{ "claim_boundary": "Development-only confirmation on an already-consumed P1 task. Passing can open a newly registered matched pilot but cannot be reported as held-out tuning evidence.", "decision": "STOP_NO_PROSPECTIVE_RESPONSE", "frozen_gate": { "expected_action_pairs": 12, "expected_repeat_pairs": 24, "minimum_efficacy_balanced_accuracy": 0.75, "minimum_efficacy_class": 4, "minimum_efficacy_delta_over_best_outcome": 0.15, "minimum_stable_response_features": 2, "response_thresholds_identical_to_phase6_v0": true }, "horizons": { "10": { "actions": [ { "delta_outcome": { "admitted_fraction": 0.0, "admitted_input_tokens_mean_over_limit": 0.0, "completed_fail_fraction_of_total": 0.0, "completed_over_admitted": 0.0, "completed_pass_rate": 0.0, "outstanding_over_admitted": 0.0, "tpot_max_over_slo": 0.04424092126093565, "tpot_mean_over_slo": 0.008821253352839065, "ttft_max_over_slo_max": 0.001164793997304514, "ttft_mean_over_slo_max": -0.0007446444380963049 }, "delta_state": { "batch_size.mean": 0.028437757317164802, "batch_tokens.mean": -0.16003443371509718, "decode_batch_size.mean": 0.02853132719514928, "graph_full_share": -0.0012762931357137575, "graph_none_share": -0.0069204281757616545, "graph_padding_fraction": 0.006029111786335234, "kv_usage_end_minus_start": 0.00036424504378218536, "kv_usage_max": 0.009254174352690048, "kv_usage_mean": 0.00277340001356393, "preemptions": 0.0, "prefill_token_fraction": -0.0003954091503666213, "queue_running_mean": 0.21264213909999974, "queue_waiting_mean": -0.073977916, "scheduler_steps_per_s": 0.20000000000000284 }, "full_action_efficacy": 0, "full_feasibility_transition": "false->false", "group": { "level": "high", "offered_rate_per_gpu": 2.9833333333333334, "replicate": 1, "request_hash": "3f0ab9be34dbee884f5dbe2135b983d31a47578a43f606a5f390b0528dbedbe3", "tp": 1 }, "kind": "matched_mns_increase", "source": { "cell": "tp1_mns8", "early_stopped": true, "full_feasible": false, "full_pass_rate": 0.20670391061452514, "mns": 8, "requests_sha256": "5f5bfa4a16846190f488d3369eed89ae895804145cea59197e6b544b703223fc", "result_sha256": "f65e3dc91d17f5caab1f4aecf97d2c464b99f107231af2cfd96f5484880136ee", "trial_id": "cells/tp1_mns8/high-rep1/result.json" }, "target": { "cell": "tp1_mns64", "early_stopped": true, "full_feasible": false, "full_pass_rate": 0.24581005586592178, "mns": 64, "requests_sha256": "e970dcac20144cb1e13a35997377f6695a9c57d788a88bad7f18cf6c64e35cec", "result_sha256": "3233686b60b920dcd6cc2f45193a81f3af2d46f781ef42c4d117630612084feb", "trial_id": "cells/tp1_mns64/high-rep1/result.json" } }, { "delta_outcome": { "admitted_fraction": 0.0, "admitted_input_tokens_mean_over_limit": 0.0, "completed_fail_fraction_of_total": 0.0, "completed_over_admitted": -0.032258064516129004, "completed_pass_rate": 0.0, "outstanding_over_admitted": 0.03225806451612906, "tpot_max_over_slo": 0.33050636819556467, "tpot_mean_over_slo": 0.2980276034718437, "ttft_max_over_slo_max": -0.14499169250484553, "ttft_mean_over_slo_max": -0.08870032973580096 }, "delta_state": { "batch_size.mean": 4.5469521321594915, "batch_tokens.mean": 177.93791776288268, "decode_batch_size.mean": 4.490206104102559, "graph_full_share": -0.049141465319289424, "graph_none_share": 0.04533918395046808, "graph_padding_fraction": 0.004209922294522723, "kv_usage_end_minus_start": 0.05537535994708087, "kv_usage_max": 0.08922043547542868, "kv_usage_mean": 0.04639671324585863, "preemptions": 0.0, "prefill_token_fraction": 0.003967110353963799, "queue_running_mean": 4.6810500217, "queue_waiting_mean": -2.3262808059, "scheduler_steps_per_s": -14.599999999999998 }, "full_action_efficacy": 0, "full_feasibility_transition": "false->false", "group": { "level": "high", "offered_rate_per_gpu": 2.9833333333333334, "replicate": 2, "request_hash": "b19c91876a8417fe06cb2775f65a1111b202f88ce10ec33576af3d84c059f02a", "tp": 1 }, "kind": "matched_mns_increase", "source": { "cell": "tp1_mns8", "early_stopped": true, "full_feasible": false, "full_pass_rate": 0.7988826815642458, "mns": 8, "requests_sha256": "e802d09057de349e5164c18314fcad09dc88c9673834fd6e5b100a57dd12fbf4", "result_sha256": "4835a985dd6bb8dc50b53737baa980ac95a49e22f0b8a7326e1fb416c9b4e798", "trial_id": "cells/tp1_mns8/high-rep2/result.json" }, "target": { "cell": "tp1_mns64", "early_stopped": true, "full_feasible": false, "full_pass_rate": 0.88268156424581, "mns": 64, "requests_sha256": "ec06878e2aefdc4853d972197cedc29ab94df7147abb2701e0d88b5dabf29728", "result_sha256": "8215477a014f061c55e94d879a25bbf1676ee2ef1c5e81efadd23d20d337e07a", "trial_id": "cells/tp1_mns64/high-rep2/result.json" } }, { "delta_outcome": { "admitted_fraction": 0.0, "admitted_input_tokens_mean_over_limit": 0.0, "completed_fail_fraction_of_total": 0.0, "completed_over_admitted": 0.0, "completed_pass_rate": 0.0, "outstanding_over_admitted": 0.0, "tpot_max_over_slo": 0.018804627718169264, "tpot_mean_over_slo": 0.011575705913082518, "ttft_max_over_slo_max": 0.0012841941691779934, "ttft_mean_over_slo_max": -0.0028507176665395206 }, "delta_state": { "batch_size.mean": 0.028484697519924218, "batch_tokens.mean": 0.386581442994256, "decode_batch_size.mean": 0.02831307423779572, "graph_full_share": -0.00017162328212849687, "graph_none_share": -0.003114247473621861, "graph_padding_fraction": 0.003200143463807169, "kv_usage_end_minus_start": 8.07298706054782e-05, "kv_usage_max": 0.0034809583857736692, "kv_usage_mean": 0.0010653256527363222, "preemptions": 0.0, "prefill_token_fraction": -0.00013286942159473902, "queue_running_mean": 0.1416205103000001, "queue_waiting_mean": -0.028138744, "scheduler_steps_per_s": -0.6000000000000085 }, "full_action_efficacy": 0, "full_feasibility_transition": "false->false", "group": { "level": "high", "offered_rate_per_gpu": 2.9833333333333334, "replicate": 3, "request_hash": "a85849f1f057a67c82523cff57a52af22d2eb9514a953e076990aa6c25c1ef7e", "tp": 1 }, "kind": "matched_mns_increase", "source": { "cell": "tp1_mns8", "early_stopped": true, "full_feasible": false, "full_pass_rate": 0.6424581005586593, "mns": 8, "requests_sha256": "cf1226185845cd8aa4e7ead0cb5c95c77eda62b84851a2ab579f984bbb88e8ac", "result_sha256": "95d5103f2678eeb99c258870324566932e2f2853c3b6e0ee94bc6c538f6ed865", "trial_id": "cells/tp1_mns8/high-rep3/result.json" }, "target": { "cell": "tp1_mns64", "early_stopped": true, "full_feasible": false, "full_pass_rate": 0.9273743016759777, "mns": 64, "requests_sha256": "18d72d1fe16136359e28a1909e395ffd553dbe3edd2c0fd93a9cc270a5ec6d60", "result_sha256": "480a3b50acaf9b9c1902e2669cf3aa7723a6703a00778da269b5b9757990ceef", "trial_id": "cells/tp1_mns64/high-rep3/result.json" } }, { "delta_outcome": { "admitted_fraction": 0.0, "admitted_input_tokens_mean_over_limit": 0.0, "completed_fail_fraction_of_total": 0.0, "completed_over_admitted": 0.0, "completed_pass_rate": 0.0, "outstanding_over_admitted": 0.0, "tpot_max_over_slo": -0.0040397713397167245, "tpot_mean_over_slo": -0.007906587129236364, "ttft_max_over_slo_max": -0.00039426567188155515, "ttft_mean_over_slo_max": -0.0011763758957386017 }, "delta_state": { "batch_size.mean": -0.08160779537149798, "batch_tokens.mean": -1.7609001853017006, "decode_batch_size.mean": -0.08094394190416265, "graph_full_share": 0.0006306607939687847, "graph_none_share": -0.005369997286859742, "graph_padding_fraction": 0.0002696603812660705, "kv_usage_end_minus_start": 9.361443848254769e-05, "kv_usage_max": 0.0003780815386457137, "kv_usage_mean": 2.4214752986723154e-05, "preemptions": 0.0, "prefill_token_fraction": -6.638371684375421e-05, "queue_running_mean": -0.0994468485000004, "queue_waiting_mean": 0.0, "scheduler_steps_per_s": 2.3000000000000114 }, "full_action_efficacy": 0, "full_feasibility_transition": "true->true", "group": { "level": "low", "offered_rate_per_gpu": 2.033333333333333, "replicate": 1, "request_hash": "c8ea9278f180ebb23cee595e167627cbed711032e70c9ca5ddc7205320cd75c0", "tp": 1 }, "kind": "matched_mns_increase", "source": { "cell": "tp1_mns8", "early_stopped": false, "full_feasible": true, "full_pass_rate": 1.0, "mns": 8, "requests_sha256": "0edc6095b16e5361a88b2765cc5d0f267ed16ff1d883e02fc3c58f57e59c1018", "result_sha256": "8bc433109d148c21b7d158df95ec5d1fa4fa619243114563068e1e98c4241a29", "trial_id": "cells/tp1_mns8/low-rep1/result.json" }, "target": { "cell": "tp1_mns64", "early_stopped": false, "full_feasible": true, "full_pass_rate": 1.0, "mns": 64, "requests_sha256": "44d729c1ace9b09fe2ea4c4e110bd0f95fee69f83edc78ee0f65256922dcd2c2", "result_sha256": "d2c08e71c5d4e0dd99662bab8457bd1ad0a4aba6bda57794dbfce62e36720f62", "trial_id": "cells/tp1_mns64/low-rep1/result.json" } }, { "delta_outcome": { "admitted_fraction": 0.0, "admitted_input_tokens_mean_over_limit": 0.0, "completed_fail_fraction_of_total": 0.0, "completed_over_admitted": 0.0, "completed_pass_rate": 0.0, "outstanding_over_admitted": 0.0, "tpot_max_over_slo": 0.08836641039618653, "tpot_mean_over_slo": 0.05249869848340616, "ttft_max_over_slo_max": -0.03703611367382109, "ttft_mean_over_slo_max": -0.011082230367607901 }, "delta_state": { "batch_size.mean": 0.2702547536010833, "batch_tokens.mean": 7.3674632893748395, "decode_batch_size.mean": 0.2679754243751127, "graph_full_share": -0.000425589040939367, "graph_none_share": -0.003064812355221179, "graph_padding_fraction": 0.008784696835623326, "kv_usage_end_minus_start": 2.2346672411965862e-05, "kv_usage_max": 0.018441793350708946, "kv_usage_mean": 0.009225433346662204, "preemptions": 0.0, "prefill_token_fraction": 2.292768411660262e-05, "queue_running_mean": 0.7380613438000001, "queue_waiting_mean": -0.2325565183, "scheduler_steps_per_s": -3.8000000000000043 }, "full_action_efficacy": 0, "full_feasibility_transition": "true->true", "group": { "level": "low", "offered_rate_per_gpu": 2.033333333333333, "replicate": 2, "request_hash": "1f27fe31ca32fd67f2f70a54eee83146329a7838ff2a33be48ce4372640a858c", "tp": 1 }, "kind": "matched_mns_increase", "source": { "cell": "tp1_mns8", "early_stopped": false, "full_feasible": true, "full_pass_rate": 1.0, "mns": 8, "requests_sha256": "ca58a878401ac44cca0a9a595eab41659dafec2e923f7f2474e4f82ea30b2f3e", "result_sha256": "42a56af136d1d3bdfc16d9cea092c2db0eaa4db2c85dceb21958c5e418122900", "trial_id": "cells/tp1_mns8/low-rep2/result.json" }, "target": { "cell": "tp1_mns64", "early_stopped": false, "full_feasible": true, "full_pass_rate": 1.0, "mns": 64, "requests_sha256": "97de3cced64fd88a5c2dcd087ca6c29d8b09a0a43d7bf0e4e844ef246462465b", "result_sha256": "f194db2f7abb995171cd24626e4e515c000481f81283c929b9a5b63f110f1506", "trial_id": "cells/tp1_mns64/low-rep2/result.json" } }, { "delta_outcome": { "admitted_fraction": 0.0, "admitted_input_tokens_mean_over_limit": 