{ "schema": "anvil-serving.benchmark-decision-summary/v1", "campaign_id": "2026-10-21-flare-inference-diagnostics", "complete": false, "wave_closed": true, "rows": [ { "name": "incumbent-before-c1-prime", "arm": "incumbent-before", "concurrency": 1, "repetition": "prime", "attempted": 52, "completed": 52, "failed": 0, "wall_ms": 118423.75193792395, "metrics": { "completion_tokens_max": 511.0, "completion_tokens_mean": 271.71153846153845, "completion_tokens_mean_ci95_high": 294.9441548961937, "completion_tokens_mean_ci95_low": 248.47892202688317, "completion_tokens_min": 123.0, "completion_tokens_p25": 199.0, "completion_tokens_p50": 261.0, "completion_tokens_p75": 310.0, "completion_tokens_p90": 388.0, "completion_tokens_p95": 443.0, "completion_tokens_p99": 511.0, "completion_tokens_samples": 52, "completion_tokens_stddev": 85.47590797789486, "content_chunks": 998, "content_chunks_s": null, "controlled_output_exact_adherence": 0, "controlled_output_exact_adherence_rate": null, "controlled_output_samples": 0, "controlled_output_unobservable": 0, "decode_tok_s_max": 217.1415587120076, "decode_tok_s_mean": 191.0658705953232, "decode_tok_s_mean_ci95_high": 193.92971405161035, "decode_tok_s_mean_ci95_low": 188.20202713903603, "decode_tok_s_min": 170.40824792056247, "decode_tok_s_p25": 184.29549946018517, "decode_tok_s_p50": 190.92309358728758, "decode_tok_s_p75": 196.73754324290587, "decode_tok_s_p90": 203.8784158035637, "decode_tok_s_p95": 212.2725287651881, "decode_tok_s_p99": 217.1415587120076, "decode_tok_s_samples": 52, "decode_tok_s_stddev": 10.536463700556535, "e2e_max_ms": 3374.4164230301976, "e2e_mean_ci95_high_ms": 2393.593406340487, "e2e_mean_ci95_low_ms": 2159.649921308071, "e2e_mean_ms": 2276.621663824279, "e2e_min_ms": 1486.192912911065, "e2e_p25_ms": 1966.2409969605505, "e2e_p50_ms": 2202.74304097984, "e2e_p75_ms": 2457.2736389236525, "e2e_p90_ms": 2904.1201189393178, "e2e_p95_ms": 3306.0056299436837, "e2e_p99_ms": 3374.4164230301976, "e2e_samples": 52, "e2e_stddev_ms": 430.3547096148562, "effective_prefill_tok_s_max": 9245.645572144542, "effective_prefill_tok_s_mean": 8925.85387387029, "effective_prefill_tok_s_mean_ci95_high": 8978.697307187083, "effective_prefill_tok_s_mean_ci95_low": 8873.010440553497, "effective_prefill_tok_s_min": 8650.384070761735, "effective_prefill_tok_s_p25": 8762.011702279337, "effective_prefill_tok_s_p50": 8858.321369878684, "effective_prefill_tok_s_p75": 9083.456173498347, "effective_prefill_tok_s_p90": 9229.441636401518, "effective_prefill_tok_s_p95": 9238.398952842388, "effective_prefill_tok_s_p99": 9245.645572144542, "effective_prefill_tok_s_samples": 52, "effective_prefill_tok_s_stddev": 194.41806979107898, "generation_max_ms": 2536.833217018284, "generation_mean_ci95_high_ms": 1531.8710774246558, "generation_mean_ci95_low_ms": 1297.3593225657266, "generation_mean_ms": 1414.6151999951912, "generation_min_ms": 617.8568389732391, "generation_p25_ms": 1094.6214459836483, "generation_p50_ms": 1326.3043519109488, "generation_p75_ms": 1592.7171770017594, "generation_p90_ms": 2062.319192918949, "generation_p95_ms": 2419.382034917362, "generation_p99_ms": 2536.833217018284, "generation_samples": 52, "generation_stddev_ms": 431.4000800218932, "mean_inter_token_latency_ms_max": 5.868260557823235, "mean_inter_token_latency_ms_mean": 5.249376301265809, "mean_inter_token_latency_ms_mean_ci95_high": 5.327892427122476, "mean_inter_token_latency_ms_mean_ci95_low": 5.170860175409141, "mean_inter_token_latency_ms_min": 4.605290695763536, "mean_inter_token_latency_ms_p25": 5.072996595266218, "mean_inter_token_latency_ms_p50": 5.219735554419458, "mean_inter_token_latency_ms_p75": 5.410059178402625, "mean_inter_token_latency_ms_p90": 5.583006559382911, "mean_inter_token_latency_ms_p95": 5.833771620040798, "mean_inter_token_latency_ms_p99": 5.868260557823235, "mean_inter_token_latency_ms_samples": 52, "mean_inter_token_latency_ms_stddev": 0.2888713446193858, "mean_time_per_output_token_max_ms": 5.868260557823235, "mean_time_per_output_token_mean_ci95_high_ms": 5.327892427122476, "mean_time_per_output_token_mean_ci95_low_ms": 5.170860175409141, "mean_time_per_output_token_mean_ms": 5.249376301265809, "mean_time_per_output_token_min_ms": 4.605290695763536, "mean_time_per_output_token_p25_ms": 5.072996595266218, "mean_time_per_output_token_p50_ms": 5.219735554419458, "mean_time_per_output_token_p75_ms": 5.410059178402625, "mean_time_per_output_token_p90_ms": 5.583006559382911, "mean_time_per_output_token_p95_ms": 5.833771620040798, "mean_time_per_output_token_p99_ms": 5.868260557823235, "mean_time_per_output_token_samples": 52, "mean_time_per_output_token_stddev_ms": 0.2888713446193858, "output_token_sources": { "usage": 52 }, "output_tokens": 14129, "prefix_cache_hit_avg": 0.0, "prompt_token_samples": 52, "prompt_tokens": 399899, "prompt_tokens_max": 7708.0, "prompt_tokens_mean": 7690.365384615385, "prompt_tokens_mean_ci95_high": 7693.3501411136585, "prompt_tokens_mean_ci95_low": 7687.380628117111, "prompt_tokens_min": 7675.0, "prompt_tokens_p25": 7680.0, "prompt_tokens_p50": 7688.0, "prompt_tokens_p75": 7701.0, "prompt_tokens_p90": 7705.0, "prompt_tokens_p95": 7707.0, "prompt_tokens_p99": 7708.0, "prompt_tokens_samples": 52, "prompt_tokens_stddev": 10.981318978876677, "reasoning_chunks": 3918, "request_canary_passed": 0, "request_canary_samples": 0, "throughput_performance_eligible": true, "throughput_tok_s": 119.30883601294967, "time_to_first_output_max_ms": 890.9430999774486, "time_to_first_output_mean_ci95_high_ms": 867.3949425773293, "time_to_first_output_mean_ci95_low_ms": 856.6179850808461, "time_to_first_output_mean_ms": 862.0064638290877, "time_to_first_output_min_ms": 830.1205080933869, "time_to_first_output_p25_ms": 842.3445200314745, "time_to_first_output_p50_ms": 864.994402974844, "time_to_first_output_p75_ms": 878.1697859521955, "time_to_first_output_p90_ms": 885.2243819274008, "time_to_first_output_p95_ms": 887.4645220348611, "time_to_first_output_p99_ms": 890.9430999774486, "time_to_first_output_samples": 52, "time_to_first_output_stddev_ms": 19.82493512605414, "tpot_max_ms": 5.868260557823235, "tpot_mean_ci95_high_ms": 5.327892427122476, "tpot_mean_ci95_low_ms": 5.170860175409141, "tpot_mean_ms": 5.249376301265809, "tpot_min_ms": 4.605290695763536, "tpot_p25_ms": 5.072996595266218, "tpot_p50_ms": 5.219735554419458, "tpot_p75_ms": 5.410059178402625, "tpot_p90_ms": 5.583006559382911, "tpot_p95_ms": 5.833771620040798, "tpot_p99_ms": 5.868260557823235, "tpot_samples": 52, "tpot_stddev_ms": 0.2888713446193858, "ttft_max_ms": 3104.2221901006997, "ttft_mean_ci95_high_ms": 2096.022363295185, "ttft_mean_ci95_low_ms": 1869.8094086410463, "ttft_mean_ms": 1982.9158859681158, "ttft_min_ms": 1213.1594179663807, "ttft_p25_ms": 1684.423672966659, "ttft_p50_ms": 1927.9364970279858, "ttft_p75_ms": 2140.088113024831, "ttft_p90_ms": 2557.28344200179, "ttft_p95_ms": 2953.112500021234, "ttft_p99_ms": 3104.2221901006997, "ttft_samples": 52, "ttft_stddev_ms": 416.13388121413686 }, "actual_start_unix_ns": 1792644667181625837, "actual_end_unix_ns": 1792644785605382414 }, { "name": "incumbent-before-c1-r1", "arm": "incumbent-before", "concurrency": 1, "repetition": "r1", "attempted": 52, "completed": 52, "failed": 0, "wall_ms": 124747.8433289798, "metrics": { "completion_tokens_max": 550.0, "completion_tokens_mean": 283.86538461538464, "completion_tokens_mean_ci95_high": 308.0869504710491, "completion_tokens_mean_ci95_low": 259.6438187597202, "completion_tokens_min": 130.0, "completion_tokens_p25": 217.0, "completion_tokens_p50": 275.0, "completion_tokens_p75": 308.0, "completion_tokens_p90": 429.0, "completion_tokens_p95": 448.0, "completion_tokens_p99": 550.0, "completion_tokens_samples": 52, "completion_tokens_stddev": 89.11438537206733, "content_chunks": 960, "content_chunks_s": null, "controlled_output_exact_adherence": 0, "controlled_output_exact_adherence_rate": null, "controlled_output_samples": 0, "controlled_output_unobservable": 0, "decode_tok_s_max": 205.64425215567385, "decode_tok_s_mean": 188.10839690127074, "decode_tok_s_mean_ci95_high": 190.6335803202845, "decode_tok_s_mean_ci95_low": 185.58321348225698, "decode_tok_s_min": 162.13698836902057, "decode_tok_s_p25": 182.36467386215634, "decode_tok_s_p50": 189.26295916528753, "decode_tok_s_p75": 196.45067102328778, "decode_tok_s_p90": 199.06589429204902, "decode_tok_s_p95": 200.2108132881529, "decode_tok_s_p99": 205.64425215567385, "decode_tok_s_samples": 52, "decode_tok_s_stddev": 9.290488058372992, "e2e_max_ms": 3649.134580977261, "e2e_mean_ci95_high_ms": 2526.628081892079, "e2e_mean_ci95_low_ms": 2269.887655450247, "e2e_mean_ms": 2398.257868671163, "e2e_min_ms": 1556.104553048499, "e2e_p25_ms": 2079.255662043579, "e2e_p50_ms": 2365.0113940238953, "e2e_p75_ms": 2545.2415060717613, "e2e_p90_ms": 3166.8293320108205, "e2e_p95_ms": 3425.1413649180904, "e2e_p99_ms": 3649.134580977261, "e2e_samples": 52, "e2e_stddev_ms": 472.29121021454876, "effective_prefill_tok_s_max": 8694.381131640846, "effective_prefill_tok_s_mean": 8629.446737245304, "effective_prefill_tok_s_mean_ci95_high": 8636.34810931114, "effective_prefill_tok_s_mean_ci95_low": 8622.545365179467, "effective_prefill_tok_s_min": 8571.817252732886, "effective_prefill_tok_s_p25": 8611.924513455768, "effective_prefill_tok_s_p50": 8625.80527763321, "effective_prefill_tok_s_p75": 8642.214594401361, "effective_prefill_tok_s_p90": 8669.732195230801, "effective_prefill_tok_s_p95": 8673.755508704413, "effective_prefill_tok_s_p99": 8694.381131640846, "effective_prefill_tok_s_samples": 52, "effective_prefill_tok_s_stddev": 25.391072300435034, "generation_max_ms": 2757.1136999176815, "generation_mean_ci95_high_ms": 1635.5069494261147, "generation_mean_ci95_low_ms": 1378.635564453496, "generation_mean_ms": 1507.0712569398054, "generation_min_ms": 661.6649420466274, "generation_p25_ms": 1185.520889936015, "generation_p50_ms": 1470.782942022197, "generation_p75_ms": 1649.9780560843647, "generation_p90_ms": 2274.6461960487068, "generation_p95_ms": 2533.930707955733, "generation_p99_ms": 2757.1136999176815, "generation_samples": 52, "generation_stddev_ms": 472.53211720317665, "mean_inter_token_latency_ms_max": 6.167624118711394, "mean_inter_token_latency_ms_mean": 5.329209475301058, "mean_inter_token_latency_ms_mean_ci95_high": 5.403012123800469, "mean_inter_token_latency_ms_mean_ci95_low": 5.2554068268016465, "mean_inter_token_latency_ms_min": 4.862766595795707, "mean_inter_token_latency_ms_p25": 5.084913084399307, "mean_inter_token_latency_ms_p50": 5.271623448108233, "mean_inter_token_latency_ms_p75": 5.464014083989998, "mean_inter_token_latency_ms_p90": 5.668748787373004, "mean_inter_token_latency_ms_p95": 5.864113492373405, "mean_inter_token_latency_ms_p99": 6.167624118711394, "mean_inter_token_latency_ms_samples": 52, "mean_inter_token_latency_ms_stddev": 0.271529830030279, "mean_time_per_output_token_max_ms": 6.167624118711394, "mean_time_per_output_token_mean_ci95_high_ms": 5.403012123800469, "mean_time_per_output_token_mean_ci95_low_ms": 5.2554068268016465, "mean_time_per_output_token_mean_ms": 5.329209475301058, "mean_time_per_output_token_min_ms": 4.862766595795707, "mean_time_per_output_token_p25_ms": 5.084913084399307, "mean_time_per_output_token_p50_ms": 5.271623448108233, "mean_time_per_output_token_p75_ms": 5.464014083989998, "mean_time_per_output_token_p90_ms": 5.668748787373004, "mean_time_per_output_token_p95_ms": 5.864113492373405, "mean_time_per_output_token_p99_ms": 6.167624118711394, "mean_time_per_output_token_samples": 52, "mean_time_per_output_token_stddev_ms": 0.271529830030279, "output_token_sources": { "usage": 52 }, "output_tokens": 14761, "prefix_cache_hit_avg": 0.0, "prompt_token_samples": 52, "prompt_tokens": 399899, "prompt_tokens_max": 7708.0, "prompt_tokens_mean": 7690.365384615385, "prompt_tokens_mean_ci95_high": 7693.3501411136585, "prompt_tokens_mean_ci95_low": 7687.380628117111, "prompt_tokens_min": 7675.0, "prompt_tokens_p25": 7680.0, "prompt_tokens_p50": 7688.0, "prompt_tokens_p75": 7701.0, "prompt_tokens_p90": 7705.0, "prompt_tokens_p95": 7707.0, "prompt_tokens_p99": 7708.0, "prompt_tokens_samples": 52, "prompt_tokens_stddev": 10.981318978876677, "reasoning_chunks": 4153, "request_canary_passed": 0, "request_canary_samples": 0, "throughput_performance_eligible": true, "throughput_tok_s": 118.32669492387862, "time_to_first_output_max_ms": 897.9426150908694, "time_to_first_output_mean_ci95_high_ms": 892.1284655146617, "time_to_first_output_mean_ci95_low_ms": 890.2447579480536, "time_to_first_output_mean_ms": 891.1866117313576, "time_to_first_output_min_ms": 882.7540320344269, "time_to_first_output_p25_ms": 888.7244560755789, "time_to_first_output_p50_ms": 891.4197089616209, "time_to_first_output_p75_ms": 893.4851429658011, "time_to_first_output_p90_ms": 895.5448750639334, "time_to_first_output_p95_ms": 895.9765649633482, "time_to_first_output_p99_ms": 897.9426150908694, "time_to_first_output_samples": 52, "time_to_first_output_stddev_ms": 3.4652062343803136, "tpot_max_ms": 6.167624118711394, "tpot_mean_ci95_high_ms": 5.403012123800469, "tpot_mean_ci95_low_ms": 5.2554068268016465, "tpot_mean_ms": 5.329209475301058, "tpot_min_ms": 4.862766595795707, "tpot_p25_ms": 5.084913084399307, "tpot_p50_ms": 5.271623448108233, "tpot_p75_ms": 5.464014083989998, "tpot_p90_ms": 5.668748787373004, "tpot_p95_ms": 5.864113492373405, "tpot_p99_ms": 6.167624118711394, "tpot_samples": 52, "tpot_stddev_ms": 0.271529830030279, "ttft_max_ms": 3385.437272954732, "ttft_mean_ci95_high_ms": 2234.6421365071255, "ttft_mean_ci95_low_ms": 1977.321674916504, "ttft_mean_ms": 2105.9819057118148, "ttft_min_ms": 1292.485455982387, "ttft_p25_ms": 1762.34822999686, "ttft_p50_ms": 2060.251028975472, "ttft_p75_ms": 2301.8081249902025, "ttft_p90_ms": 2947.548507945612, "ttft_p95_ms": 3147.705007926561, "ttft_p99_ms": 3385.437272954732, "ttft_samples": 52, "ttft_stddev_ms": 473.3582237199223 }, "actual_start_unix_ns": 1792644785973139632, "actual_end_unix_ns": 1792644910720985098 }, { "name": "incumbent-before-c1-r2", "arm": "incumbent-before", "concurrency": 1, "repetition": "r2", "attempted": 52, "completed": 52, "failed": 0, "wall_ms": 120245.3136119293, "metrics": { "completion_tokens_max": 486.0, "completion_tokens_mean": 270.25, "completion_tokens_mean_ci95_high": 293.237377843518, "completion_tokens_mean_ci95_low": 247.26262215648197, "completion_tokens_min": 57.0, "completion_tokens_p25": 216.0, "completion_tokens_p50": 267.0, "completion_tokens_p75": 325.0, "completion_tokens_p90": 375.0, "completion_tokens_p95": 405.0, "completion_tokens_p99": 486.0, "completion_tokens_samples": 52, "completion_tokens_stddev": 84.57364235027457, "content_chunks": 934, "content_chunks_s": null, "controlled_output_exact_adherence": 0, "controlled_output_exact_adherence_rate": null, "controlled_output_samples": 0, "controlled_output_unobservable": 0, "decode_tok_s_max": 213.81462968854476, "decode_tok_s_mean": 188.99668318001474, "decode_tok_s_mean_ci95_high": 192.027586293262, "decode_tok_s_mean_ci95_low": 185.96578006676748, "decode_tok_s_min": 144.3232640062857, "decode_tok_s_p25": 184.08434189339684, "decode_tok_s_p50": 190.65026445904357, "decode_tok_s_p75": 194.70150875780962, "decode_tok_s_p90": 201.63775587525603, "decode_tok_s_p95": 205.62676171285503, "decode_tok_s_p99": 213.81462968854476, "decode_tok_s_samples": 52, "decode_tok_s_stddev": 11.15109855691476, "e2e_max_ms": 3427.975403959863, "e2e_mean_ci95_high_ms": 2433.0260836769257, "e2e_mean_ci95_low_ms": 2189.789993864061, "e2e_mean_ms": 2311.4080387704935, "e2e_min_ms": 1271.9210219802335, "e2e_p25_ms": 1994.724771939218, "e2e_p50_ms": 2273.080746992491, "e2e_p75_ms": 2575.9935670066625, "e2e_p90_ms": 2874.513532035053, "e2e_p95_ms": 3022.8046079864725, "e2e_p99_ms": 3427.975403959863, "e2e_samples": 52, "e2e_stddev_ms": 447.44907850186104, "effective_prefill_tok_s_max": 8723.38103129467, "effective_prefill_tok_s_mean": 8663.948725518472, "effective_prefill_tok_s_mean_ci95_high": 8671.702121609898, "effective_prefill_tok_s_mean_ci95_low": 8656.195329427046, "effective_prefill_tok_s_min": 8592.960112029066, "effective_prefill_tok_s_p25": 8642.383791367465, "effective_prefill_tok_s_p50": 8663.741046956262, "effective_prefill_tok_s_p75": 8685.767042672715, "effective_prefill_tok_s_p90": 8695.739746619014, "effective_prefill_tok_s_p95": 8715.3760755626, "effective_prefill_tok_s_p99": 8723.38103129467, "effective_prefill_tok_s_samples": 52, "effective_prefill_tok_s_stddev": 28.525782823079652, "generation_max_ms": 2539.3614600179717, "generation_mean_ci95_high_ms": 1545.2939421299802, "generation_mean_ci95_low_ms": 1302.2521479911857, "generation_mean_ms": 1423.773045060583, "generation_min_ms": 388.0178319523111, "generation_p25_ms": 1106.4506069524214, "generation_p50_ms": 1387.149133021012, "generation_p75_ms": 1688.0862249527127, "generation_p90_ms": 1987.5241870759055, "generation_p95_ms": 2134.459304041229, "generation_p99_ms": 2539.3614600179717, "generation_samples": 52, "generation_stddev_ms": 447.09165859601416, "mean_inter_token_latency_ms_max": 6.92888985629127, "mean_inter_token_latency_ms_mean": 5.310823420767946, "mean_inter_token_latency_ms_mean_ci95_high": 5.404326972397817, "mean_inter_token_latency_ms_mean_ci95_low": 5.217319869138075, "mean_inter_token_latency_ms_min": 4.67694844574789, "mean_inter_token_latency_ms_p25": 5.129637192769556, "mean_inter_token_latency_ms_p50": 5.240114141970287, "mean_inter_token_latency_ms_p75": 5.425915079256303, "mean_inter_token_latency_ms_p90": 5.694814136409573, "mean_inter_token_latency_ms_p95": 5.807736204134681, "mean_inter_token_latency_ms_p99": 6.92888985629127, "mean_inter_token_latency_ms_samples": 52, "mean_inter_token_latency_ms_stddev": 0.3440120916729528, "mean_time_per_output_token_max_ms": 6.92888985629127, "mean_time_per_output_token_mean_ci95_high_ms": 5.404326972397817, "mean_time_per_output_token_mean_ci95_low_ms": 5.217319869138075, "mean_time_per_output_token_mean_ms": 5.310823420767946, "mean_time_per_output_token_min_ms": 4.67694844574789, "mean_time_per_output_token_p25_ms": 5.129637192769556, "mean_time_per_output_token_p50_ms": 5.240114141970287, "mean_time_per_output_token_p75_ms": 5.425915079256303, "mean_time_per_output_token_p90_ms": 5.694814136409573, "mean_time_per_output_token_p95_ms": 5.807736204134681, "mean_time_per_output_token_p99_ms": 6.92888985629127, "mean_time_per_output_token_samples": 52, "mean_time_per_output_token_stddev_ms": 0.3440120916729528, "output_token_sources": { "usage": 52 }, "output_tokens": 14053, "prefix_cache_hit_avg": 0.0, "prompt_token_samples": 52, "prompt_tokens": 399899, "prompt_tokens_max": 7708.0, "prompt_tokens_mean": 7690.365384615385, "prompt_tokens_mean_ci95_high": 7693.3501411136585, "prompt_tokens_mean_ci95_low": 7687.380628117111, "prompt_tokens_min": 7675.0, "prompt_tokens_p25": 7680.0, "prompt_tokens_p50": 7688.0, "prompt_tokens_p75": 7701.0, "prompt_tokens_p90": 7705.0, "prompt_tokens_p95": 7707.0, "prompt_tokens_p99": 7708.0, "prompt_tokens_samples": 52, "prompt_tokens_stddev": 10.981318978876677, "reasoning_chunks": 3933, "request_canary_passed": 0, "request_canary_samples": 0, "throughput_performance_eligible": true, "throughput_tok_s": 116.86941950480995, "time_to_first_output_max_ms": 893.4057530714199, "time_to_first_output_mean_ci95_high_ms": 888.2623673069901, "time_to_first_output_mean_ci95_low_ms": 887.0076201128308, "time_to_first_output_mean_ms": 887.6349937099104, "time_to_first_output_min_ms": 883.2584490301087, "time_to_first_output_p25_ms": 885.7697449857369, "time_to_first_output_p50_ms": 887.7673039678484, "time_to_first_output_p75_ms": 889.2285779584199, "time_to_first_output_p90_ms": 890.8421780215576, "time_to_first_output_p95_ms": 891.3549970602617, "time_to_first_output_p99_ms": 893.4057530714199, "time_to_first_output_samples": 52, "time_to_first_output_stddev_ms": 2.3081915032073326, "tpot_max_ms": 6.92888985629127, "tpot_mean_ci95_high_ms": 5.404326972397817, "tpot_mean_ci95_low_ms": 5.217319869138075, "tpot_mean_ms": 5.310823420767946, "tpot_min_ms": 4.67694844574789, "tpot_p25_ms": 5.129637192769556, "tpot_p50_ms": 5.240114141970287, "tpot_p75_ms": 5.425915079256303, "tpot_p90_ms": 5.694814136409573, "tpot_p95_ms": 5.807736204134681, "tpot_p99_ms": 6.92888985629127, "tpot_samples": 52, "tpot_stddev_ms": 0.3440120916729528, "ttft_max_ms": 3180.0511359469965, "ttft_mean_ci95_high_ms": 2150.2028174919997, "ttft_mean_ci95_low_ms": 1909.5907428930952, "ttft_mean_ms": 2029.8967801925476, "ttft_min_ms": 964.0104860300198, "ttft_p25_ms": 1745.0899699470028, "ttft_p50_ms": 1964.4858810352162, "ttft_p75_ms": 2305.6630169739947, "ttft_p90_ms": 2563.26951389201, "ttft_p95_ms": 2712.021376937628, "ttft_p99_ms": 3180.0511359469965, "ttft_samples": 52, "ttft_stddev_ms": 442.62202676638594 }, "actual_start_unix_ns": 1792644911083881618, "actual_end_unix_ns": 1792645031329197497 }, { "name": "incumbent-before-c1-r3", "arm": "incumbent-before", "concurrency": 1, "repetition": "r3", "attempted": 52, "completed": 52, "failed": 0, "wall_ms": 125834.0294000227, "metrics": { "completion_tokens_max": 1036.0, "completion_tokens_mean": 293.88461538461536, "completion_tokens_mean_ci95_high": 335.6795253330568, "completion_tokens_mean_ci95_low": 252.08970543617392, "completion_tokens_min": 61.0, "completion_tokens_p25": 203.0, "completion_tokens_p50": 249.0, "completion_tokens_p75": 341.0, "completion_tokens_p90": 451.0, "completion_tokens_p95": 575.0, "completion_tokens_p99": 1036.0, "completion_tokens_samples": 52, "completion_tokens_stddev": 153.76907231888327, "content_chunks": 931, "content_chunks_s": null, "controlled_output_exact_adherence": 0, "controlled_output_exact_adherence_rate": null, "controlled_output_samples": 0, "controlled_output_unobservable": 0, "decode_tok_s_max": 206.66960163658624, "decode_tok_s_mean": 191.1636786434264, "decode_tok_s_mean_ci95_high": 193.61279234158158, "decode_tok_s_mean_ci95_low": 188.71456494527123, "decode_tok_s_min": 164.56456631422057, "decode_tok_s_p25": 186.9713998472287, "decode_tok_s_p50": 190.652220554731, "decode_tok_s_p75": 197.626132568778, "decode_tok_s_p90": 202.78990635698074, "decode_tok_s_p95": 204.698066939612, "decode_tok_s_p99": 206.66960163658624, "decode_tok_s_samples": 52, "decode_tok_s_stddev": 9.010617365448665, "e2e_max_ms": 6079.2663500178605, "e2e_mean_ci95_high_ms": 2634.572369546113, "e2e_mean_ci95_low_ms": 2203.367108140729, "e2e_mean_ms": 2418.969738843421, "e2e_min_ms": 1204.4429380912334, "e2e_p25_ms": 1955.4801019839942, "e2e_p50_ms": 2142.12976093404, "e2e_p75_ms": 2745.0413530459628, "e2e_p90_ms": 3354.1153100086376, "e2e_p95_ms": 3907.997975940816, "e2e_p99_ms": 6079.2663500178605, "e2e_samples": 52, "e2e_stddev_ms": 793.2309593096753, "effective_prefill_tok_s_max": 8788.262553538609, "effective_prefill_tok_s_mean": 8689.5915344842, "effective_prefill_tok_s_mean_ci95_high": 8696.303833785461, "effective_prefill_tok_s_mean_ci95_low": 8682.879235182938, "effective_prefill_tok_s_min": 8628.947090225438, "effective_prefill_tok_s_p25": 8672.993370707252, "effective_prefill_tok_s_p50": 8686.726221643337, "effective_prefill_tok_s_p75": 8701.966725234834, "effective_prefill_tok_s_p90": 8715.940881450784, "effective_prefill_tok_s_p95": 8721.828961234996, "effective_prefill_tok_s_p99": 8788.262553538609, "effective_prefill_tok_s_samples": 52, "effective_prefill_tok_s_stddev": 24.695448272409664, "generation_max_ms": 5192.823796998709, "generation_mean_ci95_high_ms": 1749.3661387920167, "generation_mean_ci95_low_ms": 1318.539441268755, "generation_mean_ms": 1533.9527900303858, "generation_min_ms": 318.70434805750847, "generation_p25_ms": 1070.0009049614891, "generation_p50_ms": 1258.6542039643973, "generation_p75_ms": 1861.3943030359223, "generation_p90_ms": 2466.2736429600045, "generation_p95_ms": 3023.908955976367, "generation_p99_ms": 5192.823796998709, "generation_samples": 52, "generation_stddev_ms": 792.5345656933342, "mean_inter_token_latency_ms_max": 6.076642271159357, "mean_inter_token_latency_ms_mean": 5.242919340470161, "mean_inter_token_latency_ms_mean_ci95_high": 5.312471205640618, "mean_inter_token_latency_ms_mean_ci95_low": 5.173367475299704, "mean_inter_token_latency_ms_min": 4.838640961617707, "mean_inter_token_latency_ms_p25": 5.041495010266435, "mean_inter_token_latency_ms_p50": 5.213603136522294, "mean_inter_token_latency_ms_p75": 5.33802056179798, "mean_inter_token_latency_ms_p90": 5.61129745299512, "mean_inter_token_latency_ms_p95": 5.648514976905238, "mean_inter_token_latency_ms_p99": 6.076642271159357, "mean_inter_token_latency_ms_samples": 52, "mean_inter_token_latency_ms_stddev": 0.25589062875126434, "mean_time_per_output_token_max_ms": 6.076642271159357, "mean_time_per_output_token_mean_ci95_high_ms": 5.312471205640618, "mean_time_per_output_token_mean_ci95_low_ms": 5.173367475299704, "mean_time_per_output_token_mean_ms": 5.242919340470161, "mean_time_per_output_token_min_ms": 4.838640961617707, "mean_time_per_output_token_p25_ms": 5.041495010266435, "mean_time_per_output_token_p50_ms": 5.213603136522294, "mean_time_per_output_token_p75_ms": 5.33802056179798, "mean_time_per_output_token_p90_ms": 5.61129745299512, "mean_time_per_output_token_p95_ms": 5.648514976905238, "mean_time_per_output_token_p99_ms": 6.076642271159357, "mean_time_per_output_token_samples": 52, "mean_time_per_output_token_stddev_ms": 0.25589062875126434, "output_token_sources": { "usage": 52 }, "output_tokens": 15282, "prefix_cache_hit_avg": 0.0, "prompt_token_samples": 52, "prompt_tokens": 399899, "prompt_tokens_max": 7708.0, "prompt_tokens_mean": 7690.365384615385, "prompt_tokens_mean_ci95_high": 7693.3501411136585, "prompt_tokens_mean_ci95_low": 7687.380628117111, "prompt_tokens_min": 7675.0, "prompt_tokens_p25": 7680.0, "prompt_tokens_p50": 7688.0, "prompt_tokens_p75": 7701.0, "prompt_tokens_p90": 7705.0, "prompt_tokens_p95": 7707.0, "prompt_tokens_p99": 7708.0, "prompt_tokens_samples": 52, "prompt_tokens_stddev": 10.981318978876677, "reasoning_chunks": 4310, "request_canary_passed": 0, "request_canary_samples": 0, "throughput_performance_eligible": true, "throughput_tok_s": 121.44568582016053, "time_to_first_output_max_ms": 891.6499220067635, "time_to_first_output_mean_ci95_high_ms": 885.8714044514976, "time_to_first_output_mean_ci95_low_ms": 884.1624931745729, "time_to_first_output_mean_ms": 885.0169488130352, "time_to_first_output_min_ms": 873.3239310095087, "time_to_first_output_p25_ms": 883.4755569696426, "time_to_first_output_p50_ms": 885.0038300734013, "time_to_first_output_p75_ms": 887.0139960199594, "time_to_first_output_p90_ms": 887.9540630150586, "time_to_first_output_p95_ms": 889.6505670854822, "time_to_first_output_p99_ms": 891.6499220067635, "time_to_first_output_samples": 52, "time_to_first_output_stddev_ms": 3.143656752127765, "tpot_max_ms": 6.076642271159357, "tpot_mean_ci95_high_ms": 5.312471205640618, "tpot_mean_ci95_low_ms": 5.173367475299704, "tpot_mean_ms": 5.242919340470161, "tpot_min_ms": 4.838640961617707, "tpot_p25_ms": 5.041495010266435, "tpot_p50_ms": 5.213603136522294, "tpot_p75_ms": 5.33802056179798, "tpot_p90_ms": 5.61129745299512, "tpot_p95_ms": 5.648514976905238, "tpot_p99_ms": 6.076642271159357, "tpot_samples": 52, "tpot_stddev_ms": 0.25589062875126434, "ttft_max_ms": 5831.556426011957, "ttft_mean_ci95_high_ms": 2349.994374934632, "ttft_mean_ci95_low_ms": 1926.5254154914792, "ttft_mean_ms": 2138.2598952130556, "ttft_min_ms": 945.286970003508, "ttft_p25_ms": 1698.7237220164388, "ttft_p50_ms": 1914.1110540367663, "ttft_p75_ms": 2437.084650970064, "ttft_p90_ms": 2911.651541944593, "ttft_p95_ms": 3583.0599109176546, "ttft_p99_ms": 5831.556426011957, "ttft_samples": 52, "ttft_stddev_ms": 778.9995136937081 }, "actual_start_unix_ns": 1792645031672418894, "actual_end_unix_ns": 1792645157506450523 }, { "name": "incumbent-before-c4-prime", "arm": "incumbent-before", "concurrency": 4, "repetition": "prime", "attempted": 160, "completed": 160, "failed": 0, "wall_ms": 319623.5196910566, "metrics": { "completion_tokens_max": 1229.0, "completion_tokens_mean": 291.3375, "completion_tokens_mean_ci95_high": 310.5387227627131, "completion_tokens_mean_ci95_low": 