{ "base_url": "http://127.0.0.1:39274/v1", "completed": 8, "concurrency": 4, "context_distribution": { "4096": 8 }, "context_seed": 0, "context_tokens": 4096, "engine": "sglang", "failed": 0, "failures": [], "finished_at": "2026-09-05T03:37:40Z", "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": 115, "content_chunks_s": null, "decode_tok_s_max": 117.34323190150998, "decode_tok_s_mean": 95.55897301619541, "decode_tok_s_min": 79.35843991784313, "decode_tok_s_p25": 79.36884571258221, "decode_tok_s_p50": 88.69349734714402, "decode_tok_s_p75": 104.48588374141326, "decode_tok_s_p90": 117.34323190150998, "decode_tok_s_p95": 117.34323190150998, "decode_tok_s_p99": 117.34323190150998, "decode_tok_s_samples": 8, "e2e_max_ms": 797.5308999884874, "e2e_mean_ms": 691.3071875023888, "e2e_min_ms": 588.6543999658898, "e2e_p25_ms": 616.1343000130728, "e2e_p50_ms": 652.0839999429882, "e2e_p75_ms": 764.9391000159085, "e2e_p90_ms": 797.5308999884874, "e2e_p95_ms": 797.5308999884874, "e2e_p99_ms": 797.5308999884874, "e2e_samples": 8, "effective_prefill_tok_s_max": 25006.86944526338, "effective_prefill_tok_s_mean": 19453.00004836384, "effective_prefill_tok_s_min": 16704.86484036529, "effective_prefill_tok_s_p25": 18743.554147012863, "effective_prefill_tok_s_p50": 18751.65756635652, "effective_prefill_tok_s_p75": 18795.916818459213, "effective_prefill_tok_s_p90": 25006.86944526338, "effective_prefill_tok_s_p95": 25006.86944526338, "effective_prefill_tok_s_p99": 25006.86944526338, "effective_prefill_tok_s_samples": 8, "generation_max_ms": 604.8506000079215, "generation_mean_ms": 503.38633748469874, "generation_min_ms": 409.056399948895, "generation_p25_ms": 423.48040000069886, "generation_p50_ms": 468.00759993493557, "generation_p75_ms": 572.7165000280365, "generation_p90_ms": 604.8506000079215, "generation_p95_ms": 604.8506000079215, "generation_p99_ms": 604.8506000079215, "generation_samples": 8, "mean_inter_token_latency_ms_max": 12.601054166831696, "mean_inter_token_latency_ms_mean": 10.683417834025366, "mean_inter_token_latency_ms_min": 8.522008332268646, "mean_inter_token_latency_ms_p25": 9.010221276610615, "mean_inter_token_latency_ms_p50": 9.957608509253948, "mean_inter_token_latency_ms_p75": 11.931593750584094, "mean_inter_token_latency_ms_p90": 12.601054166831696, "mean_inter_token_latency_ms_p95": 12.601054166831696, "mean_inter_token_latency_ms_p99": 12.601054166831696, "mean_inter_token_latency_ms_samples": 8, "output_token_sources": { "usage": 8 }, "output_tokens": 385, "prefix_cache_hit_avg": null, "prompt_token_samples": 8, "prompt_tokens": 28904, "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": 8, "reasoning_chunks": 0, "throughput_tok_s": 265.34014056206587, "time_to_first_output_max_ms": 216.28430008422583, "time_to_first_output_mean_ms": 187.9208500176901, "time_to_first_output_min_ms": 144.4803000194952, "time_to_first_output_p25_ms": 179.59800001699477, "time_to_first_output_p50_ms": 192.65390001237392, "time_to_first_output_p75_ms": 192.69179995171726, "time_to_first_output_p90_ms": 216.28430008422583, "time_to_first_output_p95_ms": 216.28430008422583, "time_to_first_output_p99_ms": 216.28430008422583, "time_to_first_output_samples": 8, "tpot_max_ms": 12.601054166831696, "tpot_mean_ms": 10.683417834025366, "tpot_min_ms": 8.522008332268646, "tpot_p25_ms": 9.010221276610615, "tpot_p50_ms": 9.957608509253948, "tpot_p75_ms": 11.931593750584094, "tpot_p90_ms": 12.601054166831696, "tpot_p95_ms": 12.601054166831696, "tpot_p99_ms": 12.601054166831696, "tpot_samples": 8, "ttft_max_ms": 216.28480008803308, "ttft_mean_ms": 187.92123752064072, "ttft_min_ms": 144.48060002177954, "ttft_p25_ms": 179.59870002232492, "ttft_p50_ms": 192.65410001389682, "ttft_p75_ms": 192.69219995476305, "ttft_p90_ms": 216.28480008803308, "ttft_p95_ms": 216.28480008803308, "ttft_p99_ms": 216.28480008803308, "ttft_samples": 8 }, "model": "qwen38-27b-inferact-nvfp4-pro6000-dflash2-k12-compile", "request_timings": [ { "content_chunks": 13, "decode_tok_s": 117.34323190150998, "e2e_ms": 588.6543999658898, "effective_prefill_tok_s": 20117.150523157907, "generation_ms": 409.056399948895, "mean_inter_token_latency_ms": 