0.0, "completed_fail_fraction_of_total": 0.0, "completed_over_admitted": 0.0, "completed_pass_rate": 0.0, "outstanding_over_admitted": 0.0, "tpot_max_over_slo": -0.0020900686586818318, "tpot_mean_over_slo": -0.0010018654846283281, "ttft_max_over_slo_max": 0.0005588701654535982, "ttft_mean_over_slo_max": 6.347352852269228e-05 }, "delta_state": { "batch_size.mean": -0.012547944623296114, "batch_tokens.mean": -0.15241343405688212, "decode_batch_size.mean": -0.012475695661875008, "graph_full_share": 7.224896142110637e-05, "graph_none_share": -0.0058946652641868255, "graph_padding_fraction": 0.00029565151617282764, "kv_usage_end_minus_start": 0.00011133071931368477, "kv_usage_max": 0.0002683613903166071, "kv_usage_mean": -1.5294343296619328e-05, "preemptions": 0.0, "prefill_token_fraction": 9.183515573107925e-05, "queue_running_mean": -0.008641511400000024, "queue_waiting_mean": 0.0, "scheduler_steps_per_s": 0.20000000000000284 }, "full_action_efficacy": 0, "full_feasibility_transition": "true->true", "group": { "level": "low", "offered_rate_per_gpu": 2.033333333333333, "replicate": 3, "request_hash": "a46c62ebd8b7df5534831feb70aed45d23b1a5e7f8bf3d4cd82a79f182fbc69e", "tp": 1 }, "kind": "matched_mns_increase", "source": { "cell": "tp1_mns8", "early_stopped": false, "full_feasible": true, "full_pass_rate": 1.0, "mns": 8, "requests_sha256": "8a72240f84bbeff9eabd63e7d1f7a4c19910660fc5e4335013a1deaf89408127", "result_sha256": "0214f32cabbcdc1bad6c331c3945c505cad136fcc9a8336aeb6314b27474e279", "trial_id": "cells/tp1_mns8/low-rep3/result.json" }, "target": { "cell": "tp1_mns64", "early_stopped": false, "full_feasible": true, "full_pass_rate": 1.0, "mns": 64, "requests_sha256": "30f4623100c58580cfa4efe9a1998c63581c8e44511c556ef39cf5bd777883bf", "result_sha256": "0be457aa2fc9ce95bdb92a6eb25544f5f61a9739932ee5afa40535d26456e348", "trial_id": "cells/tp1_mns64/low-rep3/result.json" } }, { "delta_outcome": { "admitted_fraction": 0.0, "admitted_input_tokens_mean_over_limit": 0.0, "completed_fail_fraction_of_total": -0.008, "completed_over_admitted": -0.1610169491525424, "completed_pass_rate": 0.08823529411764708, "outstanding_over_admitted": 0.1610169491525424, "tpot_max_over_slo": 0.29320454629189446, "tpot_mean_over_slo": 0.2535434351882483, "ttft_max_over_slo_max": -0.2690165124949999, "ttft_mean_over_slo_max": -0.13813761862324808 }, "delta_state": { "batch_size.mean": 10.093788314560394, "batch_tokens.mean": 434.02936828958076, "decode_batch_size.mean": 9.924284699384675, "graph_full_share": -0.09005855324219958, "graph_none_share": 0.07778357071650377, "graph_padding_fraction": 0.0017247330951379231, "kv_usage_end_minus_start": 0.05233364294717768, "kv_usage_max": 0.049202174029810086, "kv_usage_mean": 0.019156054903148284, "preemptions": 0.0, "prefill_token_fraction": 0.013365962658000052, "queue_running_mean": 20.948292951200003, "queue_waiting_mean": -13.1590807183, "scheduler_steps_per_s": -34.5 }, "full_action_efficacy": 1, "full_feasibility_transition": "false->true", "group": { "level": "high", "offered_rate_per_gpu": 3.125, "replicate": 1, "request_hash": "5240182174ec5e25247be01c017cbefe2f1e59426bb24a57746dc40be4a82208", "tp": 4 }, "kind": "matched_mns_increase", "source": { "cell": "tp4_mns16", "early_stopped": true, "full_feasible": false, "full_pass_rate": 0.10666666666666667, "mns": 16, "requests_sha256": "8695422a0f1e0e1c097a08a7f4c082f1ba2e3bb26b58caadae0afc7438aeca90", "result_sha256": "c54a2bb5c1b78bdef3fa09c39f5974215370d6fc7c2b6d403a3d376d53a1d9eb", "trial_id": "cells/tp4_mns16/high-rep1/result.json" }, "target": { "cell": "tp4_mns64", "early_stopped": false, "full_feasible": true, "full_pass_rate": 1.0, "mns": 64, "requests_sha256": "b4aeaf5b3175bd85fdac08a2ef657fb2278c6c4b36ea32ea417275ab6d55812d", "result_sha256": "111a1042b614d05772888665a7cfbc86925b2fbeb52ffe19edb62a5bcfe2f513", "trial_id": "cells/tp4_mns64/high-rep1/result.json" } }, { "delta_outcome": { "admitted_fraction": 0.0, "admitted_input_tokens_mean_over_limit": 0.0, "completed_fail_fraction_of_total": -0.012, "completed_over_admitted": -0.1578947368421053, "completed_pass_rate": 0.13636363636363635, "outstanding_over_admitted": 0.1578947368421053, "tpot_max_over_slo": 0.4737703299230537, "tpot_mean_over_slo": 0.3241796650751651, "ttft_max_over_slo_max": -0.3133456981643879, "ttft_mean_over_slo_max": -0.1578698176905678 }, "delta_state": { "batch_size.mean": 19.42960244648318, "batch_tokens.mean": 656.7778499176663, "decode_batch_size.mean": 19.184206069160197, "graph_full_share": -0.13147494707127738, "graph_none_share": 0.11760056457304163, "graph_padding_fraction": 0.0010721073978083185, "kv_usage_end_minus_start": 0.04855346265723404, "kv_usage_max": 0.042070962023699754, "kv_usage_mean": 0.022662705875080867, "preemptions": 0.0, "prefill_token_fraction": 0.007938598543200115, "queue_running_mean": 24.0049637595, "queue_waiting_mean": -15.5760687653, "scheduler_steps_per_s": -32.900000000000006 }, "full_action_efficacy": 1, "full_feasibility_transition": "false->true", "group": { "level": "high", "offered_rate_per_gpu": 3.125, "replicate": 2, "request_hash": "51be592ba9a59c8d66eb1bcc70c3d2b011c87c66128aff36574973b40389ad96", "tp": 4 }, "kind": "matched_mns_increase", "source": { "cell": "tp4_mns16", "early_stopped": true, "full_feasible": false, "full_pass_rate": 0.092, "mns": 16, "requests_sha256": "aa0482b10414d911af227e2df172c19ecca0add2e0024dc88f85ffbd063854bb", "result_sha256": "91e2491de4ca0868b667e7d126b755f03bdd78378681b8070a2bbc0c38fb9055", "trial_id": "cells/tp4_mns16/high-rep2/result.json" }, "target": { "cell": "tp4_mns64", "early_stopped": false, "full_feasible": true, "full_pass_rate": 1.0, "mns": 64, "requests_sha256": "5d82b6885f5c654ae1ae126696df4bafe04f5be1c1109db24c7856b0456d9443", "result_sha256": "6ed574c18723e265ab1327c5e43d3c76d37b3fc5711d06c77a51718e4ba16601", "trial_id": "cells/tp4_mns64/high-rep2/result.json" } }, { "delta_outcome": { "admitted_fraction": 0.0, "admitted_input_tokens_mean_over_limit": 0.0, "completed_fail_fraction_of_total": 0.0, "completed_over_admitted": -0.12844036697247707, "completed_pass_rate": 0.0, "outstanding_over_admitted": 0.12844036697247707, "tpot_max_over_slo": 0.2467615478745419, "tpot_mean_over_slo": 0.024430165887946775, "ttft_max_over_slo_max": -0.08022667251255673, "ttft_mean_over_slo_max": -0.04385035177332515 }, "delta_state": { "batch_size.mean": 2.1439298581773514, "batch_tokens.mean": 116.96904413675719, "decode_batch_size.mean": 2.0885971125866956, "graph_full_share": -0.028466450255056452, "graph_none_share": 0.016098373855303175, "graph_padding_fraction": 0.004743554546834421, "kv_usage_end_minus_start": 0.0262231865121626, "kv_usage_max": 0.020094834170883313, "kv_usage_mean": 0.005378376045436206, "preemptions": 0.0, "prefill_token_fraction": 0.005752175072525123, "queue_running_mean": 4.9523955869, "queue_waiting_mean": -4.365988725199999, "scheduler_steps_per_s": -10.100000000000009 }, "full_action_efficacy": 1, "full_feasibility_transition": "false->true", "group": { "level": "high", "offered_rate_per_gpu": 3.125, "replicate": 3, "request_hash": "604e338cf576d187157293922dada2bceb311567268ddae74eb7a089debf0242", "tp": 4 }, "kind": "matched_mns_increase", "source": { "cell": "tp4_mns16", "early_stopped": true, "full_feasible": false, "full_pass_rate": 0.14666666666666667, "mns": 16, "requests_sha256": "3a23ffbe0e724648226152ef5433cc5825ad640c37cf47997c2f091210eceeec", "result_sha256": "136dde0451f806841ff3996b081fe0bffac0a76f60c3d78b7516f2689e106008", "trial_id": "cells/tp4_mns16/high-rep3/result.json" }, "target": { "cell": "tp4_mns64", "early_stopped": false, "full_feasible": true, "full_pass_rate": 1.0, "mns": 64, "requests_sha256": "5c5861d716181f06dc3f951ae0d6ee479b1b61fc21ebf24541c5b45053fcc167", "result_sha256": "35b8f0465f6b0c197f71b5ecaf4e771e6613f86998ae44f8701ab9e4e1e440b4", "trial_id": "cells/tp4_mns64/high-rep3/result.json" } }, { "delta_outcome": { "admitted_fraction": 0.0, "admitted_input_tokens_mean_over_limit": 0.0, "completed_fail_fraction_of_total": 0.0, "completed_over_admitted": 0.0, "completed_pass_rate": 0.0, "outstanding_over_admitted": 0.0, "tpot_max_over_slo": 0.03209297007177109, "tpot_mean_over_slo": -0.01184639150074901, "ttft_max_over_slo_max": -0.04185720401195188, "ttft_mean_over_slo_max": -0.005481651257700743 }, "delta_state": { "batch_size.mean": -0.9669213868166988, "batch_tokens.mean": -22.564692860834526, "decode_batch_size.mean": -0.9550554240524107, "graph_full_share": 0.009900102186990445, "graph_none_share": -0.015869663207727508, "graph_padding_fraction": -0.0007445684074620391, "kv_usage_end_minus_start": 5.549832249451825e-05, "kv_usage_max": 0.004800695185764425, "kv_usage_mean": -0.0003621004636635179, "preemptions": 0.0, "prefill_token_fraction": -0.00013559912520566542, "queue_running_mean": -0.46276202019999957, "queue_waiting_mean": -0.0902122614, "scheduler_steps_per_s": 11.900000000000006 }, "full_action_efficacy": 1, "full_feasibility_transition": "false->true", "group": { "level": "low", "offered_rate_per_gpu": 2.125, "replicate": 1, "request_hash": "6fb170f366b68a4e4e0fc976d370cb1f5b7b915d4dc22ebe5db3c6b9615afe68", "tp": 4 }, "kind": "matched_mns_increase", "source": { "cell": "tp4_mns16", "early_stopped": true, "full_feasible": false, "full_pass_rate": 0.6196078431372549, "mns": 16, "requests_sha256": "d7eaaefd35cc906f759dcaa7ac3c8543249a1a02f28572bec953e41e4379a29f", "result_sha256": "11d9ec9d2a3358aed988f7b4b1014af087d61c4b5a72eb9076ae19bb5cc481cd", "trial_id": "cells/tp4_mns16/low-rep1/result.json" }, "target": { "cell": "tp4_mns64", "early_stopped": false, "full_feasible": true, "full_pass_rate": 1.0, "mns": 64, "requests_sha256": "931d372ded1e3aaa9c67f06653fac8f79b4ee4f369028c1148163580386681ce", "result_sha256": "325ce57f33133670a6761b5dffb831ed3b2f464755ff743b56976b4a974c1215", "trial_id": "cells/tp4_mns64/low-rep1/result.json" } }, { "delta_outcome": { "admitted_fraction": 0.0, "admitted_input_tokens_mean_over_limit": 0.0, "completed_fail_fraction_of_total": 0.0, "completed_over_admitted": 0.05128205128205132, "completed_pass_rate": 0.0, "outstanding_over_admitted": -0.051282051282051294, "tpot_max_over_slo": 0.03807669731890995, "tpot_mean_over_slo": -0.014506958738293885, "ttft_max_over_slo_max": -0.12963614600206105, "ttft_mean_over_slo_max": -0.048670619592321884 }, "delta_state": { "batch_size.mean": -1.4718645603306726, "batch_tokens.mean": -31.86018426395526, "decode_batch_size.mean": -1.4608669627184998, "graph_full_share": 0.02356281394840931, "graph_none_share": -0.027804605789997328, "graph_padding_fraction": 