272.13627723728683, "completion_tokens_min": 81.0, "completion_tokens_p25": 223.0, "completion_tokens_p50": 272.0, "completion_tokens_p75": 322.0, "completion_tokens_p90": 396.0, "completion_tokens_p95": 458.0, "completion_tokens_p99": 834.0, "completion_tokens_samples": 160, "completion_tokens_stddev": 123.9175465111109, "content_chunks": 2925, "content_chunks_s": null, "controlled_output_exact_adherence": 0, "controlled_output_exact_adherence_rate": null, "controlled_output_samples": 0, "controlled_output_unobservable": 0, "decode_tok_s_max": 145.97459484510344, "decode_tok_s_mean": 92.32721226656574, "decode_tok_s_mean_ci95_high": 95.13941393188027, "decode_tok_s_mean_ci95_low": 89.5150106012512, "decode_tok_s_min": 57.67563154113266, "decode_tok_s_p25": 78.69397919832724, "decode_tok_s_p50": 90.26438703062831, "decode_tok_s_p75": 103.11987275004135, "decode_tok_s_p90": 118.68065543268624, "decode_tok_s_p95": 124.42446981660508, "decode_tok_s_p99": 137.60133996703956, "decode_tok_s_samples": 160, "decode_tok_s_stddev": 18.148903065535666, "e2e_max_ms": 19830.83779294975, "e2e_mean_ci95_high_ms": 8234.603347016731, "e2e_mean_ci95_low_ms": 7631.9756209893385, "e2e_mean_ms": 7933.289484003035, "e2e_min_ms": 3908.728428068571, "e2e_p25_ms": 6746.144996955991, "e2e_p50_ms": 7861.14124499727, "e2e_p75_ms": 9015.040086931549, "e2e_p90_ms": 9838.96729699336, "e2e_p95_ms": 10587.598598911427, "e2e_p99_ms": 15919.37873500865, "e2e_samples": 160, "e2e_stddev_ms": 1944.567546341319, "effective_prefill_tok_s_max": 8686.024010455256, "effective_prefill_tok_s_mean": 1920.791987567527, "effective_prefill_tok_s_mean_ci95_high": 2070.84144125474, "effective_prefill_tok_s_mean_ci95_low": 1770.7425338803143, "effective_prefill_tok_s_min": 1138.3202173297525, "effective_prefill_tok_s_p25": 1404.2042033565156, "effective_prefill_tok_s_p50": 1585.0463757705713, "effective_prefill_tok_s_p75": 2001.9861515415585, "effective_prefill_tok_s_p90": 2739.8168256754857, "effective_prefill_tok_s_p95": 4414.798012436853, "effective_prefill_tok_s_p99": 4810.970210058363, "effective_prefill_tok_s_samples": 160, "effective_prefill_tok_s_stddev": 968.3633373786588, "generation_max_ms": 14598.031910019927, "generation_mean_ci95_high_ms": 3614.378944265474, "generation_mean_ci95_low_ms": 3061.4896218010185, "generation_mean_ms": 3337.934283033246, "generation_min_ms": 1118.2558169821277, "generation_p25_ms": 2364.707329077646, "generation_p50_ms": 3097.293167025782, "generation_p75_ms": 3633.3110339473933, "generation_p90_ms": 5109.044288052246, "generation_p95_ms": 6089.99876701273, "generation_p99_ms": 11727.188186952844, "generation_samples": 160, "generation_stddev_ms": 1784.0709724234462, "mean_inter_token_latency_ms_max": 17.338345038958572, "mean_inter_token_latency_ms_mean": 11.249885839207213, "mean_inter_token_latency_ms_mean_ci95_high": 11.592502414998622, "mean_inter_token_latency_ms_mean_ci95_low": 10.907269263415804, "mean_inter_token_latency_ms_min": 6.850507110919677, "mean_inter_token_latency_ms_p25": 9.669094367669445, "mean_inter_token_latency_ms_p50": 11.013808120042086, "mean_inter_token_latency_ms_p75": 12.610608209151358, "mean_inter_token_latency_ms_p90": 14.194588031819498, "mean_inter_token_latency_ms_p95": 15.492386203496372, "mean_inter_token_latency_ms_p99": 17.193212888539872, "mean_inter_token_latency_ms_samples": 160, "mean_inter_token_latency_ms_stddev": 2.2111198849562377, "mean_time_per_output_token_max_ms": 17.338345038958572, "mean_time_per_output_token_mean_ci95_high_ms": 11.592502414998622, "mean_time_per_output_token_mean_ci95_low_ms": 10.907269263415804, "mean_time_per_output_token_mean_ms": 11.249885839207213, "mean_time_per_output_token_min_ms": 6.850507110919677, "mean_time_per_output_token_p25_ms": 9.669094367669445, "mean_time_per_output_token_p50_ms": 11.013808120042086, "mean_time_per_output_token_p75_ms": 12.610608209151358, "mean_time_per_output_token_p90_ms": 14.194588031819498, "mean_time_per_output_token_p95_ms": 15.492386203496372, "mean_time_per_output_token_p99_ms": 17.193212888539872, "mean_time_per_output_token_samples": 160, "mean_time_per_output_token_stddev_ms": 2.2111198849562377, "output_token_sources": { "usage": 160 }, "output_tokens": 46614, "prefix_cache_hit_avg": 0.0, "prompt_token_samples": 160, "prompt_tokens": 1236764, "prompt_tokens_max": 7798.0, "prompt_tokens_mean": 7729.775, "prompt_tokens_mean_ci95_high": 7735.29436497065, "prompt_tokens_mean_ci95_low": 7724.255635029349, "prompt_tokens_min": 7675.0, "prompt_tokens_p25": 7702.0, "prompt_tokens_p50": 7725.0, "prompt_tokens_p75": 7759.0, "prompt_tokens_p90": 7782.0, "prompt_tokens_p95": 7790.0, "prompt_tokens_p99": 7797.0, "prompt_tokens_samples": 160, "prompt_tokens_stddev": 35.61992764286613, "reasoning_chunks": 12936, "request_canary_passed": 0, "request_canary_samples": 0, "throughput_performance_eligible": true, "throughput_tok_s": 145.84033129056465, "time_to_first_output_max_ms": 6744.1479850094765, "time_to_first_output_mean_ci95_high_ms": 4800.136417963987, "time_to_first_output_mean_ci95_low_ms": 4390.573983975591, "time_to_first_output_mean_ms": 4595.355200969789, "time_to_first_output_min_ms": 883.6033599218354, "time_to_first_output_p25_ms": 3816.2079110043123, "time_to_first_output_p50_ms": 4866.2360270973295, "time_to_first_output_p75_ms": 5457.762661972083, "time_to_first_output_p90_ms": 6233.952653943561, "time_to_first_output_p95_ms": 6420.384382014163, "time_to_first_output_p99_ms": 6610.449168947525, "time_to_first_output_samples": 160, "time_to_first_output_stddev_ms": 1321.5817708629513, "tpot_max_ms": 17.338345038958572, "tpot_mean_ci95_high_ms": 11.592502414998622, "tpot_mean_ci95_low_ms": 10.907269263415804, "tpot_mean_ms": 11.249885839207213, "tpot_min_ms": 6.850507110919677, "tpot_p25_ms": 9.669094367669445, "tpot_p50_ms": 11.013808120042086, "tpot_p75_ms": 12.610608209151358, "tpot_p90_ms": 14.194588031819498, "tpot_p95_ms": 15.492386203496372, "tpot_p99_ms": 17.193212888539872, "tpot_samples": 160, "tpot_stddev_ms": 2.2111198849562377, "ttft_max_ms": 18989.743444952182, "ttft_mean_ci95_high_ms": 7521.864924814339, "ttft_mean_ci95_low_ms": 6938.118915314941, "ttft_mean_ms": 7229.99192006464, "ttft_min_ms": 3036.8063719943166, "ttft_p25_ms": 6109.809456043877, "ttft_p50_ms": 7140.010823961347, "ttft_p75_ms": 8134.7073869546875, "ttft_p90_ms": 9005.078271962702, "ttft_p95_ms": 9641.143500921316, "ttft_p99_ms": 15063.273850013502, "ttft_samples": 160, "ttft_stddev_ms": 1883.6397602575362 }, "actual_start_unix_ns": 1792645157879522221, "actual_end_unix_ns": 1792645477503045027 }, { "name": "incumbent-before-c4-r1", "arm": "incumbent-before", "concurrency": 4, "repetition": "r1", "attempted": 160, "completed": 159, "failed": 1, "wall_ms": 326889.64053301606, "metrics": { "completion_tokens_max": 1913.0, "completion_tokens_mean": 301.6163522012579, "completion_tokens_mean_ci95_high": 327.4991908187188, "completion_tokens_mean_ci95_low": 275.733513583797, "completion_tokens_min": 83.0, "completion_tokens_p25": 222.0, "completion_tokens_p50": 278.0, "completion_tokens_p75": 326.0, "completion_tokens_p90": 414.0, "completion_tokens_p95": 454.0, "completion_tokens_p99": 957.0, "completion_tokens_samples": 159, "completion_tokens_stddev": 166.5153962829491, "content_chunks": 2842, "content_chunks_s": null, "controlled_output_exact_adherence": 0, "controlled_output_exact_adherence_rate": null, "controlled_output_samples": 0, "controlled_output_unobservable": 0, "decode_tok_s_max": 175.49750046310896, "decode_tok_s_mean": 89.579755976334, "decode_tok_s_mean_ci95_high": 92.39455992752421, "decode_tok_s_mean_ci95_low": 86.76495202514378, "decode_tok_s_min": 55.6710218565929, "decode_tok_s_p25": 76.6130077295752, "decode_tok_s_p50": 86.89515109251876, "decode_tok_s_p75": 99.54589155268249, "decode_tok_s_p90": 113.14232572174433, "decode_tok_s_p95": 126.5997726657923, "decode_tok_s_p99": 132.13204622546164, "decode_tok_s_samples": 159, "decode_tok_s_stddev": 18.108840468334655, "e2e_max_ms": 31866.451817913912, "e2e_mean_ci95_high_ms": 8381.52561232382, "e2e_mean_ci95_low_ms": 7551.267998013014, "e2e_mean_ms": 7966.396805168417, "e2e_min_ms": 3723.3210899867117, "e2e_p25_ms": 6518.893910921179, "e2e_p50_ms": 7860.375156975351, "e2e_p75_ms": 8961.008399026468, "e2e_p90_ms": 10011.434949934483, "e2e_p95_ms": 11267.857962986454, "e2e_p99_ms": 15396.18150703609, "e2e_samples": 159, "e2e_stddev_ms": 2670.7015738728537, "effective_prefill_tok_s_max": 8380.723476519604, "effective_prefill_tok_s_mean": 1989.609693092148, "effective_prefill_tok_s_mean_ci95_high": 2139.1507531892485, "effective_prefill_tok_s_mean_ci95_low": 1840.0686329950477, "effective_prefill_tok_s_min": 1130.2643520540578, "effective_prefill_tok_s_p25": 1425.1098806644975, "effective_prefill_tok_s_p50": 1745.1522134725833, "effective_prefill_tok_s_p75": 2035.9394152042473, "effective_prefill_tok_s_p90": 2847.1789323294397, "effective_prefill_tok_s_p95": 4436.291748770445, "effective_prefill_tok_s_p99": 4816.524118988695, "effective_prefill_tok_s_samples": 159, "effective_prefill_tok_s_stddev": 962.0617448753302, "generation_max_ms": 26234.084507916123, "generation_mean_ci95_high_ms": 3900.160281452169, "generation_mean_ci95_low_ms": 3172.377474753584, "generation_mean_ms": 3536.2688781028764, "generation_min_ms": 1125.8468619780615, "generation_p25_ms": 2471.894879010506, "generation_p50_ms": 3275.4999110475183, "generation_p75_ms": 4011.9108189828694, "generation_p90_ms": 5120.398084982298, "generation_p95_ms": 6100.340380100533, "generation_p99_ms": 10711.342126945965, "generation_samples": 159, "generation_stddev_ms": 2341.069390734786, "mean_inter_token_latency_ms_max": 17.962666512139368, "mean_inter_token_latency_ms_mean": 11.58208471575117, "mean_inter_token_latency_ms_mean_ci95_high": 11.918848645770785, "mean_inter_token_latency_ms_mean_ci95_low": 11.245320785731556, "mean_inter_token_latency_ms_min": 5.698086852297981, "mean_inter_token_latency_ms_p25": 10.045617999922898, "mean_inter_token_latency_ms_p50": 11.508122000217053, "mean_inter_token_latency_ms_p75": 13.052613774540095, "mean_inter_token_latency_ms_p90": 14.403393043399474, "mean_inter_token_latency_ms_p95": 15.354203008749383, "mean_inter_token_latency_ms_p99": 16.667529666772896, "mean_inter_token_latency_ms_samples": 159, "mean_inter_token_latency_ms_stddev": 2.1665467257980584, "mean_time_per_output_token_max_ms": 17.962666512139368, "mean_time_per_output_token_mean_ci95_high_ms": 11.918848645770785, "mean_time_per_output_token_mean_ci95_low_ms": 11.245320785731556, "mean_time_per_output_token_mean_ms": 11.58208471575117, "mean_time_per_output_token_min_ms": 5.698086852297981, "mean_time_per_output_token_p25_ms": 10.045617999922898, "mean_time_per_output_token_p50_ms": 11.508122000217053, "mean_time_per_output_token_p75_ms": 13.052613774540095, "mean_time_per_output_token_p90_ms": 14.403393043399474, "mean_time_per_output_token_p95_ms": 15.354203008749383, "mean_time_per_output_token_p99_ms": 16.667529666772896, "mean_time_per_output_token_samples": 159, "mean_time_per_output_token_stddev_ms": 2.1665467257980584, "output_token_sources": { "usage": 159 }, "output_tokens": 47957, "prefix_cache_hit_avg": 0.0, "prompt_token_samples": 159, "prompt_tokens": 1228980, "prompt_tokens_max": 7798.0, "prompt_tokens_mean": 7729.433962264151, "prompt_tokens_mean_ci95_high": 7734.947268214175, "prompt_tokens_mean_ci95_low": 7723.920656314127, "prompt_tokens_min": 7675.0, "prompt_tokens_p25": 7702.0, "prompt_tokens_p50": 7725.0, "prompt_tokens_p75": 7759.0, "prompt_tokens_p90": 7782.0, "prompt_tokens_p95": 7791.0, "prompt_tokens_p99": 7797.0, "prompt_tokens_samples": 159, "prompt_tokens_stddev": 35.46946062083189, "reasoning_chunks": 13506, "request_canary_passed": 0, "request_canary_samples": 0, "throughput_performance_eligible": false, "throughput_tok_s": null, "time_to_first_output_max_ms": 6790.44684197288, "time_to_first_output_mean_ci95_high_ms": 4633.5869727956715, "time_to_first_output_mean_ci95_low_ms": 4226.668881335408, "time_to_first_output_mean_ms": 4430.12792706554, "time_to_first_output_min_ms": 915.7920579891652, "time_to_first_output_p25_ms": 3778.1084950547665, "time_to_first_output_p50_ms": 4425.974961020984, "time_to_first_output_p75_ms": 5424.143151962198, "time_to_first_output_p90_ms": 6210.49051405862, "time_to_first_output_p95_ms": 6436.243084026501, "time_to_first_output_p99_ms": 6683.0110950395465, "time_to_first_output_samples": 159, "time_to_first_output_stddev_ms": 1308.9392600179626, "tpot_max_ms": 17.962666512139368, "tpot_mean_ci95_high_ms": 11.918848645770785, "tpot_mean_ci95_low_ms": 11.245320785731556, "tpot_mean_ms": 11.58208471575117, "tpot_min_ms": 5.698086852297981, "tpot_p25_ms": 10.045617999922898, "tpot_p50_ms": 11.508122000217053, "tpot_p75_ms": 13.052613774540095, "tpot_p90_ms": 14.403393043399474, "tpot_p95_ms": 15.354203008749383, "tpot_p99_ms": 16.667529666772896, "tpot_samples": 159, "tpot_stddev_ms": 2.1665467257980584, "ttft_max_ms": 30753.56537301559, "ttft_mean_ci95_high_ms": 7661.310895237624, "ttft_mean_ci95_low_ms": 6846.518744278303, "ttft_mean_ms": 7253.914819757963, "ttft_min_ms": 3067.5516059855, "ttft_p25_ms": 5661.097837029956, "ttft_p50_ms": 7172.4844289710745, "ttft_p75_ms": 8188.204938895069, "ttft_p90_ms": 9176.58881796524, "ttft_p95_ms": 10396.475435001776, "ttft_p99_ms": 14925.850194063969, "ttft_samples": 159, "ttft_stddev_ms": 2620.953596134922 }, "actual_start_unix_ns": 1792645477981883510, "actual_end_unix_ns": 1792645804871526290 }, { "name": "incumbent-before-c4-r2", "arm": "incumbent-before", "concurrency": 4, "repetition": "r2", "attempted": 160, "completed": 160, "failed": 0, "wall_ms": 310897.74879300967, "metrics": { "completion_tokens_max": 859.0, "completion_tokens_mean": 271.36875, "completion_tokens_mean_ci95_high": 286.8048835755997, "completion_tokens_mean_ci95_low": 255.93261642440024, "completion_tokens_min": 58.0, "completion_tokens_p25": 215.0, "completion_tokens_p50": 260.0, "completion_tokens_p75": 318.0, "completion_tokens_p90": 365.0, "completion_tokens_p95": 383.0, "completion_tokens_p99": 783.0, "completion_tokens_samples": 160, "completion_tokens_stddev": 99.61906197039619, "content_chunks": 2927, "content_chunks_s": null, "controlled_output_exact_adherence": 0, "controlled_output_exact_adherence_rate": null, "controlled_output_samples": 0, "controlled_output_unobservable": 0, "decode_tok_s_max": 173.37390649393353, "decode_tok_s_mean": 95.750891581038, "decode_tok_s_mean_ci95_high": 98.50855108696328, "decode_tok_s_mean_ci95_low": 92.99323207511273, "decode_tok_s_min": 61.88298683924402, "decode_tok_s_p25": 83.09683831712199, "decode_tok_s_p50": 90.92294762165204, "decode_tok_s_p75": 106.05494707214284, "decode_tok_s_p90": 117.30914936695312, "decode_tok_s_p95": 124.93046874307848, "decode_tok_s_p99": 162.90768521993488, "decode_tok_s_samples": 160, "decode_tok_s_stddev": 17.796908265180512, "e2e_max_ms": 14282.67392993439, "e2e_mean_ci95_high_ms": 7973.211785490654, "e2e_mean_ci95_low_ms": 7456.59975056877, "e2e_mean_ms": 7714.905768029712, "e2e_min_ms": 2560.6663960497826, "e2e_p25_ms": 6540.335112018511, "e2e_p50_ms": 7601.767344982363, "e2e_p75_ms": 8785.909818019718, "e2e_p90_ms": 9587.954584974796, "e2e_p95_ms": 10279.510543914512, "e2e_p99_ms": 13154.971308074892, "e2e_samples": 160, "e2e_stddev_ms": 1667.0109153138797, "effective_prefill_tok_s_max": 8679.622294975254, "effective_prefill_tok_s_mean": 1764.036668156144, "effective_prefill_tok_s_mean_ci95_high": 1882.6946182858155, "effective_prefill_tok_s_mean_ci95_low": 1645.3787180264726, "effective_prefill_tok_s_min": 1139.5934930599037, "effective_prefill_tok_s_p25": 1433.46726071091, "effective_prefill_tok_s_p50": 1583.7853589302906, "effective_prefill_tok_s_p75": 1865.257700929466, "effective_prefill_tok_s_p90": 2431.761000498173, "effective_prefill_tok_s_p95": 2685.821538204905, "effective_prefill_tok_s_p99": 4770.104244540829, "effective_prefill_tok_s_samples": 160, "effective_prefill_tok_s_stddev": 765.774254890681, "generation_max_ms": 9572.091502021067, "generation_mean_ci95_high_ms": 3154.756394605321, "generation_mean_ci95_low_ms": 2748.948157637968, "generation_mean_ms": 2951.8522761216445, "generation_min_ms": 349.89141195546836, "generation_p25_ms": 2252.4175519356504, "generation_p50_ms": 2707.0783129893243, "generation_p75_ms": 3509.974076063372, "generation_p90_ms": 4392.381399986334, "generation_p95_ms": 5094.9288960546255, "generation_p99_ms": 9543.39235497173, "generation_samples": 160, "generation_stddev_ms": 1309.4676755858932, "mean_inter_token_latency_ms_max": 16.159530285726852, "mean_inter_token_latency_ms_mean": 10.774083805970175, "mean_inter_token_latency_ms_mean_ci95_high": 11.061283443945495, "mean_inter_token_latency_ms_mean_ci95_low": 10.486884167994855, "mean_inter_token_latency_ms_min": 5.7678806472241, "mean_inter_token_latency_ms_p25": 9.427949721910577, "mean_inter_token_latency_ms_p50": 10.968027519993484, "mean_inter_token_latency_ms_p75": 12.02045916460059, "mean_inter_token_latency_ms_p90": 12.985094404159879, "mean_inter_token_latency_ms_p95": 13.685579401304482, "mean_inter_token_latency_ms_p99": 15.475227549904957, "mean_inter_token_latency_ms_samples": 160, "mean_inter_token_latency_ms_stddev": 1.853479590158771, "mean_time_per_output_token_max_ms": 16.159530285726852, "mean_time_per_output_token_mean_ci95_high_ms": 11.061283443945495, "mean_time_per_output_token_mean_ci95_low_ms": 10.486884167994855, "mean_time_per_output_token_mean_ms": 10.774083805970175, "mean_time_per_output_token_min_ms": 5.7678806472241, "mean_time_per_output_token_p25_ms": 9.427949721910577, "mean_time_per_output_token_p50_ms": 10.968027519993484, "mean_time_per_output_token_p75_ms": 12.02045916460059, "mean_time_per_output_token_p90_ms": 12.985094404159879, "mean_time_per_output_token_p95_ms": 13.685579401304482, "mean_time_per_output_token_p99_ms": 15.475227549904957, "mean_time_per_output_token_samples": 160, "mean_time_per_output_token_stddev_ms": 1.853479590158771, "output_token_sources": { "usage": 160 }, "output_tokens": 43419, "prefix_cache_hit_avg": 0.0, "prompt_token_samples": 160, "prompt_tokens": 1236764, "prompt_tokens_max": 7798.0, "prompt_tokens_mean": 7729.775, "prompt_tokens_mean_ci95_high": 7735.29436497065, "prompt_tokens_mean_ci95_low": 7724.255635029349, "prompt_tokens_min": 7675.0, "prompt_tokens_p25": 7702.0, "prompt_tokens_p50": 7725.0, "prompt_tokens_p75": 7759.0, "prompt_tokens_p90": 7782.0, "prompt_tokens_p95": 7790.0, "prompt_tokens_p99": 7797.0, "prompt_tokens_samples": 160, "prompt_tokens_stddev": 35.61992764286613, "reasoning_chunks": 11957, "request_canary_passed": 0, "request_canary_samples": 0, "throughput_performance_eligible": true, "throughput_tok_s": 139.65684913629792, "time_to_first_output_max_ms": 6826.995808049105, "time_to_first_output_mean_ci95_high_ms": 4932.419732007202, "time_to_first_output_mean_ci95_low_ms": 4593.687251808933, "time_to_first_output_mean_ms": 4763.053491908067, "time_to_first_output_min_ms": 884.3702800804749, "time_to_first_output_p25_ms": 4134.549341979437, "time_to_first_output_p50_ms": 4875.706702005118, "time_to_first_output_p75_ms": 5407.319483929314, "time_to_first_output_p90_ms": 6114.285647054203, "time_to_first_output_p95_ms": 6441.6025599930435, "time_to_first_output_p99_ms": 6772.788490052335, "time_to_first_output_samples": 160, "time_to_first_output_stddev_ms": 1093.0266886779696, "tpot_max_ms": 16.159530285726852, "tpot_mean_ci95_high_ms": 11.061283443945495, "tpot_mean_ci95_low_ms": 10.486884167994855, "tpot_mean_ms": 10.774083805970175, "tpot_min_ms": 5.7678806472241, "tpot_p25_ms": 9.427949721910577, "tpot_p50_ms": 10.968027519993484, "tpot_p75_ms": 12.02045916460059, "tpot_p90_ms": 12.985094404159879, "tpot_p95_ms": 13.685579401304482, "tpot_p99_ms": 15.475227549904957, "tpot_samples": 160, "tpot_stddev_ms": 1.853479590158771, "ttft_max_ms": 13340.448100934736, "ttft_mean_ci95_high_ms": 7322.925764084362, "ttft_mean_ci95_low_ms": 6813.699062648079, "ttft_mean_ms": 7068.31241336622, "ttft_min_ms": 1771.9410600839183, "ttft_p25_ms": 5972.293498925865, "ttft_p50_ms": 6956.712750019506, "ttft_p75_ms": 7914.519276935607, "ttft_p90_ms": 8932.236560038291, "ttft_p95_ms": 9736.62057495676, "ttft_p99_ms": 12869.276012061164, "ttft_samples": 160, "ttft_stddev_ms": 1643.1798182787634 }, "actual_start_unix_ns": 1792646122676691293, "actual_end_unix_ns": 1792646433574443614 }, { "name": "incumbent-before-c4-r3", "arm": "incumbent-before", "concurrency": 4, "repetition": "r3", "attempted": 160, "completed": 160, "failed": 0, "wall_ms": 314291.3072439842, "metrics": { "completion_tokens_max": 816.0, "completion_tokens_mean": 279.00625, "completion_tokens_mean_ci95_high": 294.22328079739657, "completion_tokens_mean_ci95_low": 263.7892192026035, "completion_tokens_min": 84.0, "completion_tokens_p25": 222.0, "completion_tokens_p50": 269.0, "completion_tokens_p75": 329.0, "completion_tokens_p90": 382.0, "completion_tokens_p95": 419.0, "completion_tokens_p99": 788.0, "completion_tokens_samples": 160, "completion_tokens_stddev": 98.20505417285986, "content_chunks": 2924, "content_chunks_s": null, "controlled_output_exact_adherence": 0, "controlled_output_exact_adherence_rate": null, "controlled_output_samples": 0, "controlled_output_unobservable": 0, "decode_tok_s_max": 200.4505871472005, "decode_tok_s_mean": 93.57264542921948, "decode_tok_s_mean_ci95_high": 96.61651269987205, "decode_tok_s_mean_ci95_low": 90.52877815856691, "decode_tok_s_min": 55.9511152893671, "decode_tok_s_p25": 82.8768913244823, "decode_tok_s_p50": 90.98414914574784, "decode_tok_s_p75": 100.76079382773833, "decode_tok_s_p90": 109.10524861912803, "decode_tok_s_p95": 131.60242172461716, "decode_tok_s_p99": 182.43562069548292, "decode_tok_s_samples": 160, "decode_tok_s_stddev": 19.6439866744945, "e2e_max_ms": 15777.520327013917, "e2e_mean_ci95_high_ms": 8085.363562980444, "e2e_mean_ci95_low_ms": 7529.953154564476, "e2e_mean_ms": 7807.65835877246, "e2e_min_ms": 3065.6640550587326, "e2e_p25_ms": 6718.951132032089, "e2e_p50_ms": 7780.957187991589, "e2e_p75_ms": 8804.102919064462, "e2e_p90_ms": 9799.38562202733, "e2e_p95_ms": 10477.991417050362, "e2e_p99_ms": 14597.076530917548, "e2e_samples": 160, "e2e_stddev_ms": 1792.2060477130713, "effective_prefill_tok_s_max": 8626.23482328456, "effective_prefill_tok_s_mean": 1838.6869480464677, "effective_prefill_tok_s_mean_ci95_high": 1974.8981312021888, "effective_prefill_tok_s_mean_ci95_low": 1702.4757648907466, "effective_prefill_tok_s_min": 1162.2804164845995, "effective_prefill_tok_s_p25": 1432.3008250361713, "effective_prefill_tok_s_p50": 1582.6479737700256, "effective_prefill_tok_s_p75": 1866.0100145294, "effective_prefill_tok_s_p90": 2498.4787344271285, "effective_prefill_tok_s_p95": 3028.6911362379005, "effective_prefill_tok_s_p99": 4843.899587912643, "effective_prefill_tok_s_samples": 160, "effective_prefill_tok_s_stddev": 879.0562888947763, "generation_max_ms": 10825.798233971, "generation_mean_ci95_high_ms": 3348.8426605326126, "generation_mean_ci95_low_ms": 2917.711922107334, "generation_mean_ms": 3133.2772913199733, "generation_min_ms": 567.6579470746219, "generation_p25_ms": 2371.3274179026484, "generation_p50_ms": 2768.693674937822, "generation_p75_ms": 3608.3399389171973, "generation_p90_ms": 4665.939369937405, "generation_p95_ms": 5441.417008987628, "generation_p99_ms": 9114.157158997841, "generation_samples": 160, "generation_stddev_ms": 1391.178676259344, "mean_inter_token_latency_ms_max": 17.872744713455948, "mean_inter_token_latency_ms_mean": 11.08589115784861, "mean_inter_token_latency_ms_mean_ci95_high": 11.403983478965985, "mean_inter_token_latency_ms_mean_ci95_low": 10.767798836731236, "mean_inter_token_latency_ms_min": 4.988760642869317, "mean_inter_token_latency_ms_p25": 9.899305460085175, "mean_inter_token_latency_ms_p50": 10.974424313146459, "mean_inter_token_latency_ms_p75": 12.034610957391292, "mean_inter_token_latency_ms_p90": 13.429152813258817, "mean_inter_token_latency_ms_p95": 14.608567625112398, "mean_inter_token_latency_ms_p99": 17.569367659743875, "mean_inter_token_latency_ms_samples": 160, "mean_inter_token_latency_ms_stddev": 2.0528494713073018, "mean_time_per_output_token_max_ms": 17.872744713455948, "mean_time_per_output_token_mean_ci95_high_ms": 11.403983478965985, "mean_time_per_output_token_mean_ci95_low_ms": 10.767798836731236, "mean_time_per_output_token_mean_ms": 11.08589115784861, "mean_time_per_output_token_min_ms": 4.988760642869317, "mean_time_per_output_token_p25_ms": 9.899305460085175, "mean_time_per_output_token_p50_ms": 10.974424313146459, "mean_time_per_output_token_p75_ms": 12.034610957391292, "mean_time_per_output_token_p90_ms": 13.429152813258817, "mean_time_per_output_token_p95_ms": 14.608567625112398, "mean_time_per_output_token_p99_ms": 17.569367659743875, "mean_time_per_output_token_samples": 160, "mean_time_per_output_token_stddev_ms": 2.0528494713073018, "output_token_sources": { "usage": 160 }, "output_tokens": 44641, "prefix_cache_hit_avg": 0.0, "prompt_token_samples": 160, "prompt_tokens": 1236764, "prompt_tokens_max": 7798.0, "prompt_tokens_mean": 7729.775, "prompt_tokens_mean_ci95_high": 7735.29436497065, "prompt_tokens_mean_ci95_low": 7724.255635029349, "prompt_tokens_min": 7675.0, "prompt_tokens_p25": 7702.0, "prompt_tokens_p50": 7725.0, "prompt_tokens_p75": 7759.0, "prompt_tokens_p90": 7782.0, "prompt_tokens_p95": 7790.0, "prompt_tokens_p99": 7797.0, "prompt_tokens_samples": 160, "prompt_tokens_stddev": 35.61992764286613, "reasoning_chunks": 12347, "request_canary_passed": 0, "request_canary_samples": 0, "throughput_performance_eligible": true, "throughput_tok_s": 142.03701779554856, "time_to_first_output_max_ms": 6693.737491965294, "time_to_first_output_mean_ci95_high_ms": 4854.934545033868, "time_to_first_output_mean_ci95_low_ms": 4493.827589871105, "time_to_first_output_mean_ms": 4674.381067452487, "time_to_first_output_min_ms": 889.7276920033619, "time_to_first_output_p25_ms": 4101.2509720167145, "time_to_first_output_p50_ms": 4881.716481992044, "time_to_first_output_p75_ms": 5385.743023012765, "time_to_first_output_p90_ms": 6165.4987500514835, "time_to_first_output_p95_ms": 6342.5326050492, "time_to_first_output_p99_ms": 6585.534672951326, "time_to_first_output_samples": 160, "time_to_first_output_stddev_ms": 1165.2249563700302, "tpot_max_ms": 17.872744713455948, "tpot_mean_ci95_high_ms": 11.403983478965985, "tpot_mean_ci95_low_ms": 10.767798836731236, "tpot_mean_ms": 11.08589115784861, "tpot_min_ms": 4.988760642869317, "tpot_p25_ms": 9.899305460085175, "tpot_p50_ms": 10.974424313146459, "tpot_p75_ms": 12.034610957391292, "tpot_p90_ms": 13.429152813258817, "tpot_p95_ms": 14.608567625112398, "tpot_p99_ms": 17.569367659743875, "tpot_samples": 160, "tpot_stddev_ms": 2.0528494713073018, "ttft_max_ms": 14646.874143974856, "ttft_mean_ci95_high_ms": 7395.593598930899, "ttft_mean_ci95_low_ms": 6867.296178105717, "ttft_mean_ms": 7131.444888518308, "ttft_min_ms": 2780.6007070466876, "ttft_p25_ms": 6052.167239016853, "ttft_p50_ms": 7086.7376329842955, "ttft_p75_ms": 7975.664112949744, "ttft_p90_ms": 8962.057242984883, "ttft_p95_ms": 9806.650268030353, "ttft_p99_ms": 13603.577213012613, "ttft_samples": 160, "ttft_stddev_ms": 1704.7174814286202 }, "actual_start_unix_ns": 1792646433973216341, "actual_end_unix_ns": 1792646748264525971 }, { "name": "flare-control-c1-prime", "arm": "flare-control", "concurrency": 1, "repetition": "prime", "attempted": 52, "completed": 52, "failed": 0, "wall_ms": 123285.2141110925, "metrics": { "completion_tokens_max": 439.0, "completion_tokens_mean": 275.9230769230769, "completion_tokens_mean_ci95_high": 298.03939260050726, "completion_tokens_mean_ci95_low": 253.80676124564653, "completion_tokens_min": 132.0, "completion_tokens_p25": 211.0, "completion_tokens_p50": 265.0, "completion_tokens_p75": 338.0, "completion_tokens_p90": 379.0, "completion_tokens_p95": 400.0, "completion_tokens_p99": 439.0, "completion_tokens_samples": 52, "completion_tokens_stddev": 81.36888795849315, "content_chunks": 2797, "content_chunks_s": null, "controlled_output_exact_adherence": 