, "mean_inter_token_latency_ms_mean": 10.781479004647013, "mean_inter_token_latency_ms_min": 8.994842856606867, "mean_inter_token_latency_ms_p25": 8.994842856606867, "mean_inter_token_latency_ms_p50": 11.474880951619337, "mean_inter_token_latency_ms_p75": 11.874713205714833, "mean_inter_token_latency_ms_p90": 11.874713205714833, "mean_inter_token_latency_ms_p95": 11.874713205714833, "mean_inter_token_latency_ms_p99": 11.874713205714833, "mean_inter_token_latency_ms_samples": 3, "output_token_sources": { "usage": 3 }, "output_tokens": 175, "prefix_cache_hit_avg": null, "prompt_token_samples": 3, "prompt_tokens": 89384, "prompt_tokens_max": 29797.0, "prompt_tokens_mean": 29794.666666666668, "prompt_tokens_min": 29792.0, "prompt_tokens_p25": 29792.0, "prompt_tokens_p50": 29795.0, "prompt_tokens_p75": 29797.0, "prompt_tokens_p90": 29797.0, "prompt_tokens_p95": 29797.0, "prompt_tokens_p99": 29797.0, "prompt_tokens_samples": 3, "reasoning_chunks": 0, "throughput_tok_s": 12.955850680814711, "time_to_first_output_max_ms": 4352.171000093222, "time_to_first_output_mean_ms": 3882.951066712849, "time_to_first_output_min_ms": 3621.5773000149056, "time_to_first_output_p25_ms": 3621.5773000149056, "time_to_first_output_p50_ms": 3675.1049000304192, "time_to_first_output_p75_ms": 4352.171000093222, "time_to_first_output_p90_ms": 4352.171000093222, "time_to_first_output_p95_ms": 4352.171000093222, "time_to_first_output_p99_ms": 4352.171000093222, "time_to_first_output_samples": 3, "tpot_max_ms": 11.874713205714833, "tpot_mean_ms": 10.781479004647013, "tpot_min_ms": 8.994842856606867, "tpot_p25_ms": 8.994842856606867, "tpot_p50_ms": 11.474880951619337, "tpot_p75_ms": 11.874713205714833, "tpot_p90_ms": 11.874713205714833, "tpot_p95_ms": 11.874713205714833, "tpot_p99_ms": 11.874713205714833, "tpot_samples": 3, "ttft_max_ms": 4352.171900100075, "ttft_mean_ms": 3882.9517333069816, "ttft_min_ms": 3621.5777999022976, "ttft_p25_ms": 3621.5777999022976, "ttft_p50_ms": 3675.1054999185726, "ttft_p75_ms": 4352.171900100075, "ttft_p90_ms": 4352.171900100075, "ttft_p95_ms": 4352.171900100075, "ttft_p99_ms": 4352.171900100075, "ttft_samples": 3 }, "model": "qwen38-27b-inferact-nvfp4-pro6000-dflash2-k12-compile", "request_timings": [ { "content_chunks": 17, "decode_tok_s": 111.17481605200948, "e2e_ms": 4855.882200063206, "effective_prefill_tok_s": 6845.319266950189, "generation_ms": 503.7111999699846, "mean_inter_token_latency_ms": 8.994842856606867, "output_token_source": "usage", "output_tokens": 57, "planned_context_tokens": 32768, "prompt_tokens": 29792, "reasoning_chunks": 0, "request_index": 0, "time_to_first_output_ms": 4352.171000093222, "tpot_ms": 8.994842856606867, "ttft_ms": 4352.171900100075, "visible_generation_ms": 503.71029996313155 }, { "content_chunks": 21, "decode_tok_s": 84.21256014155688, "e2e_ms": 4304.464699933305, "effective_prefill_tok_s": 8107.251577976287, "generation_ms": 629.3597999028862, "mean_inter_token_latency_ms": 11.874713205714833, "output_token_source": "usage", "output_tokens": 54, "planned_context_tokens": 32768, "prompt_tokens": 29795, "reasoning_chunks": 0, "request_index": 1, "time_to_first_output_ms": 3675.1049000304192, "tpot_ms": 11.874713205714833, "ttft_ms": 3675.1054999185726, "visible_generation_ms": 629.3592000147328 }, { "content_chunks": 24, "decode_tok_s": 87.14687361169354, "e2e_ms": 4344.494799966924, "effective_prefill_tok_s": 8227.630540946167, "generation_ms": 722.9174999520183, "mean_inter_token_latency_ms": 11.474880951619337, "output_token_source": "usage", "output_tokens": 64, "planned_context_tokens": 32768, "prompt_tokens": 29797, "reasoning_chunks": 0, "request_index": 2, "time_to_first_output_ms": 3621.5773000149056, "tpot_ms": 11.474880951619337, "ttft_ms": 3621.5777999022976, "visible_generation_ms": 722.9170000646263 } ], "requests": 3, "run_id": "benchmark-20260905T033720Z", "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": "2026-09-05T03:37:20Z", "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": 13507.41099996958 }