0.0035831460549140947, "kv_usage_end_minus_start": -0.002621229743826836, "kv_usage_max": 0.0030741209471074615, "kv_usage_mean": -8.774771567193577e-05, "preemptions": 0.0, "prefill_token_fraction": 0.0001064077287293852, "queue_running_mean": -0.40493394709999997, "queue_waiting_mean": -2.0223754996, "scheduler_steps_per_s": 10.599999999999994 }, "full_action_efficacy": 1, "full_feasibility_transition": "false->true", "group": { "level": "low", "offered_rate_per_gpu": 2.125, "replicate": 2, "request_hash": "eb4c0445556d23d73d27d728b45e63fec44e17e84a4f2f341787108b7f7a4901", "tp": 4 }, "kind": "matched_mns_increase", "source": { "cell": "tp4_mns16", "early_stopped": true, "full_feasible": false, "full_pass_rate": 0.6803921568627451, "mns": 16, "requests_sha256": "7fdb90f679bba1747fbacc92fac2045c820e09965c755b20f6dafd4fd9e26a7a", "result_sha256": "800e2aa6e9f9efef4be12ee778be6817364914d63ae0624775d02729089768d6", "trial_id": "cells/tp4_mns16/low-rep2/result.json" }, "target": { "cell": "tp4_mns64", "early_stopped": false, "full_feasible": true, "full_pass_rate": 1.0, "mns": 64, "requests_sha256": "72c8f70a240f79edcaa4eb0a2fe9bade8ac07093339c5b2d09c49ddd1cf27d16", "result_sha256": "ff8b2283d02d5967893ca6225b561ae11011da6149d1f6cff7b4275c00170e1d", "trial_id": "cells/tp4_mns64/low-rep2/result.json" } }, { "delta_outcome": { "admitted_fraction": 0.0, "admitted_input_tokens_mean_over_limit": 0.0, "completed_fail_fraction_of_total": 0.0, "completed_over_admitted": 0.0, "completed_pass_rate": 0.0, "outstanding_over_admitted": 0.0, "tpot_max_over_slo": 0.04088999417529798, "tpot_mean_over_slo": -0.0032529423795618906, "ttft_max_over_slo_max": 0.0002604483355147097, "ttft_mean_over_slo_max": -0.001620827191598239 }, "delta_state": { "batch_size.mean": 0.14852247907791494, "batch_tokens.mean": 11.553519812007579, "decode_batch_size.mean": 0.14728805801238387, "graph_full_share": 0.0005649503708670123, "graph_none_share": -0.0023940560373101466, "graph_padding_fraction": -0.00035714669425012356, "kv_usage_end_minus_start": 0.0014602804352583654, "kv_usage_max": 0.003991863602817669, "kv_usage_mean": 4.656051020407101e-06, "preemptions": 0.0, "prefill_token_fraction": 0.0018350834191880327, "queue_running_mean": 0.06592412640000056, "queue_waiting_mean": -0.2173940436, "scheduler_steps_per_s": 1.1000000000000085 }, "full_action_efficacy": 1, "full_feasibility_transition": "false->true", "group": { "level": "low", "offered_rate_per_gpu": 2.125, "replicate": 3, "request_hash": "a660c44d1f8b9c0ada5a19285fa579871bcfcae7c6fa71649546efda9aa3baef", "tp": 4 }, "kind": "matched_mns_increase", "source": { "cell": "tp4_mns16", "early_stopped": true, "full_feasible": false, "full_pass_rate": 0.6294117647058823, "mns": 16, "requests_sha256": "b592fe254476957265fc9139aefdea77be4b5540f3c1b8a429c10ea6ccab9a6a", "result_sha256": "cc1b1f7f5280c2892d0c6deac9df9f34ce126b0284e6ad893ffbf0ddbf27969f", "trial_id": "cells/tp4_mns16/low-rep3/result.json" }, "target": { "cell": "tp4_mns64", "early_stopped": false, "full_feasible": true, "full_pass_rate": 1.0, "mns": 64, "requests_sha256": "8b2c779d2097c5e4038e94e1ac9b986e29f55d9acfd67eeb366c4e0e9270f6e0", "result_sha256": "f026b69356701894fab184e0f2f7504a30b537982e823373899c9ab123a33891", "trial_id": "cells/tp4_mns64/low-rep3/result.json" } } ], "efficacy": { "feasibility_transitions": { "false->false": 3, "false->true": 6, "true->true": 3 }, "minimum_balanced_accuracy": 0.75, "minimum_delta_over_best_outcome": 0.15, "outcome_delta": { "best_accuracy": 0.75, "best_balanced_accuracy": 0.75, "best_feature": "outstanding_over_admitted", "features": { "admitted_fraction": { "accuracy": 0.5, "balanced_accuracy": 0.5, "folds": [ { "direction": 1, "held_out_replicate": 1, "test_labels": [ 0, 0, 1, 1 ], "test_predictions": [ 1, 1, 1, 1 ], "threshold": 0.0, "train_balanced_accuracy": 0.5 }, { "direction": 1, "held_out_replicate": 2, "test_labels": [ 0, 0, 1, 1 ], "test_predictions": [ 1, 1, 1, 1 ], "threshold": 0.0, "train_balanced_accuracy": 0.5 }, { "direction": 1, "held_out_replicate": 3, "test_labels": [ 0, 0, 1, 1 ], "test_predictions": [ 1, 1, 1, 1 ], "threshold": 0.0, "train_balanced_accuracy": 0.5 } ] }, "admitted_input_tokens_mean_over_limit": { "accuracy": 0.5, "balanced_accuracy": 0.5, "folds": [ { "direction": 1, "held_out_replicate": 1, "test_labels": [ 0, 0, 1, 1 ], "test_predictions": [ 1, 1, 1, 1 ], "threshold": 0.0, "train_balanced_accuracy": 0.5 }, { "direction": 1, "held_out_replicate": 2, "test_labels": [ 0, 0, 1, 1 ], "test_predictions": [ 1, 1, 1, 1 ], "threshold": 0.0, "train_balanced_accuracy": 0.5 }, { "direction": 1, "held_out_replicate": 3, "test_labels": [ 0, 0, 1, 1 ], "test_predictions": [ 1, 1, 1, 1 ], "threshold": 0.0, "train_balanced_accuracy": 0.5 } ] }, "completed_fail_fraction_of_total": { "accuracy": 0.6666666666666666, "balanced_accuracy": 0.6666666666666666, "folds": [ { "direction": -1, "held_out_replicate": 1, "test_labels": [ 0, 0, 1, 1 ], "test_predictions": [ 0, 0, 1, 0 ], "threshold": -0.006, "train_balanced_accuracy": 0.625 }, { "direction": -1, "held_out_replicate": 2, "test_labels": [ 0, 0, 1, 1 ], "test_predictions": [ 0, 0, 1, 0 ], "threshold": -0.004, "train_balanced_accuracy": 0.625 }, { "direction": -1, "held_out_replicate": 3, "test_labels": [ 0, 0, 1, 1 ], "test_predictions": [ 0, 0, 0, 0 ], "threshold": -0.004, "train_balanced_accuracy": 0.75 } ] }, "completed_over_admitted": { "accuracy": 0.75, "balanced_accuracy": 0.75, "folds": [ { "direction": -1, "held_out_replicate": 1, "test_labels": [ 0, 0, 1, 1 ], "test_predictions": [ 0, 0, 1, 0 ], "threshold": -0.08034921574430304, "train_balanced_accuracy": 0.75 }, { "direction": -1, "held_out_replicate": 2, "test_labels": [ 0, 0, 1, 1 ], "test_predictions": [ 0, 0, 1, 0 ], "threshold": -0.06422018348623854, "train_balanced_accuracy": 0.75 }, { "direction": -1, "held_out_replicate": 3, "test_labels": [ 0, 0, 1, 1 ], "test_predictions": [ 0, 0, 1, 0 ], "threshold": -0.09507640067911716, "train_balanced_accuracy": 0.75 } ] }, "completed_pass_rate": { "accuracy": 0.6666666666666666, "balanced_accuracy": 0.6666666666666666, "folds": [ { "direction": 1, "held_out_replicate": 1, "test_labels": [ 0, 0, 1, 1 ], "test_predictions": [ 0, 0, 1, 0 ], "threshold": 0.06818181818181818, "train_balanced_accuracy": 0.625 }, { "direction": 1, "held_out_replicate": 2, "test_labels": [ 0, 0, 1, 1 ], "test_predictions": [ 0, 0, 1, 0 ], "threshold": 0.04411764705882354, "train_balanced_accuracy": 0.625 }, { "direction": 1, "held_out_replicate": 3, "test_labels": [ 0, 0, 1, 1 ], "test_predictions": [ 0, 0, 0, 0 ], "threshold": 0.04411764705882354, "train_balanced_accuracy": 0.75 } ] }, "outstanding_over_admitted": { "accuracy": 0.75, "balanced_accuracy": 0.75, "folds": [ { "direction": 1, "held_out_replicate": 1, "test_labels": [ 0, 0, 1, 1 ], "test_predictions": [ 0, 0, 1, 0 ], "threshold": 0.08034921574430307, "train_balanced_accuracy": 0.75 }, { "direction": 1, "held_out_replicate": 2, "test_labels": [ 0, 0, 1, 1 ], "test_predictions": [ 0, 0, 1, 0 ], "threshold": 0.06422018348623854, "train_balanced_accuracy": 0.75 }, { "direction": 1, "held_out_replicate": 3, "test_labels": [ 0, 0, 1, 1 ], "test_predictions": [ 0, 0, 1, 0 ], "threshold": 0.09507640067911718, "train_balanced_accuracy": 0.75 } ] }, "tpot_max_over_slo": { "accuracy": 0.6666666666666666, "balanced_accuracy": 0.6666666666666666, "folds": [ { "direction": 1, "held_out_replicate": 1, "test_labels": [ 0, 0, 1, 1 ], "test_predictions": [ 1, 0, 1, 1 ], "threshold": 0.028440662518539606, "train_balanced_accuracy": 0.75 }, { "direction": 1, "held_out_replicate": 2, "test_labels": [ 0, 0, 1, 1 ], "test_predictions": [ 1, 1, 1, 1 ], "threshold": 0.025448798894970176, "train_balanced_accuracy": 0.875 }, { "direction": 1, "held_out_replicate": 3, "test_labels": [ 0, 0, 1, 1 ], "test_predictions": [ 1, 0, 1, 1 ], "threshold": 0.014026599366027181, "train_balanced_accuracy": 0.625 } ] }, "tpot_mean_over_slo": { "accuracy": 0.4166666666666667, "balanced_accuracy": 0.41666666666666663, "folds": [ { "direction": -1, "held_out_replicate": 1, "test_labels": [ 0, 0, 1, 1 ], "test_predictions": [ 0, 1, 0, 1 ], "threshold": -0.0021274039320951094, "train_balanced_accuracy": 0.75 }, { "direction": 1, "held_out_replicate": 2, "test_labels": [ 0, 0, 1, 1 ], "test_predictions": [ 1, 1, 1, 0 ], "threshold": 0.018002935900514647, "train_balanced_accuracy": 0.75 }, { "direction": -1, "held_out_replicate": 3, "test_labels": [ 0, 0, 1, 1 ], "test_predictions": [ 0, 0, 0, 0 ], "threshold": -0.009876489314992687, "train_balanced_accuracy": 0.75 } ] }, "ttft_max_over_slo_max": { "accuracy": 0.6666666666666666, "balanced_accuracy": 0.6666666666666667, "folds": [ { "direction": -1, "held_out_replicate": 1, "test_labels": [ 0, 0, 1, 1 ], "test_predictions": [ 0, 1, 1, 1 ], "threshold": 0.00040965925048415397, "train_balanced_accuracy": 0.75 }, { "direction": -1, "held_out_replicate": 2, "test_labels": [ 0, 0, 1, 1 ], "test_predictions": [ 1, 1, 1, 1 ], "threshold": 0.00040965925048415397, "train_balanced_accuracy": 0.875 }, { "direction": -1, "held_out_replicate": 3, "test_labels": [ 0, 0, 1, 1 ], "test_predictions": [ 0, 0, 1, 0 ], "threshold": -0.039446658842886485, "train_balanced_accuracy": 0.875 } ] }, "ttft_mean_over_slo_max": { "accuracy": 0.6666666666666666, "balanced_accuracy": 0.6666666666666666, "folds": [ { "direction": -1, "held_out_replicate": 1, "test_labels": [ 0, 0, 1, 1 ], "test_predictions": [ 0, 0, 1, 0 ], "threshold": -0.027466291070466525, "train_balanced_accuracy": 0.75 }, { "direction": -1, "held_out_replicate": 2, "test_labels": [ 0, 0, 1, 1 ], "test_predictions": [ 1, 1, 1, 1 ], "threshold": -0.0013986015436684204, "train_balanced_accuracy": 0.875 }, { "direction": -1, "held_out_replicate": 3, "test_labels": [ 0, 0, 1, 1 ], "test_predictions": [ 0, 0, 1, 0 ], "threshold": -0.0033290135767196725, "train_balanced_accuracy": 0.75 } ] } }, "labels": { "distinct_n": 2, "max": 1.0, "min": 0.0, "n": 12 }, "negative": 6, "positive": 6 }, "telemetry_delta": { "best_accuracy": 0.75, "best_balanced_accuracy": 0.75, "best_feature": "queue_waiting_mean", "features": { "decode_batch_size.mean": { "accuracy": 0.5, "balanced_accuracy": 0.5, "folds": [ { "direction": 1, "held_out_replicate": 1, "test_labels": [ 0, 0, 1, 1 ], "test_predictions": [ 0, 0, 1, 0 ], "threshold": 0.0878005661250898, "train_balanced_accuracy": 0.625 }, { "direction": 1, "held_out_replicate": 2, "test_labels": [ 0, 0, 1, 1 ], "test_predictions": [ 1, 1, 1, 0 ], "threshold": 0.08790969260376658, "train_balanced_accuracy": 0.875 }, { "direction": -1, "held_out_replicate": 