0, "controlled_output_exact_adherence_rate": null, "controlled_output_samples": 0, "controlled_output_unobservable": 0, "decode_tok_s_max": 200.86624078897083, "decode_tok_s_mean": 189.84077392330315, "decode_tok_s_mean_ci95_high": 191.6325402640885, "decode_tok_s_mean_ci95_low": 188.0490075825178, "decode_tok_s_min": 176.00681700287646, "decode_tok_s_p25": 185.84063593761127, "decode_tok_s_p50": 189.13557145885147, "decode_tok_s_p75": 193.9802291198568, "decode_tok_s_p90": 199.5867995608083, "decode_tok_s_p95": 200.494679874495, "decode_tok_s_p99": 200.86624078897083, "decode_tok_s_samples": 52, "decode_tok_s_stddev": 6.5921483830123, "e2e_max_ms": 3182.4367879889905, "e2e_mean_ci95_high_ms": 2485.435930064189, "e2e_mean_ci95_low_ms": 2253.98781792516, "e2e_mean_ms": 2369.7118739946745, "e2e_min_ms": 1531.8582050967962, "e2e_p25_ms": 2060.5490769958124, "e2e_p50_ms": 2314.462481997907, "e2e_p75_ms": 2712.1939359931275, "e2e_p90_ms": 2907.543177017942, "e2e_p95_ms": 3111.996329971589, "e2e_p99_ms": 3182.4367879889905, "e2e_samples": 52, "e2e_stddev_ms": 425.76430404418744, "effective_prefill_tok_s_max": 8803.862675215116, "effective_prefill_tok_s_mean": 8342.38690120917, "effective_prefill_tok_s_mean_ci95_high": 8419.418444299547, "effective_prefill_tok_s_mean_ci95_low": 8265.355358118795, "effective_prefill_tok_s_min": 7973.185987398148, "effective_prefill_tok_s_p25": 8043.406478928073, "effective_prefill_tok_s_p50": 8436.909775907276, "effective_prefill_tok_s_p75": 8591.967189650572, "effective_prefill_tok_s_p90": 8684.015198301659, "effective_prefill_tok_s_p95": 8763.252266847312, "effective_prefill_tok_s_p99": 8803.862675215116, "effective_prefill_tok_s_samples": 52, "effective_prefill_tok_s_stddev": 283.40936575557714, "generation_max_ms": 2284.5492609776556, "generation_mean_ci95_high_ms": 1560.6756104402155, "generation_mean_ci95_low_ms": 1332.8901219500106, "generation_mean_ms": 1446.782866195113, "generation_min_ms": 659.8545990418643, "generation_p25_ms": 1104.7751880250871, "generation_p50_ms": 1371.0208990378305, "generation_p75_ms": 1761.3117269938812, "generation_p90_ms": 2002.8199659427628, "generation_p95_ms": 2170.464608934708, "generation_p99_ms": 2284.5492609776556, "generation_samples": 52, "generation_stddev_ms": 419.0266625296157, "mean_inter_token_latency_ms_max": 5.681598116643727, "mean_inter_token_latency_ms_mean": 5.273849970660383, "mean_inter_token_latency_ms_mean_ci95_high": 5.323992976855044, "mean_inter_token_latency_ms_mean_ci95_low": 5.223706964465723, "mean_inter_token_latency_ms_min": 4.978437372413394, "mean_inter_token_latency_ms_p25": 5.147811622567651, "mean_inter_token_latency_ms_p50": 5.271251011373741, "mean_inter_token_latency_ms_p75": 5.366094364723731, "mean_inter_token_latency_ms_p90": 5.5607581268018365, "mean_inter_token_latency_ms_p95": 5.63118492637091, "mean_inter_token_latency_ms_p99": 5.681598116643727, "mean_inter_token_latency_ms_samples": 52, "mean_inter_token_latency_ms_stddev": 0.18448283667424192, "mean_time_per_output_token_max_ms": 5.681598116643727, "mean_time_per_output_token_mean_ci95_high_ms": 5.323992976855044, "mean_time_per_output_token_mean_ci95_low_ms": 5.223706964465723, "mean_time_per_output_token_mean_ms": 5.273849970660383, "mean_time_per_output_token_min_ms": 4.978437372413394, "mean_time_per_output_token_p25_ms": 5.147811622567651, "mean_time_per_output_token_p50_ms": 5.271251011373741, "mean_time_per_output_token_p75_ms": 5.366094364723731, "mean_time_per_output_token_p90_ms": 5.5607581268018365, "mean_time_per_output_token_p95_ms": 5.63118492637091, "mean_time_per_output_token_p99_ms": 5.681598116643727, "mean_time_per_output_token_samples": 52, "mean_time_per_output_token_stddev_ms": 0.18448283667424192, "output_token_sources": { "usage": 52 }, "output_tokens": 14348, "prefix_cache_hit_avg": 0.0011702978394623165, "prompt_token_samples": 52, "prompt_tokens": 399899, "prompt_tokens_max": 7708.0, "prompt_tokens_mean": 7690.365384615385, "prompt_tokens_mean_ci95_high": 7693.3501411136585, "prompt_tokens_mean_ci95_low": 7687.380628117111, "prompt_tokens_min": 7675.0, "prompt_tokens_p25": 7680.0, "prompt_tokens_p50": 7688.0, "prompt_tokens_p75": 7701.0, "prompt_tokens_p90": 7705.0, "prompt_tokens_p95": 7707.0, "prompt_tokens_p99": 7708.0, "prompt_tokens_samples": 52, "prompt_tokens_stddev": 10.981318978876677, "reasoning_chunks": 11442, "request_canary_passed": 0, "request_canary_samples": 0, "throughput_performance_eligible": true, "throughput_tok_s": 116.38054168499878, "time_to_first_output_max_ms": 966.3640120998025, "time_to_first_output_mean_ci95_high_ms": 931.7967081876968, "time_to_first_output_mean_ci95_low_ms": 914.0613074114265, "time_to_first_output_mean_ms": 922.9290077995616, "time_to_first_output_min_ms": 872.0036060549319, "time_to_first_output_p25_ms": 890.7266779569909, "time_to_first_output_p50_ms": 910.0016739685088, "time_to_first_output_p75_ms": 956.746709998697, "time_to_first_output_p90_ms": 963.1182960001752, "time_to_first_output_p95_ms": 965.2501009404659, "time_to_first_output_p99_ms": 966.3640120998025, "time_to_first_output_samples": 52, "time_to_first_output_stddev_ms": 32.62545759680969, "tpot_max_ms": 5.681598116643727, "tpot_mean_ci95_high_ms": 5.323992976855044, "tpot_mean_ci95_low_ms": 5.223706964465723, "tpot_mean_ms": 5.273849970660383, "tpot_min_ms": 4.978437372413394, "tpot_p25_ms": 5.147811622567651, "tpot_p50_ms": 5.271251011373741, "tpot_p75_ms": 5.366094364723731, "tpot_p90_ms": 5.5607581268018365, "tpot_p95_ms": 5.63118492637091, "tpot_p99_ms": 5.681598116643727, "tpot_samples": 52, "tpot_stddev_ms": 0.18448283667424192, "ttft_max_ms": 2964.0783859649673, "ttft_mean_ci95_high_ms": 2214.001719678661, "ttft_mean_ci95_low_ms": 1984.0497134361037, "ttft_mean_ms": 2099.0257165573826, "ttft_min_ms": 1298.4473570249975, "ttft_p25_ms": 1783.4194710012525, "ttft_p50_ms": 2004.7864660155028, "ttft_p75_ms": 2454.0148170199245, "ttft_p90_ms": 2678.6721110111102, "ttft_p95_ms": 2811.0748099861667, "ttft_p99_ms": 2964.0783859649673, "ttft_samples": 52, "ttft_stddev_ms": 423.0121170425288 }, "actual_start_unix_ns": 1792647036571694646, "actual_end_unix_ns": 1792647159856911003 }, { "name": "flare-control-c1-r1", "arm": "flare-control", "concurrency": 1, "repetition": "r1", "attempted": 52, "completed": 52, "failed": 0, "wall_ms": 127162.90889203083, "metrics": { "completion_tokens_max": 479.0, "completion_tokens_mean": 280.1923076923077, "completion_tokens_mean_ci95_high": 303.33613570494026, "completion_tokens_mean_ci95_low": 257.0484796796751, "completion_tokens_min": 130.0, "completion_tokens_p25": 211.0, "completion_tokens_p50": 282.0, "completion_tokens_p75": 339.0, "completion_tokens_p90": 386.0, "completion_tokens_p95": 425.0, "completion_tokens_p99": 479.0, "completion_tokens_samples": 52, "completion_tokens_stddev": 85.1492434796597, "content_chunks": 2844, "content_chunks_s": null, "controlled_output_exact_adherence": 0, "controlled_output_exact_adherence_rate": null, "controlled_output_samples": 0, "controlled_output_unobservable": 0, "decode_tok_s_max": 197.5991438817123, "decode_tok_s_mean": 185.54234242203245, "decode_tok_s_mean_ci95_high": 187.0099579306332, "decode_tok_s_mean_ci95_low": 184.0747269134317, "decode_tok_s_min": 174.69884081825947, "decode_tok_s_p25": 181.72210243828712, "decode_tok_s_p50": 185.02512721279743, "decode_tok_s_p75": 188.20181424943678, "decode_tok_s_p90": 193.52847277530634, "decode_tok_s_p95": 194.893980532655, "decode_tok_s_p99": 197.5991438817123, "decode_tok_s_samples": 52, "decode_tok_s_stddev": 5.39955404992464, "e2e_max_ms": 3441.2143629742786, "e2e_mean_ci95_high_ms": 2565.85825548625, "e2e_mean_ci95_low_ms": 2322.6841323407984, "e2e_mean_ms": 2444.2711939135243, "e2e_min_ms": 1632.4321479769424, "e2e_p25_ms": 2078.8057239260525, "e2e_p50_ms": 2482.7584320446476, "e2e_p75_ms": 2767.90993695613, "e2e_p90_ms": 3000.4303429741412, "e2e_p95_ms": 3232.034087064676, "e2e_p99_ms": 3441.2143629742786, "e2e_samples": 52, "e2e_stddev_ms": 447.33508666679603, "effective_prefill_tok_s_max": 8328.121024005886, "effective_prefill_tok_s_mean": 8165.161615859616, "effective_prefill_tok_s_mean_ci95_high": 8204.382526751871, "effective_prefill_tok_s_mean_ci95_low": 8125.940704967361, "effective_prefill_tok_s_min": 7962.0489856929935, "effective_prefill_tok_s_p25": 8015.104850116455, "effective_prefill_tok_s_p50": 8277.832284753471, "effective_prefill_tok_s_p75": 8297.031231475525, "effective_prefill_tok_s_p90": 8305.809416943155, "effective_prefill_tok_s_p95": 8319.187761810095, "effective_prefill_tok_s_p99": 8328.121024005886, "effective_prefill_tok_s_samples": 52, "effective_prefill_tok_s_stddev": 144.29898499227502, "generation_max_ms": 2514.431406976655, "generation_mean_ci95_high_ms": 1623.2521944248047, "generation_mean_ci95_low_ms": 1380.9659652798437, "generation_mean_ms": 1502.1090798523242, "generation_min_ms": 681.8779719760641, "generation_p25_ms": 1155.6106669595465, "generation_p50_ms": 1536.9845289969817, "generation_p75_ms": 1815.7917219214141, "generation_p90_ms": 2039.6376729477197, "generation_p95_ms": 2267.4592359689996, "generation_p99_ms": 2514.431406976655, "generation_samples": 52, "generation_stddev_ms": 445.7017461841708, "mean_inter_token_latency_ms_max": 5.724136435686528, "mean_inter_token_latency_ms_mean": 5.394069483281372, "mean_inter_token_latency_ms_mean_ci95_high": 5.436609387171817, "mean_inter_token_latency_ms_mean_ci95_low": 5.351529579390927, "mean_inter_token_latency_ms_min": 5.060750671058699, "mean_inter_token_latency_ms_p25": 5.297760189474596, "mean_inter_token_latency_ms_p50": 5.404071185417128, "mean_inter_token_latency_ms_p75": 5.475245737491757, "mean_inter_token_latency_ms_p90": 5.587653773425157, "mean_inter_token_latency_ms_p95": 5.659581492177378, "mean_inter_token_latency_ms_p99": 5.724136435686528, "mean_inter_token_latency_ms_samples": 52, "mean_inter_token_latency_ms_stddev": 0.15651000482684557, "mean_time_per_output_token_max_ms": 5.724136435686528, "mean_time_per_output_token_mean_ci95_high_ms": 5.436609387171817, "mean_time_per_output_token_mean_ci95_low_ms": 5.351529579390927, "mean_time_per_output_token_mean_ms": 5.394069483281372, "mean_time_per_output_token_min_ms": 5.060750671058699, "mean_time_per_output_token_p25_ms": 5.297760189474596, "mean_time_per_output_token_p50_ms": 5.404071185417128, "mean_time_per_output_token_p75_ms": 5.475245737491757, "mean_time_per_output_token_p90_ms": 5.587653773425157, "mean_time_per_output_token_p95_ms": 5.659581492177378, "mean_time_per_output_token_p99_ms": 5.724136435686528, "mean_time_per_output_token_samples": 52, "mean_time_per_output_token_stddev_ms": 0.15651000482684557, "output_token_sources": { "usage": 52 }, "output_tokens": 14570, "prefix_cache_hit_avg": 0.0011702978394623165, "prompt_token_samples": 52, "prompt_tokens": 399899, "prompt_tokens_max": 7708.0, "prompt_tokens_mean": 7690.365384615385, "prompt_tokens_mean_ci95_high": 7693.3501411136585, "prompt_tokens_mean_ci95_low": 7687.380628117111, "prompt_tokens_min": 7675.0, "prompt_tokens_p25": 7680.0, "prompt_tokens_p50": 7688.0, "prompt_tokens_p75": 7701.0, "prompt_tokens_p90": 7705.0, "prompt_tokens_p95": 7707.0, "prompt_tokens_p99": 7708.0, "prompt_tokens_samples": 52, "prompt_tokens_stddev": 10.981318978876677, "reasoning_chunks": 11617, "request_canary_passed": 0, "request_canary_samples": 0, "throughput_performance_eligible": true, "throughput_tok_s": 114.57743556630048, "time_to_first_output_max_ms": 967.8413199726492, "time_to_first_output_mean_ci95_high_ms": 947.0433937360625, "time_to_first_output_mean_ci95_low_ms": 937.2808343863381, "time_to_first_output_mean_ms": 942.1621140612003, "time_to_first_output_min_ms": 921.5764249674976, "time_to_first_output_p25_ms": 926.0374510195106, "time_to_first_output_p50_ms": 928.4125480335206, "time_to_first_output_p75_ms": 960.2154290769249, "time_to_first_output_p90_ms": 963.7098609237, "time_to_first_output_p95_ms": 964.9624049197882, "time_to_first_output_p99_ms": 967.8413199726492, "time_to_first_output_samples": 52, "time_to_first_output_stddev_ms": 17.958881793465252, "tpot_max_ms": 5.724136435686528, "tpot_mean_ci95_high_ms": 5.436609387171817, "tpot_mean_ci95_low_ms": 5.351529579390927, "tpot_mean_ms": 5.394069483281372, "tpot_min_ms": 5.060750671058699, "tpot_p25_ms": 5.297760189474596, "tpot_p50_ms": 5.404071185417128, "tpot_p75_ms": 5.475245737491757, "tpot_p90_ms": 5.587653773425157, "tpot_p95_ms": 5.659581492177378, "tpot_p99_ms": 5.724136435686528, "tpot_samples": 52, "tpot_stddev_ms": 0.15651000482684557, "ttft_max_ms": 3137.2468679910526, "ttft_mean_ci95_high_ms": 2283.7652815819256, "ttft_mean_ci95_low_ms": 2039.9327759469813, "ttft_mean_ms": 2161.8490287644536, "ttft_min_ms": 1322.0151220448315, "ttft_p25_ms": 1834.9110659910366, "ttft_p50_ms": 2189.544945023954, "ttft_p75_ms": 2487.6664399635047, "ttft_p90_ms": 2709.416968980804, "ttft_p95_ms": 2994.8108640965074, "ttft_p99_ms": 3137.2468679910526, "ttft_samples": 52, "ttft_stddev_ms": 448.5462253528848 }, "actual_start_unix_ns": 1792647160180405851, "actual_end_unix_ns": 1792647287343316827 }, { "name": "flare-control-c1-r2", "arm": "flare-control", "concurrency": 1, "repetition": "r2", "attempted": 52, "completed": 52, "failed": 0, "wall_ms": 125083.87294597924, "metrics": { "completion_tokens_max": 532.0, "completion_tokens_mean": 273.9423076923077, "completion_tokens_mean_ci95_high": 297.10192641394707, "completion_tokens_mean_ci95_low": 250.78268897066832, "completion_tokens_min": 130.0, "completion_tokens_p25": 210.0, "completion_tokens_p50": 282.0, "completion_tokens_p75": 327.0, "completion_tokens_p90": 375.0, "completion_tokens_p95": 425.0, "completion_tokens_p99": 532.0, "completion_tokens_samples": 52, "completion_tokens_stddev": 85.20733961333319, "content_chunks": 2874, "content_chunks_s": null, "controlled_output_exact_adherence": 0, "controlled_output_exact_adherence_rate": null, "controlled_output_samples": 0, "controlled_output_unobservable": 0, "decode_tok_s_max": 197.95558257188242, "decode_tok_s_mean": 186.15197911531052, "decode_tok_s_mean_ci95_high": 187.65286999174668, "decode_tok_s_mean_ci95_low": 184.65108823887437, "decode_tok_s_min": 176.08627045875497, "decode_tok_s_p25": 181.7785282906059, "decode_tok_s_p50": 185.5687073891546, "decode_tok_s_p75": 188.92940665005807, "decode_tok_s_p90": 194.48460191746554, "decode_tok_s_p95": 196.76881236995698, "decode_tok_s_p99": 197.95558257188242, "decode_tok_s_samples": 52, "decode_tok_s_stddev": 5.521978585578226, "e2e_max_ms": 3744.8392959777266, "e2e_mean_ci95_high_ms": 2526.8968391929266, "e2e_mean_ci95_low_ms": 2281.440384733597, "e2e_mean_ms": 2404.168611963262, "e2e_min_ms": 1634.2224940890446, "e2e_p25_ms": 2060.4719219263643, "e2e_p50_ms": 2393.1181200314313, "e2e_p75_ms": 2729.361913050525, "e2e_p90_ms": 2930.6076409993693, "e2e_p95_ms": 3171.6267470037565, "e2e_p99_ms": 3744.8392959777266, "e2e_samples": 52, "e2e_stddev_ms": 451.53358798301275, "effective_prefill_tok_s_max": 8340.685750097076, "effective_prefill_tok_s_mean": 8190.525423939406, "effective_prefill_tok_s_mean_ci95_high": 8228.862883720543, "effective_prefill_tok_s_mean_ci95_low": 8152.187964158269, "effective_prefill_tok_s_min": 7994.109249084489, "effective_prefill_tok_s_p25": 8045.63821943011, "effective_prefill_tok_s_p50": 8292.85077255015, "effective_prefill_tok_s_p75": 8320.373473265645, "effective_prefill_tok_s_p90": 8323.667182006595, "effective_prefill_tok_s_p95": 8333.291042092194, "effective_prefill_tok_s_p99": 8340.685750097076, "effective_prefill_tok_s_samples": 52, "effective_prefill_tok_s_stddev": 141.04865001217289, "generation_max_ms": 2821.7048899969086, "generation_mean_ci95_high_ms": 1587.4748221595942, "generation_mean_ci95_low_ms": 1342.403317077455, "generation_mean_ms": 1464.9390696185246, "generation_min_ms": 679.960434907116, "generation_p25_ms": 1101.3369370484725, "generation_p50_ms": 1470.5534380627796, "generation_p75_ms": 1773.1490610167384, "generation_p90_ms": 2009.4603600446135, "generation_p95_ms": 2248.289212002419, "generation_p99_ms": 2821.7048899969086, "generation_samples": 52, "generation_stddev_ms": 450.82544782080925, "mean_inter_token_latency_ms_max": 5.679034472106853, "mean_inter_token_latency_ms_mean": 5.376562619108197, "mean_inter_token_latency_ms_mean_ci95_high": 5.419638831122864, "mean_inter_token_latency_ms_mean_ci95_low": 5.333486407093531, "mean_inter_token_latency_ms_min": 5.0516382867700935, "mean_inter_token_latency_ms_p25": 5.271011123310977, "mean_inter_token_latency_ms_p50": 5.382247157982206, "mean_inter_token_latency_ms_p75": 5.461166239853493, "mean_inter_token_latency_ms_p90": 5.578520950264405, "mean_inter_token_latency_ms_p95": 5.636357042219555, "mean_inter_token_latency_ms_p99": 5.679034472106853, "mean_inter_token_latency_ms_samples": 52, "mean_inter_token_latency_ms_stddev": 0.15848315425677326, "mean_time_per_output_token_max_ms": 5.679034472106853, "mean_time_per_output_token_mean_ci95_high_ms": 5.419638831122864, "mean_time_per_output_token_mean_ci95_low_ms": 5.333486407093531, "mean_time_per_output_token_mean_ms": 5.376562619108197, "mean_time_per_output_token_min_ms": 5.0516382867700935, "mean_time_per_output_token_p25_ms": 5.271011123310977, "mean_time_per_output_token_p50_ms": 5.382247157982206, "mean_time_per_output_token_p75_ms": 5.461166239853493, "mean_time_per_output_token_p90_ms": 5.578520950264405, "mean_time_per_output_token_p95_ms": 5.636357042219555, "mean_time_per_output_token_p99_ms": 5.679034472106853, "mean_time_per_output_token_samples": 52, "mean_time_per_output_token_stddev_ms": 0.15848315425677326, "output_token_sources": { "usage": 52 }, "output_tokens": 14245, "prefix_cache_hit_avg": 0.0011702978394623165, "prompt_token_samples": 52, "prompt_tokens": 399899, "prompt_tokens_max": 7708.0, "prompt_tokens_mean": 7690.365384615385, "prompt_tokens_mean_ci95_high": 7693.3501411136585, "prompt_tokens_mean_ci95_low": 7687.380628117111, "prompt_tokens_min": 7675.0, "prompt_tokens_p25": 7680.0, "prompt_tokens_p50": 7688.0, "prompt_tokens_p75": 7701.0, "prompt_tokens_p90": 7705.0, "prompt_tokens_p95": 7707.0, "prompt_tokens_p99": 7708.0, "prompt_tokens_samples": 52, "prompt_tokens_stddev": 10.981318978876677, "reasoning_chunks": 11262, "request_canary_passed": 0, "request_canary_samples": 0, "throughput_performance_eligible": true, "throughput_tok_s": 113.88358598515796, "time_to_first_output_max_ms": 964.2099901102483, "time_to_first_output_mean_ci95_high_ms": 943.9828950864045, "time_to_first_output_mean_ci95_low_ms": 934.4761896030694, "time_to_first_output_mean_ms": 939.229542344737, "time_to_first_output_min_ms": 920.8671170054004, "time_to_first_output_p25_ms": 922.9162960546091, "time_to_first_output_p50_ms": 925.8577310247347, "time_to_first_output_p75_ms": 956.9333059480414, "time_to_first_output_p90_ms": 961.5098820067942, "time_to_first_output_p95_ms": 962.544719921425, "time_to_first_output_p99_ms": 964.2099901102483, "time_to_first_output_samples": 52, "time_to_first_output_stddev_ms": 17.488221469846614, "tpot_max_ms": 5.679034472106853, "tpot_mean_ci95_high_ms": 5.419638831122864, "tpot_mean_ci95_low_ms": 5.333486407093531, "tpot_mean_ms": 5.376562619108197, "tpot_min_ms": 5.0516382867700935, "tpot_p25_ms": 5.271011123310977, "tpot_p50_ms": 5.382247157982206, "tpot_p75_ms": 5.461166239853493, "tpot_p90_ms": 5.578520950264405, "tpot_p95_ms": 5.636357042219555, "tpot_p99_ms": 5.679034472106853, "tpot_samples": 52, "tpot_stddev_ms": 0.15848315425677326, "ttft_max_ms": 3428.1929610297084, "ttft_mean_ci95_high_ms": 2241.7170532403284, "ttft_mean_ci95_low_ms": 1998.7699622210416, "ttft_mean_ms": 2120.243507730685, "ttft_min_ms": 1318.808465031907, "ttft_p25_ms": 1783.838030998595, "ttft_p50_ms": 2131.2407180666924, "ttft_p75_ms": 2442.6455959910527, "ttft_p90_ms": 2627.329553011805, "ttft_p95_ms": 2843.11183297541, "ttft_p99_ms": 3428.1929610297084, "ttft_samples": 52, "ttft_stddev_ms": 446.9174458647219 }, "actual_start_unix_ns": 1792647287672764171, "actual_end_unix_ns": 1792647412756639049 }, { "name": "flare-control-c1-r3", "arm": "flare-control", "concurrency": 1, "repetition": "r3", "attempted": 52, "completed": 52, "failed": 0, "wall_ms": 126337.54927292466, "metrics": { "completion_tokens_max": 505.0, "completion_tokens_mean": 278.8269230769231, "completion_tokens_mean_ci95_high": 302.4440770412777, "completion_tokens_mean_ci95_low": 255.2097691125685, "completion_tokens_min": 132.0, "completion_tokens_p25": 211.0, "completion_tokens_p50": 265.0, "completion_tokens_p75": 346.0, "completion_tokens_p90": 378.0, "completion_tokens_p95": 434.0, "completion_tokens_p99": 505.0, "completion_tokens_samples": 52, "completion_tokens_stddev": 86.89067306021235, "content_chunks": 2857, "content_chunks_s": null, "controlled_output_exact_adherence": 0, "controlled_output_exact_adherence_rate": null, "controlled_output_samples": 0, "controlled_output_unobservable": 0, "decode_tok_s_max": 199.55234368288882, "decode_tok_s_mean": 186.29080454755345, "decode_tok_s_mean_ci95_high": 187.92309698284834, "decode_tok_s_mean_ci95_low": 184.65851211225856, "decode_tok_s_min": 175.19018521627197, "decode_tok_s_p25": 182.063131915195, "decode_tok_s_p50": 186.06013611214968, "decode_tok_s_p75": 188.82105281922748, "decode_tok_s_p90": 195.61319740049518, "decode_tok_s_p95": 197.77889866156414, "decode_tok_s_p99": 199.55234368288882, "decode_tok_s_samples": 52, "decode_tok_s_stddev": 6.005422522456848, "e2e_max_ms": 3613.902127952315, "e2e_mean_ci95_high_ms": 2552.4697006018755, "e2e_mean_ci95_low_ms": 2304.104160844177, "e2e_mean_ms": 2428.286930723026, "e2e_min_ms": 1616.129406960681, "e2e_p25_ms": 2075.613309047185, "e2e_p50_ms": 2340.827399981208, "e2e_p75_ms": 2773.9633839810267, "e2e_p90_ms": 2968.2000529719517, "e2e_p95_ms": 3163.1907179253176, "e2e_p99_ms": 3613.902127952315, "e2e_samples": 52, "e2e_stddev_ms": 456.885045231976, "effective_prefill_tok_s_max": 8405.17200998029, "effective_prefill_tok_s_mean": 8187.984583911757, "effective_prefill_tok_s_mean_ci95_high": 8229.437704978918, "effective_prefill_tok_s_mean_ci95_low": 8146.531462844595, "effective_prefill_tok_s_min": 7964.485649854875, "effective_prefill_tok_s_p25": 8030.847463866219, "effective_prefill_tok_s_p50": 8309.102511408217, "effective_prefill_tok_s_p75": 8324.953969362057, "effective_prefill_tok_s_p90": 8332.094368458882, "effective_prefill_tok_s_p95": 8335.450081293071, "effective_prefill_tok_s_p99": 8405.17200998029, "effective_prefill_tok_s_samples": 52, "effective_prefill_tok_s_stddev": 152.5115852404785, "generation_max_ms": 2690.752713009715, "generation_mean_ci95_high_ms": 1612.3625282095213, "generation_mean_ci95_low_ms": 1365.072149659345, "generation_mean_ms": 1488.7173389344332, "generation_min_ms": 692.6848739385605, "generation_p25_ms": 1153.4460480324924, "generation_p50_ms": 1390.5691639520228, "generation_p75_ms": 1838.805104023777, "generation_p90_ms": 2010.1027969503775, "generation_p95_ms": 2245.5123179825023, "generation_p99_ms": 2690.752713009715, "generation_samples": 52, "generation_stddev_ms": 454.9072141793677, "mean_inter_token_latency_ms_max": 5.708082326447123, "mean_inter_token_latency_ms_mean": 5.373387611154373, "mean_inter_token_latency_ms_mean_ci95_high": 5.420156298869133, "mean_inter_token_latency_ms_mean_ci95_low": 5.326618923439612, "mean_inter_token_latency_ms_min": 5.011216513643722, "mean_inter_token_latency_ms_p25": 5.2716061005003425, "mean_inter_token_latency_ms_p50": 5.363845565032376, "mean_inter_token_latency_ms_p75": 5.465419479957634, "mean_inter_token_latency_ms_p90": 5.606952777558253, "mean_inter_token_latency_ms_p95": 5.650322231201407, "mean_inter_token_latency_ms_p99": 5.708082326447123, "mean_inter_token_latency_ms_samples": 52, "mean_inter_token_latency_ms_stddev": 0.17206826698136032, "mean_time_per_output_token_max_ms": 5.708082326447123, "mean_time_per_output_token_mean_ci95_high_ms": 5.420156298869133, "mean_time_per_output_token_mean_ci95_low_ms": 5.326618923439612, "mean_time_per_output_token_mean_ms": 5.373387611154373, "mean_time_per_output_token_min_ms": 5.011216513643722, "mean_time_per_output_token_p25_ms": 5.2716061005003425, "mean_time_per_output_token_p50_ms": 5.363845565032376, "mean_time_per_output_token_p75_ms": 5.465419479957634, "mean_time_per_output_token_p90_ms": 5.606952777558253, "mean_time_per_output_token_p95_ms": 5.650322231201407, "mean_time_per_output_token_p99_ms": 5.708082326447123, "mean_time_per_output_token_samples": 52, "mean_time_per_output_token_stddev_ms": 0.17206826698136032, "output_token_sources": { "usage": 52 }, "output_tokens": 14499, "prefix_cache_hit_avg": 0.0011702978394623165, "prompt_token_samples": 52, "prompt_tokens": 399899, "prompt_tokens_max": 7708.0, "prompt_tokens_mean": 7690.365384615385, "prompt_tokens_mean_ci95_high": 7693.3501411136585, "prompt_tokens_mean_ci95_low": 7687.380628117111, "prompt_tokens_min": 7675.0, "prompt_tokens_p25": 7680.0, "prompt_tokens_p50": 7688.0, "prompt_tokens_p75": 7701.0, "prompt_tokens_p90": 7705.0, "prompt_tokens_p95": 7707.0, "prompt_tokens_p99": 7708.0, "prompt_tokens_samples": 52, "prompt_tokens_stddev": 10.981318978876677, "reasoning_chunks": 11533, "request_canary_passed": 0, "request_canary_samples": 0, "throughput_performance_eligible": true, "throughput_tok_s": 114.76398017408174, "time_to_first_output_max_ms": 967.4196600681171, "time_to_first_output_mean_ci95_high_ms": 944.6884180714118, "time_to_first_output_mean_ci95_low_ms": 934.4507655057743, "time_to_first_output_mean_ms": 939.569591788593, "time_to_first_output_min_ms": 913.128248997964, "time_to_first_output_p25_ms": 922.4571470404044, "time_to_first_output_p50_ms": 924.6799709508196, "time_to_first_output_p75_ms": 958.0972560215741, "time_to_first_output_p90_ms": 962.9561329493299, "time_to_first_output_p95_ms": 966.1378480959684, "time_to_first_output_p99_ms": 967.4196600681171, "time_to_first_output_samples": 52, "time_to_first_output_stddev_ms": 18.83284758458739, "tpot_max_ms": 5.708082326447123, "tpot_mean_ci95_high_ms": 5.420156298869133, "tpot_mean_ci95_low_ms": 5.326618923439612, "tpot_mean_ms": 5.373387611154373, "tpot_min_ms": 5.011216513643722, "tpot_p25_ms": 5.2716061005003425, "tpot_p50_ms": 5.363845565032376, "tpot_p75_ms": 5.465419479957634, "tpot_p90_ms": 5.606952777558253, "tpot_p95_ms": 5.650322231201407, "tpot_p99_ms": 5.708082326447123, "tpot_samples": 52, "tpot_stddev_ms": 0.17206826698136032, "ttft_max_ms": 3363.0841079866514, "ttft_mean_ci95_high_ms": 2269.952393500991, "ttft_mean_ci95_low_ms": 2025.9261516428296, "ttft_mean_ms": 2147.9392725719104, "ttft_min_ms": 1352.0799539983273, "ttft_p25_ms": 1828.8939900230616, "ttft_p50_ms": 2011.644673999399, "ttft_p75_ms": 2483.867362025194, "ttft_p90_ms": 2704.342719982378, "ttft_p95_ms": 2845.0895429123193, "ttft_p99_ms": 3363.0841079866514, "ttft_samples": 52, "ttft_stddev_ms": 448.90261611141716 }, "actual_start_unix_ns": 1792647413160180996, "actual_end_unix_ns": 1792647539497732650 }, { "name": "flare-control-c4-prime", "arm": "flare-control", "concurrency": 4, "repetition": "prime", "attempted": 160, "completed": 160, "failed": 0, "wall_ms": 310684.7816060763, "metrics": { "completion_tokens_max": 535.0, "completion_tokens_mean": 286.95, "completion_tokens_mean_ci95_high": 299.53242223933074, "completion_tokens_mean_ci95_low": 274.36757776066923, "completion_tokens_min": 86.0, "completion_tokens_p25": 234.0, "completion_tokens_p50": 278.0, "completion_tokens_p75": 320.0, "completion_tokens_p90": 389.0, "completion_tokens_p95": 435.0, "completion_tokens_p99": 519.0, "completion_tokens_samples": 160, "completion_tokens_stddev": 81.20227093518658, "content_chunks": 9005, "content_chunks_s": null, "controlled_output_exact_adherence": 0, "controlled_output_exact_adherence_rate": null, "controlled_output_samples": 0, "controlled_output_unobservable": 0, "decode_tok_s_max": 123.75872735034974, "decode_tok_s_mean": 73.11081115358698, "decode_tok_s_mean_ci95_high": 74.68465261815314, "decode_tok_s_mean_ci95_low": 71.53696968902082, "decode_tok_s_min": 47.75734791156946, "decode_tok_s_p25": 67.05490914237737, "decode_tok_s_p50": 