{ "aggregate": { "all_treatment_deadlines_met": true, "boundary_quality": { "all_pairs_pass": true, "comparisons": 20, "minimum_cosine": 1.0, "top1_matches": 20 }, "claims_valid": true, "deadline_outcomes": [ { "all_deadlines_met": true, "deadline_misses": [], "pair": 0, "transitions": [ { "boundary_call_end_ns": 7763558369038749, "boundary_call_start_ns": 7763558302147074, "compile_duration_s": 6.780602458, "compile_ready_ns": 7763550677725071, "compile_start_ns": 7763543897122613, "consume_in_boundary_call": true, "native_deadline_miss": false, "native_ready_before_deadline": true, "new_capacity": 2048, "old_capacity": 1024, "readiness_slack_s": 7.624422003, "ready_before_boundary_call": true, "variant_consume_ns": 7763558328421512, "variant_device_check": { "capacity": 2048, "clock_domain": "linux_clock_monotonic", "compiled_device_class": "NPU", "cpu_wa": 0, "cpu_wa_matches": 1, "deadline_stored_tokens": 1023, "event": "variant_device_check", "from_capacity": 1024, "from_variant": 0, "mode": "ONE_AHEAD", "monotonic_ns": 7763558302329905, "placement_ok": 1, "stored_tokens": 1023, "variant_index": 1, "wall_time_ns": 1787347125294752540 }, "wait_us": 0.0 }, { "boundary_call_end_ns": 7763575040457148, "boundary_call_start_ns": 7763574968713120, "compile_duration_s": 6.935960068, "compile_ready_ns": 7763565274999929, "compile_start_ns": 7763558339039861, "consume_in_boundary_call": true, "native_deadline_miss": false, "native_ready_before_deadline": true, "new_capacity": 4096, "old_capacity": 2048, "readiness_slack_s": 9.693713191, "ready_before_boundary_call": true, "variant_consume_ns": 7763575003261078, "variant_device_check": { "capacity": 4096, "clock_domain": "linux_clock_monotonic", "compiled_device_class": "NPU", "cpu_wa": 0, "cpu_wa_matches": 1, "deadline_stored_tokens": 2047, "event": "variant_device_check", "from_capacity": 2048, "from_variant": 1, "mode": "ONE_AHEAD", "monotonic_ns": 7763574968858758, "placement_ok": 1, "stored_tokens": 2047, "variant_index": 2, "wall_time_ns": 1787347141961281388 }, "wait_us": 0.0 }, { "boundary_call_end_ns": 7763609574310886, "boundary_call_start_ns": 7763609487189870, "compile_duration_s": 7.113776519, "compile_ready_ns": 7763582125041528, "compile_start_ns": 7763575011265009, "consume_in_boundary_call": true, "native_deadline_miss": false, "native_ready_before_deadline": true, "new_capacity": 8192, "old_capacity": 4096, "readiness_slack_s": 27.362148342, "ready_before_boundary_call": true, "variant_consume_ns": 7763609540693361, "variant_device_check": { "capacity": 8192, "clock_domain": "linux_clock_monotonic", "compiled_device_class": "NPU", "cpu_wa": 0, "cpu_wa_matches": 1, "deadline_stored_tokens": 4095, "event": "variant_device_check", "from_capacity": 4096, "from_variant": 2, "mode": "ONE_AHEAD", "monotonic_ns": 7763609487364623, "placement_ok": 1, "stored_tokens": 4095, "variant_index": 3, "wall_time_ns": 1787347176479787251 }, "wait_us": 0.0 } ], "waited_capacities": [] }, { "all_deadlines_met": true, "deadline_misses": [], "pair": 1, "transitions": [ { "boundary_call_end_ns": 7769388185084007, "boundary_call_start_ns": 7769388124086638, "compile_duration_s": 6.805100512, "compile_ready_ns": 7769380618649151, "compile_start_ns": 7769373813548639, "consume_in_boundary_call": true, "native_deadline_miss": false, "native_ready_before_deadline": true, "new_capacity": 2048, "old_capacity": 1024, "readiness_slack_s": 7.505437487, "ready_before_boundary_call": true, "variant_consume_ns": 7769388145738666, "variant_device_check": { "capacity": 2048, "clock_domain": "linux_clock_monotonic", "compiled_device_class": "NPU", "cpu_wa": 0, "cpu_wa_matches": 1, "deadline_stored_tokens": 1023, "event": "variant_device_check", "from_capacity": 1024, "from_variant": 0, "mode": "ONE_AHEAD", "monotonic_ns": 7769388124247749, "placement_ok": 1, "stored_tokens": 1023, "variant_index": 1, "wall_time_ns": 1787352955116670377 }, "wait_us": 0.0 }, { "boundary_call_end_ns": 7769404897186935, "boundary_call_start_ns": 7769404829235732, "compile_duration_s": 7.011376585, "compile_ready_ns": 7769395164538651, "compile_start_ns": 7769388153162066, "consume_in_boundary_call": true, "native_deadline_miss": false, "native_ready_before_deadline": true, "new_capacity": 4096, "old_capacity": 2048, "readiness_slack_s": 9.664697081, "ready_before_boundary_call": true, "variant_consume_ns": 7769404860921746, "variant_device_check": { "capacity": 4096, "clock_domain": "linux_clock_monotonic", "compiled_device_class": "NPU", "cpu_wa": 0, "cpu_wa_matches": 1, "deadline_stored_tokens": 2047, "event": "variant_device_check", "from_capacity": 2048, "from_variant": 1, "mode": "ONE_AHEAD", "monotonic_ns": 7769404829369406, "placement_ok": 1, "stored_tokens": 2047, "variant_index": 2, "wall_time_ns": 1787352971821792023 }, "wait_us": 0.0 }, { "boundary_call_end_ns": 7769439095275169, "boundary_call_start_ns": 7769439010197246, "compile_duration_s": 7.014014054, "compile_ready_ns": 7769411880096132, "compile_start_ns": 7769404866082078, "consume_in_boundary_call": true, "native_deadline_miss": false, "native_ready_before_deadline": true, "new_capacity": 8192, "old_capacity": 4096, "readiness_slack_s": 27.130101114, "ready_before_boundary_call": true, "variant_consume_ns": 7769439063766458, "variant_device_check": { "capacity": 8192, "clock_domain": "linux_clock_monotonic", "compiled_device_class": "NPU", "cpu_wa": 0, "cpu_wa_matches": 1, "deadline_stored_tokens": 4095, "event": "variant_device_check", "from_capacity": 4096, "from_variant": 2, "mode": "ONE_AHEAD", "monotonic_ns": 7769439010343904, "placement_ok": 1, "stored_tokens": 4095, "variant_index": 3, "wall_time_ns": 1787353006002766537 }, "wait_us": 0.0 } ], "waited_capacities": [] }, { "all_deadlines_met": true, "deadline_misses": [], "pair": 2, "transitions": [ { "boundary_call_end_ns": 7770855248234721, "boundary_call_start_ns": 7770855190209813, "compile_duration_s": 6.751958435, "compile_ready_ns": 7770847697445140, "compile_start_ns": 7770840945486705, "consume_in_boundary_call": true, "native_deadline_miss": false, "native_ready_before_deadline": true, "new_capacity": 2048, "old_capacity": 1024, "readiness_slack_s": 7.492764673, "ready_before_boundary_call": true, "variant_consume_ns": 7770855211727787, "variant_device_check": { "capacity": 2048, "clock_domain": "linux_clock_monotonic", "compiled_device_class": "NPU", "cpu_wa": 0, "cpu_wa_matches": 1, "deadline_stored_tokens": 1023, "event": "variant_device_check", "from_capacity": 1024, "from_variant": 0, "mode": "ONE_AHEAD", "monotonic_ns": 7770855190346643, "placement_ok": 1, "stored_tokens": 1023, "variant_index": 1, "wall_time_ns": 1787354422182769291 }, "wait_us": 0.0 }, { "boundary_call_end_ns": 7770872120679842, "boundary_call_start_ns": 7770872049724053, "compile_duration_s": 6.805981231, "compile_ready_ns": 7770862025282245, "compile_start_ns": 7770855219301014, "consume_in_boundary_call": true, "native_deadline_miss": false, "native_ready_before_deadline": true, "new_capacity": 4096, "old_capacity": 2048, "readiness_slack_s": 10.024441808, "ready_before_boundary_call": true, "variant_consume_ns": 7770872083487689, "variant_device_check": { "capacity": 4096, "clock_domain": "linux_clock_monotonic", "compiled_device_class": "NPU", "cpu_wa": 0, "cpu_wa_matches": 1, "deadline_stored_tokens": 2047, "event": "variant_device_check", "from_capacity": 2048, "from_variant": 1, "mode": "ONE_AHEAD", "monotonic_ns": 7770872049887580, "placement_ok": 1, "stored_tokens": 2047, "variant_index": 2, "wall_time_ns": 1787354439042310209 }, "wait_us": 0.0 }, { "boundary_call_end_ns": 7770906292795129, "boundary_call_start_ns": 7770906206714011, "compile_duration_s": 6.946282344, "compile_ready_ns": 7770879037800094, "compile_start_ns": 7770872091517750, "consume_in_boundary_call": true, "native_deadline_miss": false, "native_ready_before_deadline": true, "new_capacity": 8192, "old_capacity": 4096, "readiness_slack_s": 27.168913917, "ready_before_boundary_call": true, "variant_consume_ns": 7770906260093318, "variant_device_check": { "capacity": 8192, "clock_domain": "linux_clock_monotonic", "compiled_device_class": "NPU", "cpu_wa": 0, "cpu_wa_matches": 1, "deadline_stored_tokens": 4095, "event": "variant_device_check", "from_capacity": 4096, "from_variant": 2, "mode": "ONE_AHEAD", "monotonic_ns": 7770906206864122, "placement_ok": 1, "stored_tokens": 4095, "variant_index": 3, "wall_time_ns": 1787354473199286757 }, "wait_us": 0.0 } ], "waited_capacities": [] }, { "all_deadlines_met": true, "deadline_misses": [], "pair": 3, "transitions": [ { "boundary_call_end_ns": 7772915488825791, "boundary_call_start_ns": 7772915437030606, "compile_duration_s": 6.776207656, "compile_ready_ns": 7772907756901412, "compile_start_ns": 7772900980693756, "consume_in_boundary_call": true, "native_deadline_miss": false, "native_ready_before_deadline": true, "new_capacity": 2048, "old_capacity": 1024, "readiness_slack_s": 7.680129194, "ready_before_boundary_call": true, "variant_consume_ns": 7772915457368301, "variant_device_check": { "capacity": 2048, "clock_domain": "linux_clock_monotonic", "compiled_device_class": "NPU", "cpu_wa": 0, "cpu_wa_matches": 1, "deadline_stored_tokens": 1023, "event": "variant_device_check", "from_capacity": 1024, "from_variant": 0, "mode": "ONE_AHEAD", "monotonic_ns": 7772915437151270, "placement_ok": 1, "stored_tokens": 1023, "variant_index": 1, "wall_time_ns": 1787356482429573888 }, "wait_us": 0.0 }, { "boundary_call_end_ns": 7772932440670060, "boundary_call_start_ns": 7772932368267546, "compile_duration_s": 6.813044537, "compile_ready_ns": 7772922275145450, "compile_start_ns": 7772915462100913, "consume_in_boundary_call": true, "native_deadline_miss": false, "native_ready_before_deadline": true, "new_capacity": 4096, "old_capacity": 2048, "readiness_slack_s": 10.093122096, "ready_before_boundary_call": true, "variant_consume_ns": 7772932401791229, "variant_device_check": { "capacity": 4096, "clock_domain": "linux_clock_monotonic", "compiled_device_class": "NPU", "cpu_wa": 0, "cpu_wa_matches": 1, "deadline_stored_tokens": 2047, "event": "variant_device_check", "from_capacity": 2048, "from_variant": 1, "mode": "ONE_AHEAD", "monotonic_ns": 7772932368435760, "placement_ok": 1, "stored_tokens": 2047, "variant_index": 2, "wall_time_ns": 1787356499360858390 }, "wait_us": 0.0 }, { "boundary_call_end_ns": 7772966983685195, "boundary_call_start_ns": 7772966897862692, "compile_duration_s": 7.042110922, "compile_ready_ns": 7772939451674690, "compile_start_ns": 7772932409563768, "consume_in_boundary_call": true, "native_deadline_miss": false, "native_ready_before_deadline": true, "new_capacity": 8192, "old_capacity": 4096, "readiness_slack_s": 27.446188002, "ready_before_boundary_call": true, "variant_consume_ns": 7772966950956039, "variant_device_check": { "capacity": 8192, "clock_domain": "linux_clock_monotonic", "compiled_device_class": "NPU", "cpu_wa": 0, "cpu_wa_matches": 1, "deadline_stored_tokens": 4095, "event": "variant_device_check", "from_capacity": 4096, "from_variant": 2, "mode": "ONE_AHEAD", "monotonic_ns": 7772966897999220, "placement_ok": 1, "stored_tokens": 4095, "variant_index": 3, "wall_time_ns": 1787356533890421874 }, "wait_us": 0.0 } ], "waited_capacities": [] }, { "all_deadlines_met": true, "deadline_misses": [], "pair": 4, "transitions": [ { "boundary_call_end_ns": 7774382060912707, "boundary_call_start_ns": 7774382004339958, "compile_duration_s": 6.704558312, "compile_ready_ns": 7774374480336701, "compile_start_ns": 7774367775778389, "consume_in_boundary_call": true, "native_deadline_miss": false, "native_ready_before_deadline": true, "new_capacity": 2048, "old_capacity": 1024, "readiness_slack_s": 7.524003257, "ready_before_boundary_call": true, "variant_consume_ns": 7774382025716870, "variant_device_check": { "capacity": 2048, "clock_domain": "linux_clock_monotonic", "compiled_device_class": "NPU", "cpu_wa": 0, "cpu_wa_matches": 1, "deadline_stored_tokens": 1023, "event": "variant_device_check", "from_capacity": 1024, "from_variant": 0, "mode": "ONE_AHEAD", "monotonic_ns": 7774382004477194, "placement_ok": 1, "stored_tokens": 1023, "variant_index": 1, "wall_time_ns": 1787357948996899829 }, "wait_us": 0.0 }, { "boundary_call_end_ns": 7774398857792654, "boundary_call_start_ns": 7774398787385059, "compile_duration_s": 6.76885075, "compile_ready_ns": 7774388802524284, "compile_start_ns": 7774382033673534, "consume_in_boundary_call": true, "native_deadline_miss": false, "native_ready_before_deadline": true, "new_capacity": 4096, "old_capacity": 2048, "readiness_slack_s": 9.984860775, "ready_before_boundary_call": true, "variant_consume_ns": 7774398820657865, "variant_device_check": { "capacity": 4096, "clock_domain": "linux_clock_monotonic", "compiled_device_class": "NPU", "cpu_wa": 0, "cpu_wa_matches": 1, "deadline_stored_tokens": 2047, "event": "variant_device_check", "from_capacity": 2048, "from_variant": 1, "mode": "ONE_AHEAD", "monotonic_ns": 7774398787515708, "placement_ok": 1, "stored_tokens": 2047, "variant_index": 2, "wall_time_ns": 1787357965779938350 }, "wait_us": 0.0 }, { "boundary_call_end_ns": 7774433482373224, "boundary_call_start_ns": 7774433395445897, "compile_duration_s": 6.995096971, "compile_ready_ns": 7774405823475301, "compile_start_ns": 7774398828378330, "consume_in_boundary_call": true, "native_deadline_miss": false, "native_ready_before_deadline": true, "new_capacity": 8192, "old_capacity": 4096, "readiness_slack_s": 27.571970596, "ready_before_boundary_call": true, "variant_consume_ns": 7774433448693403, "variant_device_check": { "capacity": 8192, "clock_domain": "linux_clock_monotonic", "compiled_device_class": "NPU", "cpu_wa": 0, "cpu_wa_matches": 1, "deadline_stored_tokens": 4095, "event": "variant_device_check", "from_capacity": 4096, "from_variant": 2, "mode": "ONE_AHEAD", "monotonic_ns": 7774433395586034, "placement_ok": 1, "stored_tokens": 4095, "variant_index": 3, "wall_time_ns": 1787358000388008663 }, "wait_us": 0.0 } ], "waited_capacities": [] } ], "errors": [], "observed_orders": [ [ "precompiled", "one_ahead_jit" ], [ "one_ahead_jit", "precompiled" ], [ "one_ahead_jit", "precompiled" ], [ "precompiled", "one_ahead_jit" ], [ "precompiled", "one_ahead_jit" ] ], "order_sensitivity": { "AB": { "mean_treatment_minus_control_s": -19.47565567833333, "pairs": 3 }, "BA": { "mean_treatment_minus_control_s": -19.429441559999987, "pairs": 2 } }, "pair_reports": [ "/data/runs/2026-08-21_21-00_wuklab-NUC16_qwen05b-fp16-jit-full-1k8k-5pairs/jit_pairs/analysis/p00/comparison.json", "/data/runs/2026-08-21_21-00_wuklab-NUC16_qwen05b-fp16-jit-full-1k8k-5pairs/jit_pairs/analysis/p01/comparison.json", "/data/runs/2026-08-21_21-00_wuklab-NUC16_qwen05b-fp16-jit-full-1k8k-5pairs/jit_pairs/analysis/p02/comparison.json", "/data/runs/2026-08-21_21-00_wuklab-NUC16_qwen05b-fp16-jit-full-1k8k-5pairs/jit_pairs/analysis/p03/comparison.json", "/data/runs/2026-08-21_21-00_wuklab-NUC16_qwen05b-fp16-jit-full-1k8k-5pairs/jit_pairs/analysis/p04/comparison.json" ], "paired_compile_by_capacity": { "1024": { "control": [ 6.61466, 6.6185, 6.503189, 6.630153, 6.539537 ], "exact_two_sided_sign_flip_p": 0.5, "geometric_mean_control_over_treatment_speedup": 0.9969164116582635, "geometric_mean_favorable_speedup": 0.9969164116582635, "geometric_mean_treatment_over_control": 1.0030931262698417, "higher_is_better": false, "inference_note": "With five pairs the minimum attainable exact two-sided sign-flip p-value is 0.0625; emphasize effect size, interval, and replication rather than a 0.05 threshold.", "mean_difference": 0.020385400000000154, "mean_percent_change": 0.3125973016070608, "median_difference": 0.015746000000000038, "median_percent_change": 0.23804700468353684, "objective": "lower", "paired_t_95pct_ci_for_mean_difference": [ -0.053500505744738705, 0.094271305744739 ], "pairs": 5, "percent_changes": [ 0.23804700468353684, 0.9191961924907495, 1.4949434808061084, -0.7132867069583412, -0.37591346298674966 ], "sample_sd_of_differences": 0.05950555758330466, "treatment": [ 6.630406, 6.679337, 6.600408, 6.582861, 6.514954 ], "treatment_minus_control": [ 0.015746000000000038, 0.06083700000000025, 0.09721899999999994, -0.04729199999999967, -0.0245829999999998 ] }, "2048": { "control": [ 6.672935, 6.717834, 6.641378, 6.758008, 6.664255 ], "exact_two_sided_sign_flip_p": 0.0625, "geometric_mean_control_over_treatment_speedup": 0.98954026593545, "geometric_mean_favorable_speedup": 0.98954026593545, "geometric_mean_treatment_over_control": 1.0105702965555041, "higher_is_better": false, "inference_note": "With five pairs the minimum attainable exact two-sided sign-flip p-value is 0.0625; emphasize effect size, interval, and replication rather than a 0.05 threshold.", "mean_difference": 0.07067900000000016, "mean_percent_change": 1.0585582789764032, "median_difference": 0.08496300000000012, "median_percent_change": 1.2647380093047866, "objective": "lower", "paired_t_95pct_ci_for_mean_difference": [ 0.019383227090895856, 0.12197477290910447 ], "pairs": 5, "percent_changes": [ 1.5789903543193502, 1.2647380093047866, 1.6325828766259138, 0.2439180302834824, 0.5725621243484831 ], "sample_sd_of_differences": 0.04131212222217598, "treatment": [ 6.7783, 6.802797, 6.749804, 6.774492, 6.702412 ], "treatment_minus_control": [ 0.10536499999999993, 0.08496300000000012, 0.10842600000000058, 0.016484000000000165, 0.038156999999999996 ] }, "4096": { "control": [ 6.541928, 6.595711, 6.530646, 6.624791, 6.545835 ], "exact_two_sided_sign_flip_p": 0.0625, "geometric_mean_control_over_treatment_speedup": 0.9568527584741136, "geometric_mean_favorable_speedup": 0.9568527584741136, "geometric_mean_treatment_over_control": 1.045092874680837, "higher_is_better": false, "inference_note": "With five pairs the minimum attainable exact two-sided sign-flip p-value is 0.0625; emphasize effect size, interval, and replication rather than a 0.05 threshold.", "mean_difference": 0.296663, "mean_percent_change": 4.518445318641944, "median_difference": 0.2728349999999997, "median_percent_change": 4.177764343680544, "objective": "lower", "paired_t_95pct_ci_for_mean_difference": [ 0.17054089519715465, 0.42278510480284537 ], "pairs": 5, "percent_changes": [ 5.981936823517465, 6.263994890012625, 4.177764343680544, 2.797975060647199, 3.370555475351884 ], "sample_sd_of_differences": 0.10157507164900259, "treatment": [ 6.933262, 7.008866, 6.803481, 6.810151, 6.766466 ], "treatment_minus_control": [ 0.3913339999999996, 0.4131550000000006, 0.2728349999999997, 0.1853600000000002, 0.22063100000000002 ] }, "8192": { "control": [ 6.614971, 6.56659, 6.589067, 6.629587, 6.586202 ], "exact_two_sided_sign_flip_p": 0.0625, "geometric_mean_control_over_treatment_speedup": 0.9398980624936215, "geometric_mean_favorable_speedup": 0.9398980624936215, "geometric_mean_treatment_over_control": 1.0639451658692896, "higher_is_better": false, "inference_note": "With five pairs the minimum attainable exact two-sided sign-flip p-value is 0.0625; emphasize effect size, interval, and replication rather than a 0.05 threshold.", "mean_difference": 0.4220418, "mean_percent_change": 6.396842730802635, "median_difference": 0.40999300000000005, "median_percent_change": 6.184291721339505, "objective": "lower", "paired_t_95pct_ci_for_mean_difference": [ 0.3572176506100426, 0.4868659493899574 ], "pairs": 5, "percent_changes": [ 7.488891485692079, 6.7681551612024, 5.372323577829758, 6.184291721339505, 6.170551707949429 ], "sample_sd_of_differences": 0.05220748281329047, "treatment": [ 7.110359, 7.011027, 6.943053, 7.03958, 6.992607 ], "treatment_minus_control": [ 0.49538800000000016, 0.44443700000000064, 0.3539859999999999, 0.40999300000000005, 0.40640499999999946 ] } }, "paired_endpoints": { "compile_model_wall_s": { "control": [ 84.069572736, 84.331265631, 83.708544242, 85.220518564, 84.096977893 ], "exact_two_sided_sign_flip_p": 0.0625, "geometric_mean_control_over_treatment_speedup": 1.2901745970368312, "geometric_mean_favorable_speedup": 1.2901745970368312, "geometric_mean_treatment_over_control": 0.7750888928496339, "higher_is_better": false, "inference_note": "With five pairs the minimum attainable exact two-sided sign-flip p-value is 0.0625; emphasize effect size, interval, and replication rather than a 0.05 threshold.", "mean_difference": -18.9567182048, "mean_percent_change": -22.487516285221286, "median_difference": -18.568237542999995, "median_percent_change": -22.086751411606457, "objective": "lower", "paired_t_95pct_ci_for_mean_difference": [ -19.963654738214576, -17.949781671385427 ], "pairs": 5, "percent_changes": [ -22.086751411606457, -21.932894090469812, -21.710838710155766, -23.67355863934215, -23.033538574532244 ], "sample_sd_of_differences": 0.8109573709340462, "treatment": [ 65.501335193, 65.834978455, 65.534717215, 65.045789129, 64.72646805 ], "treatment_minus_control": [ -18.568237542999995, -18.496287175999996, -18.173827027, -20.174729435000003, -19.370509843000008 ] }, "decode_infer_s": { "control": [ 143.221490682, 143.735924396, 143.152000032, 143.259762506, 143.207379556 ], "exact_two_sided_sign_flip_p": 0.25, "geometric_mean_control_over_treatment_speedup": 1.00265957923962, "geometric_mean_favorable_speedup": 1.00265957923962, "geometric_mean_treatment_over_control": 0.9973474753598455, "higher_is_better": false, "inference_note": "With five pairs the minimum attainable exact two-sided sign-flip p-value is 0.0625; emphasize effect size, interval, and replication rather than a 0.05 threshold.", "mean_difference": -0.3797905079999907, "mean_percent_change": -0.26469001649047996, "median_difference": -0.5051298889999885, "median_percent_change": -0.35286261378609624, "objective": "lower", "paired_t_95pct_ci_for_mean_difference": [ -1.0471918166930716, 0.2876108006930901 ], "pairs": 5, "percent_changes": [ -0.4752597146969458, -0.7301554315040866, -0.35286261378609624, 0.10784526324588041, 0.12698241428884843 ], "sample_sd_of_differences": 0.5375055852033986, "treatment": [ 142.540816634, 142.686428737, 142.646870143, 143.414261374, 143.389227744 ], "treatment_minus_control": [ -0.680674047999986, -1.0494956590000015, -0.5051298889999885, 0.15449886800001877, 0.18184818800000357 ] }, "decode_wall_s": { "control": [ 147.357470573, 147.697927908, 147.449665453, 147.753727422, 147.363224232 ], "exact_two_sided_sign_flip_p": 0.1875, "geometric_mean_control_over_treatment_speedup": 1.0030127302708374, "geometric_mean_favorable_speedup": 1.0030127302708374, "geometric_mean_treatment_over_control": 0.9969963190098057, "higher_is_better": false, "inference_note": "With five pairs the minimum attainable exact two-sided sign-flip p-value is 0.0625; emphasize effect size, interval, and replication rather than a 0.05 threshold.", "mean_difference": -0.44230946639999613, "mean_percent_change": -0.2997044120459554, "median_difference": -0.585422082000008, "median_percent_change": -0.3972802191321501, "objective": "lower", "paired_t_95pct_ci_for_mean_difference": [ -1.1877208723650616, 0.3031019395650695 ], "pairs": 5, "percent_changes": [ -0.3972802191321501, -0.8087331433275207, -0.48937388008520805, -0.04350439486133292, 0.2403695771764349 ], "sample_sd_of_differences": 0.600332646582799, "treatment": [ 146.772048491, 146.503445813, 146.728085304, 147.689448057, 147.717440591 ], "treatment_minus_control": [ -0.585422082000008, -1.1944820949999837, -0.7215801490000047, -0.06427936500000442, 0.35421635900002 ] }, "protocol_critical_path_s": { "control": [ 231.993970971, 232.85763407599998, 231.730203078, 233.54898441, 232.031900832 ], "exact_two_sided_sign_flip_p": 0.0625, "geometric_mean_control_over_treatment_speedup": 1.0913545650227277, "geometric_mean_favorable_speedup": 1.0913545650227277, "geometric_mean_treatment_over_control": 0.9162924974608732, "higher_is_better": false, "inference_note": "With five pairs the minimum attainable exact two-sided sign-flip p-value is 0.0625; emphasize effect size, interval, and replication rather than a 0.05 threshold.", "mean_difference": -19.457170030999993, "mean_percent_change": -8.370512476646716, "median_difference": -19.157789511000004, "median_percent_change": -8.257882491866477, "objective": "lower", "paired_t_95pct_ci_for_mean_difference": [ -20.208198667730215, -18.70614139426977 ], "pairs": 5, "percent_changes": [ -8.257882491866477, -8.570925045766835, -8.156394630024993, -8.669330914090274, -8.198029301485002 ], "sample_sd_of_differences": 0.6048566007170219, "treatment": [ 212.83618146, 212.899580796, 212.829373238, 213.301850105, 213.009857613 ], "treatment_minus_control": [ -19.157789511000004, -19.958053279999973, -18.90082984, -20.247134305000003, -19.022043218999983 ] }, "tokens_per_s_infer": { "control": [ 56.304399302090765, 56.10288474427073, 56.33173129399089, 56.289357590288226, 56.30994732954137 ], "exact_two_sided_sign_flip_p": 0.25, "geometric_mean_control_over_treatment_speedup": 0.9973474753598454, "geometric_mean_favorable_speedup": 1.0026595792396202, "geometric_mean_treatment_over_control": 1.0026595792396202, "higher_is_better": true, "inference_note": "With five pairs the minimum attainable exact two-sided sign-flip p-value is 0.0625; emphasize effect size, interval, and replication rather than a 0.05 threshold.", "mean_difference": 0.14978913154013754, "mean_percent_change": 0.26652336395678744, "median_difference": 0.19947750095874284, "median_percent_change": 0.35411214315016415, "objective": "higher", "paired_t_95pct_ci_for_mean_difference": [ -0.11284297488386272, 0.41242123796413777 ], "pairs": 5, "percent_changes": [ 0.477529218699349, 0.7355259139146542, 0.35411214315016415, -0.10772908253321722, -0.1268213734470129 ], "sample_sd_of_differences": 0.2115162529918741, "treatment": [ 56.5732692601714, 56.515536000018514, 56.531208794949634, 56.22871758179237, 56.23853428095076 ], "treatment_minus_control": [ 0.26886995808063574, 0.4126512557477824, 0.19947750095874284, -0.06064000849585938, -0.07141304859061393 ] }, "tokens_per_s_wall": { "control": [ 54.7240663716818, 54.59792235557298, 54.68984941556496, 54.57730333237806, 54.72192972179078 ], "exact_two_sided_sign_flip_p": 0.1875, "geometric_mean_control_over_treatment_speedup": 0.9969963190098057, "geometric_mean_favorable_speedup": 