3, "test_labels": [ 0, 0, 1, 1 ], "test_predictions": [ 0, 0, 0, 0 ], "threshold": -0.5179996829782867, "train_balanced_accuracy": 0.75 } ] }, "graph_padding_fraction": { "accuracy": 0.5, "balanced_accuracy": 0.5, "folds": [ { "direction": -1, "held_out_replicate": 1, "test_labels": [ 0, 0, 1, 1 ], "test_predictions": [ 0, 0, 0, 1 ], "threshold": -3.074758903864796e-05, "train_balanced_accuracy": 0.625 }, { "direction": -1, "held_out_replicate": 2, "test_labels": [ 0, 0, 1, 1 ], "test_predictions": [ 0, 0, 0, 0 ], "threshold": -4.3743156492026536e-05, "train_balanced_accuracy": 0.75 }, { "direction": -1, "held_out_replicate": 3, "test_labels": [ 0, 0, 1, 1 ], "test_predictions": [ 1, 1, 0, 1 ], "threshold": 0.0038965341747184087, "train_balanced_accuracy": 0.875 } ] }, "kv_usage_mean": { "accuracy": 0.5, "balanced_accuracy": 0.5, "folds": [ { "direction": -1, "held_out_replicate": 1, "test_labels": [ 0, 0, 1, 1 ], "test_predictions": [ 0, 0, 0, 1 ], "threshold": -5.152102948427755e-05, "train_balanced_accuracy": 0.625 }, { "direction": 1, "held_out_replicate": 2, "test_labels": [ 0, 0, 1, 1 ], "test_predictions": [ 1, 1, 1, 0 ], "threshold": 0.004075888029500068, "train_balanced_accuracy": 0.75 }, { "direction": -1, "held_out_replicate": 3, "test_labels": [ 0, 0, 1, 1 ], "test_predictions": [ 0, 0, 0, 0 ], "threshold": -3.176648134260631e-05, "train_balanced_accuracy": 0.75 } ] }, "prefill_token_fraction": { "accuracy": 0.6666666666666666, "balanced_accuracy": 0.6666666666666666, "folds": [ { "direction": 1, "held_out_replicate": 1, "test_labels": [ 0, 0, 1, 1 ], "test_predictions": [ 0, 0, 1, 0 ], "threshold": 9.912144223023223e-05, "train_balanced_accuracy": 0.875 }, { "direction": 1, "held_out_replicate": 2, "test_labels": [ 0, 0, 1, 1 ], "test_predictions": [ 1, 0, 1, 0 ], "threshold": 0.000963459287459556, "train_balanced_accuracy": 0.875 }, { "direction": 1, "held_out_replicate": 3, "test_labels": [ 0, 0, 1, 1 ], "test_predictions": [ 0, 1, 1, 1 ], "threshold": 6.466770642299391e-05, "train_balanced_accuracy": 0.75 } ] }, "queue_running_mean": { "accuracy": 0.5833333333333334, "balanced_accuracy": 0.5833333333333334, "folds": [ { "direction": 1, "held_out_replicate": 1, "test_labels": [ 0, 0, 1, 1 ], "test_predictions": [ 0, 0, 1, 0 ], "threshold": 4.816722804299999, "train_balanced_accuracy": 0.75 }, { "direction": 1, "held_out_replicate": 2, "test_labels": [ 0, 0, 1, 1 ], "test_predictions": [ 1, 0, 1, 0 ], "threshold": 2.5825188629999998, "train_balanced_accuracy": 0.75 }, { "direction": -1, "held_out_replicate": 3, "test_labels": [ 0, 0, 1, 1 ], "test_predictions": [ 0, 0, 0, 0 ], "threshold": -0.2521903978000002, "train_balanced_accuracy": 0.75 } ] }, "queue_waiting_mean": { "accuracy": 0.75, "balanced_accuracy": 0.75, "folds": [ { "direction": -1, "held_out_replicate": 1, "test_labels": [ 0, 0, 1, 1 ], "test_predictions": [ 0, 0, 1, 0 ], "threshold": -0.1227663938, "train_balanced_accuracy": 0.75 }, { "direction": -1, "held_out_replicate": 2, "test_labels": [ 0, 0, 1, 1 ], "test_predictions": [ 1, 1, 1, 1 ], "threshold": -0.0820950887, "train_balanced_accuracy": 1.0 }, { "direction": -1, "held_out_replicate": 3, "test_labels": [ 0, 0, 1, 1 ], "test_predictions": [ 0, 0, 1, 1 ], "threshold": -0.0820950887, "train_balanced_accuracy": 0.75 } ] }, "scheduler_steps_per_s": { "accuracy": 0.5, "balanced_accuracy": 0.5, "folds": [ { "direction": 1, "held_out_replicate": 1, "test_labels": [ 0, 0, 1, 1 ], "test_predictions": [ 0, 1, 0, 1 ], "threshold": 0.6500000000000057, "train_balanced_accuracy": 0.75 }, { "direction": -1, "held_out_replicate": 2, "test_labels": [ 0, 0, 1, 1 ], "test_predictions": [ 1, 0, 1, 0 ], "threshold": -5.3500000000000085, "train_balanced_accuracy": 0.75 }, { "direction": 1, "held_out_replicate": 3, "test_labels": [ 0, 0, 1, 1 ], "test_predictions": [ 0, 0, 0, 0 ], "threshold": 6.450000000000003, "train_balanced_accuracy": 0.75 } ] } }, "labels": { "distinct_n": 2, "max": 1.0, "min": 0.0, "n": 12 }, "negative": 6, "positive": 6 }, "telemetry_qualifying_features": [] }, "horizon_s": 10.0, "qualifying_response_features": [], "repeats": [ { "delta_outcome": { "admitted_fraction": 0.027932960893854747, "admitted_input_tokens_mean_over_limit": 0.06722636258335918, "completed_fail_fraction_of_total": 0.0, "completed_over_admitted": -0.21464019851116622, "completed_pass_rate": 0.0, "outstanding_over_admitted": 0.21464019851116628, "tpot_max_over_slo": 0.3907463973292743, "tpot_mean_over_slo": 0.45304077139972754, "ttft_max_over_slo_max": -0.026048107499567152, "ttft_mean_over_slo_max": 0.02040584175783646 }, "delta_state": { "batch_size.mean": 8.048982941677576, "batch_tokens.mean": 298.1121465228864, "decode_batch_size.mean": 7.965265224708595, "graph_full_share": -0.07611315423133669, "graph_none_share": 0.08050759417399073, "graph_padding_fraction": -0.006831997792023563, "kv_usage_end_minus_start": 0.08232590668398432, "kv_usage_max": 0.11409731241882304, "kv_usage_mean": 0.08571139047521727, "preemptions": 0.0, "prefill_token_fraction": 0.010675483227336025, "queue_running_mean": 7.2005321570000005, "queue_waiting_mean": 0.0, "scheduler_steps_per_s": -46.900000000000006 }, "group": { "cell": "tp1_mns64", "level": "high", "mns": 64, "source_replicate": 1, "target_replicate": 2, "tp": 1 }, "kind": "same_config_workload_repeat", "source": { "requests_sha256": "e970dcac20144cb1e13a35997377f6695a9c57d788a88bad7f18cf6c64e35cec", "result_sha256": "3233686b60b920dcd6cc2f45193a81f3af2d46f781ef42c4d117630612084feb", "trial_id": "cells/tp1_mns64/high-rep1/result.json" }, "target": { "requests_sha256": "ec06878e2aefdc4853d972197cedc29ab94df7147abb2701e0d88b5dabf29728", "result_sha256": "8215477a014f061c55e94d879a25bbf1676ee2ef1c5e81efadd23d20d337e07a", "trial_id": "cells/tp1_mns64/high-rep2/result.json" } }, { "delta_outcome": { "admitted_fraction": -0.03910614525139666, "admitted_input_tokens_mean_over_limit": -0.15364353631132394, "completed_fail_fraction_of_total": 0.0, "completed_over_admitted": 0.31720430107526887, "completed_pass_rate": 0.0, "outstanding_over_admitted": -0.31720430107526887, "tpot_max_over_slo": -0.4318866070898938, "tpot_mean_over_slo": -0.4648576303649653, "ttft_max_over_slo_max": -0.0451326228309578, "ttft_mean_over_slo_max": -0.03512127695333522 }, "delta_state": { "batch_size.mean": -8.639628684110797, "batch_tokens.mean": -331.31332381027744, "decode_batch_size.mean": -8.548044927723337, "graph_full_share": 0.08397919364981532, "graph_none_share": -0.08346278303674427, "graph_padding_fraction": 0.0049766447911436746, "kv_usage_end_minus_start": -0.12798481366769898, "kv_usage_max": -0.12688580277750028, "kv_usage_mean": -0.09580544440945148, "preemptions": 0.0, "prefill_token_fraction": -0.024511024899494305, "queue_running_mean": -7.6525189056, "queue_waiting_mean": 0.0, "scheduler_steps_per_s": 65.0 }, "group": { "cell": "tp1_mns64", "level": "high", "mns": 64, "source_replicate": 2, "target_replicate": 3, "tp": 1 }, "kind": "same_config_workload_repeat", "source": { "requests_sha256": "ec06878e2aefdc4853d972197cedc29ab94df7147abb2701e0d88b5dabf29728", "result_sha256": "8215477a014f061c55e94d879a25bbf1676ee2ef1c5e81efadd23d20d337e07a", "trial_id": "cells/tp1_mns64/high-rep2/result.json" }, "target": { "requests_sha256": "18d72d1fe16136359e28a1909e395ffd553dbe3edd2c0fd93a9cc270a5ec6d60", "result_sha256": "480a3b50acaf9b9c1902e2669cf3aa7723a6703a00778da269b5b9757990ceef", "trial_id": "cells/tp1_mns64/high-rep3/result.json" } }, { "delta_outcome": { "admitted_fraction": -0.008196721311475419, "admitted_input_tokens_mean_over_limit": 0.07372581845238091, "completed_fail_fraction_of_total": 0.0, "completed_over_admitted": 0.09285714285714286, "completed_pass_rate": 0.0, "outstanding_over_admitted": -0.09285714285714285, "tpot_max_over_slo": 0.1702932155907418, "tpot_mean_over_slo": 0.1739500633542023, "ttft_max_over_slo_max": 0.02082575849878293, "ttft_mean_over_slo_max": 0.029532855802237655 }, "delta_state": { "batch_size.mean": 1.3717277486910993, "batch_tokens.mean": 56.51418492510525, "decode_batch_size.mean": 1.3587752164958689, "graph_full_share": -0.010646964922293112, "graph_none_share": 0.011895900019023516, "graph_padding_fraction": 0.0086692890425959, "kv_usage_end_minus_start": -0.017684084324108285, "kv_usage_max": 0.05015486062543706, "kv_usage_mean": 0.03571575463439401, "preemptions": 0.0, "prefill_token_fraction": 0.011636868678620127, "queue_running_mean": 2.2687814585000003, "queue_waiting_mean": 0.0, "scheduler_steps_per_s": -27.10000000000001 }, "group": { "cell": "tp1_mns64", "level": "low", "mns": 64, "source_replicate": 1, "target_replicate": 2, "tp": 1 }, "kind": "same_config_workload_repeat", "source": { "requests_sha256": "44d729c1ace9b09fe2ea4c4e110bd0f95fee69f83edc78ee0f65256922dcd2c2", "result_sha256": "d2c08e71c5d4e0dd99662bab8457bd1ad0a4aba6bda57794dbfce62e36720f62", "trial_id": "cells/tp1_mns64/low-rep1/result.json" }, "target": { "requests_sha256": "97de3cced64fd88a5c2dcd087ca6c29d8b09a0a43d7bf0e4e844ef246462465b", "result_sha256": "f194db2f7abb995171cd24626e4e515c000481f81283c929b9a5b63f110f1506", "trial_id": "cells/tp1_mns64/low-rep2/result.json" } }, { "delta_outcome": { "admitted_fraction": -0.02459016393442623, "admitted_input_tokens_mean_over_limit": -0.09311379825367644, "completed_fail_fraction_of_total": 0.0, "completed_over_admitted": -0.24411764705882344, "completed_pass_rate": 0.0, "outstanding_over_admitted": 0.24411764705882355, "tpot_max_over_slo": -0.3084416280309926, "tpot_mean_over_slo": -0.21178917402407615, "ttft_max_over_slo_max": -0.07642558032724386, "ttft_mean_over_slo_max": -0.031511784150528625 }, "delta_state": { "batch_size.mean": -1.5740567152122056, "batch_tokens.mean": -51.36471646204379, "decode_batch_size.mean": -1.5621527698392743, "graph_full_share": 0.008413543976771298, "graph_none_share": -0.010745558883376392, "graph_padding_fraction": -0.007424342975302652, "kv_usage_end_minus_start": 0.022080127884903655, "kv_usage_max": -0.0773803576780897, "kv_usage_mean": -0.04564071465536516, "preemptions": 0.0, "prefill_token_fraction": -0.004854783915929306, "queue_running_mean": -3.3275810848000003, "queue_waiting_mean": 0.0, "scheduler_steps_per_s": 11.400000000000006 }, "group": { "cell": "tp1_mns64", "level": "low", "mns": 64, "source_replicate": 2, "target_replicate": 3, "tp": 1 }, "kind": "same_config_workload_repeat", "source": { "requests_sha256": "97de3cced64fd88a5c2dcd087ca6c29d8b09a0a43d7bf0e4e844ef246462465b", "result_sha256": "f194db2f7abb995171cd24626e4e515c000481f81283c929b9a5b63f110f1506", "trial_id": "cells/tp1_mns64/low-rep2/result.json" }, "target": { "requests_sha256": "30f4623100c58580cfa4efe9a1998c63581c8e44511c556ef39cf5bd777883bf", "result_sha256": "0be457aa2fc9ce95bdb92a6eb25544f5f61a9739932ee5afa40535d26456e348", "trial_id": "cells/tp1_mns64/low-rep3/result.json" } }, { "delta_outcome": { "admitted_fraction": 0.027932960893854747, "admitted_input_tokens_mean_over_limit": 0.06722636258335918, "completed_fail_fraction_of_total": 0.0, "completed_over_admitted": -0.18238213399503722, "completed_pass_rate": 0.0, "outstanding_over_admitted": 0.18238213399503722, "tpot_max_over_slo": 0.10448095039464528, "tpot_mean_over_slo": 0.1638344212807229, "ttft_max_over_slo_max": 0.1201083790025829, "ttft_mean_over_slo_max": 0.10836152705554111 }, "delta_state": { "batch_size.mean": 3.5304685668352485, "batch_tokens.mean": 120.01419432628865, "decode_batch_size.mean": 3.503590447801186, "graph_full_share": -0.028247982047761022, "graph_none_share": 0.028247982047760994, "graph_padding_fraction": -0.005012808300211052, "kv_usage_end_minus_start": 0.027314791780685632, "kv_usage_max": 0.0341310512960844, "kv_usage_mean": 0.042088077242922566, "preemptions": 0.0, "prefill_token_fraction": 0.006312963723005605, "queue_running_mean": 2.7321242744000003, "queue_waiting_mean": 2.2523028899, "scheduler_steps_per_s": -32.1 }, "group": { "cell": "tp1_mns8", "level": "high", "mns": 8, "source_replicate": 1, "target_replicate": 2, "tp": 1 }, "kind": "same_config_workload_repeat", "source": { "requests_sha256": "5f5bfa4a16846190f488d3369eed89ae895804145cea59197e6b544b703223fc", "result_sha256": "f65e3dc91d17f5caab1f4aecf97d2c464b99f107231af2cfd96f5484880136ee", "trial_id": "cells/tp1_mns8/high-rep1/result.json" }, "target": { "requests_sha256": "e802d09057de349e5164c18314fcad09dc88c9673834fd6e5b100a57dd12fbf4", "result_sha256": "4835a985dd6bb8dc50b53737baa980ac95a49e22f0b8a7326e1fb416c9b4e798", "trial_id": "cells/tp1_mns8/high-rep2/result.json" } }, { "delta_outcome": { "admitted_fraction": -0.03910614525139666, "admitted_input_tokens_mean_over_limit": -0.15364353631132394, "completed_fail_fraction_of_total": 0.0, "completed_over_admitted": 0.28494623655913986, "completed_pass_rate": 0.0, "outstanding_over_admitted": -0.28494623655913975, "tpot_max_over_slo": -0.12018486661249839, "tpot_mean_over_slo": -0.17840573280620411, "ttft_max_over_slo_max": -0.19140850950498134, "ttft_mean_over_slo_max": -0.12097088902259666 }, "delta_state": { "batch_size.mean": -4.121161249471228, "batch_tokens.mean": -153.761987490389, "decode_batch_size.mean": -4.086151897858574, "graph_full_share": 0.035009351612654394, "graph_none_share": -0.03500935161265434, "graph_padding_fraction": 0.0059864236218592284, "kv_usage_end_minus_start": -0.07269018359122359, "kv_usage_max": -0.041146325687845264, "kv_usage_mean": -0.050474056816329174, "preemptions": 0.0, "prefill_token_fraction": -0.020411045123935767, "queue_running_mean": -3.1130893942, "queue_waiting_mean": -2.2981420619, "scheduler_steps_per_s": 51.00000000000001 }, "group": { "cell": "tp1_mns8", "level": "high", "mns": 8, "source_replicate": 2, "target_replicate": 3, "tp": 1 }, "kind": "same_config_workload_repeat", "source": { "requests_sha256": "e802d09057de349e5164c18314fcad09dc88c9673834fd6e5b100a57dd12fbf4", "result_sha256": "4835a985dd6bb8dc50b53737baa980ac95a49e22f0b8a7326e1fb416c9b4e798", "trial_id": "cells/tp1_mns8/high-rep2/result.json" }, "target": { "requests_sha256": "cf1226185845cd8aa4e7ead0cb5c95c77eda62b84851a2ab579f984bbb88e8ac", "result_sha256": "95d5103f2678eeb99c258870324566932e2f2853c3b6e0ee94bc6c538f6ed865", "trial_id": "cells/tp1_mns8/high-rep3/result.json" } }, { "delta_outcome": { "admitted_fraction": -0.008196721311475419, "admitted_input_tokens_mean_over_limit": 0.07372581845238091, "completed_fail_fraction_of_total": 0.0, "completed_over_admitted": 0.09285714285714286, "completed_pass_rate": 0.0, "outstanding_over_admitted": -0.09285714285714285, "tpot_max_over_slo": 0.07788703385483853, "tpot_mean_over_slo": 0.11354477774155977, "ttft_max_over_slo_max": 0.05746760650072247, "ttft_mean_over_slo_max": 0.039438710274106954 }, "delta_state": { "batch_size.mean": 1.019865199718518, "batch_tokens.mean": 47.38582145042871, "decode_batch_size.mean": 1.0098558502165935, "graph_full_share": -0.00959071508738496, "graph_none_share": 0.009590715087384953, "graph_padding_fraction": 0.00015425258823864325, "kv_usage_end_minus_start": -0.017612816558037703, "kv_usage_max": 0.032091148813373827, "kv_usage_mean": 0.02651453604071853, "preemptions": 0.0, "prefill_token_fraction": 0.01154755727765977, "queue_running_mean": 1.4312732661999998, "queue_waiting_mean": 0.2325565183, "scheduler_steps_per_s": -20.999999999999993 }, "group": { "cell": "tp1_mns8", "level": "low", "mns": 8, "source_replicate": 1, "target_replicate": 2, "tp": 1 }, "kind": "same_config_workload_repeat", "source": { "requests_sha256": "0edc6095b16e5361a88b2765cc5d0f267ed16ff1d883e02fc3c58f57e59c1018", "result_sha256": "8bc433109d148c21b7d158df95ec5d1fa4fa619243114563068e1e98c4241a29", "trial_id": "cells/tp1_mns8/low-rep1/result.json" }, "target": { "requests_sha256": "ca58a878401ac44cca0a9a595eab41659dafec2e923f7f2474e4f82ea30b2f3e", "result_sha256": "42a56af136d1d3bdfc16d9cea092c2db0eaa4db2c85dceb21958c5e418122900", "trial_id": "cells/tp1_mns8/low-rep2/result.json" } }, { "delta_outcome": { "admitted_fraction": -0.02459016393442623, "admitted_input_tokens_mean_over_limit": -0.09311379825367644, "completed_fail_fraction_of_total": 0.0, "completed_over_admitted": -0.24411764705882344, "completed_pass_rate": 0.0, "outstanding_over_admitted": 0.24411764705882355, "tpot_max_over_slo": -0.21798514897612425, "tpot_mean_over_slo": -0.15828861005604167, "ttft_max_over_slo_max": -0.11402056416651855, "ttft_mean_over_slo_max": -0.04265748804665922 }, "delta_state": { "batch_size.mean": -1.2912540169878262, "batch_tokens.mean": -43.844839738612066, "decode_batch_size.mean": -1.2817016498022866, "graph_full_share": 0.007915705974410825, "graph_none_share": -0.007915705974410745, "graph_padding_fraction": 0.0010647023441478472, "kv_usage_end_minus_start": 0.021991143838001936, "kv_usage_max": -0.05920692571769737, "kv_usage_mean": -0.036399986965406333, "preemptions": 0.0, "prefill_token_fraction": -0.004923691387543783, "queue_running_mean": -2.5808782296, "queue_waiting_mean": -0.2325565183, "scheduler_steps_per_s": 7.399999999999999 }, "group": { "cell": "tp1_mns8", "level": "low", "mns": 8, "source_replicate": 2, "target_replicate": 3, "tp": 1 }, "kind": "same_config_workload_repeat", "source": { "requests_sha256": "ca58a878401ac44cca0a9a595eab41659dafec2e923f7f2474e4f82ea30b2f3e", "result_sha256": "42a56af136d1d3bdfc16d9cea092c2db0eaa4db2c85dceb21958c5e418122900", "trial_id": "cells/tp1_mns8/low-rep2/result.json" }, "target": { "requests_sha256": "8a72240f84bbeff9eabd63e7d1f7a4c19910660fc5e4335013a1deaf89408127", "result_sha256": "0214f32cabbcdc1bad6c331c3945c505cad136fcc9a8336aeb6314b27474e279", "trial_id": "cells/tp1_mns8/low-rep3/result.json" } }, { "delta_outcome": { "admitted_fraction": -0.014492753623188387, "admitted_input_tokens_mean_over_limit": 0.04833232120973391, "completed_fail_fraction_of_total": 0.0, "completed_over_admitted": -0.008403361344537785, "completed_pass_rate": 0.0, "outstanding_over_admitted": 0.00840336134453784, "tpot_max_over_slo": 0.09841551118728337, "tpot_mean_over_slo": 0.06345367037672023, "ttft_max_over_slo_max": 0.010166719497647136, "ttft_mean_over_slo_max": 0.006259302157370798 }, "delta_state": { "batch_size.mean": 2.0365641486137616, "batch_tokens.mean": 91.46332360534291, "decode_batch_size.mean": 2.012753956673029, "graph_full_share": -0.01772084409024577, "graph_none_share": 0.021565271074194634, "graph_padding_fraction": -0.002786564210742205, "kv_usage_end_minus_start": 0.007622555602450509, "kv_usage_max": 0.020987606952384663, "kv_usage_mean": 0.007630049119681775, "preemptions": 0.0, "prefill_token_fraction": 0.005242973565105835, "queue_running_mean": 0.8145702931999992, "queue_waiting_mean": 0.0, "scheduler_steps_per_s": -13.199999999999996 }, "group": { "cell": "tp2_mns64", "level": "high", "mns": 64, "source_replicate": 1, "target_replicate": 2, "tp": 2 }, "kind": "same_config_workload_repeat", "source": { "requests_sha256": "19537ec05f1e40543d850964db12234561954c209d55cd8cd0a1da716ab105fc", "result_sha256": "49523b3269db649aa70072058666da3309ecabb93242cd504b6413f5de0f5d54", "trial_id": "cells/tp2_mns64/high-rep1/result.json" }, "target": { "requests_sha256": "d95e0dcc3d3f25af11b4d3da0ca133f647fd602753b804337a7131cc56e03059", "result_sha256": "f37a42d9ea734adcb24892f86cbb19e2c51cee0baf8a6a63c5d51c4820bd28f2", "trial_id": "cells/tp2_mns64/high-rep2/result.json" } }, { "delta_outcome": { "admitted_fraction": 0.005797101449275366, "admitted_input_tokens_mean_over_limit": -0.08122977688274602, "completed_fail_fraction_of_total": 0.0, "completed_over_admitted": 0.10543840177580466, "completed_pass_rate": 0.0, "outstanding_over_admitted": -0.10543840177580466, "tpot_max_over_slo": -0.08790813528521968, "tpot_mean_over_slo": -0.0976607763237542, "ttft_max_over_slo_max": -0.031135886325500906, "ttft_mean_over_slo_max": -0.007596982260572371 }, "delta_state": { "batch_size.mean": -2.9554801420404466, "batch_tokens.mean": -131.8397196967789, "decode_batch_size.mean": -2.9177065756667595, "graph_full_share": 0.026547586067947715, "graph_none_share": -0.024236910368448883, "graph_padding_fraction": 0.005457793019402147, "kv_usage_end_minus_start": -0.01758559388149228, "kv_usage_max": -0.016281062553556214, "kv_usage_mean": -0.00956401468061617, "preemptions": 0.0, "prefill_token_fraction": -0.00900387036264405, "queue_running_mean": -2.3104728836999993, "queue_waiting_mean": 0.0, "scheduler_steps_per_s": 20.099999999999994 }, "group": { "cell": "tp2_mns64", "level": "high", "mns": 64, "source_replicate": 2, "target_replicate": 3, "tp": 2 }, "kind": "same_config_workload_repeat", "source": { "requests_sha256": "d95e0dcc3d3f25af11b4d3da0ca133f647fd602753b804337a7131cc56e03059", "result_sha256": "f37a42d9ea734adcb24892f86cbb19e2c51cee0baf8a6a63c5d51c4820bd28f2", "trial_id": "cells/tp2_mns64/high-rep2/result.json" }, "target": { "requests_sha256": "ccba3d0a18a7db8501b040855bda473ee55c519619d4b9fa2a105bda06300829", "result_sha256": "55447419925f30397d19863e149890241c58863d0adaafd511a67757905cb58e", "trial_id": "cells/tp2_mns64/high-rep3/result.json" } }, { "delta_outcome": { "admitted_fraction": -0.004255319148936149, "admitted_input_tokens_mean_over_limit": 0.027061719804318485, "completed_fail_fraction_of_total": 0.0, "completed_over_admitted": -0.002024291497975783, "completed_pass_rate": 