71.72274357089327, "decode_tok_s_p75": 78.11533230563282, "decode_tok_s_p90": 85.06069831255896, "decode_tok_s_p95": 90.80244084090606, "decode_tok_s_p99": 105.59798951367436, "decode_tok_s_samples": 160, "decode_tok_s_stddev": 10.156987151110702, "e2e_max_ms": 11150.089272065088, "e2e_mean_ci95_high_ms": 7937.5034845605305, "e2e_mean_ci95_low_ms": 7544.676956351376, "e2e_mean_ms": 7741.090220455953, "e2e_min_ms": 4990.432252991013, "e2e_p25_ms": 6990.52547395695, "e2e_p50_ms": 7640.071667963639, "e2e_p75_ms": 8423.601867980324, "e2e_p90_ms": 9575.06790093612, "e2e_p95_ms": 10115.27174001094, "e2e_p99_ms": 11122.513103997335, "e2e_samples": 160, "e2e_stddev_ms": 1267.5781168135825, "effective_prefill_tok_s_max": 2516.658253920352, "effective_prefill_tok_s_mean": 2061.845166738082, "effective_prefill_tok_s_mean_ci95_high": 2085.1141492232214, "effective_prefill_tok_s_mean_ci95_low": 2038.576184252942, "effective_prefill_tok_s_min": 1439.3437086419822, "effective_prefill_tok_s_p25": 1973.5326749208643, "effective_prefill_tok_s_p50": 2056.6405938995404, "effective_prefill_tok_s_p75": 2147.305914802889, "effective_prefill_tok_s_p90": 2242.2207060279543, "effective_prefill_tok_s_p95": 2301.14999462989, "effective_prefill_tok_s_p99": 2511.7449536304366, "effective_prefill_tok_s_samples": 160, "effective_prefill_tok_s_stddev": 150.16935405634013, "generation_max_ms": 7498.726202058606, "generation_mean_ci95_high_ms": 4158.139310559888, "generation_mean_ci95_low_ms": 3784.3995781038057, "generation_mean_ms": 3971.269444331847, "generation_min_ms": 1333.3686830010265, "generation_p25_ms": 3225.9992060717195, "generation_p50_ms": 3819.8554479749873, "generation_p75_ms": 4466.154893976636, "generation_p90_ms": 5686.616264982149, "generation_p95_ms": 6255.522697931156, "generation_p99_ms": 7282.1220780024305, "generation_samples": 160, "generation_stddev_ms": 1205.9885782277313, "mean_inter_token_latency_ms_max": 20.93918619291137, "mean_inter_token_latency_ms_mean": 13.92330142743511, "mean_inter_token_latency_ms_mean_ci95_high": 14.207255736413428, "mean_inter_token_latency_ms_mean_ci95_low": 13.639347118456792, "mean_inter_token_latency_ms_min": 8.080238229737857, "mean_inter_token_latency_ms_p25": 12.776411344892024, "mean_inter_token_latency_ms_p50": 13.912490768914951, "mean_inter_token_latency_ms_p75": 14.867072247695527, "mean_inter_token_latency_ms_p90": 16.050645874726566, "mean_inter_token_latency_ms_p95": 17.010825808455877, "mean_inter_token_latency_ms_p99": 18.962594726441527, "mean_inter_token_latency_ms_samples": 160, "mean_inter_token_latency_ms_stddev": 1.8325354444707813, "mean_time_per_output_token_max_ms": 20.93918619291137, "mean_time_per_output_token_mean_ci95_high_ms": 14.207255736413428, "mean_time_per_output_token_mean_ci95_low_ms": 13.639347118456792, "mean_time_per_output_token_mean_ms": 13.92330142743511, "mean_time_per_output_token_min_ms": 8.080238229737857, "mean_time_per_output_token_p25_ms": 12.776411344892024, "mean_time_per_output_token_p50_ms": 13.912490768914951, "mean_time_per_output_token_p75_ms": 14.867072247695527, "mean_time_per_output_token_p90_ms": 16.050645874726566, "mean_time_per_output_token_p95_ms": 17.010825808455877, "mean_time_per_output_token_p99_ms": 18.962594726441527, "mean_time_per_output_token_samples": 160, "mean_time_per_output_token_stddev_ms": 1.8325354444707813, "output_token_sources": { "usage": 160 }, "output_tokens": 45912, "prefix_cache_hit_avg": 0.0011643533907411787, "prompt_token_samples": 160, "prompt_tokens": 1236764, "prompt_tokens_max": 7798.0, "prompt_tokens_mean": 7729.775, "prompt_tokens_mean_ci95_high": 7735.29436497065, "prompt_tokens_mean_ci95_low": 7724.255635029349, "prompt_tokens_min": 7675.0, "prompt_tokens_p25": 7702.0, "prompt_tokens_p50": 7725.0, "prompt_tokens_p75": 7759.0, "prompt_tokens_p90": 7782.0, "prompt_tokens_p95": 7790.0, "prompt_tokens_p99": 7797.0, "prompt_tokens_samples": 160, "prompt_tokens_stddev": 35.61992764286613, "reasoning_chunks": 36565, "request_canary_passed": 0, "request_canary_samples": 0, "throughput_performance_eligible": true, "throughput_tok_s": 147.77679087678257, "time_to_first_output_max_ms": 5356.607983005233, "time_to_first_output_mean_ci95_high_ms": 3815.026655951005, "time_to_first_output_mean_ci95_low_ms": 3724.614896297208, "time_to_first_output_mean_ms": 3769.8207761241065, "time_to_first_output_min_ms": 3050.0764210009947, "time_to_first_output_p25_ms": 3601.0354380123317, "time_to_first_output_p50_ms": 3767.3213148955256, "time_to_first_output_p75_ms": 3927.582502947189, "time_to_first_output_p90_ms": 4080.426074913703, "time_to_first_output_p95_ms": 4198.083230992779, "time_to_first_output_p99_ms": 4709.075290011242, "time_to_first_output_samples": 160, "time_to_first_output_stddev_ms": 291.7419262956281, "tpot_max_ms": 20.93918619291137, "tpot_mean_ci95_high_ms": 14.207255736413428, "tpot_mean_ci95_low_ms": 13.639347118456792, "tpot_mean_ms": 13.92330142743511, "tpot_min_ms": 8.080238229737857, "tpot_p25_ms": 12.776411344892024, "tpot_p50_ms": 13.912490768914951, "tpot_p75_ms": 14.867072247695527, "tpot_p90_ms": 16.050645874726566, "tpot_p95_ms": 17.010825808455877, "tpot_p99_ms": 18.962594726441527, "tpot_samples": 160, "tpot_stddev_ms": 1.8325354444707813, "ttft_max_ms": 10328.442079946399, "ttft_mean_ci95_high_ms": 7162.704783752124, "ttft_mean_ci95_low_ms": 6791.58711652286, "ttft_mean_ms": 6977.145950137492, "ttft_min_ms": 4099.70710799098, "ttft_p25_ms": 6200.334649998695, "ttft_p50_ms": 6839.738828944974, "ttft_p75_ms": 7653.710862970911, "ttft_p90_ms": 8630.748225026764, "ttft_p95_ms": 9045.181890949607, "ttft_p99_ms": 10202.191052027047, "ttft_samples": 160, "ttft_stddev_ms": 1197.527661605006 }, "actual_start_unix_ns": 1792647539859687030, "actual_end_unix_ns": 1792647850544471175 }, { "name": "flare-control-c4-r1", "arm": "flare-control", "concurrency": 4, "repetition": "r1", "attempted": 160, "completed": 160, "failed": 0, "wall_ms": 306042.91476507206, "metrics": { "completion_tokens_max": 561.0, "completion_tokens_mean": 276.1, "completion_tokens_mean_ci95_high": 288.14029967449204, "completion_tokens_mean_ci95_low": 264.059700325508, "completion_tokens_min": 96.0, "completion_tokens_p25": 223.0, "completion_tokens_p50": 265.0, "completion_tokens_p75": 314.0, "completion_tokens_p90": 370.0, "completion_tokens_p95": 424.0, "completion_tokens_p99": 542.0, "completion_tokens_samples": 160, "completion_tokens_stddev": 77.70361363750752, "content_chunks": 8692, "content_chunks_s": null, "controlled_output_exact_adherence": 0, "controlled_output_exact_adherence_rate": null, "controlled_output_samples": 0, "controlled_output_unobservable": 0, "decode_tok_s_max": 162.51724958225282, "decode_tok_s_mean": 72.91857223969645, "decode_tok_s_mean_ci95_high": 74.82029897481677, "decode_tok_s_mean_ci95_low": 71.01684550457612, "decode_tok_s_min": 46.64587065848308, "decode_tok_s_p25": 67.20118900235644, "decode_tok_s_p50": 71.78844138810767, "decode_tok_s_p75": 77.04622383768266, "decode_tok_s_p90": 83.76669624341102, "decode_tok_s_p95": 88.95842177668641, "decode_tok_s_p99": 127.6154190235459, "decode_tok_s_samples": 160, "decode_tok_s_stddev": 12.273036673910116, "e2e_max_ms": 12683.959729038179, "e2e_mean_ci95_high_ms": 7789.1221896228735, "e2e_mean_ci95_low_ms": 7420.949838705558, "e2e_mean_ms": 7605.036014164216, "e2e_min_ms": 4528.585381922312, "e2e_p25_ms": 6823.300797026604, "e2e_p50_ms": 7540.844279923476, "e2e_p75_ms": 8230.79073394183, "e2e_p90_ms": 9063.07127699256, "e2e_p95_ms": 9584.592539002188, "e2e_p99_ms": 11314.499851083383, "e2e_samples": 160, "e2e_stddev_ms": 1188.0236738750018, "effective_prefill_tok_s_max": 2609.0597801125055, "effective_prefill_tok_s_mean": 2064.758465556689, "effective_prefill_tok_s_mean_ci95_high": 2089.523934951157, "effective_prefill_tok_s_mean_ci95_low": 2039.9929961622204, "effective_prefill_tok_s_min": 1643.838106035555, "effective_prefill_tok_s_p25": 1968.407945125882, "effective_prefill_tok_s_p50": 2049.629336490415, "effective_prefill_tok_s_p75": 2141.4922819851017, "effective_prefill_tok_s_p90": 2262.059299095451, "effective_prefill_tok_s_p95": 2369.4051899915066, "effective_prefill_tok_s_p99": 2558.4291780001868, "effective_prefill_tok_s_samples": 160, "effective_prefill_tok_s_stddev": 159.82712369328837, "generation_max_ms": 7965.125209069811, "generation_mean_ci95_high_ms": 4014.2443719174958, "generation_mean_ci95_low_ms": 3664.1902431845187, "generation_mean_ms": 3839.2173075510073, "generation_min_ms": 1292.1442040242255, "generation_p25_ms": 3162.638501962647, "generation_p50_ms": 3703.7698589265347, "generation_p75_ms": 4410.587624064647, "generation_p90_ms": 5397.667484008707, "generation_p95_ms": 5653.773958911188, "generation_p99_ms": 7343.356732977554, "generation_samples": 160, "generation_stddev_ms": 1129.5595419816327, "mean_inter_token_latency_ms_max": 21.438124873292267, "mean_inter_token_latency_ms_mean": 14.021850620574778, "mean_inter_token_latency_ms_mean_ci95_high": 14.329168890607042, "mean_inter_token_latency_ms_mean_ci95_low": 13.714532350542514, "mean_inter_token_latency_ms_min": 6.153192984563048, "mean_inter_token_latency_ms_p25": 12.910221833263295, "mean_inter_token_latency_ms_p50": 13.914949416396889, "mean_inter_token_latency_ms_p75": 14.859306313243346, "mean_inter_token_latency_ms_p90": 16.424003861505877, "mean_inter_token_latency_ms_p95": 17.316315920751467, "mean_inter_token_latency_ms_p99": 19.626923686946217, "mean_inter_token_latency_ms_samples": 160, "mean_inter_token_latency_ms_stddev": 1.9833177548665697, "mean_time_per_output_token_max_ms": 21.438124873292267, "mean_time_per_output_token_mean_ci95_high_ms": 14.329168890607042, "mean_time_per_output_token_mean_ci95_low_ms": 13.714532350542514, "mean_time_per_output_token_mean_ms": 14.021850620574778, "mean_time_per_output_token_min_ms": 6.153192984563048, "mean_time_per_output_token_p25_ms": 12.910221833263295, "mean_time_per_output_token_p50_ms": 13.914949416396889, "mean_time_per_output_token_p75_ms": 14.859306313243346, "mean_time_per_output_token_p90_ms": 16.424003861505877, "mean_time_per_output_token_p95_ms": 17.316315920751467, "mean_time_per_output_token_p99_ms": 19.626923686946217, "mean_time_per_output_token_samples": 160, "mean_time_per_output_token_stddev_ms": 1.9833177548665697, "output_token_sources": { "usage": 160 }, "output_tokens": 44176, "prefix_cache_hit_avg": 0.0011643533907411787, "prompt_token_samples": 160, "prompt_tokens": 1236764, "prompt_tokens_max": 7798.0, "prompt_tokens_mean": 7729.775, "prompt_tokens_mean_ci95_high": 7735.29436497065, "prompt_tokens_mean_ci95_low": 7724.255635029349, "prompt_tokens_min": 7675.0, "prompt_tokens_p25": 7702.0, "prompt_tokens_p50": 7725.0, "prompt_tokens_p75": 7759.0, "prompt_tokens_p90": 7782.0, "prompt_tokens_p95": 7790.0, "prompt_tokens_p99": 7797.0, "prompt_tokens_samples": 160, "prompt_tokens_stddev": 35.61992764286613, "reasoning_chunks": 35150, "request_canary_passed": 0, "request_canary_samples": 0, "throughput_performance_eligible": true, "throughput_tok_s": 144.3457693961347, "time_to_first_output_max_ms": 4718.834519968368, "time_to_first_output_mean_ci95_high_ms": 3810.8838825085104, "time_to_first_output_mean_ci95_low_ms": 3720.753530717907, "time_to_first_output_mean_ms": 3765.8187066132086, "time_to_first_output_min_ms": 2947.805205010809, "time_to_first_output_p25_ms": 3596.7922870768234, "time_to_first_output_p50_ms": 3773.049894021824, "time_to_first_output_p75_ms": 3932.7803950291127, "time_to_first_output_p90_ms": 4104.568782960996, "time_to_first_output_p95_ms": 4202.858007047325, "time_to_first_output_p99_ms": 4641.934716026299, "time_to_first_output_samples": 160, "time_to_first_output_stddev_ms": 290.8338754801459, "tpot_max_ms": 21.438124873292267, "tpot_mean_ci95_high_ms": 14.329168890607042, "tpot_mean_ci95_low_ms": 13.714532350542514, "tpot_mean_ms": 14.021850620574778, "tpot_min_ms": 6.153192984563048, "tpot_p25_ms": 12.910221833263295, "tpot_p50_ms": 13.914949416396889, "tpot_p75_ms": 14.859306313243346, "tpot_p90_ms": 16.424003861505877, "tpot_p95_ms": 17.316315920751467, "tpot_p99_ms": 19.626923686946217, "tpot_samples": 160, "tpot_stddev_ms": 1.9833177548665697, "ttft_max_ms": 11581.233770004474, "ttft_mean_ci95_high_ms": 7049.196291960223, "ttft_mean_ci95_low_ms": 6691.692952452897, "ttft_mean_ms": 6870.44462220656, "ttft_min_ms": 4016.189166926779, "ttft_p25_ms": 6092.571947956458, "ttft_p50_ms": 6692.41088302806, "ttft_p75_ms": 7476.967433001846, "ttft_p90_ms": 8320.817606057972, "ttft_p95_ms": 8924.96906989254, "ttft_p99_ms": 10635.396335972473, "ttft_samples": 160, "ttft_stddev_ms": 1153.5967591424574 }, "actual_start_unix_ns": 1792647851082883597, "actual_end_unix_ns": 1792648157125800497 }, { "name": "flare-control-c4-r2", "arm": "flare-control", "concurrency": 4, "repetition": "r2", "attempted": 160, "completed": 160, "failed": 0, "wall_ms": 311784.02114100754, "metrics": { "completion_tokens_max": 569.0, "completion_tokens_mean": 287.19375, "completion_tokens_mean_ci95_high": 299.43748005501124, "completion_tokens_mean_ci95_low": 274.9500199449888, "completion_tokens_min": 97.0, "completion_tokens_p25": 240.0, "completion_tokens_p50": 279.0, "completion_tokens_p75": 323.0, "completion_tokens_p90": 389.0, "completion_tokens_p95": 446.0, "completion_tokens_p99": 527.0, "completion_tokens_samples": 160, "completion_tokens_stddev": 79.01647761243682, "content_chunks": 8591, "content_chunks_s": null, "controlled_output_exact_adherence": 0, "controlled_output_exact_adherence_rate": null, "controlled_output_samples": 0, "controlled_output_unobservable": 0, "decode_tok_s_max": 136.94595636213518, "decode_tok_s_mean": 71.49850685538638, "decode_tok_s_mean_ci95_high": 73.09622680939464, "decode_tok_s_mean_ci95_low": 69.90078690137813, "decode_tok_s_min": 47.877920083845304, "decode_tok_s_p25": 66.10757258787987, "decode_tok_s_p50": 71.062532244414, "decode_tok_s_p75": 75.18933803202745, "decode_tok_s_p90": 79.31335561887448, "decode_tok_s_p95": 83.93953684049553, "decode_tok_s_p99": 128.26586339780792, "decode_tok_s_samples": 160, "decode_tok_s_stddev": 10.311090036256273, "e2e_max_ms": 11867.445359937847, "e2e_mean_ci95_high_ms": 7928.041835465144, "e2e_mean_ci95_low_ms": 7582.331964407324, "e2e_mean_ms": 7755.186899936234, "e2e_min_ms": 5380.423863069154, "e2e_p25_ms": 7003.988085081801, "e2e_p50_ms": 7634.084293968044, "e2e_p75_ms": 8423.397873993963, "e2e_p90_ms": 9216.13909106236, "e2e_p95_ms": 9528.854456963018, "e2e_p99_ms": 11110.757480026223, "e2e_samples": 160, "e2e_stddev_ms": 1115.541430761055, "effective_prefill_tok_s_max": 2593.9997643436823, "effective_prefill_tok_s_mean": 2087.6530369884604, "effective_prefill_tok_s_mean_ci95_high": 2110.6177788547843, "effective_prefill_tok_s_mean_ci95_low": 2064.6882951221364, "effective_prefill_tok_s_min": 1580.6698643050363, "effective_prefill_tok_s_p25": 1980.127033100884, "effective_prefill_tok_s_p50": 2080.962191948593, "effective_prefill_tok_s_p75": 2166.5447121005636, "effective_prefill_tok_s_p90": 2250.061661971901, "effective_prefill_tok_s_p95": 2360.8753480006458, "effective_prefill_tok_s_p99": 2528.029445838354, "effective_prefill_tok_s_samples": 160, "effective_prefill_tok_s_stddev": 148.2058983171618, "generation_max_ms": 7863.99481294211, "generation_mean_ci95_high_ms": 4202.247565542814, "generation_mean_ci95_low_ms": 3865.259495276666, "generation_mean_ms": 4033.75353040974, "generation_min_ms": 1606.6092300461605, "generation_p25_ms": 3389.2273301025853, "generation_p50_ms": 3974.437173921615, "generation_p75_ms": 4632.006598054431, "generation_p90_ms": 5364.8841850226745, "generation_p95_ms": 5913.253781036474, "generation_p99_ms": 7396.092047099955, "generation_samples": 160, "generation_stddev_ms": 1087.3978023937682, "mean_inter_token_latency_ms_max": 20.886454512826976, "mean_inter_token_latency_ms_mean": 14.231327984721947, "mean_inter_token_latency_ms_mean_ci95_high": 14.51276352973906, "mean_inter_token_latency_ms_mean_ci95_low": 13.949892439704835, "mean_inter_token_latency_ms_min": 7.302150618858978, "mean_inter_token_latency_ms_p25": 13.29331103701707, "mean_inter_token_latency_ms_p50": 14.061011496387747, "mean_inter_token_latency_ms_p75": 15.014266059409424, "mean_inter_token_latency_ms_p90": 16.37268530031376, "mean_inter_token_latency_ms_p95": 17.131194715179376, "mean_inter_token_latency_ms_p99": 20.459009927769465, "mean_inter_token_latency_ms_samples": 160, "mean_inter_token_latency_ms_stddev": 1.816280279152908, "mean_time_per_output_token_max_ms": 20.886454512826976, "mean_time_per_output_token_mean_ci95_high_ms": 14.51276352973906, "mean_time_per_output_token_mean_ci95_low_ms": 13.949892439704835, "mean_time_per_output_token_mean_ms": 14.231327984721947, "mean_time_per_output_token_min_ms": 7.302150618858978, "mean_time_per_output_token_p25_ms": 13.29331103701707, "mean_time_per_output_token_p50_ms": 14.061011496387747, "mean_time_per_output_token_p75_ms": 15.014266059409424, "mean_time_per_output_token_p90_ms": 16.37268530031376, "mean_time_per_output_token_p95_ms": 17.131194715179376, "mean_time_per_output_token_p99_ms": 20.459009927769465, "mean_time_per_output_token_samples": 160, "mean_time_per_output_token_stddev_ms": 1.816280279152908, "output_token_sources": { "usage": 160 }, "output_tokens": 45951, "prefix_cache_hit_avg": 0.0011643533907411787, "prompt_token_samples": 160, "prompt_tokens": 1236764, "prompt_tokens_max": 7798.0, "prompt_tokens_mean": 7729.775, "prompt_tokens_mean_ci95_high": 7735.29436497065, "prompt_tokens_mean_ci95_low": 7724.255635029349, "prompt_tokens_min": 7675.0, "prompt_tokens_p25": 7702.0, "prompt_tokens_p50": 7725.0, "prompt_tokens_p75": 7759.0, "prompt_tokens_p90": 7782.0, "prompt_tokens_p95": 7790.0, "prompt_tokens_p99": 7797.0, "prompt_tokens_samples": 160, "prompt_tokens_stddev": 35.61992764286613, "reasoning_chunks": 37022, "request_canary_passed": 0, "request_canary_samples": 0, "throughput_performance_eligible": true, "throughput_tok_s": 147.3808690767324, "time_to_first_output_max_ms": 4890.331735019572, "time_to_first_output_mean_ci95_high_ms": 3763.100096839346, "time_to_first_output_mean_ci95_low_ms": 3679.766642213642, "time_to_first_output_mean_ms": 3721.433369526494, "time_to_first_output_min_ms": 2962.6062830211595, "time_to_first_output_p25_ms": 3545.8228799980134, "time_to_first_output_p50_ms": 3707.17977208551, "time_to_first_output_p75_ms": 3895.4489519819617, "time_to_first_output_p90_ms": 4068.841736065224, "time_to_first_output_p95_ms": 4113.352214917541, "time_to_first_output_p99_ms": 4383.598159998655, "time_to_first_output_samples": 160, "time_to_first_output_stddev_ms": 268.90155296685634, "tpot_max_ms": 20.886454512826976, "tpot_mean_ci95_high_ms": 14.51276352973906, "tpot_mean_ci95_low_ms": 13.949892439704835, "tpot_mean_ms": 14.231327984721947, "tpot_min_ms": 7.302150618858978, "tpot_p25_ms": 13.29331103701707, "tpot_p50_ms": 14.061011496387747, "tpot_p75_ms": 15.014266059409424, "tpot_p90_ms": 16.37268530031376, "tpot_p95_ms": 17.131194715179376, "tpot_p99_ms": 20.459009927769465, "tpot_samples": 160, "tpot_stddev_ms": 1.816280279152908, "ttft_max_ms": 10889.315109001473, "ttft_mean_ci95_high_ms": 7213.8737124615045, "ttft_mean_ci95_low_ms": 6872.994755251561, "ttft_mean_ms": 7043.434233856533, "ttft_min_ms": 4630.693274084479, "ttft_p25_ms": 6232.0306760957465, "ttft_p50_ms": 6918.457592022605, "ttft_p75_ms": 7741.341492044739, "ttft_p90_ms": 8445.046177948825, "ttft_p95_ms": 8951.34963199962, "ttft_p99_ms": 10159.724105033092, "ttft_samples": 160, "ttft_stddev_ms": 1099.9529706188755 }, "actual_start_unix_ns": 1792648157482292145, "actual_end_unix_ns": 1792648469266316968 }, { "name": "flare-control-c4-r3", "arm": "flare-control", "concurrency": 4, "repetition": "r3", "attempted": 160, "completed": 160, "failed": 0, "wall_ms": 310849.2424610304, "metrics": { "completion_tokens_max": 666.0, "completion_tokens_mean": 288.025, "completion_tokens_mean_ci95_high": 301.0254600738618, "completion_tokens_mean_ci95_low": 275.0245399261382, "completion_tokens_min": 93.0, "completion_tokens_p25": 226.0, "completion_tokens_p50": 275.0, "completion_tokens_p75": 339.0, "completion_tokens_p90": 383.0, "completion_tokens_p95": 420.0, "completion_tokens_p99": 537.0, "completion_tokens_samples": 160, "completion_tokens_stddev": 83.90013155813105, "content_chunks": 8662, "content_chunks_s": null, "controlled_output_exact_adherence": 0, "controlled_output_exact_adherence_rate": null, "controlled_output_samples": 0, "controlled_output_unobservable": 0, "decode_tok_s_max": 147.73866883955458, "decode_tok_s_mean": 73.22676461338068, "decode_tok_s_mean_ci95_high": 74.89298593566217, "decode_tok_s_mean_ci95_low": 71.5605432910992, "decode_tok_s_min": 50.67349796647653, "decode_tok_s_p25": 66.13638310368478, "decode_tok_s_p50": 72.27617909539926, "decode_tok_s_p75": 77.33229483295234, "decode_tok_s_p90": 83.93348375733025, "decode_tok_s_p95": 90.39535302209539, "decode_tok_s_p99": 100.39315232033337, "decode_tok_s_samples": 160, "decode_tok_s_stddev": 10.753172376218266, "e2e_max_ms": 14075.848345994018, "e2e_mean_ci95_high_ms": 7944.748136653482, "e2e_mean_ci95_low_ms": 7537.194058084594, "e2e_mean_ms": 7740.971097369038, "e2e_min_ms": 4956.869779038243, "e2e_p25_ms": 6777.986389002763, "e2e_p50_ms": 7578.708763932809, "e2e_p75_ms": 8411.8400599109, "e2e_p90_ms": 9431.57409504056, "e2e_p95_ms": 9938.12012905255, "e2e_p99_ms": 11732.285027042963, "e2e_samples": 160, "e2e_stddev_ms": 1315.1011816009236, "effective_prefill_tok_s_max": 2642.687929913561, "effective_prefill_tok_s_mean": 2068.407746063856, "effective_prefill_tok_s_mean_ci95_high": 2090.469953106296, "effective_prefill_tok_s_mean_ci95_low": 2046.3455390214158, "effective_prefill_tok_s_min": 1756.3178267832484, "effective_prefill_tok_s_p25": 1966.2634417301003, "effective_prefill_tok_s_p50": 2041.95305542292, "effective_prefill_tok_s_p75": 2149.754453763539, "effective_prefill_tok_s_p90": 2245.3021251654513, "effective_prefill_tok_s_p95": 2301.8171628253167, "effective_prefill_tok_s_p99": 2543.879953579475, "effective_prefill_tok_s_samples": 160, "effective_prefill_tok_s_stddev": 142.3812744169759, "generation_max_ms": 10008.59323097393, "generation_mean_ci95_high_ms": 4182.164137159652, "generation_mean_ci95_low_ms": 3791.20096107834, "generation_mean_ms": 3986.6825491189957, "generation_min_ms": 1405.6491079973057, "generation_p25_ms": 3153.549670940265, "generation_p50_ms": 3894.075477961451, "generation_p75_ms": 4706.6212189383805, "generation_p90_ms": 5590.8735420089215, "generation_p95_ms": 5966.418721945956, "generation_p99_ms": 7578.087285975926, "generation_samples": 160, "generation_stddev_ms": 1261.5654261942955, "mean_inter_token_latency_ms_max": 19.734181379417663, "mean_inter_token_latency_ms_mean": 13.902525036483855, "mean_inter_token_latency_ms_mean_ci95_high": 14.17632542628076, "mean_inter_token_latency_ms_mean_ci95_low": 13.62872464668695, "mean_inter_token_latency_ms_min": 6.768708611325098, "mean_inter_token_latency_ms_p25": 12.848672956822377, "mean_inter_token_latency_ms_p50": 13.811244600730246, "mean_inter_token_latency_ms_p75": 15.050516136802901, "mean_inter_token_latency_ms_p90": 16.16779028255447, "mean_inter_token_latency_ms_p95": 16.664362313830495, "mean_inter_token_latency_ms_p99": 17.52092425711453, "mean_inter_token_latency_ms_samples": 160, "mean_inter_token_latency_ms_stddev": 1.7670058285717265, "mean_time_per_output_token_max_ms": 19.734181379417663, "mean_time_per_output_token_mean_ci95_high_ms": 14.17632542628076, "mean_time_per_output_token_mean_ci95_low_ms": 13.62872464668695, "mean_time_per_output_token_mean_ms": 13.902525036483855, "mean_time_per_output_token_min_ms": 6.768708611325098, "mean_time_per_output_token_p25_ms": 12.848672956822377, "mean_time_per_output_token_p50_ms": 13.811244600730246, "mean_time_per_output_token_p75_ms": 15.050516136802901, "mean_time_per_output_token_p90_ms": 16.16779028255447, "mean_time_per_output_token_p95_ms": 16.664362313830495, "mean_time_per_output_token_p99_ms": 17.52092425711453, "mean_time_per_output_token_samples": 160, "mean_time_per_output_token_stddev_ms": 1.7670058285717265, "output_token_sources": { "usage": 160 }, "output_tokens": 46084, "prefix_cache_hit_avg": 0.0011643533907411787, "prompt_token_samples": 160, "prompt_tokens": 1236764, "prompt_tokens_max": 7798.0, "prompt_tokens_mean": 7729.775, "prompt_tokens_mean_ci95_high": 7735.29436497065, "prompt_tokens_mean_ci95_low": 7724.255635029349, "prompt_tokens_min": 7675.0, "prompt_tokens_p25": 7702.0, "prompt_tokens_p50": 7725.0, "prompt_tokens_p75": 7759.0, "prompt_tokens_p90": 7782.0, "prompt_tokens_p95": 7790.0, "prompt_tokens_p99": 7797.0, "prompt_tokens_samples": 160, "prompt_tokens_stddev": 35.61992764286613, "reasoning_chunks": 37091, "request_canary_passed": 0, "request_canary_samples": 0, "throughput_performance_eligible": true, "throughput_tok_s": 148.25192956929055, "time_to_first_output_max_ms": 4419.473447022028, "time_to_first_output_mean_ci95_high_ms": 3793.5575957831993, "time_to_first_output_mean_ci95_low_ms": 3715.0195007168854, "time_to_first_output_mean_ms": 3754.2885482500424, "time_to_first_output_min_ms": 2904.618405038491, "time_to_first_output_p25_ms": 3597.319870023057, "time_to_first_output_p50_ms": 3796.5796140488237, "time_to_first_output_p75_ms": 3934.753163950518, "time_to_first_output_p90_ms": 4048.1514970306307, "time_to_first_output_p95_ms": 4125.9034250397235, "time_to_first_output_p99_ms": 4380.243528052233, "time_to_first_output_samples": 160, "time_to_first_output_stddev_ms": 253.42781989835288, "tpot_max_ms": 19.734181379417663, "tpot_mean_ci95_high_ms": 14.17632542628076, "tpot_mean_ci95_low_ms": 13.62872464668695, "tpot_mean_ms": 13.902525036483855, "tpot_min_ms": 6.768708611325098, "tpot_p25_ms": 12.848672956822377, "tpot_p50_ms": 13.811244600730246, "tpot_p75_ms": 15.050516136802901, "tpot_p90_ms": 16.16779028255447, "tpot_p95_ms": 16.664362313830495, "tpot_p99_ms": 17.52092425711453, "tpot_samples": 160, "tpot_stddev_ms": 1.7670058285717265, "ttft_max_ms": 12630.32150908839, "ttft_mean_ci95_high_ms": 7178.077595444443, "ttft_mean_ci95_low_ms": 6787.289537759392, "ttft_mean_ms": 6982.683566601918, "ttft_min_ms": 4426.362961996347, "ttft_p25_ms": 6029.178343014792, "ttft_p50_ms": 6873.991836910136, "ttft_p75_ms": 7756.285822018981, "ttft_p90_ms": 8588.229013956152, "ttft_p95_ms": 9070.46537694987, "ttft_p99_ms": 10911.082480102777, "ttft_samples": 160, "ttft_stddev_ms": 1261.0003517122723 }, "actual_start_unix_ns": 1792648469681220631, "actual_end_unix_ns": 1792648780530465553 }, { "name": "flare-instrumented-c1-prime", "arm": "flare-instrumented", "concurrency": 1, "repetition": "prime", "attempted": 52, "completed": 52, "failed": 0, "wall_ms": 124892.98763603438, "metrics": { "completion_tokens_max": 629.0, "completion_tokens_mean": 278.59615384615387, "completion_tokens_mean_ci95_high": 305.0849212048448, "completion_tokens_mean_ci95_low": 252.10738648746292, "completion_tokens_min": 130.0, "completion_tokens_p25": 203.0, "completion_tokens_p50": 272.0, "completion_tokens_p75": 338.0, "completion_tokens_p90": 379.0, "completion_tokens_p95": 477.0, "completion_tokens_p99": 629.0, "completion_tokens_samples": 52, "completion_tokens_stddev": 97.4557234036705, "content_chunks": 2833, "content_chunks_s": null, "controlled_output_exact_adherence": 0, "controlled_output_exact_adherence_rate": null, "controlled_output_samples": 0, "controlled_output_unobservable": 0, "decode_tok_s_max": 199.06904227677785, "decode_tok_s_mean": 188.57447152978654, "decode_tok_s_mean_ci95_high": 190.12982734170916, "decode_tok_s_mean_ci95_low": 187.0191157178639, "decode_tok_s_min": 175.83663286022372, "decode_tok_s_p25": 185.37093195790885, "decode_tok_s_p50": 188.34289427857126, "decode_tok_s_p75": 191.92688510528822, "decode_tok_s_p90": 196.263535557842, "decode_tok_s_p95": 197.97698417346083, "decode_tok_s_p99": 199.06904227677785, "decode_tok_s_samples": 52, "decode_tok_s_stddev": 5.722362379059164, "e2e_max_ms": 4177.928200922906, "e2e_mean_ci95_high_ms": 2537.5491996505166, "e2e_mean_ci95_low_ms": 