1.0030127302708374, "geometric_mean_treatment_over_control": 1.0030127302708374, "higher_is_better": true, "inference_note": "With five pairs the minimum attainable exact two-sided sign-flip p-value is 0.0625; emphasize effect size, interval, and replication rather than a 0.05 threshold.", "mean_difference": 0.16498301308041477, "mean_percent_change": 0.30194049210917745, "median_difference": 0.21827505441392958, "median_percent_change": 0.3988648302036285, "objective": "higher", "paired_t_95pct_ci_for_mean_difference": [ -0.11254069872957986, 0.44250672489040943 ], "pairs": 5, "percent_changes": [ 0.3988648302036285, 0.8153269627013763, 0.49178052552447027, 0.04352332942242093, -0.23979318730600882 ], "sample_sd_of_differences": 0.22350951845803152, "treatment": [ 54.94234142609573, 55.04307393761273, 54.958803444429364, 54.60105719189728, 54.59071026235554 ], "treatment_minus_control": [ 0.21827505441392958, 0.4451515820397489, 0.2689540288644068, 0.023753859519224818, -0.13121945943523627 ] } }, "paired_memory_and_cpu": { "logical_active_kv_max_bytes": { "control": [ 100651008.0, 100651008.0, 100651008.0, 100651008.0, 100651008.0 ], "exact_two_sided_sign_flip_p": 1.0, "geometric_mean_control_over_treatment_speedup": 1.0, "geometric_mean_favorable_speedup": 1.0, "geometric_mean_treatment_over_control": 1.0, "higher_is_better": false, "inference_note": "With five pairs the minimum attainable exact two-sided sign-flip p-value is 0.0625; emphasize effect size, interval, and replication rather than a 0.05 threshold.", "mean_difference": 0.0, "mean_percent_change": 0.0, "median_difference": 0.0, "median_percent_change": 0.0, "objective": "lower", "paired_t_95pct_ci_for_mean_difference": [ 0.0, 0.0 ], "pairs": 5, "percent_changes": [ 0.0, 0.0, 0.0, 0.0, 0.0 ], "sample_sd_of_differences": 0.0, "treatment": [ 100651008.0, 100651008.0, 100651008.0, 100651008.0, 100651008.0 ], "treatment_minus_control": [ 0.0, 0.0, 0.0, 0.0, 0.0 ] }, "logical_active_kv_time_weighted_mean_bytes": { "control": [ 71915170.08447982, 71523692.18363163, 71850601.73874289, 71673258.88229117, 71530203.40146793 ], "exact_two_sided_sign_flip_p": 0.6875, "geometric_mean_control_over_treatment_speedup": 1.0007308179439096, "geometric_mean_favorable_speedup": 1.0007308179439096, "geometric_mean_treatment_over_control": 0.9992697157609165, "higher_is_better": false, "inference_note": "With five pairs the minimum attainable exact two-sided sign-flip p-value is 0.0625; emphasize effect size, interval, and replication rather than a 0.05 threshold.", "mean_difference": -52502.859743666646, "mean_percent_change": -0.07258867368643622, "median_difference": -101314.08752292395, "median_percent_change": -0.14135549171736678, "objective": "lower", "paired_t_95pct_ci_for_mean_difference": [ -347770.84247729345, 242765.12298996013 ], "pairs": 5, "percent_changes": [ -0.4928660029198772, 0.19291907455625681, -0.24595039745986894, -0.14135549171736678, 0.324309449108675 ], "sample_sd_of_differences": 237800.23770383483, "treatment": [ 71560724.66019142, 71661675.02868076, 71673884.89818914, 71571944.79476824, 71762182.61006555 ], "treatment_minus_control": [ -354445.42428840697, 137982.84504912794, -176716.84055374563, -101314.08752292395, 231979.20859761536 ] }, "process_average_cpu_cores": { "control": [ 0.3687191901524845, 0.3705116771699127, 0.3679077029298738, 0.3724422942142223, 0.3681685125765197 ], "exact_two_sided_sign_flip_p": 0.0625, "geometric_mean_control_over_treatment_speedup": 0.9238121798868041, "geometric_mean_favorable_speedup": 0.9238121798868041, "geometric_mean_treatment_over_control": 1.082471114553319, "higher_is_better": false, "inference_note": "With five pairs the minimum attainable exact two-sided sign-flip p-value is 0.0625; emphasize effect size, interval, and replication rather than a 0.05 threshold.", "mean_difference": 0.030475454043219396, "mean_percent_change": 8.248166358964168, "median_difference": 0.03053198429442122, "median_percent_change": 8.292937405415811, "objective": "lower", "paired_t_95pct_ci_for_mean_difference": [ 0.02815206575185062, 0.03279884233458817 ], "pairs": 5, "percent_changes": [ 8.99685116284021, 7.646942844486023, 8.4685145898892, 7.835585792189598, 8.292937405415811 ], "sample_sd_of_differences": 0.0018711892933689206, "treatment": [ 0.4018923068993333, 0.39884449335524247, 0.39906402042981637, 0.4016253297037769, 0.39870049687094095 ], "treatment_minus_control": [ 0.03317311674684881, 0.02833281618532979, 0.031156317499942576, 0.029183035489554587, 0.03053198429442122 ] }, "runner_npu_alloc_max_bytes": { "control": [ 1643012840.0, 1643012840.0, 1643012840.0, 1643012840.0, 1643012840.0 ], "exact_two_sided_sign_flip_p": 0.0625, "geometric_mean_control_over_treatment_speedup": 0.9700327565989119, "geometric_mean_favorable_speedup": 0.9700327565989119, "geometric_mean_treatment_over_control": 1.030893022114179, "higher_is_better": false, "inference_note": "With five pairs the minimum attainable exact two-sided sign-flip p-value is 0.0625; emphasize effect size, interval, and replication rather than a 0.05 threshold.", "mean_difference": 50757632.0, "mean_percent_change": 3.0893022114178974, "median_difference": 50757632.0, "median_percent_change": 3.0893022114178974, "objective": "lower", "paired_t_95pct_ci_for_mean_difference": [ 50757632.0, 50757632.0 ], "pairs": 5, "percent_changes": [ 3.0893022114178974, 3.0893022114178974, 3.0893022114178974, 3.0893022114178974, 3.0893022114178974 ], "sample_sd_of_differences": 0.0, "treatment": [ 1693770472.0, 1693770472.0, 1693770472.0, 1693770472.0, 1693770472.0 ], "treatment_minus_control": [ 50757632.0, 50757632.0, 50757632.0, 50757632.0, 50757632.0 ] }, "runner_npu_alloc_time_weighted_mean_bytes": { "control": [ 1087559074.865844, 1087345082.0252256, 1089056427.7306561, 1083877824.5111778, 1087292973.6158886 ], "exact_two_sided_sign_flip_p": 0.0625, "geometric_mean_control_over_treatment_speedup": 0.924166510234933, "geometric_mean_favorable_speedup": 0.924166510234933, "geometric_mean_treatment_over_control": 1.082056089379163, "higher_is_better": false, "inference_note": "With five pairs the minimum attainable exact two-sided sign-flip p-value is 0.0625; emphasize effect size, interval, and replication rather than a 0.05 threshold.", "mean_difference": 89198542.30221395, "mean_percent_change": 8.20618552317838, "median_difference": 87323561.29514241, "median_percent_change": 8.029316596518145, "objective": "lower", "paired_t_95pct_ci_for_mean_difference": [ 84024690.1319798, 94372394.47244811 ], "pairs": 5, "percent_changes": [ 8.029316596518145, 7.880461421247679, 7.867393481125279, 8.72265970824795, 8.531096408752846 ], "sample_sd_of_differences": 4166869.9211323867, "treatment": [ 1174882636.1609864, 1173032891.7300575, 1174736782.1317136, 1178420798.7964487, 1180050985.4406557 ], "treatment_minus_control": [ 87323561.29514241, 85687809.70483184, 85680354.40105748, 94542974.28527093, 92758011.82476711 ] }, "runner_rss_max_bytes": { "control": [ 2203516928.0, 2166091776.0, 2173329408.0, 2175143936.0, 2116251648.0 ], "exact_two_sided_sign_flip_p": 0.0625, "geometric_mean_control_over_treatment_speedup": 0.946733181397626, "geometric_mean_favorable_speedup": 0.946733181397626, "geometric_mean_treatment_over_control": 1.0562638129190087, "higher_is_better": false, "inference_note": "With five pairs the minimum attainable exact two-sided sign-flip p-value is 0.0625; emphasize effect size, interval, and replication rather than a 0.05 threshold.", "mean_difference": 121732300.8, "mean_percent_change": 5.63580019858932, "median_difference": 110669824.0, "median_percent_change": 5.092179013113506, "objective": "lower", "paired_t_95pct_ci_for_mean_difference": [ 81803452.7544043, 161661148.8455957 ], "pairs": 5, "percent_changes": [ 4.240586800701901, 5.059650436528872, 5.092179013113506, 5.423686683325769, 8.36289805927655 ], "sample_sd_of_differences": 32157531.841333643, "treatment": [ 2296958976.0, 2275688448.0, 2283999232.0, 2293116928.0, 2293231616.0 ], "treatment_minus_control": [ 93442048.0, 109596672.0, 110669824.0, 117972992.0, 176979968.0 ] }, "runner_rss_time_weighted_mean_bytes": { "control": [ 1909567818.144264, 1877492560.2305627, 1884727299.8957303, 1882871556.6949093, 1834834792.070105 ], "exact_two_sided_sign_flip_p": 0.125, "geometric_mean_control_over_treatment_speedup": 1.0207884428301044, "geometric_mean_favorable_speedup": 1.0207884428301044, "geometric_mean_treatment_over_control": 0.9796349155634355, "higher_is_better": false, "inference_note": "With five pairs the minimum attainable exact two-sided sign-flip p-value is 0.0625; emphasize effect size, interval, and replication rather than a 0.05 threshold.", "mean_difference": -38384449.28061509, "mean_percent_change": -2.025876052213111, "median_difference": -47786623.00099349, "median_percent_change": -2.545236344112337, "objective": "lower", "paired_t_95pct_ci_for_mean_difference": [ -76410088.55315036, -358810.00807982683 ], "pairs": 5, "percent_changes": [ -3.5740798189619603, -2.545236344112337, -2.7178675403366928, -1.981985802589945, 0.6897892449353804 ], "sample_sd_of_differences": 30624742.900102295, "treatment": [ 1841318340.1265776, 1829705937.2295692, 1833502908.388, 1845553309.760212, 1847491285.128137 ], "treatment_minus_control": [ -68249478.01768637, -47786623.00099349, -51224391.507730246, -37318246.93469739, 12656493.058032036 ] }, "system_rss_fixed_interval_mean_bytes": { "control": [ 2089622869.3333333, 2061398835.2, 2069494033.0666666, 2069376068.2666667, 2013813145.6 ], "exact_two_sided_sign_flip_p": 0.0625, "geometric_mean_control_over_treatment_speedup": 1.0567845643303784, "geometric_mean_favorable_speedup": 1.0567845643303784, "geometric_mean_treatment_over_control": 0.9462666599730671, "higher_is_better": false, "inference_note": "With five pairs the minimum attainable exact two-sided sign-flip p-value is 0.0625; emphasize effect size, interval, and replication rather than a 0.05 threshold.", "mean_difference": -110868540.46476188, "mean_percent_change": -5.364186790168714, "median_difference": -119187836.34285712, "median_percent_change": -5.7818911269198106, "objective": "lower", "paired_t_95pct_ci_for_mean_difference": [ -150027787.63786408, -71709293.29165968 ], "pairs": 5, "percent_changes": [ -6.591426147361176, -5.7818911269198106, -6.102862972719575, -5.530280842695008, -2.8144728611479986 ], "sample_sd_of_differences": 31537717.702591993, "treatment": [ 1951886921.142857, 1942210998.857143, 1943195648.0, 1954933760.0, 1957134921.142857 ], "treatment_minus_control": [ -137735948.19047618, -119187836.34285712, -126298385.0666666, -114442308.26666665, -56678224.45714283 ] }, "system_rss_max_bytes": { "control": [ 2202165248.0, 2164391936.0, 2171232256.0, 2173009920.0, 2114629632.0 ], "exact_two_sided_sign_flip_p": 0.0625, "geometric_mean_control_over_treatment_speedup": 0.946616129059707, "geometric_mean_favorable_speedup": 0.946616129059707, "geometric_mean_treatment_over_control": 1.0563944235699008, "higher_is_better": false, "inference_note": "With five pairs the minimum attainable exact two-sided sign-flip p-value is 0.0625; emphasize effect size, interval, and replication rather than a 0.05 threshold.", "mean_difference": 121914163.2, "mean_percent_change": 5.648961190863399, "median_difference": 111083520.0, "median_percent_change": 5.11615096418317, "objective": "lower", "paired_t_95pct_ci_for_mean_difference": [ 81792743.08727267, 162035583.31272733 ], "pairs": 5, "percent_changes": [ 4.215475840621385, 5.076114273602616, 5.11615096418317, 5.4570986956193925, 8.379966180290431 ], "sample_sd_of_differences": 32312623.7782076, "treatment": [ 2294996992.0, 2274258944.0, 2282315776.0, 2291593216.0, 2291834880.0 ], "treatment_minus_control": [ 92831744.0, 109867008.0, 111083520.0, 118583296.0, 177205248.0 ] } }, "paired_per_band": { "0-1024": { "infer_s": { "control": [ 13.895383894, 14.186192547, 13.92561879, 14.278610629, 14.101704326 ], "exact_two_sided_sign_flip_p": 0.625, "geometric_mean_control_over_treatment_speedup": 1.0027402236370315, "geometric_mean_favorable_speedup": 1.0027402236370315, "geometric_mean_treatment_over_control": 0.9972672646689165, "higher_is_better": false, "inference_note": "With five pairs the minimum attainable exact two-sided sign-flip p-value is 0.0625; emphasize effect size, interval, and replication rather than a 0.05 threshold.", "mean_difference": -0.03901465699999953, "mean_percent_change": -0.26817784194456207, "median_difference": -0.12299945399999856, "median_percent_change": -0.8614245264884908, "objective": "lower", "paired_t_95pct_ci_for_mean_difference": [ -0.23583156765740804, 0.15780225365740896 ], "pairs": 5, "percent_changes": [ 1.4714891618668355, -1.108012537396719, 0.2777151132973117, -0.8614245264884908, -1.120656421001748 ], "sample_sd_of_differences": 0.15851061027733965, "treatment": [ 14.099852962, 14.029007755, 13.964292338, 14.155611175, 13.943672671 ], "treatment_minus_control": [ 0.20446906799999987, -0.1571847919999989, 0.038673548000000224, -0.12299945399999856, -0.15803165500000027 ] }, "tokens_per_s": { "control": [ 64.4818456859541, 63.16000554986687, 64.3418445895861, 62.75120341052057, 63.53841913618917 ], "exact_two_sided_sign_flip_p": 0.625, "geometric_mean_control_over_treatment_speedup": 0.9972672646689165, "geometric_mean_favorable_speedup": 1.0027402236370315, "geometric_mean_treatment_over_control": 1.0027402236370315, "higher_is_better": true, "inference_note": "With five pairs the minimum attainable exact two-sided sign-flip p-value is 0.0625; emphasize effect size, interval, and replication rather than a 0.05 threshold.", "mean_difference": 0.17195089723608845, "mean_percent_change": 0.27911953012557505, "median_difference": 0.5452511842772481, "median_percent_change": 0.8689095262607088, "objective": "higher", "paired_t_95pct_ci_for_mean_difference": [ -0.7223384513514485, 1.0662402458236253 ], "pairs": 5, "percent_changes": [ -1.4501503565395844, 1.1204270091302675, -0.27694599242069484, 0.8689095262607088, 1.1333574641971782 ], "sample_sd_of_differences": 0.7202346075631746, "treatment": [ 63.54676197083593, 63.86766731101575, 64.16365242954569, 63.296454594797815, 64.25853655210206 ], "treatment_minus_control": [ -0.9350837151181679, 0.7076617611488842, -0.17819216004041039, 0.5452511842772481, 0.7201174159128882 ] } }, "1024-2048": { "infer_s": { "control": [ 16.086443406, 16.429506961, 16.007269808, 16.376137428, 16.213047445 ], "exact_two_sided_sign_flip_p": 0.3125, "geometric_mean_control_over_treatment_speedup": 0.9937617718675725, "geometric_mean_favorable_speedup": 0.9937617718675725, "geometric_mean_treatment_over_control": 1.0062773879102875, "higher_is_better": false, "inference_note": "With five pairs the minimum attainable exact two-sided sign-flip p-value is 0.0625; emphasize effect size, interval, and replication rather than a 0.05 threshold.", "mean_difference": 0.10121877719999972, "mean_percent_change": 0.6331477117383826, "median_difference": 0.0959855239999996, "median_percent_change": 0.5920264177700961, "objective": "lower", "paired_t_95pct_ci_for_mean_difference": [ -0.13241008268537216, 0.3348476370853716 ], "pairs": 5, "percent_changes": [ 0.8233465512370418, -0.9848967128722173, 2.2922734944882324, 0.4429888080687598, 0.5920264177700961 ], "sample_sd_of_differences": 0.18815788254745378, "treatment": [ 16.218890583, 16.267693287, 16.374200211, 16.448681884, 16.309032969 ], "treatment_minus_control": [ 0.13244717699999953, -0.16181367400000113, 0.36693040300000135, 0.07254445599999926, 0.0959855239999996 ] }, "tokens_per_s": { "control": [ 63.65608445295394, 62.32688554992846, 63.97093397452653, 62.530007732419236, 63.159008414287655 ], "exact_two_sided_sign_flip_p": 0.3125, "geometric_mean_control_over_treatment_speedup": 1.0062773879102875, "geometric_mean_favorable_speedup": 0.9937617718675725, "geometric_mean_treatment_over_control": 0.9937617718675725, "higher_is_better": true, "inference_note": "With five pairs the minimum attainable exact two-sided sign-flip p-value is 0.0625; emphasize effect size, interval, and replication rather than a 0.05 threshold.", "mean_difference": -0.396178746597009, "mean_percent_change": -0.6184824937713841, "median_difference": -0.37171735010218043, "median_percent_change": -0.5885420930992507, "objective": "higher", "paired_t_95pct_ci_for_mean_difference": [ -1.3041694776403094, 0.5118119844462914 ], "pairs": 5, "percent_changes": [ -0.8166229146328073, 0.9946934156258677, -2.2409058046908417, -0.4410350720598884, -0.5885420930992507 ], "sample_sd_of_differences": 0.7312693021300796, "treatment": [ 63.13625428075311, 62.94684697665827, 62.53740560177642, 62.25422846775751, 62.787291064185474 ], "treatment_minus_control": [ -0.5198301722008338, 0.6199614267298088, -1.4335283727501107, -0.275779264661729, -0.37171735010218043 ] } }, "2048-4096": { "infer_s": { "control": [ 33.429199937, 33.647303438, 33.574458698, 33.092519245, 33.759629239 ], "exact_two_sided_sign_flip_p": 0.4375, "geometric_mean_control_over_treatment_speedup": 1.0040665147538372, "geometric_mean_favorable_speedup": 1.0040665147538372, "geometric_mean_treatment_over_control": 0.9959499548146626, "higher_is_better": false, "inference_note": "With five pairs the minimum attainable exact two-sided sign-flip p-value is 0.0625; emphasize effect size, interval, and replication rather than a 0.05 threshold.", "mean_difference": -0.13620262399999916, "mean_percent_change": -0.40145838128551115, "median_difference": -0.20123455099999887, "median_percent_change": -0.5960804532993138, "objective": "lower", "paired_t_95pct_ci_for_mean_difference": [ -0.5264366431887142, 0.2540313951887158 ], "pairs": 5, "percent_changes": [ -0.33680928413539546, -1.1261650155657383, -1.1186861607462693, 1.1704490073191611, -0.5960804532993138 ], "sample_sd_of_differences": 0.31428311889445826, "treatment": [ 33.316607288, 33.268379278, 33.198865875, 33.479850308, 33.558394688 ], "treatment_minus_control": [ -0.11259264899999977, -0.37892416000000395, -0.3755928229999981, 0.38733106300000486, -0.20123455099999887 ] }, "tokens_per_s": { "control": [ 61.26380541142533, 60.866690365655444, 60.99874962755535, 61.887098556554754, 60.664173338553645 ], "exact_two_sided_sign_flip_p": 0.4375, "geometric_mean_control_over_treatment_speedup": 0.9959499548146626, "geometric_mean_favorable_speedup": 1.0040665147538372, "geometric_mean_treatment_over_control": 1.0040665147538372, "higher_is_better": true, "inference_note": "With five pairs the minimum attainable exact two-sided sign-flip p-value is 0.0625; emphasize effect size, interval, and replication rather than a 0.05 threshold.", "mean_difference": 0.24764195524710858, "mean_percent_change": 0.41020573750973954, "median_difference": 0.36377567511998166, "median_percent_change": 0.5996548788192138, "objective": "higher", "paired_t_95pct_ci_for_mean_difference": [ -0.47050027755844953, 0.9657841880526667 ], "pairs": 5, "percent_changes": [ 0.33794752276759077, 1.138991944373381, 1.131342330831962, -1.1569079892434495, 0.5996548788192138 ], "sample_sd_of_differences": 0.5783708483570564, "treatment": [ 61.4708449241664, 61.55995706572695, 61.688854303370086, 61.171121769043005, 61.02794901367363 ], "treatment_minus_control": [ 0.20703951274106913, 0.6932667000715043, 0.6901046758147373, -0.7159767875117495, 0.36377567511998166 ] } }, "4096-8192": { "infer_s": { "control": [ 79.810463445, 79.47292145, 79.644652736, 79.512495204, 79.132998546 ], "exact_two_sided_sign_flip_p": 0.25, "geometric_mean_control_over_treatment_speedup": 1.0038607411035048, "geometric_mean_favorable_speedup": 1.0038607411035048, "geometric_mean_treatment_over_control": 0.996154106894089, "higher_is_better": false, "inference_note": "With five pairs the minimum attainable exact two-sided sign-flip p-value is 0.0625; emphasize effect size, interval, and replication rather than a 0.05 threshold.", "mean_difference": -0.30579200420000063, "mean_percent_change": -0.38301744773000185, "median_difference": -0.3515730330000082, "median_percent_change": -0.44238090985644546, "objective": "lower", "paired_t_95pct_ci_for_mean_difference": [ -0.9244029055594282, 0.3128188971594269 ], "pairs": 5, "percent_changes": [ -1.1339335783003621, -0.44238090985644546, -0.6719107920199556, -0.2293692287383195, 0.5625072702650735 ], "sample_sd_of_differences": 0.49821121148162384, "treatment": [ 78.905465801, 79.121348417, 79.109511719, 79.330118007, 79.578127416 ], "treatment_minus_control": [ -0.9049976439999909, -0.3515730330000082, -0.5351410170000008, -0.1823771970000081, 0.4451288700000049 ] }, "tokens_per_s": { "control": [ 51.32159147055558, 51.539567506360996, 51.42843692943338, 51.51391601396939, 51.76096034853268 ], "exact_two_sided_sign_flip_p": 0.25, "geometric_mean_control_over_treatment_speedup": 0.9961541068940889, "geometric_mean_favorable_speedup": 1.0038607411035048, "geometric_mean_treatment_over_control": 1.0038607411035048, "higher_is_better": true, "inference_note": "With five pairs the minimum attainable exact two-sided sign-flip p-value is 0.0625; emphasize effect size, interval, and replication rather than a 0.05 threshold.", "mean_difference": 0.19888612700558211, "mean_percent_change": 0.3876554817608784, "median_difference": 0.22901432331791227, "median_percent_change": 0.4443466144523766, "objective": "higher", "paired_t_95pct_ci_for_mean_difference": [ -0.20347458014975395, 0.6012468341609182 ], "pairs": 5, "percent_changes": [ 1.1469391059453435, 0.4443466144523766, 0.6764559727038164, 0.22989654066051848, -0.5593608249576631 ], "sample_sd_of_differences": 0.3240496003609715, "treatment": [ 51.910218872924894, 51.76858182967891, 51.77632766271075, 51.63234472484427, 51.471429813721116 ], "treatment_minus_control": [ 0.5886274023693119, 0.22901432331791227, 0.3478907332773673, 0.1184287108748805, -0.2895305348115613 ] } } }, "paired_thermal_maxima": { "hwmon/hwmon1/acpitz/temp1": { "control": [ 40.0, 40.0, 40.0, 40.0, 40.0 ], "exact_two_sided_sign_flip_p": 1.0, "geometric_mean_control_over_treatment_speedup": 1.0, "geometric_mean_favorable_speedup": 1.0, "geometric_mean_treatment_over_control": 1.0, "higher_is_better": false, "inference_note": "With five pairs the minimum attainable exact two-sided sign-flip p-value is 0.0625; emphasize effect size, interval, and replication rather than a 0.05 threshold.", "mean_difference": 0.0, "mean_percent_change": 0.0, "median_difference": 0.0, "median_percent_change": 0.0, "objective": "lower", "paired_t_95pct_ci_for_mean_difference": [ 0.0, 0.0 ], "pairs": 5, "percent_changes": [ 0.0, 0.0, 0.0, 0.0, 0.0 ], "sample_sd_of_differences": 0.0, "treatment": [ 40.0, 40.0, 40.0, 40.0, 40.0 ], "treatment_minus_control": [ 0.0, 0.0, 0.0, 0.0, 0.0 ] }, "hwmon/hwmon1/acpitz/temp2": { "control": [ 27.8, 27.8, 27.8, 27.8, 27.8 ], "exact_two_sided_sign_flip_p": 1.0, "geometric_mean_control_over_treatment_speedup": 1.0, "geometric_mean_favorable_speedup": 1.0, "geometric_mean_treatment_over_control": 1.0, "higher_is_better": false, "inference_note": "With five pairs the minimum attainable exact two-sided sign-flip p-value is 0.0625; emphasize effect size, interval, and replication rather than a 0.05 threshold.", "mean_difference": 0.0, "mean_percent_change": 0.0, "median_difference": 0.0, "median_percent_change": 0.0, "objective": "lower", "paired_t_95pct_ci_for_mean_difference": [ 0.0, 0.0 ], "pairs": 5, "percent_changes": [ 0.0, 0.0, 0.0, 0.0, 0.0 ], "sample_sd_of_differences": 0.0, "treatment": [ 27.8, 27.8, 27.8, 27.8, 27.8 ], "treatment_minus_control": [ 0.0, 0.0, 0.0, 0.0, 0.0 ] }, "hwmon/hwmon2/nvme/Composite": { "control": [ 39.85, 42.85, 42.85, 42.85, 39.85 ], "exact_two_sided_sign_flip_p": 0.625, "geometric_mean_control_over_treatment_speedup": 0.9864483656414745, "geometric_mean_favorable_speedup": 0.9864483656414745, "geometric_mean_treatment_over_control": 1.0137378040559812, "higher_is_better": false, "inference_note": "With five pairs the minimum attainable exact two-sided sign-flip p-value is 0.0625; emphasize effect size, interval, and replication rather than a 0.05 threshold.", "mean_difference": 0.6, "mean_percent_change": 1.4353709725355732, "median_difference": 1.0, "median_percent_change": 2.509410288582183, "objective": "lower", "paired_t_95pct_ci_for_mean_difference": [ -1.477701188644315, 2.677701188644315 ], "pairs": 5, "percent_changes": [ 2.509410288582183, 4.667444574095683, 4.667444574095683, -4.667444574095683, 0.0 ], "sample_sd_of_differences": 1.6733200530681511, "treatment": [ 40.85, 44.85, 44.85, 40.85, 39.85 ], "treatment_minus_control": [ 1.0, 2.0, 2.0, -2.0, 0.0 ] }, "hwmon/hwmon2/nvme/Sensor 2": { "control": [ 57.85, 57.85, 57.85, 57.85, 57.85 ], "exact_two_sided_sign_flip_p": 1.0, "geometric_mean_control_over_treatment_speedup": 1.0, "geometric_mean_favorable_speedup": 1.0, "geometric_mean_treatment_over_control": 1.0, "higher_is_better": false, "inference_note": "With five pairs the minimum attainable exact two-sided sign-flip p-value is 0.0625; emphasize effect size, interval, and replication rather than a 0.05 threshold.", "mean_difference": 0.0, "mean_percent_change": 0.0, "median_difference": 0.0, "median_percent_change": 0.0, "objective": "lower", "paired_t_95pct_ci_for_mean_difference": [ 0.0, 0.0 ], "pairs": 5, "percent_changes": [ 0.0, 0.0, 0.0, 0.0, 0.0 ], "sample_sd_of_differences": 0.0, "treatment": [ 57.85, 57.85, 57.85, 57.85, 57.85 ], "treatment_minus_control": [ 0.0, 0.0, 0.0, 0.0, 0.0 ] }, "hwmon/hwmon6/coretemp/Core 0": { "control": [ 66.0, 60.0, 60.0, 61.0, 58.0 ], "exact_two_sided_sign_flip_p": 0.1875, "geometric_mean_control_over_treatment_speedup": 0.7728214341878408, "geometric_mean_favorable_speedup": 0.7728214341878408, "geometric_mean_treatment_over_control": 1.2939599702626021, "higher_is_better": false, "inference_note": "With five pairs the minimum attainable exact two-sided sign-flip p-value is 0.0625; emphasize effect size, interval, and replication rather than a 0.05 threshold.", "mean_difference": 19.6, "mean_percent_change": 33.16127927094575, "median_difference": 29.0, "median_percent_change": 47.540983606557376, "objective": "lower", "paired_t_95pct_ci_for_mean_difference": [ -5.7342048569510276, 44.93420485695103 ], "pairs": 5, "percent_changes": [ -9.090909090909092, 3.3333333333333335, 53.333333333333336, 47.540983606557376, 70.6896551724138 ], "sample_sd_of_differences": 20.403431084011334, "treatment": [ 60.0, 62.0, 92.0, 90.0, 99.0 ], "treatment_minus_control": [ -6.0, 2.0, 32.0, 29.0, 41.0 ] }, "hwmon/hwmon6/coretemp/Core 12": { "control": [ 64.0, 61.0, 59.0, 82.0, 57.0 ], "exact_two_sided_sign_flip_p": 0.8125, "geometric_mean_control_over_treatment_speedup": 0.9525092512030016, "geometric_mean_favorable_speedup": 0.9525092512030016, "geometric_mean_treatment_over_control": 1.049858569601312, "higher_is_better": false, "inference_note": "With five pairs the minimum attainable exact two-sided sign-flip p-value is 0.0625; emphasize effect size, interval, and replication rather than a 0.05 threshold.", "mean_difference": 4.0, "mean_percent_change": 9.102722500057366, "median_difference": -2.0, "median_percent_change": -3.389830508474576, "objective": "lower", "paired_t_95pct_ci_for_mean_difference": [ -23.485403219280773, 31.485403219280773 ], "pairs": 5, "percent_changes": [ -9.375, 52.459016393442624, -3.389830508474576, -29.26829268292683, 35.08771929824562 ], "sample_sd_of_differences": 22.135943621178654, "treatment": [ 58.0, 93.0, 57.0, 58.0, 77.0 ], "treatment_minus_control": [ -6.0, 32.0, -2.0, -24.0, 20.0 ] }, "hwmon/hwmon6/coretemp/Core 16": { "control": [ 65.0, 64.0, 63.0, 74.0, 71.0 ], "exact_two_sided_sign_flip_p": 0.25, "geometric_mean_control_over_treatment_speedup": 1.043042479093756, "geometric_mean_favorable_speedup": 1.043042479093756, "geometric_mean_treatment_over_control": 0.9587337237394652, "higher_is_better": false, "inference_note": "With five pairs the minimum attainable exact two-sided sign-flip p-value is 0.0625; emphasize effect size, interval, and replication rather than a 0.05 threshold.", "mean_difference": -2.8, "mean_percent_change": -3.9630585546078505, "median_difference": -2.0, "median_percent_change": -2.7027027027027026, "objective": "lower", "paired_t_95pct_ci_for_mean_difference": [ -8.169389087616345, 2.5693890876163463 ], "pairs": 5, "percent_changes": [ -4.615384615384615, 0.0, 1.5873015873015872, -2.7027027027027026, -14.084507042253522 ], "sample_sd_of_differences": 4.324349662087931, "treatment": [ 62.0, 64.0, 64.0, 72.0, 61.0 ], "treatment_minus_control": [ -3.0, 0.0, 1.0, -2.0, -10.0 ] }, "hwmon/hwmon6/coretemp/Core 17": { "control": [ 65.0, 64.0, 63.0, 74.0, 71.0 ], "exact_two_sided_sign_flip_p": 0.25, "geometric_mean_control_over_treatment_speedup": 1.043042479093756, "geometric_mean_favorable_speedup": 1.043042479093756, "geometric_mean_treatment_over_control": 0.9587337237394652, "higher_is_better": false, "inference_note": "With five pairs the minimum attainable exact two-sided sign-flip p-value is 0.0625; emphasize effect size, interval, and replication rather than a 0.05 threshold.", "mean_difference": -2.8, "mean_percent_change": -3.9630585546078505, "median_difference": -2.0, "median_percent_change": -2.7027027027027026, "objective": "lower", "paired_t_95pct_ci_for_mean_difference": [ -8.169389087616345, 2.5693890876163463 ], "pairs": 5, "percent_changes": [ -4.615384615384615, 0.0, 1.5873015873015872, -2.7027027027027026, -14.084507042253522 ], "sample_sd_of_differences": 4.324349662087931, "treatment": [ 62.0, 64.0, 64.0, 72.0, 61.0 ], "treatment_minus_control": [ -3.0, 0.0, 1.0, -2.0, -10.0 ] }, "hwmon/hwmon6/coretemp/Core 18": { "control": [ 65.0, 64.0, 63.0, 74.0, 71.0 ], "exact_two_sided_sign_flip_p": 0.25, "geometric_mean_control_over_treatment_speedup": 1.043042479093756, "geometric_mean_favorable_speedup": 1.043042479093756, "geometric_mean_treatment_over_control": 0.9587337237394652, "higher_is_better": false, "inference_note": "With five pairs the minimum attainable exact two-sided sign-flip p-value is 0.0625; emphasize effect size, interval, and replication rather than a 0.05 threshold.", "mean_difference": -2.8, "mean_percent_change": -3.9630585546078505, "median_difference": -2.0, "median_percent_change": -2.7027027027027026, "objective": "lower", "paired_t_95pct_ci_for_mean_difference": [ -8.169389087616345, 2.5693890876163463 ], "pairs": 5, "percent_changes": [ -4.615384615384615, 0.0, 1.5873015873015872, -2.7027027027027026, -14.084507042253522 ], "sample_sd_of_differences": 4.324349662087931, "treatment": [ 62.0, 64.0, 64.0, 72.0, 61.0 ], "treatment_minus_control": [ -3.0, 0.0, 1.0, -2.0, -10.0 ] }, "hwmon/hwmon6/coretemp/Core 19": { "control": [ 65.0, 64.0, 64.0, 74.0, 71.0 ], "exact_two_sided_sign_flip_p": 0.25, "geometric_mean_control_over_treatment_speedup": 1.046332899303502, "geometric_mean_favorable_speedup": 1.046332899303502, "geometric_mean_treatment_over_control": 0.9557187780921886, "higher_is_better": false, "inference_note": "With five pairs the minimum attainable exact two-sided sign-flip p-value is 0.0625; emphasize effect size, interval, and replication rather than a 0.05 threshold.", "mean_difference": -3.0, "mean_percent_change": -4.280518872068168, "median_difference": -2.0, "median_percent_change": -2.7027027027027026, "objective": "lower", "paired_t_95pct_ci_for_mean_difference": [ -8.119511622145708, 2.1195116221457093 ], "pairs": 5, "percent_changes": [ -4.615384615384615, 0.0, 0.0, -2.7027027027027026, -14.084507042253522 ], "sample_sd_of_differences": 4.123105625617661, "treatment": [ 62.0, 64.0, 64.0, 72.0, 61.0 ], "treatment_minus_control": [ -3.0, 0.0, 0.0, -2.0, -10.0 ] }, "hwmon/hwmon6/coretemp/Core 20": { "control": [ 62.0, 61.0, 55.0, 61.0, 61.0 ], "exact_two_sided_sign_flip_p": 0.5, "geometric_mean_control_over_treatment_speedup": 0.9773796758970468, "geometric_mean_favorable_speedup": 0.9773796758970468, "geometric_mean_treatment_over_control": 1.0231438453865864, "higher_is_better": false, "inference_note": "With five pairs the minimum attainable exact two-sided sign-flip p-value is 0.0625; emphasize effect size, interval, and replication rather than a 0.05 threshold.", "mean_difference": 1.4, "mean_percent_change": 2.464593048411134, "median_difference": 3.0, "median_percent_change": 4.918032786885246, "objective": "lower", "paired_t_95pct_ci_for_mean_difference": [ -3.1282447580874533, 5.9282447580874535 ], "pairs": 5, "percent_changes": [ -8.064516129032258, 4.918032786885246, 7.2727272727272725, 3.278688524590164, 4.918032786885246 ], "sample_sd_of_differences": 3.646916505762094, "treatment": [ 57.0, 64.0, 59.0, 63.0, 64.0 ], "treatment_minus_control": [ -5.0, 3.0, 4.0, 2.0, 3.0 ] }, "hwmon/hwmon6/coretemp/Core 21": { "control": [ 62.0, 61.0, 55.0, 61.0, 61.0 ], "exact_two_sided_sign_flip_p": 0.5, "geometric_mean_control_over_treatment_speedup": 0.9773796758970468, "geometric_mean_favorable_speedup": 0.9773796758970468, "geometric_mean_treatment_over_control": 1.0231438453865864, "higher_is_better": false, "inference_note": "With five pairs the minimum attainable exact two-sided sign-flip p-value is 0.0625; emphasize effect size, interval, and replication rather than a 0.05 threshold.", "mean_difference": 1.4, "mean_percent_change": 2.464593048411134, "median_difference": 3.0, "median_percent_change": 4.918032786885246, "objective": "lower", "paired_t_95pct_ci_for_mean_difference": [ -3.1282447580874533, 5.9282447580874535 ], "pairs": 5, "percent_changes": [ -8.064516129032258, 4.918032786885246, 7.2727272727272725, 3.278688524590164, 4.918032786885246 ], "sample_sd_of_differences": 3.646916505762094, "treatment": [ 57.0, 64.0, 59.0, 63.0, 64.0 ], "treatment_minus_control": [ -5.0, 3.0, 4.0, 2.0, 3.0 ] }, "hwmon/hwmon6/coretemp/Core 22": { "control": [ 62.0, 61.0, 55.0, 61.0, 61.0 ], "exact_two_sided_sign_flip_p": 0.5, "geometric_mean_control_over_treatment_speedup": 0.9773796758970468, "geometric_mean_favorable_speedup": 0.9773796758970468, "geometric_mean_treatment_over_control": 1.0231438453865864, "higher_is_better": false, "inference_note": "With five pairs the minimum attainable exact two-sided sign-flip p-value is 0.0625; emphasize effect size, interval, and replication rather than a 0.05 threshold.", "mean_difference": 1.4, "mean_percent_change": 2.464593048411134, "median_difference": 3.0, "median_percent_change": 4.918032786885246, "objective": "lower", "paired_t_95pct_ci_for_mean_difference": [ -3.1282447580874533, 5.9282447580874535 ], "pairs": 5, "percent_changes": [ -8.064516129032258, 4.918032786885246, 7.2727272727272725, 3.278688524590164, 4.918032786885246 ], "sample_sd_of_differences": 3.646916505762094, "treatment": [ 57.0, 64.0, 59.0, 63.0, 64.0 ], "treatment_minus_control": [ -5.0, 3.0, 4.0, 2.0, 3.0 ] }, "hwmon/hwmon6/coretemp/Core 23": { "control": [ 62.0, 61.0, 55.0, 61.0, 61.0 ], "exact_two_sided_sign_flip_p": 0.5, "geometric_mean_control_over_treatment_speedup": 0.9773796758970468, "geometric_mean_favorable_speedup": 0.9773796758970468, "geometric_mean_treatment_over_control": 1.0231438453865864, "higher_is_better": false, "inference_note": "With five pairs the minimum attainable exact two-sided sign-flip p-value is 0.0625; emphasize effect size, interval, and replication rather than a 0.05 threshold.", "mean_difference": 1.4, "mean_percent_change": 2.464593048411134, "median_difference": 3.0, "median_percent_change": 4.918032786885246, "objective": "lower", "paired_t_95pct_ci_for_mean_difference": [ -3.1282447580874533, 5.9282447580874535 ], "pairs": 5, "percent_changes": [ -8.064516129032258, 4.918032786885246, 7.2727272727272725, 3.278688524590164, 4.918032786885246 ], "sample_sd_of_differences": 3.646916505762094, "treatment": [ 57.0, 64.0, 59.0, 63.0, 64.0 ], "treatment_minus_control": [ -5.0, 3.0, 4.0, 2.0, 3.0 ] }, "hwmon/hwmon6/coretemp/Core 32": { "control": [ 60.0, 56.0, 57.0, 55.0, 53.0 ], "exact_two_sided_sign_flip_p": 0.75, "geometric_mean_control_over_treatment_speedup": 0.9757357401538619, "geometric_mean_favorable_speedup": 0.9757357401538619, "geometric_mean_treatment_over_control": 1.0248676550910312, "higher_is_better": false, "inference_note": "With five pairs the minimum attainable exact two-sided sign-flip p-value is 0.0625; emphasize effect size, interval, and replication rather than a 0.05 threshold.", "mean_difference": 1.4, "mean_percent_change": 2.685306748861863, "median_difference": -1.0, "median_percent_change": -1.6666666666666667, "objective": "lower", "paired_t_95pct_ci_for_mean_difference": [ -3.4567951701051274, 6.256795170105127 ], "pairs": 5, "percent_changes": [ -1.6666666666666667, 3.5714285714285716, -1.7543859649122806, -1.8181818181818181, 15.09433962264151 ], "sample_sd_of_differences": 3.9115214431215892, "treatment": [ 59.0, 58.0, 56.0, 54.0, 61.0 ], "treatment_minus_control": [ -1.0, 2.0, -1.0, -1.0, 8.0 ] }, "hwmon/hwmon6/coretemp/Core 33": { "control": [ 60.0, 56.0, 57.0, 55.0, 53.0 ], "exact_two_sided_sign_flip_p": 0.75, "geometric_mean_control_over_treatment_speedup": 0.9757357401538619, "geometric_mean_favorable_speedup": 0.9757357401538619, "geometric_mean_treatment_over_control": 1.0248676550910312, "higher_is_better": false, "inference_note": "With five pairs the minimum attainable exact two-sided sign-flip p-value is 0.0625; emphasize effect size, interval, and replication rather than a 0.05 threshold.", "mean_difference": 1.4, "mean_percent_change": 2.685306748861863, "median_difference": -1.0, "median_percent_change": -1.6666666666666667, "objective": "lower", "paired_t_95pct_ci_for_mean_difference": [ -3.4567951701051274, 6.256795170105127 ], "pairs": 5, "percent_changes": [ -1.6666666666666667, 3.5714285714285716, -1.7543859649122806, -1.8181818181818181, 15.09433962264151 ], "sample_sd_of_differences": 3.9115214431215892, "treatment": [ 59.0, 58.0, 56.0, 54.0, 61.0 ], "treatment_minus_control": [ -1.0, 2.0, -1.0, -1.0, 8.0 ] }, "hwmon/hwmon6/coretemp/Core 34": { "control": [ 60.0, 56.0, 57.0, 55.0, 53.0 ], "exact_two_sided_sign_flip_p": 0.75, "geometric_mean_control_over_treatment_speedup": 0.9757357401538619, "geometric_mean_favorable_speedup": 0.9757357401538619, "geometric_mean_treatment_over_control": 1.0248676550910312, "higher_is_better": false, "inference_note": "With five pairs the minimum attainable exact two-sided sign-flip p-value is 0.0625; emphasize effect size, interval, and replication rather than a 0.05 threshold.", "mean_difference": 1.4, "mean_percent_change": 2.685306748861863, "median_difference": -1.0, "median_percent_change": -1.6666666666666667, "objective": "lower", "paired_t_95pct_ci_for_mean_difference": [ -3.4567951701051274, 6.256795170105127 ], "pairs": 5, "percent_changes": [ -1.6666666666666667, 3.5714285714285716, -1.7543859649122806, -1.8181818181818181, 15.09433962264151 ], "sample_sd_of_differences": 3.9115214431215892, "treatment": [ 59.0, 58.0, 56.0, 54.0, 61.0 ], "treatment_minus_control": [ -1.0, 2.0, -1.0, -1.0, 8.0 ] }, "hwmon/hwmon6/coretemp/Core 35": { "control": [ 60.0, 56.0, 57.0, 55.0, 53.0 ], "exact_two_sided_sign_flip_p": 0.75, "geometric_mean_control_over_treatment_speedup": 0.9757357401538619, "geometric_mean_favorable_speedup": 0.9757357401538619, "geometric_mean_treatment_over_control": 