0.0, "outstanding_over_admitted": 0.0020242914979757, "tpot_max_over_slo": 0.09163424472588544, "tpot_mean_over_slo": 0.0001830762350235604, "ttft_max_over_slo_max": 0.0006185861614843163, "ttft_mean_over_slo_max": 0.0017585336856889636 }, "delta_state": { "batch_size.mean": -0.4199832128448815, "batch_tokens.mean": 6.494131535483177, "decode_batch_size.mean": -0.4188308558741509, "graph_full_share": 0.0010917066038501622, "graph_none_share": 0.00011588552243251699, "graph_padding_fraction": -0.0006875875614091621, "kv_usage_end_minus_start": 0.009093350727084282, "kv_usage_max": 0.011843098134008634, "kv_usage_mean": 0.0004732259215823806, "preemptions": 0.0, "prefill_token_fraction": 0.005759524366025781, "queue_running_mean": -0.24529797560000066, "queue_waiting_mean": 0.0, "scheduler_steps_per_s": 2.799999999999997 }, "group": { "cell": "tp2_mns64", "level": "low", "mns": 64, "source_replicate": 1, "target_replicate": 2, "tp": 2 }, "kind": "same_config_workload_repeat", "source": { "requests_sha256": "8685f475baf247f7ce52a6a2df1183cd77a6f580cd57a3a87c407b4e96c0895f", "result_sha256": "7a8f4e462d496bd56b73b0a686653cbe499a238df62a8c9a50db687d051f1c68", "trial_id": "cells/tp2_mns64/low-rep1/result.json" }, "target": { "requests_sha256": "016a76ab3dc015700f9ecb9f60a20fcc3b456c45270472fdd1d04e84ebd7021a", "result_sha256": "368869326f44a0032985eff9d3abfa2306b93c25370caa05cee255a8c0208c0e", "trial_id": "cells/tp2_mns64/low-rep2/result.json" } }, { "delta_outcome": { "admitted_fraction": -0.008510638297872353, "admitted_input_tokens_mean_over_limit": -0.08059763768960165, "completed_fail_fraction_of_total": 0.0, "completed_over_admitted": -0.2543859649122807, "completed_pass_rate": 0.0, "outstanding_over_admitted": 0.2543859649122807, "tpot_max_over_slo": -0.1787865552757903, "tpot_mean_over_slo": -0.07545848129810456, "ttft_max_over_slo_max": -0.024053132161498063, "ttft_mean_over_slo_max": -0.008335181439306037 }, "delta_state": { "batch_size.mean": -1.151692859112214, "batch_tokens.mean": -39.27803694513372, "decode_batch_size.mean": -1.1449765646539838, "graph_full_share": 0.006457127102288407, "graph_none_share": -0.008497380755445271, "graph_padding_fraction": 6.114154911974513e-05, "kv_usage_end_minus_start": 0.012533732366445394, "kv_usage_max": -0.022906035375820144, "kv_usage_mean": -0.00690787067830519, "preemptions": 0.0, "prefill_token_fraction": -0.0041771367400077475, "queue_running_mean": -1.9580360157999994, "queue_waiting_mean": 0.0, "scheduler_steps_per_s": 14.099999999999994 }, "group": { "cell": "tp2_mns64", "level": "low", "mns": 64, "source_replicate": 2, "target_replicate": 3, "tp": 2 }, "kind": "same_config_workload_repeat", "source": { "requests_sha256": "016a76ab3dc015700f9ecb9f60a20fcc3b456c45270472fdd1d04e84ebd7021a", "result_sha256": "368869326f44a0032985eff9d3abfa2306b93c25370caa05cee255a8c0208c0e", "trial_id": "cells/tp2_mns64/low-rep2/result.json" }, "target": { "requests_sha256": "5455d1cdeabd6e2937130ed8fc8d86c1aaa6f04cbf26c3cb69ad064be5f67d73", "result_sha256": "2de5c4524f57b487b94b4fdba35defc2a566236fc3e31f7260b7a9f85bb7007c", "trial_id": "cells/tp2_mns64/low-rep3/result.json" } }, { "delta_outcome": { "admitted_fraction": -0.01785714285714285, "admitted_input_tokens_mean_over_limit": 0.0482287348533163, "completed_fail_fraction_of_total": 0.0, "completed_over_admitted": 0.007421150278293154, "completed_pass_rate": 0.0, "outstanding_over_admitted": -0.007421150278293098, "tpot_max_over_slo": 0.045055894329630586, "tpot_mean_over_slo": 0.03517161669315627, "ttft_max_over_slo_max": 0.04909342716564424, "ttft_mean_over_slo_max": 0.07044418369208602 }, "delta_state": { "batch_size.mean": 0.9503756445719365, "batch_tokens.mean": 50.369704265273356, "decode_batch_size.mean": 0.9415787316873265, "graph_full_share": -0.009959703582283153, "graph_none_share": 0.009959703582283236, "graph_padding_fraction": -0.001694114329886807, "kv_usage_end_minus_start": 0.0035348314893507427, "kv_usage_max": 0.009660171254290972, "kv_usage_mean": 0.004883298444633271, "preemptions": 0.0, "prefill_token_fraction": 0.007370470138318375, "queue_running_mean": 0.27710616229999996, "queue_waiting_mean": 1.1825361130999998, "scheduler_steps_per_s": -17.900000000000006 }, "group": { "cell": "tp2_mns8", "level": "high", "mns": 8, "source_replicate": 1, "target_replicate": 2, "tp": 2 }, "kind": "same_config_workload_repeat", "source": { "requests_sha256": "5d41a316f329e2efcc0f728cdd078377eb72fb9258738636db704bcfcca290aa", "result_sha256": "466b3991190a232c984ba1b54fb6793cac1893dce196ee96f7238d70d49fcb74", "trial_id": "cells/tp2_mns8/high-rep1/result.json" }, "target": { "requests_sha256": "1b7d2cdcd639c00c443ae1b98287a7cc57323eb80c7f0e1bfc72f1ab5bf48760", "result_sha256": "e9156d6871cc8cb3a20dce629cc2443df7473f3836b46eb0944d491ea5c4d44c", "trial_id": "cells/tp2_mns8/high-rep2/result.json" } }, { "delta_outcome": { "admitted_fraction": 0.005952380952380931, "admitted_input_tokens_mean_over_limit": -0.08086076227365946, "completed_fail_fraction_of_total": 0.0, "completed_over_admitted": 0.04961984793917562, "completed_pass_rate": 0.0, "outstanding_over_admitted": -0.049619847939175676, "tpot_max_over_slo": -0.03665214141383477, "tpot_mean_over_slo": -0.03040908016559296, "ttft_max_over_slo_max": -0.06188676766275117, "ttft_mean_over_slo_max": -0.061940220978646765 }, "delta_state": { "batch_size.mean": -0.9308857183911066, "batch_tokens.mean": -38.259866334430285, "decode_batch_size.mean": -0.9247474057174099, "graph_full_share": 0.009718264940999921, "graph_none_share": -0.009718264941000004, "graph_padding_fraction": 0.0021664497846041342, "kv_usage_end_minus_start": 0.0029988642599186877, "kv_usage_max": -0.006508173500248904, "kv_usage_mean": -0.0037595414789287837, "preemptions": 0.0, "prefill_token_fraction": -0.0037648024150247483, "queue_running_mean": -0.21647643550000062, "queue_waiting_mean": -1.8036835434999998, "scheduler_steps_per_s": 15.700000000000003 }, "group": { "cell": "tp2_mns8", "level": "high", "mns": 8, "source_replicate": 2, "target_replicate": 3, "tp": 2 }, "kind": "same_config_workload_repeat", "source": { "requests_sha256": "1b7d2cdcd639c00c443ae1b98287a7cc57323eb80c7f0e1bfc72f1ab5bf48760", "result_sha256": "e9156d6871cc8cb3a20dce629cc2443df7473f3836b46eb0944d491ea5c4d44c", "trial_id": "cells/tp2_mns8/high-rep2/result.json" }, "target": { "requests_sha256": "5a68fda7c29477e5b3e66d70ad2b05b77be6c200cc1beea89d663a5729f31a2b", "result_sha256": "f2116797b35057949acbf1f57898befac9ff18b94c808462f0969bdcf2446dd6", "trial_id": "cells/tp2_mns8/high-rep3/result.json" } }, { "delta_outcome": { "admitted_fraction": -0.008733624454148492, "admitted_input_tokens_mean_over_limit": 0.03586033261326227, "completed_fail_fraction_of_total": 0.0, "completed_over_admitted": 0.07276507276507282, "completed_pass_rate": 0.0, "outstanding_over_admitted": -0.07276507276507277, "tpot_max_over_slo": -0.00577808629105403, "tpot_mean_over_slo": -0.014899153875042898, "ttft_max_over_slo_max": 0.014347533161829545, "ttft_mean_over_slo_max": 0.00013980660489499447 }, "delta_state": { "batch_size.mean": -0.6370388590554663, "batch_tokens.mean": -0.24447251544799542, "decode_batch_size.mean": -0.6327331911137266, "graph_full_share": 0.004305667941739033, "graph_none_share": -0.004305667941739005, "graph_padding_fraction": 0.0015483374551458753, "kv_usage_end_minus_start": -0.0016334239373173576, "kv_usage_max": 0.0027181195206921194, "kv_usage_mean": -0.0003114508043975048, "preemptions": 0.0, "prefill_token_fraction": 0.005116159306570833, "queue_running_mean": -0.5131508089000008, "queue_waiting_mean": -0.13343640950000002, "scheduler_steps_per_s": 7.5 }, "group": { "cell": "tp2_mns8", "level": "low", "mns": 8, "source_replicate": 1, "target_replicate": 2, "tp": 2 }, "kind": "same_config_workload_repeat", "source": { "requests_sha256": "455a1c21388700f0e3234c49d15ba3b0426b0d3b551c096b297793afa0d4602f", "result_sha256": "13de25a73cc8c752245f169cfca1d64220f697c15b835a646ba969a1005047fb", "trial_id": "cells/tp2_mns8/low-rep1/result.json" }, "target": { "requests_sha256": "cae4b764ebb8714a6f5d2009bb79497420b388986a992c51410f1681810c14c6", "result_sha256": "92f92a576e990946c4767a2107d2ce0e6ba1b066337911399360c199fae3f5e4", "trial_id": "cells/tp2_mns8/low-rep2/result.json" } }, { "delta_outcome": { "admitted_fraction": -0.008733624454148464, "admitted_input_tokens_mean_over_limit": -0.08428101999879345, "completed_fail_fraction_of_total": 0.0, "completed_over_admitted": -0.23320463320463325, "completed_pass_rate": 0.0, "outstanding_over_admitted": 0.2332046332046332, "tpot_max_over_slo": -0.02744940897875181, "tpot_mean_over_slo": -0.059641868091682704, "ttft_max_over_slo_max": -0.10803754650017558, "ttft_mean_over_slo_max": -0.032639687448619975 }, "delta_state": { "batch_size.mean": -1.203120302435217, "batch_tokens.mean": -43.54127643086966, "decode_batch_size.mean": -1.1934092223734076, "graph_full_share": 0.010598391508127913, "graph_none_share": -0.0105983915081279, "graph_padding_fraction": 0.004406513335831962, "kv_usage_end_minus_start": 0.00967293237880118, "kv_usage_max": -0.008881742659163083, "kv_usage_mean": -0.005494017047481575, "preemptions": 0.0, "prefill_token_fraction": -0.0054281680100684015, "queue_running_mean": -1.7641761531999998, "queue_waiting_mean": -0.32204582449999997, "scheduler_steps_per_s": 15.600000000000009 }, "group": { "cell": "tp2_mns8", "level": "low", "mns": 8, "source_replicate": 2, "target_replicate": 3, "tp": 2 }, "kind": "same_config_workload_repeat", "source": { "requests_sha256": "cae4b764ebb8714a6f5d2009bb79497420b388986a992c51410f1681810c14c6", "result_sha256": "92f92a576e990946c4767a2107d2ce0e6ba1b066337911399360c199fae3f5e4", "trial_id": "cells/tp2_mns8/low-rep2/result.json" }, "target": { "requests_sha256": "cc8c01e4560411cd5b84cdc9bebdda9a86b88615ee1c39a996affe669d26c167", "result_sha256": "c9a7d026bd89cc027c45b8f1899d56e21d53d4234e95db365a9492f670a78d36", "trial_id": "cells/tp2_mns8/low-rep3/result.json" } }, { "delta_outcome": { "admitted_fraction": -0.005333333333333329, "admitted_input_tokens_mean_over_limit": 0.00313201278155667, "completed_fail_fraction_of_total": 0.004, "completed_over_admitted": 0.00267618198037467, "completed_pass_rate": -0.048128342245989275, "outstanding_over_admitted": -0.00267618198037467, "tpot_max_over_slo": 0.01678296282344799, "tpot_mean_over_slo": 0.03333551369285431, "ttft_max_over_slo_max": 0.04254181500679499, "ttft_mean_over_slo_max": 0.027600657575446386 }, "delta_state": { "batch_size.mean": 2.617504152501592, "batch_tokens.mean": 83.36000248373932, "decode_batch_size.mean": 