2263.0867236899344, "e2e_mean_ms": 2400.3179616702255, "e2e_min_ms": 1584.592030965723, "e2e_p25_ms": 2010.7692730380222, "e2e_p50_ms": 2337.6500299200416, "e2e_p75_ms": 2664.008849998936, "e2e_p90_ms": 2970.648267073557, "e2e_p95_ms": 3373.3882879605517, "e2e_p99_ms": 4177.928200922906, "e2e_samples": 52, "e2e_stddev_ms": 504.89210727891873, "effective_prefill_tok_s_max": 8663.578206080092, "effective_prefill_tok_s_mean": 8264.409908607488, "effective_prefill_tok_s_mean_ci95_high": 8327.69880332272, "effective_prefill_tok_s_mean_ci95_low": 8201.121013892254, "effective_prefill_tok_s_min": 7979.735010000045, "effective_prefill_tok_s_p25": 8021.514025554984, "effective_prefill_tok_s_p50": 8342.713403556952, "effective_prefill_tok_s_p75": 8428.26259440696, "effective_prefill_tok_s_p90": 8573.074297419276, "effective_prefill_tok_s_p95": 8617.886746400332, "effective_prefill_tok_s_p99": 8663.578206080092, "effective_prefill_tok_s_samples": 52, "effective_prefill_tok_s_stddev": 232.84832149307803, "generation_max_ms": 3263.8986069941893, "generation_mean_ci95_high_ms": 1605.3838406562077, "generation_mean_ci95_low_ms": 1332.658743698244, "generation_mean_ms": 1469.0212921772259, "generation_min_ms": 679.1601929580793, "generation_p25_ms": 1089.6061420207843, "generation_p50_ms": 1419.9973249342293, "generation_p75_ms": 1766.297618043609, "generation_p90_ms": 2008.9914090931416, "generation_p95_ms": 2425.3104309318587, "generation_p99_ms": 3263.8986069941893, "generation_samples": 52, "generation_stddev_ms": 501.6960822386867, "mean_inter_token_latency_ms_max": 5.687097072627188, "mean_inter_token_latency_ms_mean": 5.307783998669215, "mean_inter_token_latency_ms_mean_ci95_high": 5.352012501144406, "mean_inter_token_latency_ms_mean_ci95_low": 5.263555496194023, "mean_inter_token_latency_ms_min": 5.023382785001994, "mean_inter_token_latency_ms_p25": 5.204359610293676, "mean_inter_token_latency_ms_p50": 5.306614519695178, "mean_inter_token_latency_ms_p75": 5.379723082802704, "mean_inter_token_latency_ms_p90": 5.553732113797439, "mean_inter_token_latency_ms_p95": 5.637078033601802, "mean_inter_token_latency_ms_p99": 5.687097072627188, "mean_inter_token_latency_ms_samples": 52, "mean_inter_token_latency_ms_stddev": 0.16272258521560096, "mean_time_per_output_token_max_ms": 5.687097072627188, "mean_time_per_output_token_mean_ci95_high_ms": 5.352012501144406, "mean_time_per_output_token_mean_ci95_low_ms": 5.263555496194023, "mean_time_per_output_token_mean_ms": 5.307783998669215, "mean_time_per_output_token_min_ms": 5.023382785001994, "mean_time_per_output_token_p25_ms": 5.204359610293676, "mean_time_per_output_token_p50_ms": 5.306614519695178, "mean_time_per_output_token_p75_ms": 5.379723082802704, "mean_time_per_output_token_p90_ms": 5.553732113797439, "mean_time_per_output_token_p95_ms": 5.637078033601802, "mean_time_per_output_token_p99_ms": 5.687097072627188, "mean_time_per_output_token_samples": 52, "mean_time_per_output_token_stddev_ms": 0.16272258521560096, "output_token_sources": { "usage": 52 }, "output_tokens": 14487, "prefix_cache_hit_avg": 0.0011702978394623165, "prompt_token_samples": 52, "prompt_tokens": 399899, "prompt_tokens_max": 7708.0, "prompt_tokens_mean": 7690.365384615385, "prompt_tokens_mean_ci95_high": 7693.3501411136585, "prompt_tokens_mean_ci95_low": 7687.380628117111, "prompt_tokens_min": 7675.0, "prompt_tokens_p25": 7680.0, "prompt_tokens_p50": 7688.0, "prompt_tokens_p75": 7701.0, "prompt_tokens_p90": 7705.0, "prompt_tokens_p95": 7707.0, "prompt_tokens_p99": 7708.0, "prompt_tokens_samples": 52, "prompt_tokens_stddev": 10.981318978876677, "reasoning_chunks": 11542, "request_canary_passed": 0, "request_canary_samples": 0, "throughput_performance_eligible": true, "throughput_tok_s": 115.99530345305136, "time_to_first_output_max_ms": 965.9468629397452, "time_to_first_output_mean_ci95_high_ms": 938.740989105713, "time_to_first_output_mean_ci95_low_ms": 923.852349880286, "time_to_first_output_mean_ms": 931.2966694929995, "time_to_first_output_min_ms": 886.1234720097855, "time_to_first_output_p25_ms": 908.1299679819494, "time_to_first_output_p50_ms": 919.3097380921245, "time_to_first_output_p75_ms": 958.8334020227194, "time_to_first_output_p90_ms": 963.1539880065247, "time_to_first_output_p95_ms": 964.7389389574528, "time_to_first_output_p99_ms": 965.9468629397452, "time_to_first_output_samples": 52, "time_to_first_output_stddev_ms": 27.388649055694483, "tpot_max_ms": 5.687097072627188, "tpot_mean_ci95_high_ms": 5.352012501144406, "tpot_mean_ci95_low_ms": 5.263555496194023, "tpot_mean_ms": 5.307783998669215, "tpot_min_ms": 5.023382785001994, "tpot_p25_ms": 5.204359610293676, "tpot_p50_ms": 5.306614519695178, "tpot_p75_ms": 5.379723082802704, "tpot_p90_ms": 5.553732113797439, "tpot_p95_ms": 5.637078033601802, "tpot_p99_ms": 5.687097072627188, "tpot_samples": 52, "tpot_stddev_ms": 0.16272258521560096, "ttft_max_ms": 3929.495814954862, "ttft_mean_ci95_high_ms": 2263.7084551994917, "ttft_mean_ci95_low_ms": 1988.9749930206813, "ttft_mean_ms": 2126.3417241100865, "ttft_min_ms": 1315.4565829318017, "ttft_p25_ms": 1722.8170779999346, "ttft_p50_ms": 2062.785184942186, "ttft_p75_ms": 2389.2068520653993, "ttft_p90_ms": 2707.1162490174174, "ttft_p95_ms": 3018.404060974717, "ttft_p99_ms": 3929.495814954862, "ttft_samples": 52, "ttft_stddev_ms": 505.3906045772693 }, "actual_start_unix_ns": 1792649210470175352, "actual_end_unix_ns": 1792649335363165496 }, { "name": "flare-instrumented-c1-r1", "arm": "flare-instrumented", "concurrency": 1, "repetition": "r1", "attempted": 52, "completed": 52, "failed": 0, "wall_ms": 125008.95848998334, "metrics": { "completion_tokens_max": 528.0, "completion_tokens_mean": 273.63461538461536, "completion_tokens_mean_ci95_high": 297.7858083110585, "completion_tokens_mean_ci95_low": 249.48342245817227, "completion_tokens_min": 130.0, "completion_tokens_p25": 195.0, "completion_tokens_p50": 265.0, "completion_tokens_p75": 336.0, "completion_tokens_p90": 379.0, "completion_tokens_p95": 425.0, "completion_tokens_p99": 528.0, "completion_tokens_samples": 52, "completion_tokens_stddev": 88.85547393868752, "content_chunks": 2836, "content_chunks_s": null, "controlled_output_exact_adherence": 0, "controlled_output_exact_adherence_rate": null, "controlled_output_samples": 0, "controlled_output_unobservable": 0, "decode_tok_s_max": 198.2982348374093, "decode_tok_s_mean": 186.16601422168222, "decode_tok_s_mean_ci95_high": 187.74659524681599, "decode_tok_s_mean_ci95_low": 184.58543319654845, "decode_tok_s_min": 174.3802676674369, "decode_tok_s_p25": 182.7413994976534, "decode_tok_s_p50": 185.28105371377796, "decode_tok_s_p75": 189.07208525233753, "decode_tok_s_p90": 194.62096103473306, "decode_tok_s_p95": 197.43933136429698, "decode_tok_s_p99": 198.2982348374093, "decode_tok_s_samples": 52, "decode_tok_s_stddev": 5.81516931749512, "e2e_max_ms": 3719.012539018877, "e2e_mean_ci95_high_ms": 2529.2003478689835, "e2e_mean_ci95_low_ms": 2276.5446895647797, "e2e_mean_ms": 2402.8725187168816, "e2e_min_ms": 1634.3378799501806, "e2e_p25_ms": 2004.8764690291137, "e2e_p50_ms": 2345.615235972218, "e2e_p75_ms": 2746.305259061046, "e2e_p90_ms": 2946.351006976329, "e2e_p95_ms": 3169.015981024131, "e2e_p99_ms": 3719.012539018877, "e2e_samples": 52, "e2e_stddev_ms": 464.7770056387333, "effective_prefill_tok_s_max": 8359.712820485478, "effective_prefill_tok_s_mean": 8169.045640210578, "effective_prefill_tok_s_mean_ci95_high": 8206.525996486367, "effective_prefill_tok_s_mean_ci95_low": 8131.5652839347895, "effective_prefill_tok_s_min": 7965.926816428873, "effective_prefill_tok_s_p25": 8020.962871629952, "effective_prefill_tok_s_p50": 8259.682471999888, "effective_prefill_tok_s_p75": 8293.985774091525, "effective_prefill_tok_s_p90": 8303.588282721197, "effective_prefill_tok_s_p95": 8313.066929708832, "effective_prefill_tok_s_p99": 8359.712820485478, "effective_prefill_tok_s_samples": 52, "effective_prefill_tok_s_stddev": 137.89525140307822, "generation_max_ms": 2792.7816049195826, "generation_mean_ci95_high_ms": 1587.4267744409246, "generation_mean_ci95_low_ms": 1334.9423791471543, "generation_mean_ms": 1461.1845767940395, "generation_min_ms": 681.3535450492054, "generation_p25_ms": 1079.3727779528126, "generation_p50_ms": 1392.45021995157, "generation_p75_ms": 1817.8596650250256, "generation_p90_ms": 2016.8042939621955, "generation_p95_ms": 2250.9222420630977, "generation_p99_ms": 2792.7816049195826, "generation_samples": 52, "generation_stddev_ms": 464.46195586031035, "mean_inter_token_latency_ms_max": 5.734593789631716, "mean_inter_token_latency_ms_mean": 5.3766682628269855, "mean_inter_token_latency_ms_mean_ci95_high": 5.4221135603989286, "mean_inter_token_latency_ms_mean_ci95_low": 5.3312229652550425, "mean_inter_token_latency_ms_min": 5.042909236282059, "mean_inter_token_latency_ms_p25": 5.288859729952147, "mean_inter_token_latency_ms_p50": 5.373642613303472, "mean_inter_token_latency_ms_p75": 5.459667157204905, "mean_inter_token_latency_ms_p90": 5.589499257924892, "mean_inter_token_latency_ms_p95": 5.646960123274953, "mean_inter_token_latency_ms_p99": 5.734593789631716, "mean_inter_token_latency_ms_samples": 52, "mean_inter_token_latency_ms_stddev": 0.16719933737179657, "mean_time_per_output_token_max_ms": 5.734593789631716, "mean_time_per_output_token_mean_ci95_high_ms": 5.4221135603989286, "mean_time_per_output_token_mean_ci95_low_ms": 5.3312229652550425, "mean_time_per_output_token_mean_ms": 5.3766682628269855, "mean_time_per_output_token_min_ms": 5.042909236282059, "mean_time_per_output_token_p25_ms": 5.288859729952147, "mean_time_per_output_token_p50_ms": 5.373642613303472, "mean_time_per_output_token_p75_ms": 5.459667157204905, "mean_time_per_output_token_p90_ms": 5.589499257924892, "mean_time_per_output_token_p95_ms": 5.646960123274953, "mean_time_per_output_token_p99_ms": 5.734593789631716, "mean_time_per_output_token_samples": 52, "mean_time_per_output_token_stddev_ms": 0.16719933737179657, "output_token_sources": { "usage": 52 }, "output_tokens": 14229, "prefix_cache_hit_avg": 0.0011702978394623165, "prompt_token_samples": 52, "prompt_tokens": 399899, "prompt_tokens_max": 7708.0, "prompt_tokens_mean": 7690.365384615385, "prompt_tokens_mean_ci95_high": 7693.3501411136585, "prompt_tokens_mean_ci95_low": 7687.380628117111, "prompt_tokens_min": 7675.0, "prompt_tokens_p25": 7680.0, "prompt_tokens_p50": 7688.0, "prompt_tokens_p75": 7701.0, "prompt_tokens_p90": 7705.0, "prompt_tokens_p95": 7707.0, "prompt_tokens_p99": 7708.0, "prompt_tokens_samples": 52, "prompt_tokens_stddev": 10.981318978876677, "reasoning_chunks": 11285, "request_canary_passed": 0, "request_canary_samples": 0, "throughput_performance_eligible": true, "throughput_tok_s": 113.82384248197808, "time_to_first_output_max_ms": 967.2446380136535, "time_to_first_output_mean_ci95_high_ms": 946.365564206585, "time_to_first_output_mean_ci95_low_ms": 937.0103196390991, "time_to_first_output_mean_ms": 941.687941922842, "time_to_first_output_min_ms": 918.0937389610335, "time_to_first_output_p25_ms": 926.7232760321349, "time_to_first_output_p50_ms": 929.5467130141333, "time_to_first_output_p75_ms": 960.1818720111623, "time_to_first_output_p90_ms": 962.788843899034, "time_to_first_output_p95_ms": 965.1440000161529, "time_to_first_output_p99_ms": 967.2446380136535, "time_to_first_output_samples": 52, "time_to_first_output_stddev_ms": 17.209598970702228, "tpot_max_ms": 5.734593789631716, "tpot_mean_ci95_high_ms": 5.4221135603989286, "tpot_mean_ci95_low_ms": 5.3312229652550425, "tpot_mean_ms": 5.3766682628269855, "tpot_min_ms": 5.042909236282059, "tpot_p25_ms": 5.288859729952147, "tpot_p50_ms": 5.373642613303472, "tpot_p75_ms": 5.459667157204905, "tpot_p90_ms": 5.589499257924892, "tpot_p95_ms": 5.646960123274953, "tpot_p99_ms": 5.734593789631716, "tpot_samples": 52, "tpot_stddev_ms": 0.16719933737179657, "ttft_max_ms": 3374.9553579837084, "ttft_mean_ci95_high_ms": 2247.974913801071, "ttft_mean_ci95_low_ms": 1995.585144308463, "ttft_mean_ms": 2121.780029054767, "ttft_min_ms": 1321.7975939624012, "ttft_p25_ms": 1727.1484499797225, "ttft_p50_ms": 2019.771832972765, "ttft_p75_ms": 2476.637826068327, "ttft_p90_ms": 2634.7313419682905, "ttft_p95_ms": 2882.733780075796, "ttft_p99_ms": 3374.9553579837084, "ttft_samples": 52, "ttft_stddev_ms": 464.28788536129406 }, "actual_start_unix_ns": 1792649335883111185, "actual_end_unix_ns": 1792649460892071949 }, { "name": "flare-instrumented-c1-r2", "arm": "flare-instrumented", "concurrency": 1, "repetition": "r2", "attempted": 52, "completed": 52, "failed": 0, "wall_ms": 124169.57050701603, "metrics": { "completion_tokens_max": 505.0, "completion_tokens_mean": 270.53846153846155, "completion_tokens_mean_ci95_high": 293.76746915298895, "completion_tokens_mean_ci95_low": 247.30945392393414, "completion_tokens_min": 130.0, "completion_tokens_p25": 210.0, "completion_tokens_p50": 265.0, "completion_tokens_p75": 327.0, "completion_tokens_p90": 379.0, "completion_tokens_p95": 402.0, "completion_tokens_p99": 505.0, "completion_tokens_samples": 52, "completion_tokens_stddev": 85.46263064522654, "content_chunks": 2854, "content_chunks_s": null, "controlled_output_exact_adherence": 0, "controlled_output_exact_adherence_rate": null, "controlled_output_samples": 0, "controlled_output_unobservable": 0, "decode_tok_s_max": 198.0386229095764, "decode_tok_s_mean": 185.89944785907588, "decode_tok_s_mean_ci95_high": 187.4148369623858, "decode_tok_s_mean_ci95_low": 184.38405875576595, "decode_tok_s_min": 175.25520725380292, "decode_tok_s_p25": 181.88913716632024, "decode_tok_s_p50": 185.64956033358416, "decode_tok_s_p75": 189.15961940013003, "decode_tok_s_p90": 193.17141259908627, "decode_tok_s_p95": 195.82991638356287, "decode_tok_s_p99": 198.0386229095764, "decode_tok_s_samples": 52, "decode_tok_s_stddev": 5.5753195043503965, "e2e_max_ms": 3613.9932069927454, "e2e_mean_ci95_high_ms": 2508.6707258294928, "e2e_mean_ci95_low_ms": 2264.465266976339, "e2e_mean_ms": 2386.567996402916, "e2e_min_ms": 1628.1259040115401, "e2e_p25_ms": 2061.053253011778, "e2e_p50_ms": 2340.7689130399376, "e2e_p75_ms": 2735.7375220162794, "e2e_p90_ms": 2967.421191977337, "e2e_p95_ms": 3133.365849032998, "e2e_p99_ms": 3613.9932069927454, "e2e_samples": 52, "e2e_stddev_ms": 449.2322977771723, "effective_prefill_tok_s_max": 8403.827724227123, "effective_prefill_tok_s_mean": 8191.167574074469, "effective_prefill_tok_s_mean_ci95_high": 8229.530493833743, "effective_prefill_tok_s_mean_ci95_low": 8152.804654315194, "effective_prefill_tok_s_min": 7996.141273825119, "effective_prefill_tok_s_p25": 8039.569925840841, "effective_prefill_tok_s_p50": 8291.043094092745, "effective_prefill_tok_s_p75": 8319.493903482353, "effective_prefill_tok_s_p90": 8330.842012724217, "effective_prefill_tok_s_p95": 8332.873479949882, "effective_prefill_tok_s_p99": 8403.827724227123, "effective_prefill_tok_s_samples": 52, "effective_prefill_tok_s_stddev": 141.1423206822198, "generation_max_ms": 2689.832004951313, "generation_mean_ci95_high_ms": 1569.4310453335688, "generation_mean_ci95_low_ms": 1325.392588235115, "generation_mean_ms": 1447.411816784342, "generation_min_ms": 679.5629339758307, "generation_p25_ms": 1102.1079430356622, "generation_p50_ms": 1390.1689850026742, "generation_p75_ms": 1785.899844020605, "generation_p90_ms": 2008.4838969632983, "generation_p95_ms": 2171.6843590838835, "generation_p99_ms": 2689.832004951313, "generation_samples": 52, "generation_stddev_ms": 448.9250868640784, "mean_inter_token_latency_ms_max": 5.705964551180551, "mean_inter_token_latency_ms_mean": 5.383983426272962, "mean_inter_token_latency_ms_mean_ci95_high": 5.4277316839705545, "mean_inter_token_latency_ms_mean_ci95_low": 5.340235168575369, "mean_inter_token_latency_ms_min": 5.0495200648642955, "mean_inter_token_latency_ms_p25": 5.285027452030933, "mean_inter_token_latency_ms_p50": 5.370277799367108, "mean_inter_token_latency_ms_p75": 5.49238732473187, "mean_inter_token_latency_ms_p90": 5.608862371912274, "mean_inter_token_latency_ms_p95": 5.645236181554391, "mean_inter_token_latency_ms_p99": 5.705964551180551, "mean_inter_token_latency_ms_samples": 52, "mean_inter_token_latency_ms_stddev": 0.16095570034783957, "mean_time_per_output_token_max_ms": 5.705964551180551, "mean_time_per_output_token_mean_ci95_high_ms": 5.4277316839705545, "mean_time_per_output_token_mean_ci95_low_ms": 5.340235168575369, "mean_time_per_output_token_mean_ms": 5.383983426272962, "mean_time_per_output_token_min_ms": 5.0495200648642955, "mean_time_per_output_token_p25_ms": 5.285027452030933, "mean_time_per_output_token_p50_ms": 5.370277799367108, "mean_time_per_output_token_p75_ms": 5.49238732473187, "mean_time_per_output_token_p90_ms": 5.608862371912274, "mean_time_per_output_token_p95_ms": 5.645236181554391, "mean_time_per_output_token_p99_ms": 5.705964551180551, "mean_time_per_output_token_samples": 52, "mean_time_per_output_token_stddev_ms": 0.16095570034783957, "output_token_sources": { "usage": 52 }, "output_tokens": 14068, "prefix_cache_hit_avg": 0.0011702978394623165, "prompt_token_samples": 52, "prompt_tokens": 399899, "prompt_tokens_max": 7708.0, "prompt_tokens_mean": 7690.365384615385, "prompt_tokens_mean_ci95_high": 7693.3501411136585, "prompt_tokens_mean_ci95_low": 7687.380628117111, "prompt_tokens_min": 7675.0, "prompt_tokens_p25": 7680.0, "prompt_tokens_p50": 7688.0, "prompt_tokens_p75": 7701.0, "prompt_tokens_p90": 7705.0, "prompt_tokens_p95": 7707.0, "prompt_tokens_p99": 7708.0, "prompt_tokens_samples": 52, "prompt_tokens_stddev": 10.981318978876677, "reasoning_chunks": 11105, "request_canary_passed": 0, "request_canary_samples": 0, "throughput_performance_eligible": true, "throughput_tok_s": 113.29667923112537, "time_to_first_output_max_ms": 963.7148389592767, "time_to_first_output_mean_ci95_high_ms": 943.910577492346, "time_to_first_output_mean_ci95_low_ms": 934.4017817448021, "time_to_first_output_mean_ms": 939.156179618574, "time_to_first_output_min_ms": 913.2743140216917, "time_to_first_output_p25_ms": 923.2727559283376, "time_to_first_output_p50_ms": 926.5356649411842, "time_to_first_output_p75_ms": 957.2668579639867, "time_to_first_output_p90_ms": 961.2318660365418, "time_to_first_output_p95_ms": 962.6687540439889, "time_to_first_output_p99_ms": 963.7148389592767, "time_to_first_output_samples": 52, "time_to_first_output_stddev_ms": 17.492066650858938, "tpot_max_ms": 5.705964551180551, "tpot_mean_ci95_high_ms": 5.4277316839705545, "tpot_mean_ci95_low_ms": 5.340235168575369, "tpot_mean_ms": 5.383983426272962, "tpot_min_ms": 5.0495200648642955, "tpot_p25_ms": 5.285027452030933, "tpot_p50_ms": 5.370277799367108, "tpot_p75_ms": 5.49238732473187, "tpot_p90_ms": 5.608862371912274, "tpot_p95_ms": 5.645236181554391, "tpot_p99_ms": 5.705964551180551, "tpot_samples": 52, "tpot_stddev_ms": 0.16095570034783957, "ttft_max_ms": 3363.6926850304008, "ttft_mean_ci95_high_ms": 2225.589192572808, "ttft_mean_ci95_low_ms": 1983.9581546462082, "ttft_mean_ms": 2104.773673609508, "ttft_min_ms": 1317.3748960252851, "ttft_p25_ms": 1746.4797560824081, "ttft_p50_ms": 2011.7093609878793, "ttft_p75_ms": 2420.000536018051, "ttft_p90_ms": 2704.089603968896, "ttft_p95_ms": 2816.5018169675022, "ttft_p99_ms": 3363.6926850304008, "ttft_samples": 52, "ttft_stddev_ms": 444.4964780550708 }, "actual_start_unix_ns": 1792649461283237876, "actual_end_unix_ns": 1792649585452811314 }, { "name": "flare-instrumented-c1-r3", "arm": "flare-instrumented", "concurrency": 1, "repetition": "r3", "attempted": 52, "completed": 52, "failed": 0, "wall_ms": 126810.8699589502, "metrics": { "completion_tokens_max": 509.0, "completion_tokens_mean": 280.4230769230769, "completion_tokens_mean_ci95_high": 305.17055756308577, "completion_tokens_mean_ci95_low": 255.67559628306805, "completion_tokens_min": 130.0, "completion_tokens_p25": 210.0, "completion_tokens_p50": 275.0, "completion_tokens_p75": 339.0, "completion_tokens_p90": 386.0, "completion_tokens_p95": 453.0, "completion_tokens_p99": 509.0, "completion_tokens_samples": 52, "completion_tokens_stddev": 91.04929631235136, "content_chunks": 2858, "content_chunks_s": null, "controlled_output_exact_adherence": 0, "controlled_output_exact_adherence_rate": null, "controlled_output_samples": 0, "controlled_output_unobservable": 0, "decode_tok_s_max": 197.9559910335571, "decode_tok_s_mean": 186.39077793157273, "decode_tok_s_mean_ci95_high": 187.95237760121822, "decode_tok_s_mean_ci95_low": 184.82917826192724, "decode_tok_s_min": 175.10690576673997, "decode_tok_s_p25": 183.09481835814094, "decode_tok_s_p50": 186.5845954060821, "decode_tok_s_p75": 189.17022801526974, "decode_tok_s_p90": 194.59893750902262, "decode_tok_s_p95": 197.03059616906614, "decode_tok_s_p99": 197.9559910335571, "decode_tok_s_samples": 52, "decode_tok_s_stddev": 5.745334368014709, "e2e_max_ms": 3667.6837200066075, "e2e_mean_ci95_high_ms": 2567.9708928078026, "e2e_mean_ci95_low_ms": 2307.1223825464767, "e2e_mean_ms": 2437.5466376771396, "e2e_min_ms": 1631.550679099746, "e2e_p25_ms": 2078.955200035125, "e2e_p50_ms": 2391.8770679738373, "e2e_p75_ms": 2757.534471922554, "e2e_p90_ms": 2995.38640701212, "e2e_p95_ms": 3282.7746969414875, "e2e_p99_ms": 3667.6837200066075, "e2e_samples": 52, "e2e_stddev_ms": 479.84830554877874, "effective_prefill_tok_s_max": 8393.817184232748, "effective_prefill_tok_s_mean": 8177.198576185015, "effective_prefill_tok_s_mean_ci95_high": 8217.858633418093, "effective_prefill_tok_s_mean_ci95_low": 8136.538518951936, "effective_prefill_tok_s_min": 7947.418845808043, "effective_prefill_tok_s_p25": 8031.826782608199, "effective_prefill_tok_s_p50": 8277.329905283017, "effective_prefill_tok_s_p75": 8311.456717974634, "effective_prefill_tok_s_p90": 8326.43488611325, "effective_prefill_tok_s_p95": 8341.186466174517, "effective_prefill_tok_s_p99": 8393.817184232748, "effective_prefill_tok_s_samples": 52, "effective_prefill_tok_s_stddev": 149.59379716037387, "generation_max_ms": 2746.5917349327356, "generation_mean_ci95_high_ms": 1627.0129099451733, "generation_mean_ci95_low_ms": 1366.4861596637897, "generation_mean_ms": 1496.7495348044815, "generation_min_ms": 680.2740480052307, "generation_p25_ms": 1153.5810030763969, "generation_p50_ms": 1435.1206630235538, "generation_p75_ms": 1814.4228999735788, "generation_p90_ms": 2037.2928510187194, "generation_p95_ms": 2331.446134019643, "generation_p99_ms": 2746.5917349327356, "generation_samples": 52, "generation_stddev_ms": 479.25640651506683, "mean_inter_token_latency_ms_max": 5.710797044932658, "mean_inter_token_latency_ms_mean": 5.37008041121803, "mean_inter_token_latency_ms_mean_ci95_high": 5.41514032741241, "mean_inter_token_latency_ms_mean_ci95_low": 5.32502049502365, "mean_inter_token_latency_ms_min": 5.051627863237956, "mean_inter_token_latency_ms_p25": 5.273442232598688, "mean_inter_token_latency_ms_p50": 5.355573533355103, "mean_inter_token_latency_ms_p75": 5.454260579614137, "mean_inter_token_latency_ms_p90": 5.58697665753409, "mean_inter_token_latency_ms_p95": 5.682543199509382, "mean_inter_token_latency_ms_p99": 5.710797044932658, "mean_inter_token_latency_ms_samples": 52, "mean_inter_token_latency_ms_stddev": 0.16578146766015167, "mean_time_per_output_token_max_ms": 5.710797044932658, "mean_time_per_output_token_mean_ci95_high_ms": 5.41514032741241, "mean_time_per_output_token_mean_ci95_low_ms": 5.32502049502365, "mean_time_per_output_token_mean_ms": 5.37008041121803, "mean_time_per_output_token_min_ms": 5.051627863237956, "mean_time_per_output_token_p25_ms": 5.273442232598688, "mean_time_per_output_token_p50_ms": 5.355573533355103, "mean_time_per_output_token_p75_ms": 5.454260579614137, "mean_time_per_output_token_p90_ms": 5.58697665753409, "mean_time_per_output_token_p95_ms": 5.682543199509382, "mean_time_per_output_token_p99_ms": 5.710797044932658, "mean_time_per_output_token_samples": 52, "mean_time_per_output_token_stddev_ms": 0.16578146766015167, "output_token_sources": { "usage": 52 }, "output_tokens": 14582, "prefix_cache_hit_avg": 0.0011702978394623165, "prompt_token_samples": 52, "prompt_tokens": 399899, "prompt_tokens_max": 7708.0, "prompt_tokens_mean": 7690.365384615385, "prompt_tokens_mean_ci95_high": 7693.3501411136585, "prompt_tokens_mean_ci95_low": 7687.380628117111, "prompt_tokens_min": 7675.0, "prompt_tokens_p25": 7680.0, "prompt_tokens_p50": 7688.0, "prompt_tokens_p75": 7701.0, "prompt_tokens_p90": 7705.0, "prompt_tokens_p95": 7707.0, "prompt_tokens_p99": 7708.0, "prompt_tokens_samples": 52, "prompt_tokens_stddev": 10.981318978876677, "reasoning_chunks": 11612, "request_canary_passed": 0, "request_canary_samples": 0, "throughput_performance_eligible": true, "throughput_tok_s": 114.99014244378516, "time_to_first_output_max_ms": 969.6229869732633, "time_to_first_output_mean_ci95_high_ms": 945.8370985346196, "time_to_first_output_mean_ci95_low_ms": 935.7571072106971, "time_to_first_output_mean_ms": 940.7971028726583, "time_to_first_output_min_ms": 914.3634929787368, "time_to_first_output_p25_ms": 924.4535949546844, "time_to_first_output_p50_ms": 927.6075719390064, "time_to_first_output_p75_ms": 958.0935559934005, "time_to_first_output_p90_ms": 964.9988430319354, "time_to_first_output_p95_ms": 967.736704973504, "time_to_first_output_p99_ms": 969.6229869732633, "time_to_first_output_samples": 52, "time_to_first_output_stddev_ms": 18.542819170732056, "tpot_max_ms": 5.710797044932658, "tpot_mean_ci95_high_ms": 5.41514032741241, "tpot_mean_ci95_low_ms": 5.32502049502365, "tpot_mean_ms": 5.37008041121803, "tpot_min_ms": 5.051627863237956, "tpot_p25_ms": 5.273442232598688, "tpot_p50_ms": 5.355573533355103, "tpot_p75_ms": 5.454260579614137, "tpot_p90_ms": 5.58697665753409, "tpot_p95_ms": 5.682543199509382, "tpot_p99_ms": 5.710797044932658, "tpot_samples": 52, "tpot_stddev_ms": 0.16578146766015167, "ttft_max_ms": 3364.548353943974, "ttft_mean_ci95_high_ms": 2283.6477029810585, "ttft_mean_ci95_low_ms": 2027.425594362624, "ttft_mean_ms": 2155.5366486718412, "ttft_min_ms": 1320.8591049769893, "ttft_p25_ms": 1828.137345961295, "ttft_p50_ms": 2115.0715600233525, "ttft_p75_ms": 2478.293061023578, "ttft_p90_ms": 2731.6495520062745, "ttft_p95_ms": 3056.870927917771, "ttft_p99_ms": 3364.548353943974, "ttft_samples": 52, "ttft_stddev_ms": 471.33772986289205 }, "actual_start_unix_ns": 1792649585872226145, "actual_end_unix_ns": 1792649712683098818 }, { "name": "flare-instrumented-c4-prime", "arm": "flare-instrumented", "concurrency": 4, "repetition": "prime", "attempted": 160, "completed": 160, "failed": 0, "wall_ms": 311172.6718719583, "metrics": { "completion_tokens_max": 1045.0, "completion_tokens_mean": 287.1125, "completion_tokens_mean_ci95_high": 303.06107686514963, "completion_tokens_mean_ci95_low": 271.1639231348504, "completion_tokens_min": 99.0, "completion_tokens_p25": 222.0, "completion_tokens_p50": 280.0, "completion_tokens_p75": 329.0, "completion_tokens_p90": 390.0, "completion_tokens_p95": 431.0, "completion_tokens_p99": 618.0, "completion_tokens_samples": 160, "completion_tokens_stddev": 102.92618026967513, "content_chunks": 8849, "content_chunks_s": null, "controlled_output_exact_adherence": 0, "controlled_output_exact_adherence_rate": null, "controlled_output_samples": 0, "controlled_output_unobservable": 0, "decode_tok_s_max": 146.65668289689614, "decode_tok_s_mean": 72.54423576455419, "decode_tok_s_mean_ci95_high": 74.32853993863202, "decode_tok_s_mean_ci95_low": 70.75993159047636, "decode_tok_s_min": 47.68503680822973, "decode_tok_s_p25": 66.63194421765562, "decode_tok_s_p50": 71.66128445832521, "decode_tok_s_p75": 75.51802662496789, "decode_tok_s_p90": 82.28932054638662, "decode_tok_s_p95": 88.52137731919717, "decode_tok_s_p99": 134.24125393838864, "decode_tok_s_samples": 160, "decode_tok_s_stddev": 11.515235160472466, "e2e_max_ms": 17649.57372704521, "e2e_mean_ci95_high_ms": 7985.454189898064, "e2e_mean_ci95_low_ms": 7513.235763608918, "e2e_mean_ms": 7749.344976753491, "e2e_min_ms": 4412.860458018258, "e2e_p25_ms": 6902.44943194557, "e2e_p50_ms": 7588.231888017617, "e2e_p75_ms": 8372.854460962117, "e2e_p90_ms": 9411.134581896476, "e2e_p95_ms": 9853.125813999213, "e2e_p99_ms": 12257.305486011319, "e2e_samples": 160, "e2e_stddev_ms": 1523.7610001775843, "effective_prefill_tok_s_max": 