1.0248676550910312, "higher_is_better": false, "inference_note": "With five pairs the minimum attainable exact two-sided sign-flip p-value is 0.0625; emphasize effect size, interval, and replication rather than a 0.05 threshold.", "mean_difference": 1.4, "mean_percent_change": 2.685306748861863, "median_difference": -1.0, "median_percent_change": -1.6666666666666667, "objective": "lower", "paired_t_95pct_ci_for_mean_difference": [ -3.4567951701051274, 6.256795170105127 ], "pairs": 5, "percent_changes": [ -1.6666666666666667, 3.5714285714285716, -1.7543859649122806, -1.8181818181818181, 15.09433962264151 ], "sample_sd_of_differences": 3.9115214431215892, "treatment": [ 59.0, 58.0, 56.0, 54.0, 61.0 ], "treatment_minus_control": [ -1.0, 2.0, -1.0, -1.0, 8.0 ] }, "hwmon/hwmon6/coretemp/Core 4": { "control": [ 62.0, 56.0, 56.0, 85.0, 75.0 ], "exact_two_sided_sign_flip_p": 1.0, "geometric_mean_control_over_treatment_speedup": 0.9462273981298371, "geometric_mean_favorable_speedup": 0.9462273981298371, "geometric_mean_treatment_over_control": 1.0568284135255872, "higher_is_better": false, "inference_note": "With five pairs the minimum attainable exact two-sided sign-flip p-value is 0.0625; emphasize effect size, interval, and replication rather than a 0.05 threshold.", "mean_difference": 3.8, "mean_percent_change": 7.776904310111141, "median_difference": 0.0, "median_percent_change": 0.0, "objective": "lower", "paired_t_95pct_ci_for_mean_difference": [ -14.750309831629117, 22.35030983162912 ], "pairs": 5, "percent_changes": [ -6.451612903225806, 53.57142857142857, 0.0, -8.235294117647058, 0.0 ], "sample_sd_of_differences": 14.939879517586478, "treatment": [ 58.0, 86.0, 56.0, 78.0, 75.0 ], "treatment_minus_control": [ -4.0, 30.0, 0.0, -7.0, 0.0 ] }, "hwmon/hwmon6/coretemp/Core 8": { "control": [ 99.0, 98.0, 99.0, 98.0, 98.0 ], "exact_two_sided_sign_flip_p": 0.5, "geometric_mean_control_over_treatment_speedup": 0.9959472859157004, "geometric_mean_favorable_speedup": 0.9959472859157004, "geometric_mean_treatment_over_control": 1.0040692054103781, "higher_is_better": false, "inference_note": "With five pairs the minimum attainable exact two-sided sign-flip p-value is 0.0625; emphasize effect size, interval, and replication rather than a 0.05 threshold.", "mean_difference": 0.4, "mean_percent_change": 0.40816326530612246, "median_difference": 0.0, "median_percent_change": 0.0, "objective": "lower", "paired_t_95pct_ci_for_mean_difference": [ -0.280087354890164, 1.0800873548901642 ], "pairs": 5, "percent_changes": [ 0.0, 1.0204081632653061, 0.0, 0.0, 1.0204081632653061 ], "sample_sd_of_differences": 0.5477225575051661, "treatment": [ 99.0, 99.0, 99.0, 98.0, 99.0 ], "treatment_minus_control": [ 0.0, 1.0, 0.0, 0.0, 1.0 ] }, "hwmon/hwmon6/coretemp/Package id 0": { "control": [ 99.0, 98.0, 99.0, 98.0, 98.0 ], "exact_two_sided_sign_flip_p": 0.5, "geometric_mean_control_over_treatment_speedup": 0.9959472859157004, "geometric_mean_favorable_speedup": 0.9959472859157004, "geometric_mean_treatment_over_control": 1.0040692054103781, "higher_is_better": false, "inference_note": "With five pairs the minimum attainable exact two-sided sign-flip p-value is 0.0625; emphasize effect size, interval, and replication rather than a 0.05 threshold.", "mean_difference": 0.4, "mean_percent_change": 0.40816326530612246, "median_difference": 0.0, "median_percent_change": 0.0, "objective": "lower", "paired_t_95pct_ci_for_mean_difference": [ -0.280087354890164, 1.0800873548901642 ], "pairs": 5, "percent_changes": [ 0.0, 1.0204081632653061, 0.0, 0.0, 1.0204081632653061 ], "sample_sd_of_differences": 0.5477225575051661, "treatment": [ 99.0, 99.0, 99.0, 98.0, 99.0 ], "treatment_minus_control": [ 0.0, 1.0, 0.0, 0.0, 1.0 ] }, "hwmon/hwmon7/iwlwifi_1/temp1": { "control": [ 56.0, 57.0, 57.0, 57.0, 55.0 ], "exact_two_sided_sign_flip_p": 0.375, "geometric_mean_control_over_treatment_speedup": 0.979454401815342, "geometric_mean_favorable_speedup": 0.979454401815342, "geometric_mean_treatment_over_control": 1.0209765744546946, "higher_is_better": false, "inference_note": "With five pairs the minimum attainable exact two-sided sign-flip p-value is 0.0625; emphasize effect size, interval, and replication rather than a 0.05 threshold.", "mean_difference": 1.2, "mean_percent_change": 2.1430849851902485, "median_difference": 1.0, "median_percent_change": 1.8181818181818181, "objective": "lower", "paired_t_95pct_ci_for_mean_difference": [ -1.1883882976054163, 3.588388297605416 ], "pairs": 5, "percent_changes": [ 7.142857142857143, 3.508771929824561, 0.0, -1.7543859649122806, 1.8181818181818181 ], "sample_sd_of_differences": 1.9235384061671346, "treatment": [ 60.0, 59.0, 57.0, 56.0, 56.0 ], "treatment_minus_control": [ 4.0, 2.0, 0.0, -1.0, 1.0 ] }, "thermal/thermal_zone0/acpitz": { "control": [ 40.0, 40.0, 40.0, 40.0, 40.0 ], "exact_two_sided_sign_flip_p": 1.0, "geometric_mean_control_over_treatment_speedup": 1.0, "geometric_mean_favorable_speedup": 1.0, "geometric_mean_treatment_over_control": 1.0, "higher_is_better": false, "inference_note": "With five pairs the minimum attainable exact two-sided sign-flip p-value is 0.0625; emphasize effect size, interval, and replication rather than a 0.05 threshold.", "mean_difference": 0.0, "mean_percent_change": 0.0, "median_difference": 0.0, "median_percent_change": 0.0, "objective": "lower", "paired_t_95pct_ci_for_mean_difference": [ 0.0, 0.0 ], "pairs": 5, "percent_changes": [ 0.0, 0.0, 0.0, 0.0, 0.0 ], "sample_sd_of_differences": 0.0, "treatment": [ 40.0, 40.0, 40.0, 40.0, 40.0 ], "treatment_minus_control": [ 0.0, 0.0, 0.0, 0.0, 0.0 ] }, "thermal/thermal_zone1/acpitz": { "control": [ 27.8, 27.8, 27.8, 27.8, 27.8 ], "exact_two_sided_sign_flip_p": 1.0, "geometric_mean_control_over_treatment_speedup": 1.0, "geometric_mean_favorable_speedup": 1.0, "geometric_mean_treatment_over_control": 1.0, "higher_is_better": false, "inference_note": "With five pairs the minimum attainable exact two-sided sign-flip p-value is 0.0625; emphasize effect size, interval, and replication rather than a 0.05 threshold.", "mean_difference": 0.0, "mean_percent_change": 0.0, "median_difference": 0.0, "median_percent_change": 0.0, "objective": "lower", "paired_t_95pct_ci_for_mean_difference": [ 0.0, 0.0 ], "pairs": 5, "percent_changes": [ 0.0, 0.0, 0.0, 0.0, 0.0 ], "sample_sd_of_differences": 0.0, "treatment": [ 27.8, 27.8, 27.8, 27.8, 27.8 ], "treatment_minus_control": [ 0.0, 0.0, 0.0, 0.0, 0.0 ] }, "thermal/thermal_zone2/INT3400 Thermal": { "control": [ 20.0, 20.0, 20.0, 20.0, 20.0 ], "exact_two_sided_sign_flip_p": 1.0, "geometric_mean_control_over_treatment_speedup": 1.0, "geometric_mean_favorable_speedup": 1.0, "geometric_mean_treatment_over_control": 1.0, "higher_is_better": false, "inference_note": "With five pairs the minimum attainable exact two-sided sign-flip p-value is 0.0625; emphasize effect size, interval, and replication rather than a 0.05 threshold.", "mean_difference": 0.0, "mean_percent_change": 0.0, "median_difference": 0.0, "median_percent_change": 0.0, "objective": "lower", "paired_t_95pct_ci_for_mean_difference": [ 0.0, 0.0 ], "pairs": 5, "percent_changes": [ 0.0, 0.0, 0.0, 0.0, 0.0 ], "sample_sd_of_differences": 0.0, "treatment": [ 20.0, 20.0, 20.0, 20.0, 20.0 ], "treatment_minus_control": [ 0.0, 0.0, 0.0, 0.0, 0.0 ] }, "thermal/thermal_zone3/TCPU": { "control": [ 98.05, 97.05, 97.05, 97.05, 97.05 ], "exact_two_sided_sign_flip_p": 0.25, "geometric_mean_control_over_treatment_speedup": 0.9938681312980404, "geometric_mean_favorable_speedup": 0.9938681312980404, "geometric_mean_treatment_over_control": 1.0061697004953274, "higher_is_better": false, "inference_note": "With five pairs the minimum attainable exact two-sided sign-flip p-value is 0.0625; emphasize effect size, interval, and replication rather than a 0.05 threshold.", "mean_difference": 0.6, "mean_percent_change": 0.6182380216383309, "median_difference": 1.0, "median_percent_change": 1.0303967027305514, "objective": "lower", "paired_t_95pct_ci_for_mean_difference": [ -0.08008735489016405, 1.280087354890164 ], "pairs": 5, "percent_changes": [ 0.0, 1.0303967027305514, 1.0303967027305514, 0.0, 1.0303967027305514 ], "sample_sd_of_differences": 0.5477225575051661, "treatment": [ 98.05, 98.05, 98.05, 97.05, 98.05 ], "treatment_minus_control": [ 0.0, 1.0, 1.0, 0.0, 1.0 ] }, "thermal/thermal_zone4/TCPU_PCI": { "control": [ 98.0, 97.0, 97.0, 97.0, 97.0 ], "exact_two_sided_sign_flip_p": 0.25, "geometric_mean_control_over_treatment_speedup": 0.9938649963607276, "geometric_mean_favorable_speedup": 0.9938649963607276, "geometric_mean_treatment_over_control": 1.0061728742452316, "higher_is_better": false, "inference_note": "With five pairs the minimum attainable exact two-sided sign-flip p-value is 0.0625; emphasize effect size, interval, and replication rather than a 0.05 threshold.", "mean_difference": 0.6, "mean_percent_change": 0.6185567010309277, "median_difference": 1.0, "median_percent_change": 1.0309278350515463, "objective": "lower", "paired_t_95pct_ci_for_mean_difference": [ -0.08008735489016405, 1.280087354890164 ], "pairs": 5, "percent_changes": [ 0.0, 1.0309278350515463, 1.0309278350515463, 0.0, 1.0309278350515463 ], "sample_sd_of_differences": 0.5477225575051661, "treatment": [ 98.0, 98.0, 98.0, 97.0, 98.0 ], "treatment_minus_control": [ 0.0, 1.0, 1.0, 0.0, 1.0 ] }, "thermal/thermal_zone5/x86_pkg_temp": { "control": [ 99.0, 98.0, 98.0, 98.0, 98.0 ], "exact_two_sided_sign_flip_p": 0.25, "geometric_mean_control_over_treatment_speedup": 0.9939270922244149, "geometric_mean_favorable_speedup": 0.9939270922244149, "geometric_mean_treatment_over_control": 1.0061100133230034, "higher_is_better": false, "inference_note": "With five pairs the minimum attainable exact two-sided sign-flip p-value is 0.0625; emphasize effect size, interval, and replication rather than a 0.05 threshold.", "mean_difference": 0.6, "mean_percent_change": 0.6122448979591837, "median_difference": 1.0, "median_percent_change": 1.0204081632653061, "objective": "lower", "paired_t_95pct_ci_for_mean_difference": [ -0.08008735489016405, 1.280087354890164 ], "pairs": 5, "percent_changes": [ 0.0, 1.0204081632653061, 1.0204081632653061, 0.0, 1.0204081632653061 ], "sample_sd_of_differences": 0.5477225575051661, "treatment": [ 99.0, 99.0, 99.0, 98.0, 99.0 ], "treatment_minus_control": [ 0.0, 1.0, 1.0, 0.0, 1.0 ] }, "thermal/thermal_zone6/iwlwifi_1": { "control": [ 56.0, 57.0, 57.0, 57.0, 55.0 ], "exact_two_sided_sign_flip_p": 0.375, "geometric_mean_control_over_treatment_speedup": 0.979454401815342, "geometric_mean_favorable_speedup": 0.979454401815342, "geometric_mean_treatment_over_control": 1.0209765744546946, "higher_is_better": false, "inference_note": "With five pairs the minimum attainable exact two-sided sign-flip p-value is 0.0625; emphasize effect size, interval, and replication rather than a 0.05 threshold.", "mean_difference": 1.2, "mean_percent_change": 2.1430849851902485, "median_difference": 1.0, "median_percent_change": 1.8181818181818181, "objective": "lower", "paired_t_95pct_ci_for_mean_difference": [ -1.1883882976054163, 3.588388297605416 ], "pairs": 5, "percent_changes": [ 7.142857142857143, 3.508771929824561, 0.0, -1.7543859649122806, 1.8181818181818181 ], "sample_sd_of_differences": 1.9235384061671346, "treatment": [ 60.0, 59.0, 57.0, 56.0, 56.0 ], "treatment_minus_control": [ 4.0, 2.0, 0.0, -1.0, 1.0 ] } }, "pairs": 5, "primary_endpoint": "protocol_critical_path_s", "schema_version": 1, "small_n_limit": "N=5 estimates paired effect magnitude; minimum exact two-sided sign-flip p is 0.0625.", "status": "complete", "sweep_root": "/data/runs/2026-08-21_21-00_wuklab-NUC16_qwen05b-fp16-jit-full-1k8k-5pairs/jit_pairs" }, "arms": [ { "band": "4096-8192", "context_after": 8192, "decoded_tokens": 8064, "expected_decoded_tokens": 8064, "key": "direct", "label": "B0", "memory": { "npu_alloc": { "latest_bytes": 1540615472.0, "max_bytes": 1540615472.0, "time_weighted_mean_bytes": 1137634844.9661405 }, "process_rss": { "latest_bytes": 1739489280.0, "max_bytes": 1739489280.0, "time_weighted_mean_bytes": 1558882824.2995179 }, "samples": 131 }, "metrics": { "compile_model_wall_s": 65.122588684, "decode_infer_s": 155.474822801, "decode_wall_s": 159.242971728, "protocol_critical_path_s": 224.94137566499998, "tokens_per_s": 51.86691873784296 }, "pair": null, "phase": "complete", "position": 0, "progress_fraction": 1.0, "proof": { "boundary_logits_complete": true, "deadline_misses": [], "deadlines_met": true, "ok": true, "waited_capacities": [] }, "state": "complete", "timeline": { "boundaries": [ { "context_after": 1024, "infer_end_monotonic_ns": 7775260457008074, "infer_start_monotonic_ns": 7775260437634206, "latency_ms": 19.373868 }, { "context_after": 2048, "infer_end_monotonic_ns": 7775280727041040, "infer_start_monotonic_ns": 7775280707600158, "latency_ms": 19.440882 }, { "context_after": 4096, "infer_end_monotonic_ns": 7775321162556803, "infer_start_monotonic_ns": 7775321143435966, "latency_ms": 19.120837 } ], "compiles": [ { "capacity": 1024, "duration_s": null, "ready_offset_s": null, "start_offset_s": null, "state": "pending" }, { "capacity": 