8.522008332268646, "output_token_source": "usage", "output_tokens": 49, "planned_context_tokens": 4096, "prompt_tokens": 3613, "reasoning_chunks": 0, "request_index": 0, "time_to_first_output_ms": 179.59800001699477, "tpot_ms": 8.522008332268646, "ttft_ms": 179.59870002232492, "visible_generation_ms": 409.05569994356483 }, { "content_chunks": 16, "decode_tok_s": 79.36884571258221, "e2e_ms": 797.5308999884874, "effective_prefill_tok_s": 18743.554147012863, "generation_ms": 604.7712999861687, "mean_inter_token_latency_ms": 12.599402083045183, "output_token_source": "usage", "output_tokens": 49, "planned_context_tokens": 4096, "prompt_tokens": 3613, "reasoning_chunks": 0, "request_index": 1, "time_to_first_output_ms": 192.75960000231862, "tpot_ms": 12.599402083045183, "ttft_ms": 192.75980000384152, "visible_generation_ms": 604.7710999846458 }, { "content_chunks": 16, "decode_tok_s": 79.35843991784313, "e2e_ms": 797.5269000744447, "effective_prefill_tok_s": 18751.65756635652, "generation_ms": 604.8506000079215, "mean_inter_token_latency_ms": 12.601054166831696, "output_token_source": "usage", "output_tokens": 49, "planned_context_tokens": 4096, "prompt_tokens": 3613, "reasoning_chunks": 0, "request_index": 2, "time_to_first_output_ms": 192.67630006652325, "tpot_ms": 12.601054166831696, "ttft_ms": 192.67650006804615, "visible_generation_ms": 604.8504000063986 }, { "content_chunks": 15, "decode_tok_s": 83.81110025230673, "e2e_ms": 764.9391000159085, "effective_prefill_tok_s": 18795.916818459213, "generation_ms": 572.7165000280365, "mean_inter_token_latency_ms": 11.931593750584094, "output_token_source": "usage", "output_tokens": 49, "planned_context_tokens": 4096, "prompt_tokens": 3613, "reasoning_chunks": 0, "request_index": 3, "time_to_first_output_ms": 192.22259998787194, "tpot_ms": 11.931593750584094, "ttft_ms": 192.22319999244064, "visible_generation_ms": 572.7159000234678 }, { "content_chunks": 11, "decode_tok_s": 88.69349734714402, "e2e_ms": 629.2959999991581, "effective_prefill_tok_s": 25006.86944526338, "generation_ms": 484.8156999796629, "mean_inter_token_latency_ms": 11.274783720457277, "output_token_source": "usage", "output_tokens": 44, "planned_context_tokens": 4096, "prompt_tokens": 3613, "reasoning_chunks": 0, "request_index": 4, "time_to_first_output_ms": 144.4803000194952, "tpot_ms": 11.274783720457277, "ttft_ms": 144.48060002177954, "visible_generation_ms": 484.81539997737855 }, { "content_chunks": 15, "decode_tok_s": 100.42571959629319, "e2e_ms": 684.2919000191614, "effective_prefill_tok_s": 16704.86484036529, "generation_ms": 468.00759993493557, "mean_inter_token_latency_ms": 9.957608509253948, "output_token_source": "usage", "output_tokens": 48, "planned_context_tokens": 4096, "prompt_tokens": 3613, "reasoning_chunks": 0, "request_index": 5, "time_to_first_output_ms": 216.28430008422583, "tpot_ms": 9.957608509253948, "ttft_ms": 216.28480008803308, "visible_generation_ms": 468.0070999311283 }, { "content_chunks": 15, "decode_tok_s": 104.48588374141326, "e2e_ms": 652.0839999429882, "effective_prefill_tok_s": 18750.149206688133, "generation_ms": 459.3921999912709, "mean_inter_token_latency_ms": 9.570670833151476, "output_token_source": "usage", "output_tokens": 49, "planned_context_tokens": 4096, "prompt_tokens": 3613, "reasoning_chunks": 0, "request_index": 6, "time_to_first_output_ms": 192.69179995171726, "tpot_ms": 9.570670833151476, "ttft_ms": 192.69219995476305, "visible_generation_ms": 459.3917999882251 }, { "content_chunks": 14, "decode_tok_s": 110.9850656604708, "e2e_ms": 616.1343000130728, "effective_prefill_tok_s": 18753.837839607408, "generation_ms": 423.48040000069886, "mean_inter_token_latency_ms": 9.010221276610615, "output_token_source": "usage", "output_tokens": 48, "planned_context_tokens": 4096, "prompt_tokens": 3613, "reasoning_chunks": 0, "request_index": 7, "time_to_first_output_ms": 192.65390001237392, "tpot_ms": 9.010221276610615, "ttft_ms": 192.65410001389682, "visible_generation_ms": 423.48019999917597 } ], "requests": 8, "run_id": "benchmark-20260905T033738Z", "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": "2026-09-05T03:37:38Z", "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": 1450.9677999885753 }