2609.9790463532086, "effective_prefill_tok_s_mean": 2079.7275299389617, "effective_prefill_tok_s_mean_ci95_high": 2102.541835831474, "effective_prefill_tok_s_mean_ci95_low": 2056.913224046449, "effective_prefill_tok_s_min": 1733.2407215874325, "effective_prefill_tok_s_p25": 1981.9286866140876, "effective_prefill_tok_s_p50": 2062.655862546485, "effective_prefill_tok_s_p75": 2174.414514831066, "effective_prefill_tok_s_p90": 2260.51082511699, "effective_prefill_tok_s_p95": 2334.5468614296487, "effective_prefill_tok_s_p99": 2505.9768338071, "effective_prefill_tok_s_samples": 160, "effective_prefill_tok_s_stddev": 147.23504052273455, "generation_max_ms": 13928.432433051057, "generation_mean_ci95_high_ms": 4241.21397801425, "generation_mean_ci95_low_ms": 3787.2860899551856, "generation_mean_ms": 4014.2500339847174, "generation_min_ms": 1398.0043879710138, "generation_p25_ms": 3101.3634420232847, "generation_p50_ms": 3956.9407110102475, "generation_p75_ms": 4675.855781999417, "generation_p90_ms": 5574.658486060798, "generation_p95_ms": 6078.491974039935, "generation_p99_ms": 8579.156560008414, "generation_samples": 160, "generation_stddev_ms": 1464.7408364659086, "mean_inter_token_latency_ms_max": 20.970939039464362, "mean_inter_token_latency_ms_mean": 14.067820653885798, "mean_inter_token_latency_ms_mean_ci95_high": 14.36470277577339, "mean_inter_token_latency_ms_mean_ci95_low": 13.770938531998206, "mean_inter_token_latency_ms_min": 6.8186459713058465, "mean_inter_token_latency_ms_p25": 13.221828881604246, "mean_inter_token_latency_ms_p50": 13.929049919564537, "mean_inter_token_latency_ms_p75": 15.007592864609492, "mean_inter_token_latency_ms_p90": 16.10175264209918, "mean_inter_token_latency_ms_p95": 17.55526033536482, "mean_inter_token_latency_ms_p99": 20.252751610994416, "mean_inter_token_latency_ms_samples": 160, "mean_inter_token_latency_ms_stddev": 1.9159667382622807, "mean_time_per_output_token_max_ms": 20.970939039464362, "mean_time_per_output_token_mean_ci95_high_ms": 14.36470277577339, "mean_time_per_output_token_mean_ci95_low_ms": 13.770938531998206, "mean_time_per_output_token_mean_ms": 14.067820653885798, "mean_time_per_output_token_min_ms": 6.8186459713058465, "mean_time_per_output_token_p25_ms": 13.221828881604246, "mean_time_per_output_token_p50_ms": 13.929049919564537, "mean_time_per_output_token_p75_ms": 15.007592864609492, "mean_time_per_output_token_p90_ms": 16.10175264209918, "mean_time_per_output_token_p95_ms": 17.55526033536482, "mean_time_per_output_token_p99_ms": 20.252751610994416, "mean_time_per_output_token_samples": 160, "mean_time_per_output_token_stddev_ms": 1.9159667382622807, "output_token_sources": { "usage": 160 }, "output_tokens": 45938, "prefix_cache_hit_avg": 0.0011643533907411787, "prompt_token_samples": 160, "prompt_tokens": 1236764, "prompt_tokens_max": 7798.0, "prompt_tokens_mean": 7729.775, "prompt_tokens_mean_ci95_high": 7735.29436497065, "prompt_tokens_mean_ci95_low": 7724.255635029349, "prompt_tokens_min": 7675.0, "prompt_tokens_p25": 7702.0, "prompt_tokens_p50": 7725.0, "prompt_tokens_p75": 7759.0, "prompt_tokens_p90": 7782.0, "prompt_tokens_p95": 7790.0, "prompt_tokens_p99": 7797.0, "prompt_tokens_samples": 160, "prompt_tokens_stddev": 35.61992764286613, "reasoning_chunks": 36756, "request_canary_passed": 0, "request_canary_samples": 0, "throughput_performance_eligible": true, "throughput_tok_s": 147.62864529087767, "time_to_first_output_max_ms": 4465.623212978244, "time_to_first_output_mean_ci95_high_ms": 3775.8093629416585, "time_to_first_output_mean_ci95_low_ms": 3694.380522595889, "time_to_first_output_mean_ms": 3735.0949427687738, "time_to_first_output_min_ms": 2987.7634500153363, "time_to_first_output_p25_ms": 3560.1693119388074, "time_to_first_output_p50_ms": 3744.6867120452225, "time_to_first_output_p75_ms": 3901.7562749795616, "time_to_first_output_p90_ms": 4051.0003719246015, "time_to_first_output_p95_ms": 4112.464047968388, "time_to_first_output_p99_ms": 4329.316287999973, "time_to_first_output_samples": 160, "time_to_first_output_stddev_ms": 262.75571706004666, "tpot_max_ms": 20.970939039464362, "tpot_mean_ci95_high_ms": 14.36470277577339, "tpot_mean_ci95_low_ms": 13.770938531998206, "tpot_mean_ms": 14.067820653885798, "tpot_min_ms": 6.8186459713058465, "tpot_p25_ms": 13.221828881604246, "tpot_p50_ms": 13.929049919564537, "tpot_p75_ms": 15.007592864609492, "tpot_p90_ms": 16.10175264209918, "tpot_p95_ms": 17.55526033536482, "tpot_p99_ms": 20.252751610994416, "tpot_samples": 160, "tpot_stddev_ms": 1.9159667382622807, "ttft_max_ms": 16839.03915202245, "ttft_mean_ci95_high_ms": 7235.035596185567, "ttft_mean_ci95_low_ms": 6779.153397749092, "ttft_mean_ms": 7007.094496967329, "ttft_min_ms": 4078.119897050783, "ttft_p25_ms": 6127.325617009774, "ttft_p50_ms": 6821.941460017115, "ttft_p75_ms": 7599.440390011296, "ttft_p90_ms": 8460.275069926865, "ttft_p95_ms": 9163.59857597854, "ttft_p99_ms": 11699.505915981717, "ttft_samples": 160, "ttft_stddev_ms": 1471.047032432766 }, "actual_start_unix_ns": 1792649713180300697, "actual_end_unix_ns": 1792650024352976199 }, { "name": "flare-instrumented-c4-r1", "arm": "flare-instrumented", "concurrency": 4, "repetition": "r1", "attempted": 160, "completed": 160, "failed": 0, "wall_ms": 308192.1920859022, "metrics": { "completion_tokens_max": 593.0, "completion_tokens_mean": 279.1, "completion_tokens_mean_ci95_high": 290.64494528615785, "completion_tokens_mean_ci95_low": 267.5550547138422, "completion_tokens_min": 129.0, "completion_tokens_p25": 224.0, "completion_tokens_p50": 267.0, "completion_tokens_p75": 316.0, "completion_tokens_p90": 362.0, "completion_tokens_p95": 390.0, "completion_tokens_p99": 476.0, "completion_tokens_samples": 160, "completion_tokens_stddev": 74.50678074751647, "content_chunks": 8864, "content_chunks_s": null, "controlled_output_exact_adherence": 0, "controlled_output_exact_adherence_rate": null, "controlled_output_samples": 0, "controlled_output_unobservable": 0, "decode_tok_s_max": 114.22396378224572, "decode_tok_s_mean": 70.47643539940618, "decode_tok_s_mean_ci95_high": 71.81478917428144, "decode_tok_s_mean_ci95_low": 69.13808162453093, "decode_tok_s_min": 50.28067156953741, "decode_tok_s_p25": 65.63878665652325, "decode_tok_s_p50": 70.00808179218119, "decode_tok_s_p75": 74.75773865208288, "decode_tok_s_p90": 77.82474775074878, "decode_tok_s_p95": 81.95117473363337, "decode_tok_s_p99": 110.96333393956336, "decode_tok_s_samples": 160, "decode_tok_s_stddev": 8.637237232020482, "e2e_max_ms": 12325.81490604207, "e2e_mean_ci95_high_ms": 7841.310889755892, "e2e_mean_ci95_low_ms": 7514.935949914456, "e2e_mean_ms": 7678.123419835174, "e2e_min_ms": 4879.950057948008, "e2e_p25_ms": 6962.566312984563, "e2e_p50_ms": 7568.677988951094, "e2e_p75_ms": 8251.413619960658, "e2e_p90_ms": 8833.302228013054, "e2e_p95_ms": 9608.672662987374, "e2e_p99_ms": 10903.771782992408, "e2e_samples": 160, "e2e_stddev_ms": 1053.151205203441, "effective_prefill_tok_s_max": 2632.090474934855, "effective_prefill_tok_s_mean": 2096.7713106312576, "effective_prefill_tok_s_mean_ci95_high": 2120.655829071758, "effective_prefill_tok_s_mean_ci95_low": 2072.886792190757, "effective_prefill_tok_s_min": 1650.8363562541856, "effective_prefill_tok_s_p25": 1995.3910619969586, "effective_prefill_tok_s_p50": 2083.3693765461753, "effective_prefill_tok_s_p75": 2162.71687704709, "effective_prefill_tok_s_p90": 2283.1146621276994, "effective_prefill_tok_s_p95": 2393.7136719703135, "effective_prefill_tok_s_p99": 2614.3784169912233, "effective_prefill_tok_s_samples": 160, "effective_prefill_tok_s_stddev": 154.14179405770338, "generation_max_ms": 9118.664406007156, "generation_mean_ci95_high_ms": 4135.781137526988, "generation_mean_ci95_low_ms": 3808.699156422904, "generation_mean_ms": 3972.2401469749457, "generation_min_ms": 1617.4760430585593, "generation_p25_ms": 3202.546023996547, "generation_p50_ms": 3846.039461088367, "generation_p75_ms": 4592.373894993216, "generation_p90_ms": 5253.75850405544, "generation_p95_ms": 5811.506358091719, "generation_p99_ms": 7006.438159034587, "generation_samples": 160, "generation_stddev_ms": 1055.4326958051638, "mean_inter_token_latency_ms_max": 19.88835806651896, "mean_inter_token_latency_ms_mean": 14.384330495844063, "mean_inter_token_latency_ms_mean_ci95_high": 14.64070026254027, "mean_inter_token_latency_ms_mean_ci95_low": 14.127960729147857, "mean_inter_token_latency_ms_min": 8.75473032879843, "mean_inter_token_latency_ms_p25": 13.346120103504683, "mean_inter_token_latency_ms_p50": 14.280297470072078, "mean_inter_token_latency_ms_p75": 15.221824288057784, "mean_inter_token_latency_ms_p90": 16.482556866093535, "mean_inter_token_latency_ms_p95": 17.235854207619763, "mean_inter_token_latency_ms_p99": 18.78322468800648, "mean_inter_token_latency_ms_samples": 160, "mean_inter_token_latency_ms_stddev": 1.6545150733999903, "mean_time_per_output_token_max_ms": 19.88835806651896, "mean_time_per_output_token_mean_ci95_high_ms": 14.64070026254027, "mean_time_per_output_token_mean_ci95_low_ms": 14.127960729147857, "mean_time_per_output_token_mean_ms": 14.384330495844063, "mean_time_per_output_token_min_ms": 8.75473032879843, "mean_time_per_output_token_p25_ms": 13.346120103504683, "mean_time_per_output_token_p50_ms": 14.280297470072078, "mean_time_per_output_token_p75_ms": 15.221824288057784, "mean_time_per_output_token_p90_ms": 16.482556866093535, "mean_time_per_output_token_p95_ms": 17.235854207619763, "mean_time_per_output_token_p99_ms": 18.78322468800648, "mean_time_per_output_token_samples": 160, "mean_time_per_output_token_stddev_ms": 1.6545150733999903, "output_token_sources": { "usage": 160 }, "output_tokens": 44656, "prefix_cache_hit_avg": 0.0011643533907411787, "prompt_token_samples": 160, "prompt_tokens": 1236764, "prompt_tokens_max": 7798.0, "prompt_tokens_mean": 7729.775, "prompt_tokens_mean_ci95_high": 7735.29436497065, "prompt_tokens_mean_ci95_low": 7724.255635029349, "prompt_tokens_min": 7675.0, "prompt_tokens_p25": 7702.0, "prompt_tokens_p50": 7725.0, "prompt_tokens_p75": 7759.0, "prompt_tokens_p90": 7782.0, "prompt_tokens_p95": 7790.0, "prompt_tokens_p99": 7797.0, "prompt_tokens_samples": 160, "prompt_tokens_stddev": 35.61992764286613, "reasoning_chunks": 35451, "request_canary_passed": 0, "request_canary_samples": 0, "throughput_performance_eligible": true, "throughput_tok_s": 144.89659746977972, "time_to_first_output_max_ms": 4683.686526957899, "time_to_first_output_mean_ci95_high_ms": 3747.3235069474345, "time_to_first_output_mean_ci95_low_ms": 3664.443038773022, "time_to_first_output_mean_ms": 3705.8832728602283, "time_to_first_output_min_ms": 2939.130750950426, "time_to_first_output_p25_ms": 3570.774797932245, "time_to_first_output_p50_ms": 3708.07266398333, "time_to_first_output_p75_ms": 3879.7955729532987, "time_to_first_output_p90_ms": 4016.8946949997917, "time_to_first_output_p95_ms": 4085.437330068089, "time_to_first_output_p99_ms": 4329.68736696057, "time_to_first_output_samples": 160, "time_to_first_output_stddev_ms": 267.43984997167377, "tpot_max_ms": 19.88835806651896, "tpot_mean_ci95_high_ms": 14.64070026254027, "tpot_mean_ci95_low_ms": 14.127960729147857, "tpot_mean_ms": 14.384330495844063, "tpot_min_ms": 8.75473032879843, "tpot_p25_ms": 13.346120103504683, "tpot_p50_ms": 14.280297470072078, "tpot_p75_ms": 15.221824288057784, "tpot_p90_ms": 16.482556866093535, "tpot_p95_ms": 17.235854207619763, "tpot_p99_ms": 18.78322468800648, "tpot_samples": 160, "tpot_stddev_ms": 1.6545150733999903, "ttft_max_ms": 11514.063915004954, "ttft_mean_ci95_high_ms": 7107.670344876204, "ttft_mean_ci95_low_ms": 6781.098169158542, "ttft_mean_ms": 6944.384257017373, "ttft_min_ms": 4418.01361402031, "ttft_p25_ms": 6279.8589039593935, "ttft_p50_ms": 6872.495774994604, "ttft_p75_ms": 7528.884617961012, "ttft_p90_ms": 8130.599186057225, "ttft_p95_ms": 8847.234047949314, "ttft_p99_ms": 10295.419191941619, "ttft_samples": 160, "ttft_stddev_ms": 1053.7876486781063 }, "actual_start_unix_ns": 1792650024955002542, "actual_end_unix_ns": 1792650333147197092 }, { "name": "flare-instrumented-c4-r2", "arm": "flare-instrumented", "concurrency": 4, "repetition": "r2", "attempted": 160, "completed": 160, "failed": 0, "wall_ms": 308939.46647597477, "metrics": { "completion_tokens_max": 547.0, "completion_tokens_mean": 285.29375, "completion_tokens_mean_ci95_high": 298.23511158349754, "completion_tokens_mean_ci95_low": 272.35238841650244, "completion_tokens_min": 103.0, "completion_tokens_p25": 235.0, "completion_tokens_p50": 274.0, "completion_tokens_p75": 333.0, "completion_tokens_p90": 393.0, "completion_tokens_p95": 435.0, "completion_tokens_p99": 507.0, "completion_tokens_samples": 160, "completion_tokens_stddev": 83.51873189317487, "content_chunks": 8433, "content_chunks_s": null, "controlled_output_exact_adherence": 0, "controlled_output_exact_adherence_rate": null, "controlled_output_samples": 0, "controlled_output_unobservable": 0, "decode_tok_s_max": 160.2202145085153, "decode_tok_s_mean": 74.33248254721356, "decode_tok_s_mean_ci95_high": 76.30518727060357, "decode_tok_s_mean_ci95_low": 72.35977782382355, "decode_tok_s_min": 48.355783207474566, "decode_tok_s_p25": 65.83335043206912, "decode_tok_s_p50": 72.39415382709356, "decode_tok_s_p75": 82.0295238576152, "decode_tok_s_p90": 89.1342084212935, "decode_tok_s_p95": 94.3530384097979, "decode_tok_s_p99": 108.63697424107245, "decode_tok_s_samples": 160, "decode_tok_s_stddev": 12.731102197724264, "e2e_max_ms": 11704.11118795164, "e2e_mean_ci95_high_ms": 7892.636263040822, "e2e_mean_ci95_low_ms": 7489.9744314891905, "e2e_mean_ms": 7691.305347265006, "e2e_min_ms": 5067.173109971918, "e2e_p25_ms": 6877.315504010767, "e2e_p50_ms": 7588.639554916881, "e2e_p75_ms": 8278.48375600297, "e2e_p90_ms": 9421.20139405597, "e2e_p95_ms": 10014.430190087296, "e2e_p99_ms": 11083.179373992607, "e2e_samples": 160, "e2e_stddev_ms": 1299.314810732862, "effective_prefill_tok_s_max": 2600.806963726078, "effective_prefill_tok_s_mean": 2060.9331556944853, "effective_prefill_tok_s_mean_ci95_high": 2085.315845643315, "effective_prefill_tok_s_mean_ci95_low": 2036.5504657456559, "effective_prefill_tok_s_min": 1640.550501579586, "effective_prefill_tok_s_p25": 1954.0599992964803, "effective_prefill_tok_s_p50": 2051.7443720580404, "effective_prefill_tok_s_p75": 2143.564130682789, "effective_prefill_tok_s_p90": 2240.1973377659015, "effective_prefill_tok_s_p95": 2375.7096442545694, "effective_prefill_tok_s_p99": 2478.7586745594117, "effective_prefill_tok_s_samples": 160, "effective_prefill_tok_s_stddev": 157.3568075918273, "generation_max_ms": 8108.21676091291, "generation_mean_ci95_high_ms": 4117.209107624209, "generation_mean_ci95_low_ms": 3720.8298550477907, "generation_mean_ms": 3919.0194813359994, "generation_min_ms": 1172.8482199832797, "generation_p25_ms": 3116.729845991358, "generation_p50_ms": 3794.3771040299907, "generation_p75_ms": 4513.355867937207, "generation_p90_ms": 5661.775717046112, "generation_p95_ms": 6320.52287296392, "generation_p99_ms": 7135.270694969222, "generation_samples": 160, "generation_stddev_ms": 1279.0420973231078, "mean_inter_token_latency_ms_max": 20.68004969973117, "mean_inter_token_latency_ms_mean": 13.785325039930282, "mean_inter_token_latency_ms_mean_ci95_high": 14.104208267078075, "mean_inter_token_latency_ms_mean_ci95_low": 13.466441812782488, "mean_inter_token_latency_ms_min": 6.2414096939487775, "mean_inter_token_latency_ms_p25": 12.149209961764344, "mean_inter_token_latency_ms_p50": 13.785467485740515, "mean_inter_token_latency_ms_p75": 15.145660469910803, "mean_inter_token_latency_ms_p90": 16.342288331916997, "mean_inter_token_latency_ms_p95": 16.740040071622982, "mean_inter_token_latency_ms_p99": 18.993898224541727, "mean_inter_token_latency_ms_samples": 160, "mean_inter_token_latency_ms_stddev": 2.0579536845140165, "mean_time_per_output_token_max_ms": 20.68004969973117, "mean_time_per_output_token_mean_ci95_high_ms": 14.104208267078075, "mean_time_per_output_token_mean_ci95_low_ms": 13.466441812782488, "mean_time_per_output_token_mean_ms": 13.785325039930282, "mean_time_per_output_token_min_ms": 6.2414096939487775, "mean_time_per_output_token_p25_ms": 12.149209961764344, "mean_time_per_output_token_p50_ms": 13.785467485740515, "mean_time_per_output_token_p75_ms": 15.145660469910803, "mean_time_per_output_token_p90_ms": 16.342288331916997, "mean_time_per_output_token_p95_ms": 16.740040071622982, "mean_time_per_output_token_p99_ms": 18.993898224541727, "mean_time_per_output_token_samples": 160, "mean_time_per_output_token_stddev_ms": 2.0579536845140165, "output_token_sources": { "usage": 160 }, "output_tokens": 45647, "prefix_cache_hit_avg": 0.0011643533907411787, "prompt_token_samples": 160, "prompt_tokens": 1236764, "prompt_tokens_max": 7798.0, "prompt_tokens_mean": 7729.775, "prompt_tokens_mean_ci95_high": 7735.29436497065, "prompt_tokens_mean_ci95_low": 7724.255635029349, "prompt_tokens_min": 7675.0, "prompt_tokens_p25": 7702.0, "prompt_tokens_p50": 7725.0, "prompt_tokens_p75": 7759.0, "prompt_tokens_p90": 7782.0, "prompt_tokens_p95": 7790.0, "prompt_tokens_p99": 7797.0, "prompt_tokens_samples": 160, "prompt_tokens_stddev": 35.61992764286613, "reasoning_chunks": 36870, "request_canary_passed": 0, "request_canary_samples": 0, "throughput_performance_eligible": true, "throughput_tok_s": 147.75386427861855, "time_to_first_output_max_ms": 4692.9369090357795, "time_to_first_output_mean_ci95_high_ms": 3816.8466646364777, "time_to_first_output_mean_ci95_low_ms": 3727.725067221536, "time_to_first_output_mean_ms": 3772.285865929007, "time_to_first_output_min_ms": 2952.160659013316, "time_to_first_output_p25_ms": 3607.3491550050676, "time_to_first_output_p50_ms": 3758.506709942594, "time_to_first_output_p75_ms": 3955.041677923873, "time_to_first_output_p90_ms": 4102.263931068592, "time_to_first_output_p95_ms": 4191.691818996333, "time_to_first_output_p99_ms": 4525.646558962762, "time_to_first_output_samples": 160, "time_to_first_output_stddev_ms": 287.5788127997853, "tpot_max_ms": 20.68004969973117, "tpot_mean_ci95_high_ms": 14.104208267078075, "tpot_mean_ci95_low_ms": 13.466441812782488, "tpot_mean_ms": 13.785325039930282, "tpot_min_ms": 6.2414096939487775, "tpot_p25_ms": 12.149209961764344, "tpot_p50_ms": 13.785467485740515, "tpot_p75_ms": 15.145660469910803, "tpot_p90_ms": 16.342288331916997, "tpot_p95_ms": 16.740040071622982, "tpot_p99_ms": 18.993898224541727, "tpot_samples": 160, "tpot_stddev_ms": 2.0579536845140165, "ttft_max_ms": 10913.0585109815, "ttft_mean_ci95_high_ms": 7169.868364038817, "ttft_mean_ci95_low_ms": 6770.825279744926, "ttft_mean_ms": 6970.3468218918715, "ttft_min_ms": 4330.412313924171, "ttft_p25_ms": 6096.379638998769, "ttft_p50_ms": 6793.6773230321705, "ttft_p75_ms": 7604.460985050537, "ttft_p90_ms": 8682.951243943535, "ttft_p95_ms": 9410.384244984016, "ttft_p99_ms": 10497.312855906785, "ttft_samples": 160, "ttft_stddev_ms": 1287.637786640062 }, "actual_start_unix_ns": 1792650333573826369, "actual_end_unix_ns": 1792650642513297764 }, { "name": "flare-instrumented-c4-r3", "arm": "flare-instrumented", "concurrency": 4, "repetition": "r3", "attempted": 160, "completed": 160, "failed": 0, "wall_ms": 307029.7308770241, "metrics": { "completion_tokens_max": 570.0, "completion_tokens_mean": 278.4, "completion_tokens_mean_ci95_high": 290.6411731217325, "completion_tokens_mean_ci95_low": 266.15882687826746, "completion_tokens_min": 91.0, "completion_tokens_p25": 224.0, "completion_tokens_p50": 274.0, "completion_tokens_p75": 323.0, "completion_tokens_p90": 390.0, "completion_tokens_p95": 425.0, "completion_tokens_p99": 464.0, "completion_tokens_samples": 160, "completion_tokens_stddev": 78.99997611654761, "content_chunks": 8614, "content_chunks_s": null, "controlled_output_exact_adherence": 0, "controlled_output_exact_adherence_rate": null, "controlled_output_samples": 0, "controlled_output_unobservable": 0, "decode_tok_s_max": 150.54213492811644, "decode_tok_s_mean": 71.89031374806426, "decode_tok_s_mean_ci95_high": 73.75314511835803, "decode_tok_s_mean_ci95_low": 70.02748237777048, "decode_tok_s_min": 41.53938600219219, "decode_tok_s_p25": 66.21596888395183, "decode_tok_s_p50": 70.61276444932882, "decode_tok_s_p75": 75.53013092358583, "decode_tok_s_p90": 81.15821161434195, "decode_tok_s_p95": 88.17891514033363, "decode_tok_s_p99": 143.99062678418423, "decode_tok_s_samples": 160, "decode_tok_s_stddev": 12.022020463144647, "e2e_max_ms": 11241.064281901345, "e2e_mean_ci95_high_ms": 7810.610829575823, "e2e_mean_ci95_low_ms": 7467.64184512161, "e2e_mean_ms": 7639.126337348716, "e2e_min_ms": 4753.880725940689, "e2e_p25_ms": 6848.746907082386, "e2e_p50_ms": 7623.2144919922575, "e2e_p75_ms": 8176.195395062678, "e2e_p90_ms": 9048.64171904046, "e2e_p95_ms": 9730.077469022945, "e2e_p99_ms": 10604.4541799929, "e2e_samples": 160, "e2e_stddev_ms": 1106.6970996634627, "effective_prefill_tok_s_max": 2559.24166987439, "effective_prefill_tok_s_mean": 2072.3184932552685, "effective_prefill_tok_s_mean_ci95_high": 2092.6685018823478, "effective_prefill_tok_s_mean_ci95_low": 2051.9684846281893, "effective_prefill_tok_s_min": 1704.140637604374, "effective_prefill_tok_s_p25": 1984.8206483537965, "effective_prefill_tok_s_p50": 2061.431056857124, "effective_prefill_tok_s_p75": 2144.781674255641, "effective_prefill_tok_s_p90": 2237.22769445135, "effective_prefill_tok_s_p95": 2301.727541569707, "effective_prefill_tok_s_p99": 2466.944056255096, "effective_prefill_tok_s_samples": 160, "effective_prefill_tok_s_stddev": 131.33138299111442, "generation_max_ms": 7419.7721069213, "generation_mean_ci95_high_ms": 4060.3527805166923, "generation_mean_ci95_low_ms": 3728.256864223371, "generation_mean_ms": 3894.3048223700316, "generation_min_ms": 1431.46238301415, "generation_p25_ms": 3176.578928017989, "generation_p50_ms": 3762.8586509963498, "generation_p75_ms": 4566.740776994266, "generation_p90_ms": 5369.139460963197, "generation_p95_ms": 5817.295919987373, "generation_p99_ms": 6611.835722927935, "generation_samples": 160, "generation_stddev_ms": 1071.6117317627734, "mean_inter_token_latency_ms_max": 24.073538302834475, "mean_inter_token_latency_ms_mean": 14.217553988476112, "mean_inter_token_latency_ms_mean_ci95_high": 14.530186730850774, "mean_inter_token_latency_ms_mean_ci95_low": 13.90492124610145, "mean_inter_token_latency_ms_min": 6.642658551889794, "mean_inter_token_latency_ms_p25": 13.233942085667989, "mean_inter_token_latency_ms_p50": 14.153103166475917, "mean_inter_token_latency_ms_p75": 15.074351975474404, "mean_inter_token_latency_ms_p90": 16.422296474414942, "mean_inter_token_latency_ms_p95": 17.555355462395855, "mean_inter_token_latency_ms_p99": 21.59624324542367, "mean_inter_token_latency_ms_samples": 160, "mean_inter_token_latency_ms_stddev": 2.017615381731771, "mean_time_per_output_token_max_ms": 24.073538302834475, "mean_time_per_output_token_mean_ci95_high_ms": 14.530186730850774, "mean_time_per_output_token_mean_ci95_low_ms": 13.90492124610145, "mean_time_per_output_token_mean_ms": 14.217553988476112, "mean_time_per_output_token_min_ms": 6.642658551889794, "mean_time_per_output_token_p25_ms": 13.233942085667989, "mean_time_per_output_token_p50_ms": 14.153103166475917, "mean_time_per_output_token_p75_ms": 15.074351975474404, "mean_time_per_output_token_p90_ms": 16.422296474414942, "mean_time_per_output_token_p95_ms": 17.555355462395855, "mean_time_per_output_token_p99_ms": 21.59624324542367, "mean_time_per_output_token_samples": 160, "mean_time_per_output_token_stddev_ms": 2.017615381731771, "output_token_sources": { "usage": 160 }, "output_tokens": 44544, "prefix_cache_hit_avg": 0.0011643533907411787, "prompt_token_samples": 160, "prompt_tokens": 1236764, "prompt_tokens_max": 7798.0, "prompt_tokens_mean": 7729.775, "prompt_tokens_mean_ci95_high": 7735.29436497065, "prompt_tokens_mean_ci95_low": 7724.255635029349, "prompt_tokens_min": 7675.0, "prompt_tokens_p25": 7702.0, "prompt_tokens_p50": 7725.0, "prompt_tokens_p75": 7759.0, "prompt_tokens_p90": 7782.0, "prompt_tokens_p95": 7790.0, "prompt_tokens_p99": 7797.0, "prompt_tokens_samples": 160, "prompt_tokens_stddev": 35.61992764286613, "reasoning_chunks": 35598, "request_canary_passed": 0, "request_canary_samples": 0, "throughput_performance_eligible": true, "throughput_tok_s": 145.08041248240352, "time_to_first_output_max_ms": 4529.555735993199, "time_to_first_output_mean_ci95_high_ms": 3781.4635450475103, "time_to_first_output_mean_ci95_low_ms": 3708.179484909859, "time_to_first_output_mean_ms": 3744.8215149786847, "time_to_first_output_min_ms": 3000.1074499450624, "time_to_first_output_p25_ms": 3593.8663079869, "time_to_first_output_p50_ms": 3753.467648057267, "time_to_first_output_p75_ms": 3897.5054359762, "time_to_first_output_p90_ms": 4038.366975961253, "time_to_first_output_p95_ms": 4113.581227022223, "time_to_first_output_p99_ms": 4178.568747011013, "time_to_first_output_samples": 160, "time_to_first_output_stddev_ms": 236.47402675482687, "tpot_max_ms": 24.073538302834475, "tpot_mean_ci95_high_ms": 14.530186730850774, "tpot_mean_ci95_low_ms": 13.90492124610145, "tpot_mean_ms": 14.217553988476112, "tpot_min_ms": 6.642658551889794, "tpot_p25_ms": 13.233942085667989, "tpot_p50_ms": 14.153103166475917, "tpot_p75_ms": 15.074351975474404, "tpot_p90_ms": 16.422296474414942, "tpot_p95_ms": 17.555355462395855, "tpot_p99_ms": 21.59624324542367, "tpot_samples": 160, "tpot_stddev_ms": 2.017615381731771, "ttft_max_ms": 10791.71229491476, "ttft_mean_ci95_high_ms": 7068.673620124178, "ttft_mean_ci95_low_ms": 6725.000629489654, "ttft_mean_ms": 6896.837124806916, "ttft_min_ms": 4130.328460014425, "ttft_p25_ms": 6119.9216190725565, "ttft_p50_ms": 6785.473134019412, "ttft_p75_ms": 7551.601937040687, "ttft_p90_ms": 8348.446596064605, "ttft_p95_ms": 8814.640001044609, "ttft_p99_ms": 9889.887569006532, "ttft_samples": 160, "ttft_stddev_ms": 1108.9687966192 }, "actual_start_unix_ns": 1792650642875912832, "actual_end_unix_ns": 1792650949905646312 }, { "name": "incumbent-after-c1-prime", "arm": "incumbent-after", "concurrency": 1, "repetition": "prime", "attempted": 52, "completed": 52, "failed": 0, "wall_ms": 128166.13242903259, "metrics": { "completion_tokens_max": 592.0, "completion_tokens_mean": 306.13461538461536, "completion_tokens_mean_ci95_high": 331.6749215717437, "completion_tokens_mean_ci95_low": 280.594309197487, "completion_tokens_min": 161.0, "completion_tokens_p25": 223.0, "completion_tokens_p50": 292.0, "completion_tokens_p75": 368.0, "completion_tokens_p90": 418.0, "completion_tokens_p95": 490.0, "completion_tokens_p99": 592.0, "completion_tokens_samples": 52, "completion_tokens_stddev": 93.96620770279733, "content_chunks": 961, "content_chunks_s": null, "controlled_output_exact_adherence": 0, "controlled_output_exact_adherence_rate": null, "controlled_output_samples": 0, "controlled_output_unobservable": 0, "decode_tok_s_max": 214.68861174403867, "decode_tok_s_mean": 191.98675162582796, "decode_tok_s_mean_ci95_high": 194.26300276737226, "decode_tok_s_mean_ci95_low": 189.71050048428367, "decode_tok_s_min": 170.78966678231558, "decode_tok_s_p25": 185.6104193356094, "decode_tok_s_p50": 191.60083366383822, "decode_tok_s_p75": 197.19162747908769, "decode_tok_s_p90": 201.8953970902862, "decode_tok_s_p95": 205.915172780486, "decode_tok_s_p99": 214.68861174403867, "decode_tok_s_samples": 52, "decode_tok_s_stddev": 8.374632863950351, "e2e_max_ms": 3879.4931969605386, "e2e_mean_ci95_high_ms": 2598.2306114498488, "e2e_mean_ci95_low_ms": 2329.5240491259456, "e2e_mean_ms": 2463.877330287897, "e2e_min_ms": 1702.2739000385627, "e2e_p25_ms": 2087.6769059104845, "e2e_p50_ms": 2403.8551079574972, "e2e_p75_ms": 2768.85183097329, "e2e_p90_ms": 3104.8659899970517, "e2e_p95_ms": 3329.721161047928, "e2e_p99_ms": 3879.4931969605386, "e2e_samples": 52, "e2e_stddev_ms": 494.3037186288239, "effective_prefill_tok_s_max": 9184.661150618227, "effective_prefill_tok_s_mean": 8813.145321948208, "effective_prefill_tok_s_mean_ci95_high": 8860.429846769433, "effective_prefill_tok_s_mean_ci95_low": 8765.860797126983, "effective_prefill_tok_s_min": 8610.477108136978, "effective_prefill_tok_s_p25": 8664.03975535565, "effective_prefill_tok_s_p50": 