2048, "duration_s": null, "ready_offset_s": null, "start_offset_s": null, "state": "pending" }, { "capacity": 4096, "duration_s": null, "ready_offset_s": null, "start_offset_s": null, "state": "pending" }, { "capacity": 8192, "duration_s": 64.886313, "ready_offset_s": 64.886313134, "start_offset_s": 0.0, "state": "ready" } ], "decode_end_offset_s": 284.691730182, "decode_start_offset_s": 125.497605925, "origin_monotonic_ns": 7775117161343277 } }, { "band": "4096-8192", "context_after": 8192, "decoded_tokens": 8064, "expected_decoded_tokens": 8064, "key": "pyramid", "label": "B1", "memory": { "npu_alloc": { "latest_bytes": 1555617448.0, "max_bytes": 1555617448.0, "time_weighted_mean_bytes": 1060519485.8069065 }, "process_rss": { "latest_bytes": 2210148352.0, "max_bytes": 2210148352.0, "time_weighted_mean_bytes": 1916558931.4975204 }, "samples": 131 }, "metrics": { "compile_model_wall_s": 84.358093351, "decode_infer_s": 144.601785912, "decode_wall_s": 149.467930885, "protocol_critical_path_s": 234.41458535399997, "tokens_per_s": 55.76694609364985 }, "pair": null, "phase": "complete", "position": 1, "progress_fraction": 1.0, "proof": { "boundary_logits_complete": true, "deadline_misses": [], "deadlines_met": true, "ok": true, "waited_capacities": [] }, "state": "complete", "timeline": { "boundaries": [ { "context_after": 1024, "infer_end_monotonic_ns": 7776162322249056, "infer_start_monotonic_ns": 7776162292212847, "latency_ms": 30.036209 }, { "context_after": 2048, "infer_end_monotonic_ns": 7776179900046381, "infer_start_monotonic_ns": 7776179873946282, "latency_ms": 26.100099 }, { "context_after": 4096, "infer_end_monotonic_ns": 7776214746693945, "infer_start_monotonic_ns": 7776214711990143, "latency_ms": 34.703802 } ], "compiles": [ { "capacity": 1024, "duration_s": 83.802075, "ready_offset_s": 83.802075876, "start_offset_s": 0.0, "state": "ready" }, { "capacity": 2048, "duration_s": 77.240195, "ready_offset_s": 83.802089423, "start_offset_s": 6.561894398, "state": "ready" }, { "capacity": 4096, "duration_s": 70.554623, "ready_offset_s": 83.802097934, "start_offset_s": 13.24747472, "state": "ready" }, { "capacity": 8192, "duration_s": 64.050744, "ready_offset_s": 83.802107604, "start_offset_s": 19.751363382, "state": "ready" } ], "decode_end_offset_s": 293.878005087, "decode_start_offset_s": 144.455571083, "origin_monotonic_ns": 7776003019752829 } }, { "band": "4096-8192", "context_after": 8192, "decoded_tokens": 8064, "expected_decoded_tokens": 8064, "key": "precompiled", "label": "I1", "memory": { "npu_alloc": { "latest_bytes": 1643012840.0, "max_bytes": 1643012840.0, "time_weighted_mean_bytes": 1087559074.865844 }, "process_rss": { "latest_bytes": 2203516928.0, "max_bytes": 2203516928.0, "time_weighted_mean_bytes": 1909567818.1442642 }, "samples": 68 }, "metrics": { "compile_model_wall_s": 84.069572736, "decode_infer_s": 143.221490682, "decode_wall_s": 147.357470573, "protocol_critical_path_s": 231.993970971, "tokens_per_s": 56.304399302090765 }, "pair": 0, "phase": "complete", "position": 0, "progress_fraction": 1.0, "proof": { "boundary_logits_complete": true, "deadline_misses": [], "deadlines_met": true, "ok": true, "waited_capacities": [] }, "state": "complete", "timeline": { "boundaries": [ { "context_after": 1024, "infer_end_monotonic_ns": 7762984015727394, "infer_start_monotonic_ns": 7762983966982493, "latency_ms": 48.744901 }, { "context_after": 2048, "infer_end_monotonic_ns": 7763000541952727, "infer_start_monotonic_ns": 7763000483353637, "latency_ms": 58.59909 }, { "context_after": 4096, "infer_end_monotonic_ns": 7763034981596304, "infer_start_monotonic_ns": 7763034896281934, "latency_ms": 85.31437 } ], "compiles": [ { "capacity": 1024, "duration_s": 83.527895, "ready_offset_s": 83.527895638, "start_offset_s": 0.0, "state": "ready" }, { "capacity": 2048, "duration_s": 76.913245, "ready_offset_s": 83.527907442, "start_offset_s": 6.614662059, "state": "ready" }, { "capacity": 4096, "duration_s": 70.240319, "ready_offset_s": 83.527918185, "start_offset_s": 13.287598916, "state": "ready" }, { "capacity": 8192, "duration_s": 63.698396, "ready_offset_s": 83.527926039, "start_offset_s": 19.829529506, "state": "ready" } ], "decode_end_offset_s": 291.470434122, "decode_start_offset_s": 144.153630586, "origin_monotonic_ns": 7762825615884516 } }, { "band": "4096-8192", "context_after": 8192, "decoded_tokens": 8064, "expected_decoded_tokens": 8064, "key": "one_ahead_jit", "label": "I2", "memory": { "npu_alloc": { "latest_bytes": 1693770472.0, "max_bytes": 1693770472.0, "time_weighted_mean_bytes": 1174882636.1609864 }, "process_rss": { "latest_bytes": 2296958976.0, "max_bytes": 2296958976.0, "time_weighted_mean_bytes": 1841318340.1265788 }, "samples": 68 }, "metrics": { "compile_model_wall_s": 65.501335193, "decode_infer_s": 142.540816634, "decode_wall_s": 146.772048491, "protocol_critical_path_s": 212.83618146, "tokens_per_s": 56.5732692601714 }, "pair": 0, "phase": "complete", "position": 1, "progress_fraction": 1.0, "proof": { "boundary_logits_complete": true, "deadline_misses": [], "deadlines_met": true, "ok": true, "waited_capacities": [] }, "state": "complete", "timeline": { "boundaries": [ { "context_after": 1024, "infer_end_monotonic_ns": 7763558369038749, "infer_start_monotonic_ns": 7763558302147074, "latency_ms": 66.891675 }, { "context_after": 2048, "infer_end_monotonic_ns": 7763575040457148, "infer_start_monotonic_ns": 7763574968713120, "latency_ms": 71.744028 }, { "context_after": 4096, "infer_end_monotonic_ns": 7763609574310886, "infer_start_monotonic_ns": 7763609487189870, "latency_ms": 87.121016 } ], "compiles": [ { "capacity": 1024, "duration_s": 64.964188, "ready_offset_s": 64.964188804, "start_offset_s": 0.0, "state": "ready" }, { "capacity": 2048, "duration_s": 6.780602, "ready_offset_s": 132.348392899, "start_offset_s": 125.567790441, "state": "ready" }, { "capacity": 4096, "duration_s": 6.93596, "ready_offset_s": 146.945667757, "start_offset_s": 140.009707689, "state": "ready" }, { "capacity": 8192, "duration_s": 7.113776, "ready_offset_s": 163.795709356, "start_offset_s": 156.681932837, "state": "ready" } ], "decode_end_offset_s": 272.295810629, "decode_start_offset_s": 125.567600328, "origin_monotonic_ns": 7763418329332172 } }, { "band": "4096-8192", "context_after": 8192, "decoded_tokens": 8064, "expected_decoded_tokens": 8064, "key": "precompiled", "label": "I1", "memory": { "npu_alloc": { "latest_bytes": 1643012840.0, "max_bytes": 1643012840.0, "time_weighted_mean_bytes": 1087345082.0252256 }, "process_rss": { "latest_bytes": 2166091776.0, "max_bytes": 2166091776.0, "time_weighted_mean_bytes": 1877492560.230563 }, "samples": 68 }, "metrics": { "compile_model_wall_s": 84.331265631, "decode_infer_s": 143.735924396, "decode_wall_s": 147.697927908, "protocol_critical_path_s": 232.85763407599998, "tokens_per_s": 56.10288474427073 }, "pair": 1, "phase": "complete", "position": 1, "progress_fraction": 1.0, "proof": { "boundary_logits_complete": true, "deadline_misses": [], "deadlines_met": true, "ok": true, "waited_capacities": [] }, "state": "complete", "timeline": { "boundaries": [ { "context_after": 1024, "infer_end_monotonic_ns": 7769980799996225, "infer_start_monotonic_ns": 7769980738978464, "latency_ms": 61.017761 }, { "context_after": 2048, "infer_end_monotonic_ns": 7769997699218915, "infer_start_monotonic_ns": 7769997632716747, "latency_ms": 66.502168 }, { "context_after": 4096, "infer_end_monotonic_ns": 7770032367253645, "infer_start_monotonic_ns": 7770032283459853, "latency_ms": 83.793792 } ], "compiles": [ { "capacity": 1024, "duration_s": 83.797482, "ready_offset_s": 83.797482764, "start_offset_s": 0.0, "state": "ready" }, { "capacity": 2048, "duration_s": 77.178993, "ready_offset_s": 83.797495367, "start_offset_s": 6.61850199, "state": "ready" }, { "capacity": 4096, "duration_s": 70.461166, "ready_offset_s": 83.797504724, "start_offset_s": 13.336337743, "state": "ready" }, { "capacity": 8192, "duration_s": 63.86546, "ready_offset_s": 83.797511236, "start_offset_s": 19.932051158, "state": "ready" } ], "decode_end_offset_s": 292.339587917, "decode_start_offset_s": 144.679903839, "origin_monotonic_ns": 7769821554363754 } }, { "band": "4096-8192", "context_after": 8192, "decoded_tokens": 8064, "expected_decoded_tokens": 8064, "key": "one_ahead_jit", "label": "I2", "memory": { "npu_alloc": { "latest_bytes": 1693770472.0, "max_bytes": 1693770472.0, "time_weighted_mean_bytes": 1173032891.7300577 }, "process_rss": { "latest_bytes": 2275688448.0, "max_bytes": 2275688448.0, "time_weighted_mean_bytes": 1829705937.2295692 }, "samples": 68 }, "metrics": { "compile_model_wall_s": 65.834978455, "decode_infer_s": 142.686428737, "decode_wall_s": 146.503445813, "protocol_critical_path_s": 212.899580796, "tokens_per_s": 56.515536000018514 }, "pair": 1, "phase": "complete", "position": 0, "progress_fraction": 1.0, "proof": { "boundary_logits_complete": true, "deadline_misses": [], "deadlines_met": true, "ok": true, "waited_capacities": [] }, "state": "complete", "timeline": { "boundaries": [ { "context_after": 1024, "infer_end_monotonic_ns": 7769388185084007, "infer_start_monotonic_ns": 7769388124086638, "latency_ms": 60.997369 }, { "context_after": 2048, "infer_end_monotonic_ns": 7769404897186935, "infer_start_monotonic_ns": 7769404829235732, "latency_ms": 67.951203 }, { "context_after": 4096, "infer_end_monotonic_ns": 7769439095275169, "infer_start_monotonic_ns": 7769439010197246, "latency_ms": 85.077923 } ], "compiles": [ { "capacity": 1024, "duration_s": 65.294471, "ready_offset_s": 65.294471674, "start_offset_s": 0.0, "state": "ready" }, { "capacity": 2048, "duration_s": 6.8051, "ready_offset_s": 132.692743721, "start_offset_s": 125.887643209, "state": "ready" }, { "capacity": 4096, "duration_s": 7.011376, "ready_offset_s": 147.238633221, "start_offset_s": 140.227256636, "state": "ready" }, { "capacity": 8192, "duration_s": 7.014014, "ready_offset_s": 163.954190702, "start_offset_s": 156.940176648, "state": "ready" } ], "decode_end_offset_s": 272.349669797, "decode_start_offset_s": 125.887439319, "origin_monotonic_ns": 7769247925905430 } }, { "band": "4096-8192", "context_after": 8192, "decoded_tokens": 8064, "expected_decoded_tokens": 8064, "key": "precompiled", "label": "I1", "memory": { "npu_alloc": { "latest_bytes": 1643012840.0, "max_bytes": 1643012840.0, "time_weighted_mean_bytes": 1089056427.7306566 }, "process_rss": { "latest_bytes": 2173329408.0, "max_bytes": 2173329408.0, "time_weighted_mean_bytes": 1884727299.8957305 }, "samples": 68 }, "metrics": { "compile_model_wall_s": 83.708544242, "decode_infer_s": 143.152000032, "decode_wall_s": 147.449665453, "protocol_critical_path_s": 231.730203078, "tokens_per_s": 56.33173129399089 }, "pair": 2, "phase": "complete", "position": 1, "progress_fraction": 1.0, "proof": { "boundary_logits_complete": true, "deadline_misses": [], "deadlines_met": true, "ok": true, "waited_capacities": [] }, "state": "complete", "timeline": { "boundaries": [ { "context_after": 1024, "infer_end_monotonic_ns": 7771446992908236, "infer_start_monotonic_ns": 7771446944576227, "latency_ms": 48.332009 }, { "context_after": 2048, "infer_end_monotonic_ns": 7771463476228705, "infer_start_monotonic_ns": 7771463405974215, "latency_ms": 70.25449 }, { "context_after": 4096, "infer_end_monotonic_ns": 7771498106608495, "infer_start_monotonic_ns": 7771498011512335, "latency_ms": 95.09616 } ], "compiles": [ { "capacity": 1024, "duration_s": 83.145903, "ready_offset_s": 83.145903722, "start_offset_s": 0.0, "state": "ready" }, { "capacity": 2048, "duration_s": 76.642736, "ready_offset_s": 83.14592758, "start_offset_s": 6.503190809, "state": "ready" }, { "capacity": 4096, "duration_s": 70.001368, "ready_offset_s": 83.145939575, "start_offset_s": 13.144571161, "state": "ready" }, { "capacity": 8192, "duration_s": 63.470734, "ready_offset_s": 83.145954144, "start_offset_s": 19.675219188, "state": "ready" } ], "decode_end_offset_s": 291.208510186, "decode_start_offset_s": 143.804145913, "origin_monotonic_ns": 7771288878215960 } }, { "band": "4096-8192", "context_after": 8192, "decoded_tokens": 8064, "expected_decoded_tokens": 8064, "key": "one_ahead_jit", "label": "I2", "memory": { "npu_alloc": { "latest_bytes": 1693770472.0, "max_bytes": 1693770472.0, "time_weighted_mean_bytes": 1174736782.1317136 }, "process_rss": { "latest_bytes": 2283999232.0, "max_bytes": 2283999232.0, "time_weighted_mean_bytes": 1833502908.3880005 }, "samples": 68 }, "metrics": { "compile_model_wall_s": 65.534717215, "decode_infer_s": 142.646870143, "decode_wall_s": 146.728085304, "protocol_critical_path_s": 212.829373238, "tokens_per_s": 56.531208794949634 }, "pair": 2, "phase": "complete", "position": 0, "progress_fraction": 1.0, "proof": { "boundary_logits_complete": true, "deadline_misses": [], "deadlines_met": true, "ok": true, "waited_capacities": [] }, "state": "complete", "timeline": { "boundaries": [ { "context_after": 1024, "infer_end_monotonic_ns": 7770855248234721, "infer_start_monotonic_ns": 7770855190209813, "latency_ms": 58.024908 }, { "context_after": 2048, "infer_end_monotonic_ns": 