2.6009989288874404, "graph_full_share": -0.030922553904903838, "graph_none_share": 0.027604433474596013, "graph_padding_fraction": -8.22833741022888e-05, "kv_usage_end_minus_start": -0.001912904196656684, "kv_usage_max": 0.0016425470701958433, "kv_usage_mean": 0.002711903470305097, "preemptions": 0.0, "prefill_token_fraction": 0.0031462566114718316, "queue_running_mean": 1.2096441719000008, "queue_waiting_mean": 2.3781169752999993, "scheduler_steps_per_s": -13.399999999999991 }, "group": { "cell": "tp4_mns16", "level": "high", "mns": 16, "source_replicate": 1, "target_replicate": 2, "tp": 4 }, "kind": "same_config_workload_repeat", "source": { "requests_sha256": "8695422a0f1e0e1c097a08a7f4c082f1ba2e3bb26b58caadae0afc7438aeca90", "result_sha256": "c54a2bb5c1b78bdef3fa09c39f5974215370d6fc7c2b6d403a3d376d53a1d9eb", "trial_id": "cells/tp4_mns16/high-rep1/result.json" }, "target": { "requests_sha256": "aa0482b10414d911af227e2df172c19ecca0add2e0024dc88f85ffbd063854bb", "result_sha256": "91e2491de4ca0868b667e7d126b755f03bdd78378681b8070a2bbc0c38fb9055", "trial_id": "cells/tp4_mns16/high-rep2/result.json" } }, { "delta_outcome": { "admitted_fraction": -0.006666666666666654, "admitted_input_tokens_mean_over_limit": -0.045669445146517385, "completed_fail_fraction_of_total": -0.012, "completed_over_admitted": 0.09077740222114916, "completed_pass_rate": 0.13636363636363635, "outstanding_over_admitted": -0.09077740222114916, "tpot_max_over_slo": 0.006141297796037171, "tpot_mean_over_slo": -0.021391237509928385, "ttft_max_over_slo_max": -0.2410895283295152, "ttft_mean_over_slo_max": -0.12306385203762282 }, "delta_state": { "batch_size.mean": 0.023181238824859562, "batch_tokens.mean": 2.5905518549044473, "decode_batch_size.mean": 0.011653673102646422, "graph_full_share": 0.0012768372268986417, "graph_none_share": 0.001837564686330409, "graph_padding_fraction": 0.0006975685594821312, "kv_usage_end_minus_start": 0.001387493177308352, "kv_usage_max": 0.0022138677902641124, "kv_usage_mean": -0.0011780510256150529, "preemptions": 0.0, "prefill_token_fraction": 0.0002680116189619497, "queue_running_mean": 0.1377117206999987, "queue_waiting_mean": -11.247484751, "scheduler_steps_per_s": 2.5 }, "group": { "cell": "tp4_mns16", "level": "high", "mns": 16, "source_replicate": 2, "target_replicate": 3, "tp": 4 }, "kind": "same_config_workload_repeat", "source": { "requests_sha256": "aa0482b10414d911af227e2df172c19ecca0add2e0024dc88f85ffbd063854bb", "result_sha256": "91e2491de4ca0868b667e7d126b755f03bdd78378681b8070a2bbc0c38fb9055", "trial_id": "cells/tp4_mns16/high-rep2/result.json" }, "target": { "requests_sha256": "3a23ffbe0e724648226152ef5433cc5825ad640c37cf47997c2f091210eceeec", "result_sha256": "136dde0451f806841ff3996b081fe0bffac0a76f60c3d78b7516f2689e106008", "trial_id": "cells/tp4_mns16/high-rep3/result.json" } }, { "delta_outcome": { "admitted_fraction": 0.0215686274509804, "admitted_input_tokens_mean_over_limit": -0.007677676072761208, "completed_fail_fraction_of_total": 0.0, "completed_over_admitted": 0.12342135476463834, "completed_pass_rate": 0.0, "outstanding_over_admitted": -0.12342135476463834, "tpot_max_over_slo": 0.042659409131532156, "tpot_mean_over_slo": 0.06205778281240598, "ttft_max_over_slo_max": 0.09863472549477593, "ttft_mean_over_slo_max": 0.04373689836276399 }, "delta_state": { "batch_size.mean": 3.8253490040280305, "batch_tokens.mean": 77.16682826085241, "decode_batch_size.mean": 3.80033895365794, "graph_full_share": -0.03715089743892919, "graph_none_share": 0.034228013337432306, "graph_padding_fraction": -0.0048637698153220395, "kv_usage_end_minus_start": -0.007677122175915652, "kv_usage_max": 0.005958058937853616, "kv_usage_mean": 0.004083928262059737, "preemptions": 0.0, "prefill_token_fraction": -0.0015885934328132567, "queue_running_mean": 4.248843562199999, "queue_waiting_mean": 1.9316373561999998, "scheduler_steps_per_s": -12.799999999999997 }, "group": { "cell": "tp4_mns16", "level": "low", "mns": 16, "source_replicate": 1, "target_replicate": 2, "tp": 4 }, "kind": "same_config_workload_repeat", "source": { "requests_sha256": "d7eaaefd35cc906f759dcaa7ac3c8543249a1a02f28572bec953e41e4379a29f", "result_sha256": "11d9ec9d2a3358aed988f7b4b1014af087d61c4b5a72eb9076ae19bb5cc481cd", "trial_id": "cells/tp4_mns16/low-rep1/result.json" }, "target": { "requests_sha256": "7fdb90f679bba1747fbacc92fac2045c820e09965c755b20f6dafd4fd9e26a7a", "result_sha256": "800e2aa6e9f9efef4be12ee778be6817364914d63ae0624775d02729089768d6", "trial_id": "cells/tp4_mns16/low-rep2/result.json" } }, { "delta_outcome": { "admitted_fraction": -0.023529411764705882, "admitted_input_tokens_mean_over_limit": 0.013229832504734862, "completed_fail_fraction_of_total": 0.0, "completed_over_admitted": -0.1573426573426573, "completed_pass_rate": 0.0, "outstanding_over_admitted": 0.15734265734265732, "tpot_max_over_slo": -0.05231123732945586, "tpot_mean_over_slo": -0.08222875707353122, "ttft_max_over_slo_max": -0.11858330133448666, "ttft_mean_over_slo_max": -0.04432077170326809 }, "delta_state": { "batch_size.mean": -6.136759360232643, "batch_tokens.mean": -119.22330840733238, "decode_batch_size.mean": -6.093614010489692, "graph_full_share": 0.04765150334943136, "graph_none_share": -0.049295061536064805, "graph_padding_fraction": 0.001850754011979509, "kv_usage_end_minus_start": 0.00735065319301953, "kv_usage_max": -0.003932931028326259, "kv_usage_mean": -0.00419858552745247, "preemptions": 0.0, "prefill_token_fraction": 0.0044424417100876745, "queue_running_mean": -5.4834562212, "queue_waiting_mean": -1.804526599, "scheduler_steps_per_s": 31.299999999999997 }, "group": { "cell": "tp4_mns16", "level": "low", "mns": 16, "source_replicate": 2, "target_replicate": 3, "tp": 4 }, "kind": "same_config_workload_repeat", "source": { "requests_sha256": "7fdb90f679bba1747fbacc92fac2045c820e09965c755b20f6dafd4fd9e26a7a", "result_sha256": "800e2aa6e9f9efef4be12ee778be6817364914d63ae0624775d02729089768d6", "trial_id": "cells/tp4_mns16/low-rep2/result.json" }, "target": { "requests_sha256": "b592fe254476957265fc9139aefdea77be4b5540f3c1b8a429c10ea6ccab9a6a", "result_sha256": "cc1b1f7f5280c2892d0c6deac9df9f34ce126b0284e6ad893ffbf0ddbf27969f", "trial_id": "cells/tp4_mns16/low-rep3/result.json" } }, { "delta_outcome": { "admitted_fraction": -0.005333333333333329, "admitted_input_tokens_mean_over_limit": 0.00313201278155667, "completed_fail_fraction_of_total": 0.0, "completed_over_admitted": 0.005798394290811748, "completed_pass_rate": 0.0, "outstanding_over_admitted": -0.005798394290811748, "tpot_max_over_slo": 0.1973487464546072, "tpot_mean_over_slo": 0.10397174357977113, "ttft_max_over_slo_max": -0.001787370662592977, "ttft_mean_over_slo_max": 0.007868458508126674 }, "delta_state": { "batch_size.mean": 11.95331828442438, "batch_tokens.mean": 306.10848411182496, "decode_batch_size.mean": 11.860920298662965, "graph_full_share": -0.07233894773398164, "graph_none_share": 0.06742142733113388, "graph_padding_fraction": -0.0007349090714318936, "kv_usage_end_minus_start": -0.0056930844866003305, "kv_usage_max": -0.005488664935914489, "kv_usage_mean": 0.006218554442237678, "preemptions": 0.0, "prefill_token_fraction": -0.002281107503328106, "queue_running_mean": 4.266314980200001, "queue_waiting_mean": -0.038871071699999996, "scheduler_steps_per_s": -11.799999999999997 }, "group": { "cell": "tp4_mns64", "level": "high", "mns": 64, "source_replicate": 1, "target_replicate": 2, "tp": 4 }, "kind": "same_config_workload_repeat", "source": { "requests_sha256": "b4aeaf5b3175bd85fdac08a2ef657fb2278c6c4b36ea32ea417275ab6d55812d", "result_sha256": "111a1042b614d05772888665a7cfbc86925b2fbeb52ffe19edb62a5bcfe2f513", "trial_id": "cells/tp4_mns64/high-rep1/result.json" }, "target": { "requests_sha256": "5d82b6885f5c654ae1ae126696df4bafe04f5be1c1109db24c7856b0456d9443", "result_sha256": "6ed574c18723e265ab1327c5e43d3c76d37b3fc5711d06c77a51718e4ba16601", "trial_id": "cells/tp4_mns64/high-rep2/result.json" } }, { "delta_outcome": { "admitted_fraction": -0.006666666666666654, "admitted_input_tokens_mean_over_limit": -0.045669445146517385, "completed_fail_fraction_of_total": 0.0, "completed_over_admitted": 0.1202317720907774, "completed_pass_rate": 0.0, "outstanding_over_admitted": -0.1202317720907774, "tpot_max_over_slo": -0.2208674842524746, "tpot_mean_over_slo": -0.3211407366971467, "ttft_max_over_slo_max": -0.007970502677684024, "ttft_mean_over_slo_max": -0.009044386120380173 }, "delta_state": { "batch_size.mean": -17.26249134948097, "batch_tokens.mean": -537.2182539260048, "decode_batch_size.mean": -17.083955283470857, "graph_full_share": 0.10428533404311957, "graph_none_share": -0.09966462603140805, "graph_padding_fraction": 0.004369015708508233, "kv_usage_end_minus_start": -0.020942782967763085, "kv_usage_max": -0.019762260062552328, "kv_usage_mean": -0.018462380855259714, "preemptions": 0.0, "prefill_token_fraction": -0.0019184118517130422, "queue_running_mean": -18.914856451900004, "queue_waiting_mean": -0.0374047109, "scheduler_steps_per_s": 25.299999999999997 }, "group": { "cell": "tp4_mns64", "level": "high", "mns": 64, "source_replicate": 2, "target_replicate": 3, "tp": 4 }, "kind": "same_config_workload_repeat", "source": { "requests_sha256": "5d82b6885f5c654ae1ae126696df4bafe04f5be1c1109db24c7856b0456d9443", "result_sha256": "6ed574c18723e265ab1327c5e43d3c76d37b3fc5711d06c77a51718e4ba16601", "trial_id": "cells/tp4_mns64/high-rep2/result.json" }, "target": { "requests_sha256": "5c5861d716181f06dc3f951ae0d6ee479b1b61fc21ebf24541c5b45053fcc167", "result_sha256": "35b8f0465f6b0c197f71b5ecaf4e771e6613f86998ae44f8701ab9e4e1e440b4", "trial_id": "cells/tp4_mns64/high-rep3/result.json" } }, { "delta_outcome": { "admitted_fraction": 0.0215686274509804, "admitted_input_tokens_mean_over_limit": -0.007677676072761208, "completed_fail_fraction_of_total": 0.0, "completed_over_admitted": 0.17470340604668966, "completed_pass_rate": 0.0, "outstanding_over_admitted": -0.17470340604668963, "tpot_max_over_slo": 0.048643136378671015, "tpot_mean_over_slo": 0.059397215574861106, "ttft_max_over_slo_max": 0.01085578350466676, "ttft_mean_over_slo_max": 0.0005479300281428442 }, "delta_state": { "batch_size.mean": 3.3204058305140567, "batch_tokens.mean": 67.87133685773168, "decode_batch_size.mean": 3.294527414991851, "graph_full_share": -0.02348818567751032, "graph_none_share": 0.022293070755162486, "graph_padding_fraction": -0.0005360553529459057, "kv_usage_end_minus_start": -0.010353850242237006, "kv_usage_max": 0.004231484699196653, "kv_usage_mean": 0.0043582810100513195, "preemptions": 0.0, "prefill_token_fraction": -0.001346586578878206, "queue_running_mean": 