8747.9109079138, "effective_prefill_tok_s_p75": 8919.791541622493, "effective_prefill_tok_s_p90": 9073.40546953393, "effective_prefill_tok_s_p95": 9154.767101178279, "effective_prefill_tok_s_p99": 9184.661150618227, "effective_prefill_tok_s_samples": 52, "effective_prefill_tok_s_stddev": 173.9661007947707, "generation_max_ms": 3015.1780359447002, "generation_mean_ci95_high_ms": 1724.3572820135018, "generation_mean_ci95_low_ms": 1457.4918425592396, "generation_mean_ms": 1590.9245622863707, "generation_min_ms": 841.0430109361187, "generation_p25_ms": 1202.5939140003175, "generation_p50_ms": 1516.7532609775662, "generation_p75_ms": 1880.0890089478344, "generation_p90_ms": 2216.513136983849, "generation_p95_ms": 2471.979864058085, "generation_p99_ms": 3015.1780359447002, "generation_samples": 52, "generation_stddev_ms": 490.9168497967208, "mean_inter_token_latency_ms_max": 5.855155167405861, "mean_inter_token_latency_ms_mean": 5.218416027786229, "mean_inter_token_latency_ms_mean_ci95_high": 5.280316340745069, "mean_inter_token_latency_ms_mean_ci95_low": 5.156515714827389, "mean_inter_token_latency_ms_min": 4.6579089215605185, "mean_inter_token_latency_ms_p25": 5.055173546131052, "mean_inter_token_latency_ms_p50": 5.218903887824631, "mean_inter_token_latency_ms_p75": 5.38542267056918, "mean_inter_token_latency_ms_p90": 5.472871943169998, "mean_inter_token_latency_ms_p95": 5.6057355406509295, "mean_inter_token_latency_ms_p99": 5.855155167405861, "mean_inter_token_latency_ms_samples": 52, "mean_inter_token_latency_ms_stddev": 0.2277395432044537, "mean_time_per_output_token_max_ms": 5.855155167405861, "mean_time_per_output_token_mean_ci95_high_ms": 5.280316340745069, "mean_time_per_output_token_mean_ci95_low_ms": 5.156515714827389, "mean_time_per_output_token_mean_ms": 5.218416027786229, "mean_time_per_output_token_min_ms": 4.6579089215605185, "mean_time_per_output_token_p25_ms": 5.055173546131052, "mean_time_per_output_token_p50_ms": 5.218903887824631, "mean_time_per_output_token_p75_ms": 5.38542267056918, "mean_time_per_output_token_p90_ms": 5.472871943169998, "mean_time_per_output_token_p95_ms": 5.6057355406509295, "mean_time_per_output_token_p99_ms": 5.855155167405861, "mean_time_per_output_token_samples": 52, "mean_time_per_output_token_stddev_ms": 0.2277395432044537, "output_token_sources": { "usage": 52 }, "output_tokens": 15919, "prefix_cache_hit_avg": 0.0, "prompt_token_samples": 52, "prompt_tokens": 399899, "prompt_tokens_max": 7708.0, "prompt_tokens_mean": 7690.365384615385, "prompt_tokens_mean_ci95_high": 7693.3501411136585, "prompt_tokens_mean_ci95_low": 7687.380628117111, "prompt_tokens_min": 7675.0, "prompt_tokens_p25": 7680.0, "prompt_tokens_p50": 7688.0, "prompt_tokens_p75": 7701.0, "prompt_tokens_p90": 7705.0, "prompt_tokens_p95": 7707.0, "prompt_tokens_p99": 7708.0, "prompt_tokens_samples": 52, "prompt_tokens_stddev": 10.981318978876677, "reasoning_chunks": 4517, "request_canary_passed": 0, "request_canary_samples": 0, "throughput_performance_eligible": true, "throughput_tok_s": 124.20597936677676, "time_to_first_output_max_ms": 895.1884899288416, "time_to_first_output_mean_ci95_high_ms": 877.8861746075524, "time_to_first_output_mean_ci95_low_ms": 868.0193613955003, "time_to_first_output_mean_ms": 872.9527680015264, "time_to_first_output_min_ms": 835.6323520420119, "time_to_first_output_p25_ms": 859.6957820700482, "time_to_first_output_p50_ms": 878.708316013217, "time_to_first_output_p75_ms": 888.7628220254555, "time_to_first_output_p90_ms": 891.8897110270336, "time_to_first_output_p95_ms": 893.6810870654881, "time_to_first_output_p99_ms": 895.1884899288416, "time_to_first_output_samples": 52, "time_to_first_output_stddev_ms": 18.15066375585705, "tpot_max_ms": 5.855155167405861, "tpot_mean_ci95_high_ms": 5.280316340745069, "tpot_mean_ci95_low_ms": 5.156515714827389, "tpot_mean_ms": 5.218416027786229, "tpot_min_ms": 4.6579089215605185, "tpot_p25_ms": 5.055173546131052, "tpot_p50_ms": 5.218903887824631, "tpot_p75_ms": 5.38542267056918, "tpot_p90_ms": 5.472871943169998, "tpot_p95_ms": 5.6057355406509295, "tpot_p99_ms": 5.855155167405861, "tpot_samples": 52, "tpot_stddev_ms": 0.2277395432044537, "ttft_max_ms": 3531.778384000063, "ttft_mean_ci95_high_ms": 2309.838294988999, "ttft_mean_ci95_low_ms": 2043.8082088350968, "ttft_mean_ms": 2176.823251912048, "ttft_min_ms": 1480.6958480039611, "ttft_p25_ms": 1766.1895810160786, "ttft_p50_ms": 2108.550664037466, "ttft_p75_ms": 2452.6240779086947, "ttft_p90_ms": 2875.3456610720605, "ttft_p95_ms": 3066.8923259945586, "ttft_p99_ms": 3531.778384000063, "ttft_samples": 52, "ttft_stddev_ms": 489.3801614510192 }, "actual_start_unix_ns": 1792651268673257164, "actual_end_unix_ns": 1792651396839392066 }, { "name": "incumbent-after-c1-r1", "arm": "incumbent-after", "concurrency": 1, "repetition": "r1", "attempted": 52, "completed": 52, "failed": 0, "wall_ms": 117554.56486693583, "metrics": { "completion_tokens_max": 487.0, "completion_tokens_mean": 259.8076923076923, "completion_tokens_mean_ci95_high": 280.2014171091456, "completion_tokens_mean_ci95_low": 239.41396750623903, "completion_tokens_min": 112.0, "completion_tokens_p25": 200.0, "completion_tokens_p50": 262.0, "completion_tokens_p75": 308.0, "completion_tokens_p90": 338.0, "completion_tokens_p95": 396.0, "completion_tokens_p99": 487.0, "completion_tokens_samples": 52, "completion_tokens_stddev": 75.03124537687903, "content_chunks": 957, "content_chunks_s": null, "controlled_output_exact_adherence": 0, "controlled_output_exact_adherence_rate": null, "controlled_output_samples": 0, "controlled_output_unobservable": 0, "decode_tok_s_max": 208.84168820199298, "decode_tok_s_mean": 189.0479885856821, "decode_tok_s_mean_ci95_high": 191.76946439066126, "decode_tok_s_mean_ci95_low": 186.32651278070296, "decode_tok_s_min": 163.84618487632426, "decode_tok_s_p25": 182.81564586255928, "decode_tok_s_p50": 186.88504690972053, "decode_tok_s_p75": 197.5835590129131, "decode_tok_s_p90": 201.16763168213507, "decode_tok_s_p95": 203.9741464549559, "decode_tok_s_p99": 208.84168820199298, "decode_tok_s_samples": 52, "decode_tok_s_stddev": 10.012674040598927, "e2e_max_ms": 3342.2569528920576, "e2e_mean_ci95_high_ms": 2362.7229559822126, "e2e_mean_ci95_low_ms": 2156.78895407022, "e2e_mean_ms": 2259.7559550262163, "e2e_min_ms": 1441.990991937928, "e2e_p25_ms": 1977.2204669425264, "e2e_p50_ms": 2249.1183610400185, "e2e_p75_ms": 2526.095572975464, "e2e_p90_ms": 2656.449705013074, "e2e_p95_ms": 2955.837599001825, "e2e_p99_ms": 3342.2569528920576, "e2e_samples": 52, "e2e_stddev_ms": 378.82938941591345, "effective_prefill_tok_s_max": 8667.317934987817, "effective_prefill_tok_s_mean": 8622.691492000838, "effective_prefill_tok_s_mean_ci95_high": 8629.558517498224, "effective_prefill_tok_s_mean_ci95_low": 8615.824466503453, "effective_prefill_tok_s_min": 8572.38915848042, "effective_prefill_tok_s_p25": 8603.28866575819, "effective_prefill_tok_s_p50": 8626.110670473021, "effective_prefill_tok_s_p75": 8643.12618352234, "effective_prefill_tok_s_p90": 8651.868161216396, "effective_prefill_tok_s_p95": 8657.20897758008, "effective_prefill_tok_s_p99": 8667.317934987817, "effective_prefill_tok_s_samples": 52, "effective_prefill_tok_s_stddev": 25.264706674228606, "generation_max_ms": 2452.5043868925422, "generation_mean_ci95_high_ms": 1470.73814997614, "generation_mean_ci95_low_ms": 1265.0108872840856, "generation_mean_ms": 1367.8745186301128, "generation_min_ms": 551.7786289565265, "generation_p25_ms": 1088.5282770032063, "generation_p50_ms": 1358.681737096049, "generation_p75_ms": 1635.61557198409, "generation_p90_ms": 1764.5528730936348, "generation_p95_ms": 2062.765095033683, "generation_p99_ms": 2452.5043868925422, "generation_samples": 52, "generation_stddev_ms": 378.449078774006, "mean_inter_token_latency_ms_max": 6.103285229099648, "mean_inter_token_latency_ms_mean": 5.304464072057214, "mean_inter_token_latency_ms_mean_ci95_high": 5.382095541450338, "mean_inter_token_latency_ms_mean_ci95_low": 5.226832602664089, "mean_inter_token_latency_ms_min": 4.788316013959788, "mean_inter_token_latency_ms_p25": 5.047370971269196, "mean_inter_token_latency_ms_p50": 5.346305149675579, "mean_inter_token_latency_ms_p75": 5.465800656742458, "mean_inter_token_latency_ms_p90": 5.6062912719210045, "mean_inter_token_latency_ms_p95": 5.894307619584185, "mean_inter_token_latency_ms_p99": 6.103285229099648, "mean_inter_token_latency_ms_samples": 52, "mean_inter_token_latency_ms_stddev": 0.2856165749862492, "mean_time_per_output_token_max_ms": 6.103285229099648, "mean_time_per_output_token_mean_ci95_high_ms": 5.382095541450338, "mean_time_per_output_token_mean_ci95_low_ms": 5.226832602664089, "mean_time_per_output_token_mean_ms": 5.304464072057214, "mean_time_per_output_token_min_ms": 4.788316013959788, "mean_time_per_output_token_p25_ms": 5.047370971269196, "mean_time_per_output_token_p50_ms": 5.346305149675579, "mean_time_per_output_token_p75_ms": 5.465800656742458, "mean_time_per_output_token_p90_ms": 5.6062912719210045, "mean_time_per_output_token_p95_ms": 5.894307619584185, "mean_time_per_output_token_p99_ms": 6.103285229099648, "mean_time_per_output_token_samples": 52, "mean_time_per_output_token_stddev_ms": 0.2856165749862492, "output_token_sources": { "usage": 52 }, "output_tokens": 13510, "prefix_cache_hit_avg": 0.0, "prompt_token_samples": 52, "prompt_tokens": 399899, "prompt_tokens_max": 7708.0, "prompt_tokens_mean": 7690.365384615385, "prompt_tokens_mean_ci95_high": 7693.3501411136585, "prompt_tokens_mean_ci95_low": 7687.380628117111, "prompt_tokens_min": 7675.0, "prompt_tokens_p25": 7680.0, "prompt_tokens_p50": 7688.0, "prompt_tokens_p75": 7701.0, "prompt_tokens_p90": 7705.0, "prompt_tokens_p95": 7707.0, "prompt_tokens_p99": 7708.0, "prompt_tokens_samples": 52, "prompt_tokens_stddev": 10.981318978876677, "reasoning_chunks": 3691, "request_canary_passed": 0, "request_canary_samples": 0, "throughput_performance_eligible": true, "throughput_tok_s": 114.92535415610995, "time_to_first_output_max_ms": 896.7390570323914, "time_to_first_output_mean_ci95_high_ms": 892.5465015060613, "time_to_first_output_mean_ci95_low_ms": 891.2163712861459, "time_to_first_output_mean_ms": 891.8814363961036, "time_to_first_output_min_ms": 885.856508044526, "time_to_first_output_p25_ms": 890.2123629814014, "time_to_first_output_p50_ms": 891.8968319194391, "time_to_first_output_p75_ms": 893.7716269865632, "time_to_first_output_p90_ms": 894.7637969395146, "time_to_first_output_p95_ms": 896.4695199392736, "time_to_first_output_p99_ms": 896.7390570323914, "time_to_first_output_samples": 52, "time_to_first_output_stddev_ms": 2.4468636280353304, "tpot_max_ms": 6.103285229099648, "tpot_mean_ci95_high_ms": 5.382095541450338, "tpot_mean_ci95_low_ms": 5.226832602664089, "tpot_mean_ms": 5.304464072057214, "tpot_min_ms": 4.788316013959788, "tpot_p25_ms": 5.047370971269196, "tpot_p50_ms": 5.346305149675579, "tpot_p75_ms": 5.465800656742458, "tpot_p90_ms": 5.6062912719210045, "tpot_p95_ms": 5.894307619584185, "tpot_p99_ms": 6.103285229099648, "tpot_samples": 52, "tpot_stddev_ms": 0.2856165749862492, "ttft_max_ms": 2988.383215968497, "ttft_mean_ci95_high_ms": 2072.3675387119465, "ttft_mean_ci95_low_ms": 1867.4267729193095, "ttft_mean_ms": 1969.897155815628, "ttft_min_ms": 1194.6830139495432, "ttft_p25_ms": 1677.4655799381435, "ttft_p50_ms": 1984.3046190217137, "ttft_p75_ms": 2193.948501953855, "ttft_p90_ms": 2380.8432869846, "ttft_p95_ms": 2684.311341959983, "ttft_p99_ms": 2988.383215968497, "ttft_samples": 52, "ttft_stddev_ms": 377.00226505010625 }, "actual_start_unix_ns": 1792651397288562055, "actual_end_unix_ns": 1792651514843129967 }, { "name": "incumbent-after-c1-r2", "arm": "incumbent-after", "concurrency": 1, "repetition": "r2", "attempted": 52, "completed": 52, "failed": 0, "wall_ms": 123765.10754495393, "metrics": { "completion_tokens_max": 924.0, "completion_tokens_mean": 283.86538461538464, "completion_tokens_mean_ci95_high": 319.9974616934938, "completion_tokens_mean_ci95_low": 247.7333075372755, "completion_tokens_min": 91.0, "completion_tokens_p25": 214.0, "completion_tokens_p50": 265.0, "completion_tokens_p75": 330.0, "completion_tokens_p90": 379.0, "completion_tokens_p95": 444.0, "completion_tokens_p99": 924.0, "completion_tokens_samples": 52, "completion_tokens_stddev": 132.93475162667303, "content_chunks": 979, "content_chunks_s": null, "controlled_output_exact_adherence": 0, "controlled_output_exact_adherence_rate": null, "controlled_output_samples": 0, "controlled_output_unobservable": 0, "decode_tok_s_max": 215.50397362948257, "decode_tok_s_mean": 190.04781487947204, "decode_tok_s_mean_ci95_high": 192.66062583533582, "decode_tok_s_mean_ci95_low": 187.43500392360826, "decode_tok_s_min": 166.5191584425972, "decode_tok_s_p25": 183.9634571514003, "decode_tok_s_p50": 189.64519254964014, "decode_tok_s_p75": 196.24527986041852, "decode_tok_s_p90": 200.47965842322307, "decode_tok_s_p95": 203.67410728293362, "decode_tok_s_p99": 215.50397362948257, "decode_tok_s_samples": 52, "decode_tok_s_stddev": 9.612881504551925, "e2e_max_ms": 5908.1313939532265, "e2e_mean_ci95_high_ms": 2570.1307585802706, "e2e_mean_ci95_low_ms": 2188.071561375541, "e2e_mean_ms": 2379.101159977906, "e2e_min_ms": 1359.8372020060197, "e2e_p25_ms": 1993.2663990184665, "e2e_p50_ms": 2286.3336770096794, "e2e_p75_ms": 2627.852967940271, "e2e_p90_ms": 2851.659342995845, "e2e_p95_ms": 3217.716268962249, "e2e_p99_ms": 5908.1313939532265, "e2e_samples": 52, "e2e_stddev_ms": 702.8234825429901, "effective_prefill_tok_s_max": 8718.349536439551, "effective_prefill_tok_s_mean": 8649.500711656056, "effective_prefill_tok_s_mean_ci95_high": 8657.805830255025, "effective_prefill_tok_s_mean_ci95_low": 8641.195593057086, "effective_prefill_tok_s_min": 8582.565886123777, "effective_prefill_tok_s_p25": 8625.070108961298, "effective_prefill_tok_s_p50": 8651.476518294125, "effective_prefill_tok_s_p75": 8668.4830259501, "effective_prefill_tok_s_p90": 8690.634802535576, "effective_prefill_tok_s_p95": 8698.548787518126, "effective_prefill_tok_s_p99": 8718.349536439551, "effective_prefill_tok_s_samples": 52, "effective_prefill_tok_s_stddev": 30.555643834075028, "generation_max_ms": 5017.300795996562, "generation_mean_ci95_high_ms": 1681.1266623279548, "generation_mean_ci95_low_ms": 1298.829065143836, "generation_mean_ms": 1489.9778637358954, "generation_min_ms": 468.3035110356286, "generation_p25_ms": 1105.3795319749042, "generation_p50_ms": 1404.1067069629207, "generation_p75_ms": 1739.3424059264362, "generation_p90_ms": 1968.1088260840625, "generation_p95_ms": 2332.6959429541603, "generation_p99_ms": 5017.300795996562, "generation_samples": 52, "generation_stddev_ms": 703.2620352724582, "mean_inter_token_latency_ms_max": 6.005315000103859, "mean_inter_token_latency_ms_mean": 5.2751579980508865, "mean_inter_token_latency_ms_mean_ci95_high": 5.348373960899434, "mean_inter_token_latency_ms_mean_ci95_low": 5.201942035202339, "mean_inter_token_latency_ms_min": 4.640285666933022, "mean_inter_token_latency_ms_p25": 5.091115838488337, "mean_inter_token_latency_ms_p50": 5.2656793294676305, "mean_inter_token_latency_ms_p75": 5.430533920651104, "mean_inter_token_latency_ms_p90": 5.589780246373266, "mean_inter_token_latency_ms_p95": 5.813306495903882, "mean_inter_token_latency_ms_p99": 6.005315000103859, "mean_inter_token_latency_ms_samples": 52, "mean_inter_token_latency_ms_stddev": 0.26937133493153737, "mean_time_per_output_token_max_ms": 6.005315000103859, "mean_time_per_output_token_mean_ci95_high_ms": 5.348373960899434, "mean_time_per_output_token_mean_ci95_low_ms": 5.201942035202339, "mean_time_per_output_token_mean_ms": 5.2751579980508865, "mean_time_per_output_token_min_ms": 4.640285666933022, "mean_time_per_output_token_p25_ms": 5.091115838488337, "mean_time_per_output_token_p50_ms": 5.2656793294676305, "mean_time_per_output_token_p75_ms": 5.430533920651104, "mean_time_per_output_token_p90_ms": 5.589780246373266, "mean_time_per_output_token_p95_ms": 5.813306495903882, "mean_time_per_output_token_p99_ms": 6.005315000103859, "mean_time_per_output_token_samples": 52, "mean_time_per_output_token_stddev_ms": 0.26937133493153737, "output_token_sources": { "usage": 52 }, "output_tokens": 14761, "prefix_cache_hit_avg": 0.0, "prompt_token_samples": 52, "prompt_tokens": 399899, "prompt_tokens_max": 7708.0, "prompt_tokens_mean": 7690.365384615385, "prompt_tokens_mean_ci95_high": 7693.3501411136585, "prompt_tokens_mean_ci95_low": 7687.380628117111, "prompt_tokens_min": 7675.0, "prompt_tokens_p25": 7680.0, "prompt_tokens_p50": 7688.0, "prompt_tokens_p75": 7701.0, "prompt_tokens_p90": 7705.0, "prompt_tokens_p95": 7707.0, "prompt_tokens_p99": 7708.0, "prompt_tokens_samples": 52, "prompt_tokens_stddev": 10.981318978876677, "reasoning_chunks": 4092, "request_canary_passed": 0, "request_canary_samples": 0, "throughput_performance_eligible": true, "throughput_tok_s": 119.26624791756039, "time_to_first_output_max_ms": 896.9345650402829, "time_to_first_output_mean_ci95_high_ms": 890.1479340228714, "time_to_first_output_mean_ci95_low_ms": 888.0986584611492, "time_to_first_output_mean_ms": 889.1232962420103, "time_to_first_output_min_ms": 880.3271729266271, "time_to_first_output_p25_ms": 887.2877350077033, "time_to_first_output_p50_ms": 888.3131350157782, "time_to_first_output_p75_ms": 891.7489689774811, "time_to_first_output_p90_ms": 894.0924080088735, "time_to_first_output_p95_ms": 895.4698629677296, "time_to_first_output_p99_ms": 896.9345650402829, "time_to_first_output_samples": 52, "time_to_first_output_stddev_ms": 3.769779650686163, "tpot_max_ms": 6.005315000103859, "tpot_mean_ci95_high_ms": 5.348373960899434, "tpot_mean_ci95_low_ms": 5.201942035202339, "tpot_mean_ms": 5.2751579980508865, "tpot_min_ms": 4.640285666933022, "tpot_p25_ms": 5.091115838488337, "tpot_p50_ms": 5.2656793294676305, "tpot_p75_ms": 5.430533920651104, "tpot_p90_ms": 5.589780246373266, "tpot_p95_ms": 5.813306495903882, "tpot_p99_ms": 6.005315000103859, "tpot_samples": 52, "tpot_stddev_ms": 0.26937133493153737, "ttft_max_ms": 5627.845167997293, "ttft_mean_ci95_high_ms": 2272.9948931644976, "ttft_mean_ci95_low_ms": 1893.0572114900094, "ttft_mean_ms": 2083.0260523272536, "ttft_min_ms": 1084.8240159684792, "ttft_p25_ms": 1714.2094100126997, "ttft_p50_ms": 1999.3418819503859, "ttft_p75_ms": 2244.2715090001, "ttft_p90_ms": 2526.6853700159118, "ttft_p95_ms": 2938.210212974809, "ttft_p99_ms": 5627.845167997293, "ttft_samples": 52, "ttft_stddev_ms": 698.9208126317767 }, "actual_start_unix_ns": 1792651517172334087, "actual_end_unix_ns": 1792651640937444199 }, { "name": "incumbent-after-c1-r3", "arm": "incumbent-after", "concurrency": 1, "repetition": "r3", "attempted": 52, "completed": 52, "failed": 0, "wall_ms": 123002.86813196726, "metrics": { "completion_tokens_max": 869.0, "completion_tokens_mean": 277.88461538461536, "completion_tokens_mean_ci95_high": 308.6795279315735, "completion_tokens_mean_ci95_low": 247.08970283765723, "completion_tokens_min": 121.0, "completion_tokens_p25": 209.0, "completion_tokens_p50": 266.0, "completion_tokens_p75": 317.0, "completion_tokens_p90": 371.0, "completion_tokens_p95": 432.0, "completion_tokens_p99": 869.0, "completion_tokens_samples": 52, "completion_tokens_stddev": 113.2986083790683, "content_chunks": 927, "content_chunks_s": null, "controlled_output_exact_adherence": 0, "controlled_output_exact_adherence_rate": null, "controlled_output_samples": 0, "controlled_output_unobservable": 0, "decode_tok_s_max": 204.7392152903618, "decode_tok_s_mean": 187.43139475503597, "decode_tok_s_mean_ci95_high": 189.72847385673725, "decode_tok_s_mean_ci95_low": 185.13431565333468, "decode_tok_s_min": 168.32521041689077, "decode_tok_s_p25": 181.66236924607304, "decode_tok_s_p50": 187.36052326159074, "decode_tok_s_p75": 192.8741066332046, "decode_tok_s_p90": 196.6539174302171, "decode_tok_s_p95": 201.70767417725398, "decode_tok_s_p99": 204.7392152903618, "decode_tok_s_samples": 52, "decode_tok_s_stddev": 8.4512617193681, "e2e_max_ms": 5173.1006099144, "e2e_mean_ci95_high_ms": 2519.3734518758038, "e2e_mean_ci95_low_ms": 2209.711475162164, "e2e_mean_ms": 2364.542463518984, "e2e_min_ms": 1503.5699639702216, "e2e_p25_ms": 1981.4227320021018, "e2e_p50_ms": 2311.8904531002045, "e2e_p75_ms": 2584.8292720038444, "e2e_p90_ms": 2924.810722935945, "e2e_p95_ms": 3204.4155430048704, "e2e_p99_ms": 5173.1006099144, "e2e_samples": 52, "e2e_stddev_ms": 569.6439464809005, "effective_prefill_tok_s_max": 8693.866702065254, "effective_prefill_tok_s_mean": 8640.31571104029, "effective_prefill_tok_s_mean_ci95_high": 8647.001609680965, "effective_prefill_tok_s_mean_ci95_low": 8633.629812399615, "effective_prefill_tok_s_min": 8578.533253689498, "effective_prefill_tok_s_p25": 8623.605110119606, "effective_prefill_tok_s_p50": 8640.399481287277, "effective_prefill_tok_s_p75": 8659.598598188522, "effective_prefill_tok_s_p90": 8670.558645695524, "effective_prefill_tok_s_p95": 8677.512839638644, "effective_prefill_tok_s_p99": 8693.866702065254, "effective_prefill_tok_s_samples": 52, "effective_prefill_tok_s_stddev": 24.598316705622185, "generation_max_ms": 4283.788835979067, "generation_mean_ci95_high_ms": 1629.3095127392444, "generation_mean_ci95_low_ms": 1319.6490554058828, "generation_mean_ms": 1474.4792840725636, "generation_min_ms": 612.1227329131216, "generation_p25_ms": 1087.6694230828434, "generation_p50_ms": 1413.7182630365714, "generation_p75_ms": 1692.336029955186, "generation_p90_ms": 2036.7454279912636, "generation_p95_ms": 2313.7093120021746, "generation_p99_ms": 4283.788835979067, "generation_samples": 52, "generation_stddev_ms": 569.6411514791139, "mean_inter_token_latency_ms_max": 5.940880736303859, "mean_inter_token_latency_ms_mean": 5.346091839234551, "mean_inter_token_latency_ms_mean_ci95_high": 5.412635023152121, "mean_inter_token_latency_ms_mean_ci95_low": 5.27954865531698, "mean_inter_token_latency_ms_min": 4.884262150667115, "mean_inter_token_latency_ms_p25": 5.184170790525015, "mean_inter_token_latency_ms_p50": 5.312076305401408, "mean_inter_token_latency_ms_p75": 5.494569006259553, "mean_inter_token_latency_ms_p90": 5.654538385866948, "mean_inter_token_latency_ms_p95": 5.873556214711429, "mean_inter_token_latency_ms_p99": 5.940880736303859, "mean_inter_token_latency_ms_samples": 52, "mean_inter_token_latency_ms_stddev": 0.24482128739533673, "mean_time_per_output_token_max_ms": 5.940880736303859, "mean_time_per_output_token_mean_ci95_high_ms": 5.412635023152121, "mean_time_per_output_token_mean_ci95_low_ms": 5.27954865531698, "mean_time_per_output_token_mean_ms": 5.346091839234551, "mean_time_per_output_token_min_ms": 4.884262150667115, "mean_time_per_output_token_p25_ms": 5.184170790525015, "mean_time_per_output_token_p50_ms": 5.312076305401408, "mean_time_per_output_token_p75_ms": 5.494569006259553, "mean_time_per_output_token_p90_ms": 5.654538385866948, "mean_time_per_output_token_p95_ms": 5.873556214711429, "mean_time_per_output_token_p99_ms": 5.940880736303859, "mean_time_per_output_token_samples": 52, "mean_time_per_output_token_stddev_ms": 0.24482128739533673, "output_token_sources": { "usage": 52 }, "output_tokens": 14450, "prefix_cache_hit_avg": 0.0, "prompt_token_samples": 52, "prompt_tokens": 399899, "prompt_tokens_max": 7708.0, "prompt_tokens_mean": 7690.365384615385, "prompt_tokens_mean_ci95_high": 7693.3501411136585, "prompt_tokens_mean_ci95_low": 7687.380628117111, "prompt_tokens_min": 7675.0, "prompt_tokens_p25": 7680.0, "prompt_tokens_p50": 7688.0, "prompt_tokens_p75": 7701.0, "prompt_tokens_p90": 7705.0, "prompt_tokens_p95": 7707.0, "prompt_tokens_p99": 7708.0, "prompt_tokens_samples": 52, "prompt_tokens_stddev": 10.981318978876677, "reasoning_chunks": 4073, "request_canary_passed": 0, "request_canary_samples": 0, "throughput_performance_eligible": true, "throughput_tok_s": 117.47693545240661, "time_to_first_output_max_ms": 898.172190063633, "time_to_first_output_mean_ci95_high_ms": 890.8371200489075, "time_to_first_output_mean_ci95_low_ms": 889.2892388439327, "time_to_first_output_mean_ms": 890.0631794464201, "time_to_first_output_min_ms": 882.8062659595162, "time_to_first_output_p25_ms": 888.3961800020188, "time_to_first_output_p50_ms": 889.9725020164624, "time_to_first_output_p75_ms": 891.8476529652253, "time_to_first_output_p90_ms": 893.5937930364162, "time_to_first_output_p95_ms": 893.7533089192584, "time_to_first_output_p99_ms": 898.172190063633, "time_to_first_output_samples": 52, "time_to_first_output_stddev_ms": 2.84743114942046, "tpot_max_ms": 5.940880736303859, "tpot_mean_ci95_high_ms": 5.412635023152121, "tpot_mean_ci95_low_ms": 5.27954865531698, "tpot_mean_ms": 5.346091839234551, "tpot_min_ms": 4.884262150667115, "tpot_p25_ms": 5.184170790525015, "tpot_p50_ms": 5.312076305401408, "tpot_p75_ms": 5.494569006259553, "tpot_p90_ms": 5.654538385866948, "tpot_p95_ms": 5.873556214711429, "tpot_p99_ms": 5.940880736303859, "tpot_samples": 52, "tpot_stddev_ms": 0.24482128739533673, "ttft_max_ms": 4877.5246639270335, "ttft_mean_ci95_high_ms": 2235.7756407461416, "ttft_mean_ci95_low_ms": 1930.553698055704, "ttft_mean_ms": 2083.164669400923, "ttft_min_ms": 1224.845606018789, "ttft_p25_ms": 1668.766421964392, "ttft_p50_ms": 2018.5454610036686, "ttft_p75_ms": 2265.097007039003, "ttft_p90_ms": 2661.4051999058574, "ttft_p95_ms": 2850.3170320764184, "ttft_p99_ms": 4877.5246639270335, "ttft_samples": 52, "ttft_stddev_ms": 561.4762065138282 }, "actual_start_unix_ns": 1792651641393351977, "actual_end_unix_ns": 1792651764396223016 }, { "name": "incumbent-after-c4-prime", "arm": "incumbent-after", "concurrency": 4, "repetition": "prime", "attempted": 160, "completed": 160, "failed": 0, "wall_ms": 312920.8453310421, "metrics": { "completion_tokens_max": 1313.0, "completion_tokens_mean": 272.64375, "completion_tokens_mean_ci95_high": 290.5355083072948, "completion_tokens_mean_ci95_low": 254.75199169270525, "completion_tokens_min": 68.0, "completion_tokens_p25": 214.0, "completion_tokens_p50": 260.0, "completion_tokens_p75": 306.0, "completion_tokens_p90": 373.0, "completion_tokens_p95": 389.0, "completion_tokens_p99": 579.0, "completion_tokens_samples": 160, "completion_tokens_stddev": 115.46675019651067, "content_chunks": 2917, "content_chunks_s": null, "controlled_output_exact_adherence": 0, "controlled_output_exact_adherence_rate": null, "controlled_output_samples": 0, "controlled_output_unobservable": 0, "decode_tok_s_max": 182.77885325455478, "decode_tok_s_mean": 93.63452964040582, "decode_tok_s_mean_ci95_high": 96.30904737114166, "decode_tok_s_mean_ci95_low": 90.96001190966997, "decode_tok_s_min": 60.0660437702385, "decode_tok_s_p25": 80.37728181473761, "decode_tok_s_p50": 90.75844525997441, "decode_tok_s_p75": 101.66265896423718, "decode_tok_s_p90": 117.17015335585036, "decode_tok_s_p95": 126.91566590480127, "decode_tok_s_p99": 136.9728281400206, "decode_tok_s_samples": 160, "decode_tok_s_stddev": 17.260342187000415, "e2e_max_ms": 21587.168070022017, "e2e_mean_ci95_high_ms": 8060.829104727349, "e2e_mean_ci95_low_ms": 7467.090357453231, "e2e_mean_ms": 7763.95973109029, "e2e_min_ms": 2816.313463030383, "e2e_p25_ms": 6708.156854030676, "e2e_p50_ms": 7779.85363395419, "e2e_p75_ms": 8675.733880954795, "e2e_p90_ms": 9487.581423018128, "e2e_p95_ms": 10168.961353017949, "e2e_p99_ms": 13903.148271958344, "e2e_samples": 160, "e2e_stddev_ms": 1915.8844657972472, "effective_prefill_tok_s_max": 8668.246511862306, "effective_prefill_tok_s_mean": 1799.8788205407752, "effective_prefill_tok_s_mean_ci95_high": 1927.3402805332405, "effective_prefill_tok_s_mean_ci95_low": 1672.41736054831, "effective_prefill_tok_s_min": 1142.6895372402632, "effective_prefill_tok_s_p25": 1383.2497408441461, "effective_prefill_tok_s_p50": 1568.5088115032538, "effective_prefill_tok_s_p75": 1901.9623958823593, "effective_prefill_tok_s_p90": 2410.707699585135, "effective_prefill_tok_s_p95": 2625.869719943121, "effective_prefill_tok_s_p99": 4811.717885600054, "effective_prefill_tok_s_samples": 160, "effective_prefill_tok_s_stddev": 822.5888315645273, "generation_max_ms": 16293.79877797328, "generation_mean_ci95_high_ms": 3268.217703279068, "generation_mean_ci95_low_ms": 2793.4792327655414, "generation_mean_ms": 