7770872120679842, "infer_start_monotonic_ns": 7770872049724053, "latency_ms": 70.955789 }, { "context_after": 4096, "infer_end_monotonic_ns": 7770906292795129, "infer_start_monotonic_ns": 7770906206714011, "latency_ms": 86.081118 } ], "compiles": [ { "capacity": 1024, "duration_s": 65.003948, "ready_offset_s": 65.003948063, "start_offset_s": 0.0, "state": "ready" }, { "capacity": 2048, "duration_s": 6.751958, "ready_offset_s": 132.355152545, "start_offset_s": 125.60319411, "state": "ready" }, { "capacity": 4096, "duration_s": 6.805981, "ready_offset_s": 146.68298965, "start_offset_s": 139.877008419, "state": "ready" }, { "capacity": 8192, "duration_s": 6.946282, "ready_offset_s": 163.695507499, "start_offset_s": 156.749225155, "state": "ready" } ], "decode_end_offset_s": 272.288762549, "decode_start_offset_s": 125.603030396, "origin_monotonic_ns": 7770715342292595 } }, { "band": "4096-8192", "context_after": 8192, "decoded_tokens": 8064, "expected_decoded_tokens": 8064, "key": "precompiled", "label": "I1", "memory": { "npu_alloc": { "latest_bytes": 1643012840.0, "max_bytes": 1643012840.0, "time_weighted_mean_bytes": 1083877824.5111783 }, "process_rss": { "latest_bytes": 2175143936.0, "max_bytes": 2175143936.0, "time_weighted_mean_bytes": 1882871556.69491 }, "samples": 68 }, "metrics": { "compile_model_wall_s": 85.220518564, "decode_infer_s": 143.259762506, "decode_wall_s": 147.753727422, "protocol_critical_path_s": 233.54898441, "tokens_per_s": 56.289357590288226 }, "pair": 3, "phase": "complete", "position": 0, "progress_fraction": 1.0, "proof": { "boundary_logits_complete": true, "deadline_misses": [], "deadlines_met": true, "ok": true, "waited_capacities": [] }, "state": "complete", "timeline": { "boundaries": [ { "context_after": 1024, "infer_end_monotonic_ns": 7772341588250317, "infer_start_monotonic_ns": 7772341532709532, "latency_ms": 55.540785 }, { "context_after": 2048, "infer_end_monotonic_ns": 7772358480588039, "infer_start_monotonic_ns": 7772358418712541, "latency_ms": 61.875498 }, { "context_after": 4096, "infer_end_monotonic_ns": 7772392524178503, "infer_start_monotonic_ns": 7772392437096076, "latency_ms": 87.082427 } ], "compiles": [ { "capacity": 1024, "duration_s": 84.646907, "ready_offset_s": 84.646907086, "start_offset_s": 0.0, "state": "ready" }, { "capacity": 2048, "duration_s": 78.016774, "ready_offset_s": 84.646929875, "start_offset_s": 6.630155353, "state": "ready" }, { "capacity": 4096, "duration_s": 71.258774, "ready_offset_s": 84.646940579, "start_offset_s": 13.388165688, "state": "ready" }, { "capacity": 8192, "duration_s": 64.633991, "ready_offset_s": 84.646950342, "start_offset_s": 20.012958923, "state": "ready" } ], "decode_end_offset_s": 293.007855794, "decode_start_offset_s": 145.299774135, "origin_monotonic_ns": 7772181594356643 } }, { "band": "4096-8192", "context_after": 8192, "decoded_tokens": 8064, "expected_decoded_tokens": 8064, "key": "one_ahead_jit", "label": "I2", "memory": { "npu_alloc": { "latest_bytes": 1693770472.0, "max_bytes": 1693770472.0, "time_weighted_mean_bytes": 1178420798.7964487 }, "process_rss": { "latest_bytes": 2293116928.0, "max_bytes": 2293116928.0, "time_weighted_mean_bytes": 1845553309.7602115 }, "samples": 68 }, "metrics": { "compile_model_wall_s": 65.045789129, "decode_infer_s": 143.414261374, "decode_wall_s": 147.689448057, "protocol_critical_path_s": 213.301850105, "tokens_per_s": 56.22871758179237 }, "pair": 3, "phase": "complete", "position": 1, "progress_fraction": 1.0, "proof": { "boundary_logits_complete": true, "deadline_misses": [], "deadlines_met": true, "ok": true, "waited_capacities": [] }, "state": "complete", "timeline": { "boundaries": [ { "context_after": 1024, "infer_end_monotonic_ns": 7772915488825791, "infer_start_monotonic_ns": 7772915437030606, "latency_ms": 51.795185 }, { "context_after": 2048, "infer_end_monotonic_ns": 7772932440670060, "infer_start_monotonic_ns": 7772932368267546, "latency_ms": 72.402514 }, { "context_after": 4096, "infer_end_monotonic_ns": 7772966983685195, "infer_start_monotonic_ns": 7772966897862692, "latency_ms": 85.822503 } ], "compiles": [ { "capacity": 1024, "duration_s": 64.506586, "ready_offset_s": 64.506586727, "start_offset_s": 0.0, "state": "ready" }, { "capacity": 2048, "duration_s": 6.776207, "ready_offset_s": 131.886416676, "start_offset_s": 125.11020902, "state": "ready" }, { "capacity": 4096, "duration_s": 6.813044, "ready_offset_s": 146.404660714, "start_offset_s": 139.591616177, "state": "ready" }, { "capacity": 8192, "duration_s": 7.04211, "ready_offset_s": 163.581189954, "start_offset_s": 156.539079032, "state": "ready" } ], "decode_end_offset_s": 272.752836288, "decode_start_offset_s": 125.109857508, "origin_monotonic_ns": 7772775870484736 } }, { "band": "4096-8192", "context_after": 8192, "decoded_tokens": 8064, "expected_decoded_tokens": 8064, "key": "precompiled", "label": "I1", "memory": { "npu_alloc": { "latest_bytes": 1643012840.0, "max_bytes": 1643012840.0, "time_weighted_mean_bytes": 1087292973.6158884 }, "process_rss": { "latest_bytes": 2116251648.0, "max_bytes": 2116251648.0, "time_weighted_mean_bytes": 1834834792.0701056 }, "samples": 68 }, "metrics": { "compile_model_wall_s": 84.096977893, "decode_infer_s": 143.207379556, "decode_wall_s": 147.363224232, "protocol_critical_path_s": 232.031900832, "tokens_per_s": 56.30994732954137 }, "pair": 4, "phase": "complete", "position": 0, "progress_fraction": 1.0, "proof": { "boundary_logits_complete": true, "deadline_misses": [], "deadlines_met": true, "ok": true, "waited_capacities": [] }, "state": "complete", "timeline": { "boundaries": [ { "context_after": 1024, "infer_end_monotonic_ns": 7773808861996258, "infer_start_monotonic_ns": 7773808815399368, "latency_ms": 46.59689 }, { "context_after": 2048, "infer_end_monotonic_ns": 7773825572195687, "infer_start_monotonic_ns": 7773825512502096, "latency_ms": 59.693591 }, { "context_after": 4096, "infer_end_monotonic_ns": 7773860431732712, "infer_start_monotonic_ns": 7773860337175542, "latency_ms": 94.55717 } ], "compiles": [ { "capacity": 1024, "duration_s": 83.546544, "ready_offset_s": 83.546544601, "start_offset_s": 0.0, "state": "ready" }, { "capacity": 2048, "duration_s": 77.00702, "ready_offset_s": 83.546559489, "start_offset_s": 6.539539385, "state": "ready" }, { "capacity": 4096, "duration_s": 70.342771, "ready_offset_s": 83.546567004, "start_offset_s": 13.203795898, "state": "ready" }, { "capacity": 8192, "duration_s": 63.796945, "ready_offset_s": 83.546578238, "start_offset_s": 19.749632566, "state": "ready" } ], "decode_end_offset_s": 291.504325247, "decode_start_offset_s": 144.185320409, "origin_monotonic_ns": 7773650183856139 } }, { "band": "4096-8192", "context_after": 8192, "decoded_tokens": 8064, "expected_decoded_tokens": 8064, "key": "one_ahead_jit", "label": "I2", "memory": { "npu_alloc": { "latest_bytes": 1693770472.0, "max_bytes": 1693770472.0, "time_weighted_mean_bytes": 1180050985.4406555 }, "process_rss": { "latest_bytes": 2293231616.0, "max_bytes": 2293231616.0, "time_weighted_mean_bytes": 1847491285.1281364 }, "samples": 68 }, "metrics": { "compile_model_wall_s": 64.72646805, "decode_infer_s": 143.389227744, "decode_wall_s": 147.717440591, "protocol_critical_path_s": 213.009857613, "tokens_per_s": 56.23853428095076 }, "pair": 4, "phase": "complete", "position": 1, "progress_fraction": 1.0, "proof": { "boundary_logits_complete": true, "deadline_misses": [], "deadlines_met": true, "ok": true, "waited_capacities": [] }, "state": "complete", "timeline": { "boundaries": [ { "context_after": 1024, "infer_end_monotonic_ns": 7774382060912707, "infer_start_monotonic_ns": 7774382004339958, "latency_ms": 56.572749 }, { "context_after": 2048, "infer_end_monotonic_ns": 7774398857792654, "infer_start_monotonic_ns": 7774398787385059, "latency_ms": 70.407595 }, { "context_after": 4096, "infer_end_monotonic_ns": 7774433482373224, "infer_start_monotonic_ns": 7774433395445897, "latency_ms": 86.927327 } ], "compiles": [ { "capacity": 1024, "duration_s": 64.202456, "ready_offset_s": 64.202456847, "start_offset_s": 0.0, "state": "ready" }, { "capacity": 2048, "duration_s": 6.704558, "ready_offset_s": 131.509407922, "start_offset_s": 124.80484961, "state": "ready" }, { "capacity": 4096, "duration_s": 6.76885, "ready_offset_s": 145.831595505, "start_offset_s": 139.062744755, "state": "ready" }, { "capacity": 8192, "duration_s": 6.995096, "ready_offset_s": 162.852546522, "start_offset_s": 155.857449551, "state": "ready" } ], "decode_end_offset_s": 272.474897044, "decode_start_offset_s": 124.804638784, "origin_monotonic_ns": 7774242970928779 } } ], "arms_complete": 12, "arms_planned": 12, "campaign_complete_at": "2026-08-22T00:51:04.325774665Z", "campaign_events": [ { "event": "preflight_complete", "wall_time_ns": 1787346391761009770 }, { "event": "jit_pairs_start", "wall_time_ns": 1787346391764402002 }, { "event": "jit_pairs_complete", "wall_time_ns": 1787358083684446243 }, { "event": "b0_start", "wall_time_ns": 1787358683688941354 }, { "event": "b0_complete", "wall_time_ns": 1787358969234707932 }, { "event": "b1_start", "wall_time_ns": 1787359569239573014 }, { "event": "b1_complete", "wall_time_ns": 1787359864321710774 }, { "event": "campaign_complete", "wall_time_ns": 1787359864325774665 } ], "capacities": [ 1024, 2048, 4096, 8192 ], "completed_pair_reports": 5, "current_arm": null, "failure": null, "final_effects": { "b1_minus_b0": { "compile_s": 19.235504667, "decode_infer_s": -10.87303688899999, "decode_wall_s": -9.775040843, "prefill_s": 0.007575420999999971, "protocol_critical_path_s": 9.473209688999987, "read_model_s": 0.001977965000000005, "request_create_s": 0.003192478999999998 }, "background_successor_compile_s": { "capacities": [ 2048, 4096, 8192 ], "deadline_wait_s": 0.0, "mean": 20.6453314 }, "compile_saved_s": { "ci95": [ 17.949781671385427, 19.963654738214576 ], "mean": 18.9567182048, "percent": 22.487516285221286, "signflip_p": 0.0625 }, "decode_infer_i2_minus_i1_s": { "ci95": [ -1.0471918166930716, 0.2876108006930901 ], "mean": -0.3797905079999907, "percent": -0.26469001649047996, "signflip_p": 0.25 }, "decode_wall_i2_minus_i1_s": { "ci95": [ -1.1877208723650616, 0.3031019395650695 ], "mean": -0.44230946639999613, "percent": -0.2997044120459554, "signflip_p": 0.1875 }, "first_token_ready_saved_s": { "ci95": [ 18.043832646125868, 19.98467241787412 ], "mean": 19.014252531999993, "percent": 22.38312755747306, "signflip_p": 0.0625 }, "launcher_e2e_saved_s": { "ci95": [ 18.71938589567684, 20.167110051523185 ], "mean": 19.443247973600013, "percent": 6.632460508417545, "signflip_p": 0.0625 }, "protocol_critical_path_saved_s": { "ci95": [ 18.70614139426977, 20.208198667730215 ], "mean": 19.457170030999993, "percent": 8.370512476646716, "signflip_p": 0.0625 } }, "generated_at": "2026-08-22T00:57:43.941838+00:00", "latest_pair": 4, "measurement_scope": { "B0_B1": "one refreshed reference run each", "I1_I2": "five counterbalanced paired runs", "memory": "runner telemetry over each arm; RSS trapezoid, NPU allocation left-hold", "trajectory": "128-token prompt through context 8192" }, "memory_comparison": { "npu_alloc_time_weighted_mean_bytes": { "B0": { "mean_bytes": 1137634844.9661405, "runs": 1 }, "B1": { "mean_bytes": 1060519485.8069065, "runs": 1 }, "I1": { "mean_bytes": 1087026276.5497587, "runs": 5 }, "I2": { "mean_bytes": 1176224818.8519723, "runs": 5 } }, "process_rss_time_weighted_mean_bytes": { "B0": { "mean_bytes": 1558882824.2995179, "runs": 1 }, "B1": { "mean_bytes": 1916558931.4975204, "runs": 1 }, "I1": { "mean_bytes": 1877898805.4071147, "runs": 5 }, "I2": { "mean_bytes": 1839514356.1264992, "runs": 5 } } }, "overall_progress_fraction": 1.0, "pairs_planned": 5, "partial_results": { "compile_model_wall_s": { "i1_mean_s": 84.2853758132, "i2_mean_s": 65.3286576084, "i2_minus_i1_mean_s": -18.9567182048, "mean_percent_change": -22.487516285221286, "pairs": 5 }, "decode_infer_s": { "i1_mean_s": 143.3153114344, "i2_mean_s": 142.93552092640002, "i2_minus_i1_mean_s": -0.3797905079999907, "mean_percent_change": -0.26469001649047996, "pairs": 5 }, "decode_wall_s": { "i1_mean_s": 147.5244031176, "i2_mean_s": 147.0820936512, "i2_minus_i1_mean_s": -0.44230946639999613, "mean_percent_change": -0.2997044120459554, "pairs": 5 }, "protocol_critical_path_s": { "i1_mean_s": 232.4325386734, "i2_mean_s": 212.9753686424, "i2_minus_i1_mean_s": -19.457170030999993, "mean_percent_change": -8.370512476646716, "pairs": 5 } }, "prompt_length": 128, "quality": { "all_five_pairs_valid": true, "all_measured_arms_complete": true, "b0_ok": true, "b1_ok": true, "boundary_comparisons": 20, "i1_i2_boundary_logits_exact": true, "i2_all_deadlines_met": true, "minimum_cosine": 1.0, "postrun_model_hash_ok": true }, "reference_run_id": "2026-08-21_21-00_wuklab-NUC16_qwen05b-fp16-jit-full-1k8k-5pairs/references", "run_id": "2026-08-21_21-00_wuklab-NUC16_qwen05b-fp16-jit-full-1k8k-5pairs", "schema_version": 2, "state": "complete", "uncertainty": { "pairs": 5, "statement": "N=5 paired, counterbalanced runs. Values show paired means with paired-t 95% CIs; the minimum attainable exact two-sided sign-flip p-value is 0.0625." } }