3030.8484680223046, "generation_min_ms": 541.8944959528744, "generation_p25_ms": 2273.2523758895695, "generation_p50_ms": 2742.0305049745366, "generation_p75_ms": 3591.0804179729894, "generation_p90_ms": 4389.671180048026, "generation_p95_ms": 4649.474120000377, "generation_p99_ms": 7513.509158976376, "generation_samples": 160, "generation_stddev_ms": 1531.8927140075828, "mean_inter_token_latency_ms_max": 16.64834134615471, "mean_inter_token_latency_ms_mean": 11.001402945932009, "mean_inter_token_latency_ms_mean_ci95_high": 11.283258956492585, "mean_inter_token_latency_ms_mean_ci95_low": 10.719546935371433, "mean_inter_token_latency_ms_min": 5.471092427783794, "mean_inter_token_latency_ms_p25": 9.835053511063157, "mean_inter_token_latency_ms_p50": 10.96800114355336, "mean_inter_token_latency_ms_p75": 12.435329121949083, "mean_inter_token_latency_ms_p90": 13.12574332935665, "mean_inter_token_latency_ms_p95": 13.56393549585624, "mean_inter_token_latency_ms_p99": 15.434758112168135, "mean_inter_token_latency_ms_samples": 160, "mean_inter_token_latency_ms_stddev": 1.8189938073059004, "mean_time_per_output_token_max_ms": 16.64834134615471, "mean_time_per_output_token_mean_ci95_high_ms": 11.283258956492585, "mean_time_per_output_token_mean_ci95_low_ms": 10.719546935371433, "mean_time_per_output_token_mean_ms": 11.001402945932009, "mean_time_per_output_token_min_ms": 5.471092427783794, "mean_time_per_output_token_p25_ms": 9.835053511063157, "mean_time_per_output_token_p50_ms": 10.96800114355336, "mean_time_per_output_token_p75_ms": 12.435329121949083, "mean_time_per_output_token_p90_ms": 13.12574332935665, "mean_time_per_output_token_p95_ms": 13.56393549585624, "mean_time_per_output_token_p99_ms": 15.434758112168135, "mean_time_per_output_token_samples": 160, "mean_time_per_output_token_stddev_ms": 1.8189938073059004, "output_token_sources": { "usage": 160 }, "output_tokens": 43623, "prefix_cache_hit_avg": 0.0, "prompt_token_samples": 160, "prompt_tokens": 1236764, "prompt_tokens_max": 7798.0, "prompt_tokens_mean": 7729.775, "prompt_tokens_mean_ci95_high": 7735.29436497065, "prompt_tokens_mean_ci95_low": 7724.255635029349, "prompt_tokens_min": 7675.0, "prompt_tokens_p25": 7702.0, "prompt_tokens_p50": 7725.0, "prompt_tokens_p75": 7759.0, "prompt_tokens_p90": 7782.0, "prompt_tokens_p95": 7790.0, "prompt_tokens_p99": 7797.0, "prompt_tokens_samples": 160, "prompt_tokens_stddev": 35.61992764286613, "reasoning_chunks": 12060, "request_canary_passed": 0, "request_canary_samples": 0, "throughput_performance_eligible": true, "throughput_tok_s": 139.40586142112326, "time_to_first_output_max_ms": 6801.497473032214, "time_to_first_output_mean_ci95_high_ms": 4914.410881859196, "time_to_first_output_mean_ci95_low_ms": 4551.811644276775, "time_to_first_output_mean_ms": 4733.111263067985, "time_to_first_output_min_ms": 885.6462480034679, "time_to_first_output_p25_ms": 4036.3636260153726, "time_to_first_output_p50_ms": 4912.2122609987855, "time_to_first_output_p75_ms": 5548.504524980672, "time_to_first_output_p90_ms": 6169.10787706729, "time_to_first_output_p95_ms": 6389.639112981968, "time_to_first_output_p99_ms": 6621.352856978774, "time_to_first_output_samples": 160, "time_to_first_output_stddev_ms": 1170.0402740826266, "tpot_max_ms": 16.64834134615471, "tpot_mean_ci95_high_ms": 11.283258956492585, "tpot_mean_ci95_low_ms": 10.719546935371433, "tpot_mean_ms": 11.001402945932009, "tpot_min_ms": 5.471092427783794, "tpot_p25_ms": 9.835053511063157, "tpot_p50_ms": 10.96800114355336, "tpot_p75_ms": 12.435329121949083, "tpot_p90_ms": 13.12574332935665, "tpot_p95_ms": 13.56393549585624, "tpot_p99_ms": 15.434758112168135, "tpot_samples": 160, "tpot_stddev_ms": 1.8189938073059004, "ttft_max_ms": 21310.01257605385, "ttft_mean_ci95_high_ms": 7401.30899300725, "ttft_mean_ci95_low_ms": 6814.236550254146, "ttft_mean_ms": 7107.772771630698, "ttft_min_ms": 2438.4037969866768, "ttft_p25_ms": 5968.190626939759, "ttft_p50_ms": 7050.392123986967, "ttft_p75_ms": 7954.145606025122, "ttft_p90_ms": 8827.411673031747, "ttft_p95_ms": 9230.664256028831, "ttft_p99_ms": 13395.00528795179, "ttft_samples": 160, "ttft_stddev_ms": 1894.37354144757 }, "actual_start_unix_ns": 1792651764881156296, "actual_end_unix_ns": 1792652077802004132 }, { "name": "incumbent-after-c4-r1", "arm": "incumbent-after", "concurrency": 4, "repetition": "r1", "attempted": 160, "completed": 160, "failed": 0, "wall_ms": 316325.98559500184, "metrics": { "completion_tokens_max": 634.0, "completion_tokens_mean": 284.30625, "completion_tokens_mean_ci95_high": 299.22893601151776, "completion_tokens_mean_ci95_low": 269.3835639884822, "completion_tokens_min": 57.0, "completion_tokens_p25": 213.0, "completion_tokens_p50": 273.0, "completion_tokens_p75": 330.0, "completion_tokens_p90": 421.0, "completion_tokens_p95": 445.0, "completion_tokens_p99": 606.0, "completion_tokens_samples": 160, "completion_tokens_stddev": 96.30546245699983, "content_chunks": 2849, "content_chunks_s": null, "controlled_output_exact_adherence": 0, "controlled_output_exact_adherence_rate": null, "controlled_output_samples": 0, "controlled_output_unobservable": 0, "decode_tok_s_max": 176.63521046350485, "decode_tok_s_mean": 93.18163753442653, "decode_tok_s_mean_ci95_high": 95.85384010140564, "decode_tok_s_mean_ci95_low": 90.50943496744742, "decode_tok_s_min": 59.52770405436609, "decode_tok_s_p25": 81.49273307594396, "decode_tok_s_p50": 90.71266136521189, "decode_tok_s_p75": 100.36135947294089, "decode_tok_s_p90": 114.31250233266222, "decode_tok_s_p95": 121.7402936145685, "decode_tok_s_p99": 173.57568733675654, "decode_tok_s_samples": 160, "decode_tok_s_stddev": 17.245400981638014, "e2e_max_ms": 12646.047619055025, "e2e_mean_ci95_high_ms": 8097.834549778587, "e2e_mean_ci95_low_ms": 7597.180012980438, "e2e_mean_ms": 7847.507281379512, "e2e_min_ms": 3274.446403956972, "e2e_p25_ms": 6732.513804920018, "e2e_p50_ms": 7725.7369350409135, "e2e_p75_ms": 8831.3524100231, "e2e_p90_ms": 9931.067563011311, "e2e_p95_ms": 10722.883736947551, "e2e_p99_ms": 12039.700660970993, "e2e_samples": 160, "e2e_stddev_ms": 1615.5190379374817, "effective_prefill_tok_s_max": 8281.434108011003, "effective_prefill_tok_s_mean": 1819.1056085577115, "effective_prefill_tok_s_mean_ci95_high": 1941.2559864355655, "effective_prefill_tok_s_mean_ci95_low": 1696.9552306798576, "effective_prefill_tok_s_min": 1130.1726554862341, "effective_prefill_tok_s_p25": 1404.3266523365191, "effective_prefill_tok_s_p50": 1626.1171334510589, "effective_prefill_tok_s_p75": 1938.1949977579266, "effective_prefill_tok_s_p90": 2512.489904532653, "effective_prefill_tok_s_p95": 2726.0854914956785, "effective_prefill_tok_s_p99": 4779.791616565452, "effective_prefill_tok_s_samples": 160, "effective_prefill_tok_s_stddev": 788.3130839678851, "generation_max_ms": 7497.431288938969, "generation_mean_ci95_high_ms": 3383.9864351666233, "generation_mean_ci95_low_ms": 2978.589528668492, "generation_mean_ms": 3181.2879819175578, "generation_min_ms": 424.00046391412616, "generation_p25_ms": 2343.121360987425, "generation_p50_ms": 3133.9458590373397, "generation_p75_ms": 3629.1450150310993, "generation_p90_ms": 4953.121689031832, "generation_p95_ms": 5711.98787400499, "generation_p99_ms": 7126.824147999287, "generation_samples": 160, "generation_stddev_ms": 1308.1403886942956, "mean_inter_token_latency_ms_max": 16.79890087960909, "mean_inter_token_latency_ms_mean": 11.053030623694607, "mean_inter_token_latency_ms_mean_ci95_high": 11.335622157306656, "mean_inter_token_latency_ms_mean_ci95_low": 10.770439090082558, "mean_inter_token_latency_ms_min": 5.6613853906926055, "mean_inter_token_latency_ms_p25": 9.878409763118434, "mean_inter_token_latency_ms_p50": 10.99029959915952, "mean_inter_token_latency_ms_p75": 12.244186171115338, "mean_inter_token_latency_ms_p90": 13.33733112731064, "mean_inter_token_latency_ms_p95": 14.013890259699009, "mean_inter_token_latency_ms_p99": 15.343635249315057, "mean_inter_token_latency_ms_samples": 160, "mean_inter_token_latency_ms_stddev": 1.8237405993757227, "mean_time_per_output_token_max_ms": 16.79890087960909, "mean_time_per_output_token_mean_ci95_high_ms": 11.335622157306656, "mean_time_per_output_token_mean_ci95_low_ms": 10.770439090082558, "mean_time_per_output_token_mean_ms": 11.053030623694607, "mean_time_per_output_token_min_ms": 5.6613853906926055, "mean_time_per_output_token_p25_ms": 9.878409763118434, "mean_time_per_output_token_p50_ms": 10.99029959915952, "mean_time_per_output_token_p75_ms": 12.244186171115338, "mean_time_per_output_token_p90_ms": 13.33733112731064, "mean_time_per_output_token_p95_ms": 14.013890259699009, "mean_time_per_output_token_p99_ms": 15.343635249315057, "mean_time_per_output_token_samples": 160, "mean_time_per_output_token_stddev_ms": 1.8237405993757227, "output_token_sources": { "usage": 160 }, "output_tokens": 45489, "prefix_cache_hit_avg": 0.0, "prompt_token_samples": 160, "prompt_tokens": 1236764, "prompt_tokens_max": 7798.0, "prompt_tokens_mean": 7729.775, "prompt_tokens_mean_ci95_high": 7735.29436497065, "prompt_tokens_mean_ci95_low": 7724.255635029349, "prompt_tokens_min": 7675.0, "prompt_tokens_p25": 7702.0, "prompt_tokens_p50": 7725.0, "prompt_tokens_p75": 7759.0, "prompt_tokens_p90": 7782.0, "prompt_tokens_p95": 7790.0, "prompt_tokens_p99": 7797.0, "prompt_tokens_samples": 160, "prompt_tokens_stddev": 35.61992764286613, "reasoning_chunks": 12645, "request_canary_passed": 0, "request_canary_samples": 0, "throughput_performance_eligible": true, "throughput_tok_s": 143.804183252401, "time_to_first_output_max_ms": 6802.147180074826, "time_to_first_output_mean_ci95_high_ms": 4847.075209519256, "time_to_first_output_mean_ci95_low_ms": 4485.363389404653, "time_to_first_output_mean_ms": 4666.2192994619545, "time_to_first_output_min_ms": 926.892600953579, "time_to_first_output_p25_ms": 3967.077050008811, "time_to_first_output_p50_ms": 4729.489141027443, "time_to_first_output_p75_ms": 5500.464620999992, "time_to_first_output_p90_ms": 6228.007539990358, "time_to_first_output_p95_ms": 6398.55719206389, "time_to_first_output_p99_ms": 6790.997784934007, "time_to_first_output_samples": 160, "time_to_first_output_stddev_ms": 1167.176743027804, "tpot_max_ms": 16.79890087960909, "tpot_mean_ci95_high_ms": 11.335622157306656, "tpot_mean_ci95_low_ms": 10.770439090082558, "tpot_mean_ms": 11.053030623694607, "tpot_min_ms": 5.6613853906926055, "tpot_p25_ms": 9.878409763118434, "tpot_p50_ms": 10.99029959915952, "tpot_p75_ms": 12.244186171115338, "tpot_p90_ms": 13.33733112731064, "tpot_p95_ms": 14.013890259699009, "tpot_p99_ms": 15.343635249315057, "tpot_samples": 160, "tpot_stddev_ms": 1.8237405993757227, "ttft_max_ms": 11462.475485983305, "ttft_mean_ci95_high_ms": 7402.063929042676, "ttft_mean_ci95_low_ms": 6925.403060912025, "ttft_mean_ms": 7163.73349497735, "ttft_min_ms": 3042.5673309946433, "ttft_p25_ms": 6166.947760968469, "ttft_p50_ms": 6988.728904980235, "ttft_p75_ms": 8043.252532952465, "ttft_p90_ms": 9108.418730087578, "ttft_p95_ms": 9846.545220934786, "ttft_p99_ms": 11201.362219057046, "ttft_samples": 160, "ttft_stddev_ms": 1538.0959334347226 }, "actual_start_unix_ns": 1792652078175855515, "actual_end_unix_ns": 1792652394501843540 }, { "name": "incumbent-after-c4-r2", "arm": "incumbent-after", "concurrency": 4, "repetition": "r2", "attempted": 160, "completed": 160, "failed": 0, "wall_ms": 317302.8066039551, "metrics": { "completion_tokens_max": 702.0, "completion_tokens_mean": 279.54375, "completion_tokens_mean_ci95_high": 294.31152447473767, "completion_tokens_mean_ci95_low": 264.7759755252623, "completion_tokens_min": 81.0, "completion_tokens_p25": 211.0, "completion_tokens_p50": 276.0, "completion_tokens_p75": 329.0, "completion_tokens_p90": 397.0, "completion_tokens_p95": 448.0, "completion_tokens_p99": 520.0, "completion_tokens_samples": 160, "completion_tokens_stddev": 95.30572104462816, "content_chunks": 2903, "content_chunks_s": null, "controlled_output_exact_adherence": 0, "controlled_output_exact_adherence_rate": null, "controlled_output_samples": 0, "controlled_output_unobservable": 0, "decode_tok_s_max": 137.71858131122082, "decode_tok_s_mean": 93.52693591040553, "decode_tok_s_mean_ci95_high": 96.0332188652663, "decode_tok_s_mean_ci95_low": 91.02065295554475, "decode_tok_s_min": 57.165697833529975, "decode_tok_s_p25": 82.90825972243191, "decode_tok_s_p50": 91.44000884451475, "decode_tok_s_p75": 101.5060296211474, "decode_tok_s_p90": 118.53854935590324, "decode_tok_s_p95": 124.93942090599512, "decode_tok_s_p99": 136.6168090090537, "decode_tok_s_samples": 160, "decode_tok_s_stddev": 16.174617547381683, "e2e_max_ms": 15151.096167042851, "e2e_mean_ci95_high_ms": 8172.02254681273, "e2e_mean_ci95_low_ms": 7596.389064344623, "e2e_mean_ms": 7884.2058055786765, "e2e_min_ms": 2868.5552490642294, "e2e_p25_ms": 6775.599709013477, "e2e_p50_ms": 7936.403442057781, "e2e_p75_ms": 8829.265380045399, "e2e_p90_ms": 9949.523736955598, "e2e_p95_ms": 10549.22072798945, "e2e_p99_ms": 12617.196012055501, "e2e_samples": 160, "e2e_stddev_ms": 1857.4621449528788, "effective_prefill_tok_s_max": 8582.8960279936, "effective_prefill_tok_s_mean": 1794.3944272820627, "effective_prefill_tok_s_mean_ci95_high": 1927.7421698957494, "effective_prefill_tok_s_mean_ci95_low": 1661.046684668376, "effective_prefill_tok_s_min": 1128.9518741982818, "effective_prefill_tok_s_p25": 1412.8160288305191, "effective_prefill_tok_s_p50": 1561.9850025604283, "effective_prefill_tok_s_p75": 1841.6028677577467, "effective_prefill_tok_s_p90": 2508.3261692123865, "effective_prefill_tok_s_p95": 2919.97505142375, "effective_prefill_tok_s_p99": 4827.233887759795, "effective_prefill_tok_s_samples": 160, "effective_prefill_tok_s_stddev": 860.5767091860098, "generation_max_ms": 10921.151685994118, "generation_mean_ci95_high_ms": 3312.8679472910853, "generation_mean_ci95_low_ms": 2896.52310930964, "generation_mean_ms": 3104.6955283003626, "generation_min_ms": 943.2481749681756, "generation_p25_ms": 2301.2896860018373, "generation_p50_ms": 2799.4603669503704, "generation_p75_ms": 3568.1120839435607, "generation_p90_ms": 4703.656404977664, "generation_p95_ms": 5259.857314056717, "generation_p99_ms": 7743.470582994632, "generation_samples": 160, "generation_stddev_ms": 1343.4673266072946, "mean_inter_token_latency_ms_max": 17.493007833334975, "mean_inter_token_latency_ms_mean": 11.007584297646849, "mean_inter_token_latency_ms_mean_ci95_high": 11.301120245425723, "mean_inter_token_latency_ms_mean_ci95_low": 10.714048349867975, "mean_inter_token_latency_ms_min": 7.261184296839135, "mean_inter_token_latency_ms_p25": 9.84546871601772, "mean_inter_token_latency_ms_p50": 10.933297412795207, "mean_inter_token_latency_ms_p75": 12.054472412146104, "mean_inter_token_latency_ms_p90": 13.302252520241835, "mean_inter_token_latency_ms_p95": 14.652445888429076, "mean_inter_token_latency_ms_p99": 16.405658014819135, "mean_inter_token_latency_ms_samples": 160, "mean_inter_token_latency_ms_stddev": 1.894371775749967, "mean_time_per_output_token_max_ms": 17.493007833334975, "mean_time_per_output_token_mean_ci95_high_ms": 11.301120245425723, "mean_time_per_output_token_mean_ci95_low_ms": 10.714048349867975, "mean_time_per_output_token_mean_ms": 11.007584297646849, "mean_time_per_output_token_min_ms": 7.261184296839135, "mean_time_per_output_token_p25_ms": 9.84546871601772, "mean_time_per_output_token_p50_ms": 10.933297412795207, "mean_time_per_output_token_p75_ms": 12.054472412146104, "mean_time_per_output_token_p90_ms": 13.302252520241835, "mean_time_per_output_token_p95_ms": 14.652445888429076, "mean_time_per_output_token_p99_ms": 16.405658014819135, "mean_time_per_output_token_samples": 160, "mean_time_per_output_token_stddev_ms": 1.894371775749967, "output_token_sources": { "usage": 160 }, "output_tokens": 44727, "prefix_cache_hit_avg": 0.0, "prompt_token_samples": 160, "prompt_tokens": 1236764, "prompt_tokens_max": 7798.0, "prompt_tokens_mean": 7729.775, "prompt_tokens_mean_ci95_high": 7735.29436497065, "prompt_tokens_mean_ci95_low": 7724.255635029349, "prompt_tokens_min": 7675.0, "prompt_tokens_p25": 7702.0, "prompt_tokens_p50": 7725.0, "prompt_tokens_p75": 7759.0, "prompt_tokens_p90": 7782.0, "prompt_tokens_p95": 7790.0, "prompt_tokens_p99": 7797.0, "prompt_tokens_samples": 160, "prompt_tokens_stddev": 35.61992764286613, "reasoning_chunks": 12331, "request_canary_passed": 0, "request_canary_samples": 0, "throughput_performance_eligible": true, "throughput_tok_s": 140.95998859482665, "time_to_first_output_max_ms": 6800.112719996832, "time_to_first_output_mean_ci95_high_ms": 4961.084573776022, "time_to_first_output_mean_ci95_low_ms": 4597.935980780606, "time_to_first_output_mean_ms": 4779.510277278314, "time_to_first_output_min_ms": 894.336826982908, "time_to_first_output_p25_ms": 4171.539091039449, "time_to_first_output_p50_ms": 4922.462269081734, "time_to_first_output_p75_ms": 5424.863247084431, "time_to_first_output_p90_ms": 6230.258041061461, "time_to_first_output_p95_ms": 6383.294756989926, "time_to_first_output_p99_ms": 6670.5980219412595, "time_to_first_output_samples": 160, "time_to_first_output_stddev_ms": 1171.8129417867153, "tpot_max_ms": 17.493007833334975, "tpot_mean_ci95_high_ms": 11.301120245425723, "tpot_mean_ci95_low_ms": 10.714048349867975, "tpot_mean_ms": 11.007584297646849, "tpot_min_ms": 7.261184296839135, "tpot_p25_ms": 9.84546871601772, "tpot_p50_ms": 10.933297412795207, "tpot_p75_ms": 12.054472412146104, "tpot_p90_ms": 13.302252520241835, "tpot_p95_ms": 14.652445888429076, "tpot_p99_ms": 16.405658014819135, "tpot_samples": 160, "tpot_stddev_ms": 1.894371775749967, "ttft_max_ms": 14171.816257061437, "ttft_mean_ci95_high_ms": 7470.431763892023, "ttft_mean_ci95_low_ms": 6923.812061978194, "ttft_mean_ms": 7197.121912935108, "ttft_min_ms": 2526.6853920184076, "ttft_p25_ms": 6194.8361420072615, "ttft_p50_ms": 7373.2240500394255, "ttft_p75_ms": 8092.205481021665, "ttft_p90_ms": 9228.856744011864, "ttft_p95_ms": 9633.107953006402, "ttft_p99_ms": 11491.849357960746, "ttft_samples": 160, "ttft_stddev_ms": 1763.8400734387742 }, "actual_start_unix_ns": 1792652395062567877, "actual_end_unix_ns": 1792652712365377885 }, { "name": "incumbent-after-c4-r3", "arm": "incumbent-after", "concurrency": 4, "repetition": "r3", "attempted": 160, "completed": 159, "failed": 1, "wall_ms": 321584.09613894764, "metrics": { "completion_tokens_max": 678.0, "completion_tokens_mean": 276.70440251572325, "completion_tokens_mean_ci95_high": 291.0332906974089, "completion_tokens_mean_ci95_low": 262.3755143340376, "completion_tokens_min": 88.0, "completion_tokens_p25": 215.0, "completion_tokens_p50": 260.0, "completion_tokens_p75": 321.0, "completion_tokens_p90": 404.0, "completion_tokens_p95": 442.0, "completion_tokens_p99": 543.0, "completion_tokens_samples": 159, "completion_tokens_stddev": 92.18388018143578, "content_chunks": 2965, "content_chunks_s": null, "controlled_output_exact_adherence": 0, "controlled_output_exact_adherence_rate": null, "controlled_output_samples": 0, "controlled_output_unobservable": 0, "decode_tok_s_max": 191.4557672251022, "decode_tok_s_mean": 92.30910485034713, "decode_tok_s_mean_ci95_high": 95.35907931860311, "decode_tok_s_mean_ci95_low": 89.25913038209114, "decode_tok_s_min": 60.47253670488209, "decode_tok_s_p25": 77.776957862738, "decode_tok_s_p50": 90.56547299097501, "decode_tok_s_p75": 104.1239505805, "decode_tok_s_p90": 117.28931381821862, "decode_tok_s_p95": 124.93541420964948, "decode_tok_s_p99": 162.43806132404657, "decode_tok_s_samples": 159, "decode_tok_s_stddev": 19.62179321753014, "e2e_max_ms": 15952.96392205637, "e2e_mean_ci95_high_ms": 8111.898490565179, "e2e_mean_ci95_low_ms": 7499.939178871599, "e2e_mean_ms": 7805.918834718389, "e2e_min_ms": 3232.8199370531365, "e2e_p25_ms": 6470.670843031257, "e2e_p50_ms": 7716.997058014385, "e2e_p75_ms": 9100.88771500159, "e2e_p90_ms": 10127.813244936988, "e2e_p95_ms": 10906.153489951976, "e2e_p99_ms": 14364.884545095265, "e2e_samples": 159, "e2e_stddev_ms": 1968.4982934397647, "effective_prefill_tok_s_max": 8608.629171868104, "effective_prefill_tok_s_mean": 1915.4720698134918, "effective_prefill_tok_s_mean_ci95_high": 2069.9015702621055, "effective_prefill_tok_s_mean_ci95_low": 1761.0425693648779, "effective_prefill_tok_s_min": 1127.2163347153669, "effective_prefill_tok_s_p25": 1307.3133465016092, "effective_prefill_tok_s_p50": 1582.7682964665728, "effective_prefill_tok_s_p75": 1992.0116509456016, "effective_prefill_tok_s_p90": 2758.6991640740525, "effective_prefill_tok_s_p95": 4448.31416491737, "effective_prefill_tok_s_p99": 4823.196536859402, "effective_prefill_tok_s_samples": 159, "effective_prefill_tok_s_stddev": 993.511177233521, "generation_max_ms": 10030.05252208095, "generation_mean_ci95_high_ms": 3365.8201269596825, "generation_mean_ci95_low_ms": 2935.3701777039355, "generation_mean_ms": 3150.595152331809, "generation_min_ms": 535.5887609766796, "generation_p25_ms": 2356.084101018496, "generation_p50_ms": 2761.532446020283, "generation_p75_ms": 3685.613527079113, "generation_p90_ms": 4951.200520037673, "generation_p95_ms": 5876.204559928738, "generation_p99_ms": 7961.859040078707, "generation_samples": 159, "generation_stddev_ms": 1384.634524436244, "mean_inter_token_latency_ms_max": 16.53643214737621, "mean_inter_token_latency_ms_mean": 11.282483521654148, "mean_inter_token_latency_ms_mean_ci95_high": 11.628227797854748, "mean_inter_token_latency_ms_mean_ci95_low": 10.936739245453548, "mean_inter_token_latency_ms_min": 5.223138558287775, "mean_inter_token_latency_ms_p25": 9.603938329509338, "mean_inter_token_latency_ms_p50": 11.041735519889038, "mean_inter_token_latency_ms_p75": 12.857278395547635, "mean_inter_token_latency_ms_p90": 14.553431004443599, "mean_inter_token_latency_ms_p95": 14.977349924679467, "mean_inter_token_latency_ms_p99": 16.276736671770973, "mean_inter_token_latency_ms_samples": 159, "mean_inter_token_latency_ms_stddev": 2.2243211424756812, "mean_time_per_output_token_max_ms": 16.53643214737621, "mean_time_per_output_token_mean_ci95_high_ms": 11.628227797854748, "mean_time_per_output_token_mean_ci95_low_ms": 10.936739245453548, "mean_time_per_output_token_mean_ms": 11.282483521654148, "mean_time_per_output_token_min_ms": 5.223138558287775, "mean_time_per_output_token_p25_ms": 9.603938329509338, "mean_time_per_output_token_p50_ms": 11.041735519889038, "mean_time_per_output_token_p75_ms": 12.857278395547635, "mean_time_per_output_token_p90_ms": 14.553431004443599, "mean_time_per_output_token_p95_ms": 14.977349924679467, "mean_time_per_output_token_p99_ms": 16.276736671770973, "mean_time_per_output_token_samples": 159, "mean_time_per_output_token_stddev_ms": 2.2243211424756812, "output_token_sources": { "usage": 159 }, "output_tokens": 43996, "prefix_cache_hit_avg": 0.0, "prompt_token_samples": 159, "prompt_tokens": 1229040, "prompt_tokens_max": 7798.0, "prompt_tokens_mean": 7729.811320754717, "prompt_tokens_mean_ci95_high": 7735.365046551788, "prompt_tokens_mean_ci95_low": 7724.257594957646, "prompt_tokens_min": 7675.0, "prompt_tokens_p25": 7702.0, "prompt_tokens_p50": 7727.0, "prompt_tokens_p75": 7760.0, "prompt_tokens_p90": 7783.0, "prompt_tokens_p95": 7791.0, "prompt_tokens_p99": 7797.0, "prompt_tokens_samples": 159, "prompt_tokens_stddev": 35.729498824070426, "reasoning_chunks": 12208, "request_canary_passed": 0, "request_canary_samples": 0, "throughput_performance_eligible": false, "throughput_tok_s": null, "time_to_first_output_max_ms": 6865.788924973458, "time_to_first_output_mean_ci95_high_ms": 4871.575693967738, "time_to_first_output_mean_ci95_low_ms": 4439.071670805422, "time_to_first_output_mean_ms": 4655.32368238658, "time_to_first_output_min_ms": 891.5472889784724, "time_to_first_output_p25_ms": 3881.0013969196007, "time_to_first_output_p50_ms": 4866.150947986171, "time_to_first_output_p75_ms": 5919.230339932255, "time_to_first_output_p90_ms": 6371.518168947659, "time_to_first_output_p95_ms": 6450.200736057013, "time_to_first_output_p99_ms": 6837.1968739666045, "time_to_first_output_samples": 159, "time_to_first_output_stddev_ms": 1391.2418934269854, "tpot_max_ms": 16.53643214737621, "tpot_mean_ci95_high_ms": 11.628227797854748, "tpot_mean_ci95_low_ms": 10.936739245453548, "tpot_mean_ms": 11.282483521654148, "tpot_min_ms": 5.223138558287775, "tpot_p25_ms": 9.603938329509338, "tpot_p50_ms": 11.041735519889038, "tpot_p75_ms": 12.857278395547635, "tpot_p90_ms": 14.553431004443599, "tpot_p95_ms": 14.977349924679467, "tpot_p99_ms": 16.276736671770973, "tpot_samples": 159, "tpot_stddev_ms": 2.2243211424756812, "ttft_max_ms": 14870.62311603222, "ttft_mean_ci95_high_ms": 7385.66037010461, "ttft_mean_ci95_low_ms": 6796.978303726179, "ttft_mean_ms": 7091.319336915394, "ttft_min_ms": 2592.2248440328985, "ttft_p25_ms": 5816.197164007463, "ttft_p50_ms": 7069.392979028635, "ttft_p75_ms": 8253.18918703124, "ttft_p90_ms": 9339.586162008345, "ttft_p95_ms": 10142.57255999837, "ttft_p99_ms": 13720.84102302324, "ttft_samples": 159, "ttft_stddev_ms": 1893.6220446381258 }, "actual_start_unix_ns": 1792652712770036625, "actual_end_unix_ns": 1792653034354136555 } ], "headline_results": [ { "arm": "incumbent-before", "concurrency": 1, "repetitions": 3, "attempted": 156, "failed": 0, "mean_block_throughput_tok_s": 118.8806000829497 }, { "arm": "incumbent-before", "concurrency": 4, "repetitions": 3, "attempted": 480, "failed": 1, "mean_block_throughput_tok_s": null }, { "arm": "flare-control", "concurrency": 1, "repetitions": 3, "attempted": 156, "failed": 0, "mean_block_throughput_tok_s": 114.4083339085134 }, { "arm": "flare-control", "concurrency": 4, "repetitions": 3, "attempted": 480, "failed": 0, "mean_block_throughput_tok_s": 146.65952268071922 }, { "arm": "flare-instrumented", "concurrency": 1, "repetitions": 3, "attempted": 156, "failed": 0, "mean_block_throughput_tok_s": 114.0368880522962 }, { "arm": "flare-instrumented", "concurrency": 4, "repetitions": 3, "attempted": 480, "failed": 0, "mean_block_throughput_tok_s": 145.91029141026726 }, { "arm": "incumbent-after", "concurrency": 1, "repetitions": 3, "attempted": 156, "failed": 0, "mean_block_throughput_tok_s": 117.22284584202565 }, { "arm": "incumbent-after", "concurrency": 4, "repetitions": 3, "attempted": 480, "failed": 1, "mean_block_throughput_tok_s": null } ], "identity": { "hardware": "2x RTX PRO 6000 Blackwell Max-Q 96GB, PCIe TP2, native Linux", "driver": "615.71.09", "power_limit_watts_per_gpu": 300, "flare_source": "47581ccac82660953a5b59cb04b2750d64edc97d", "flare_checkpoint": "local-inference-lab/GLM-5.3-Flash-NVFP4-MXFP8-CSF-QAD@dec48abd33efa73c3bb7c95b74eee10cad34f9be", "flare_control_image": "sha256:a540b65396a83b8b5148bbb51e7ee62cb16498ecee679433017673053b1b9854", "flare_instrumented_image": "sha256:47ce7fde879d731aca62cd2c9004f04e2e689498d04fae09261fcb654db8fa51", "flare_package": "d5aec1543dc32413175462d17d374e54d27def24129f878045358c4db128c357", "incumbent_checkpoint": "nvidia/GLM-5.3-Flash-FP4-Local@76348ddb2be2e35e39ecfc00e75dc07703f4f3d5", "incumbent_image": "sha256:745feb916ee816f5409fd58386114be4f22fbede7f9c0e0a7b8c4fe8dac0fb29" }, "outage_minutes": 72.34172201156616, "metrics_coverage": { "profile": "observational-inference-diagnostics/v1", "status": "partial", "missing_required_metrics": [ "repeated trace-off/on", "new Nsight node/hardware capture", "causal cache probe", "strict semantic/output quality", "no-speculation control" ] }, "comparison_evidence": null, "limitations": [ "Natural variable output; no request-canary or semantic oracle", "Different checkpoints/quantization/stacks versus incumbent", "Failed concurrency arms and overall comparison are ineligible", "Capture-OFF executable-image observations are not tracing overhead", "Three unrandomized repetitions; workload seeds are not API seeds", "Common ledgers consolidated during execution; not a formal pre-frozen comparison", "Raw host samplers and source snapshots remain private; this bundle is sanitized selected evidence" ], "decision": { "evidence_labels": [ "functional", "capacity", "compatibility-only" ], "decision_labels": [ "no-promotion" ], "promotion_authorized": false, "upstream_submission": false, "next_gate": "forward-prime/reverse-replay cache probe; repeated trace-off/on" } }