4.306671635299999, "queue_waiting_mean": -0.000525882, "scheduler_steps_per_s": -14.100000000000009 }, "group": { "cell": "tp4_mns64", "level": "low", "mns": 64, "source_replicate": 1, "target_replicate": 2, "tp": 4 }, "kind": "same_config_workload_repeat", "source": { "requests_sha256": "931d372ded1e3aaa9c67f06653fac8f79b4ee4f369028c1148163580386681ce", "result_sha256": "325ce57f33133670a6761b5dffb831ed3b2f464755ff743b56976b4a974c1215", "trial_id": "cells/tp4_mns64/low-rep1/result.json" }, "target": { "requests_sha256": "72c8f70a240f79edcaa4eb0a2fe9bade8ac07093339c5b2d09c49ddd1cf27d16", "result_sha256": "ff8b2283d02d5967893ca6225b561ae11011da6149d1f6cff7b4275c00170e1d", "trial_id": "cells/tp4_mns64/low-rep2/result.json" } }, { "delta_outcome": { "admitted_fraction": -0.023529411764705882, "admitted_input_tokens_mean_over_limit": 0.013229832504734862, "completed_fail_fraction_of_total": 0.0, "completed_over_admitted": -0.2086247086247086, "completed_pass_rate": 0.0, "outstanding_over_admitted": 0.2086247086247086, "tpot_max_over_slo": -0.049497940473067825, "tpot_mean_over_slo": -0.07097474071479923, "ttft_max_over_slo_max": 0.0113132930030891, "ttft_mean_over_slo_max": 0.0027290206974555573 }, "delta_state": { "batch_size.mean": -4.516372320824056, "batch_tokens.mean": -75.80960433136954, "decode_batch_size.mean": -4.485458989758809, "graph_full_share": 0.024653639771889058, "graph_none_share": -0.023884511783377624, "graph_padding_fraction": -0.0020895387371847093, "kv_usage_end_minus_start": 0.011432163372104731, "kv_usage_max": -0.003015188372616051, "kv_usage_mean": -0.004106181760760127, "preemptions": 0.0, "prefill_token_fraction": 0.006171117400546322, "queue_running_mean": -5.0125981476999995, "queue_waiting_mean": 0.000454857, "scheduler_steps_per_s": 21.80000000000001 }, "group": { "cell": "tp4_mns64", "level": "low", "mns": 64, "source_replicate": 2, "target_replicate": 3, "tp": 4 }, "kind": "same_config_workload_repeat", "source": { "requests_sha256": "72c8f70a240f79edcaa4eb0a2fe9bade8ac07093339c5b2d09c49ddd1cf27d16", "result_sha256": "ff8b2283d02d5967893ca6225b561ae11011da6149d1f6cff7b4275c00170e1d", "trial_id": "cells/tp4_mns64/low-rep2/result.json" }, "target": { "requests_sha256": "8b2c779d2097c5e4038e94e1ac9b986e29f55d9acfd67eeb366c4e0e9270f6e0", "result_sha256": "f026b69356701894fab184e0f2f7504a30b537982e823373899c9ab123a33891", "trial_id": "cells/tp4_mns64/low-rep3/result.json" } } ], "response_statistics": { "batch_size.mean": { "action_above_repeat_p95_fraction": 0.08333333333333333, "action_abs_median": 0.6185880702088911, "action_delta": { "distinct_n": 12, "max": 19.42960244648318, "min": -1.4718645603306726, "n": 12 }, "action_signs": { "consistency": 0.6666666666666666, "negative": 4, "positive": 8, "zero": 0 }, "effect_to_repeat_median": 0.2658268122367889, "effect_to_repeat_median_is_infinite": false, "gate_feature": false, "qualifies": false, "repeat_abs_median": 2.327034150557677, "repeat_abs_p95": 11.456264844377335, "repeat_delta": { "distinct_n": 24, "max": 11.95331828442438, "min": -17.26249134948097, "n": 24 } }, "batch_tokens.mean": { "action_above_repeat_p95_fraction": 0.16666666666666666, "action_abs_median": 17.059106336421053, "action_delta": { "distinct_n": 12, "max": 656.7778499176663, "min": -31.86018426395526, "n": 12 }, "action_signs": { "consistency": 0.5833333333333334, "negative": 5, "positive": 7, "zero": 0 }, "effect_to_repeat_median": 0.2374581652269279, "effect_to_repeat_median_is_infinite": false, "gate_feature": false, "qualifies": false, "repeat_abs_median": 71.84047059455061, "repeat_abs_p95": 327.5325978555095, "repeat_delta": { "distinct_n": 24, "max": 306.10848411182496, "min": -537.2182539260048, "n": 24 } }, "decode_batch_size.mean": { "action_above_repeat_p95_fraction": 0.08333333333333333, "action_abs_median": 0.6115154242137617, "action_delta": { "distinct_n": 12, "max": 19.184206069160197, "min": -1.4608669627184998, "n": 12 }, "action_signs": { "consistency": 0.6666666666666666, "negative": 4, "positive": 8, "zero": 0 }, "effect_to_repeat_median": 0.2650837352505827, "effect_to_repeat_median_is_infinite": false, "gate_feature": true, "qualifies": false, "repeat_abs_median": 2.3068764427802346, "repeat_abs_p95": 11.363988993022012, "repeat_delta": { "distinct_n": 24, "max": 11.860920298662965, "min": -17.083955283470857, "n": 24 } }, "graph_full_share": { "action_above_repeat_p95_fraction": 0.16666666666666666, "action_abs_median": 0.005588197661352101, "action_delta": { "distinct_n": 12, "max": 0.02356281394840931, "min": -0.13147494707127738, "n": 12 }, "action_signs": { "consistency": 0.5833333333333334, "negative": 7, "positive": 5, "zero": 0 }, "effect_to_repeat_median": 0.27121228977463446, "effect_to_repeat_median_is_infinite": false, "gate_feature": false, "qualifies": false, "repeat_abs_median": 0.020604514883878045, "repeat_abs_p95": 0.08279928773704351, "repeat_delta": { "distinct_n": 24, "max": 0.10428533404311957, "min": -0.07611315423133669, "n": 24 } }, "graph_none_share": { "action_above_repeat_p95_fraction": 0.08333333333333333, "action_abs_median": 0.011395045691744581, "action_delta": { "distinct_n": 12, "max": 0.11760056457304163, "min": -0.027804605789997328, "n": 12 }, "action_signs": { "consistency": 0.6666666666666666, "negative": 8, "positive": 4, "zero": 0 }, "effect_to_repeat_median": 0.5196295717736035, "effect_to_repeat_median_is_infinite": false, "gate_feature": false, "qualifies": false, "repeat_abs_median": 0.02192917091467856, "repeat_abs_p95": 0.08301950470733124, "repeat_delta": { "distinct_n": 24, "max": 0.08050759417399073, "min": -0.09966462603140805, "n": 24 } }, "graph_padding_fraction": { "action_above_repeat_p95_fraction": 0.08333333333333333, "action_abs_median": 0.0024624382794725458, "action_delta": { "distinct_n": 12, "max": 0.008784696835623326, "min": -0.0007445684074620391, "n": 12 }, "action_signs": { "consistency": 0.8333333333333334, "negative": 2, "positive": 10, "zero": 0 }, "effect_to_repeat_median": 1.15716396642797, "effect_to_repeat_median_is_infinite": false, "gate_feature": true, "qualifies": false, "repeat_abs_median": 0.0021279942608944218, "repeat_abs_p95": 0.007335491197810787, "repeat_delta": { "distinct_n": 24, "max": 0.0086692890425959, "min": -0.007424342975302652, "n": 24 } }, "kv_usage_end_minus_start": { "action_above_repeat_p95_fraction": 0.0, "action_abs_median": 0.0009122627395202754, "action_delta": { "distinct_n": 12, "max": 0.05537535994708087, "min": -0.002621229743826836, "n": 12 }, "action_signs": { "consistency": 0.9166666666666666, "negative": 1, "positive": 11, "zero": 0 }, "effect_to_repeat_median": 0.08374755984910728, "effect_to_repeat_median_is_infinite": false, "gate_feature": false, "qualifies": false, "repeat_abs_median": 0.010893006807170869, "repeat_abs_p95": 0.08088054822007018, "repeat_delta": { "distinct_n": 24, "max": 0.08232590668398432, "min": -0.12798481366769898, "n": 24 } }, "kv_usage_max": { "action_above_repeat_p95_fraction": 0.0, "action_abs_median": 0.0070274347692272365, "action_delta": { "distinct_n": 12, "max": 0.08922043547542868, "min": 0.0002683613903166071, "n": 12 }, "action_signs": { "consistency": 1.0, "negative": 0, "positive": 12, "zero": 0 }, "effect_to_repeat_median": 0.49974360815926006, "effect_to_repeat_median_is_infinite": false, "gate_feature": false, "qualifies": false, "repeat_abs_median": 0.014062080343782424, "repeat_abs_p95": 0.10858976920771296, "repeat_delta": { "distinct_n": 24, "max": 0.11409731241882304, "min": -0.12688580277750028, "n": 24 } }, "kv_usage_mean": { "action_above_repeat_p95_fraction": 0.0, "action_abs_median": 0.0019193628331501261, "action_delta": { "distinct_n": 12, "max": 0.04639671324585863, "min": -0.0003621004636635179, "n": 12 }, "action_signs": { "consistency": 0.75, "negative": 3, "positive": 9, "zero": 0 }, "effect_to_repeat_median": 0.29244258288516367, "effect_to_repeat_median_is_infinite": false, "gate_feature": true, "qualifies": false, "repeat_abs_median": 0.006563212560271434, "repeat_abs_p95": 0.08042579042638397, "repeat_delta": { "distinct_n": 24, "max": 0.08571139047521727, "min": -0.09580544440945148, "n": 24 } }, "preemptions": { "action_above_repeat_p95_fraction": 0.0, "action_abs_median": 0.0, "action_delta": { "distinct_n": 1, "max": 0.0, "min": 0.0, "n": 12 }, "action_signs": { "consistency": 0.0, "negative": 0, "positive": 0, "zero": 12 }, "effect_to_repeat_median": 0.0, "effect_to_repeat_median_is_infinite": false, "gate_feature": false, "qualifies": false, "repeat_abs_median": 0.0, "repeat_abs_p95": 0.0, "repeat_delta": { "distinct_n": 1, "max": 0.0, "min": 0.0, "n": 24 } }, "prefill_token_fraction": { "action_above_repeat_p95_fraction": 0.0, "action_abs_median": 0.00026550413778614335, "action_delta": { "distinct_n": 12, "max": 0.013365962658000052, "min": -0.0003954091503666213, "n": 12 }, "action_signs": { "consistency": 0.6666666666666666, "negative": 4, "positive": 8, "zero": 0 }, "effect_to_repeat_median": 0.05125991549197503, "effect_to_repeat_median_is_infinite": false, "gate_feature": true, "qualifies": false, "repeat_abs_median": 0.005179566435838334, "repeat_abs_p95": 0.0190949186571384, "repeat_delta": { "distinct_n": 24, "max": 0.011636868678620127, "min": -0.024511024899494305, "n": 24 } }, "queue_running_mean": { "action_above_repeat_p95_fraction": 0.16666666666666666, "action_abs_median": 0.43384798364999977, "action_delta": { "distinct_n": 12, "max": 24.0049637595, "min": -0.46276202019999957, "n": 12 }, "action_signs": { "consistency": 0.6666666666666666, "negative": 4, "positive": 8, "zero": 0 }, "effect_to_repeat_median": 0.1773939239284337, "effect_to_repeat_median_is_infinite": false, "gate_feature": true, "qualifies": false, "repeat_abs_median": 2.4456755566499995, "repeat_abs_p95": 7.584720893309999, "repeat_delta": { "distinct_n": 24, "max": 7.2005321570000005, "min": -18.914856451900004, "n": 24 } }, "queue_waiting_mean": { "action_above_repeat_p95_fraction": 0.25, "action_abs_median": 0.22497528095, "action_delta": { "distinct_n": 11, "max": 0.0, "min": -15.5760687653, "n": 12 }, "action_signs": { "consistency": 1.0, "negative": 10, "positive": 0, "zero": 2 }, "effect_to_repeat_median": 2.611323424650003, "effect_to_repeat_median_is_infinite": false, "gate_feature": true, "qualifies": false, "repeat_abs_median": 0.0861537406, "repeat_abs_p95": 2.3661207382899994, "repeat_delta": { "distinct_n": 17, "max": 2.3781169752999993, "min": -11.247484751, "n": 24 } }, "scheduler_steps_per_s": { "action_above_repeat_p95_fraction": 0.0, "action_abs_median": 6.950000000000006, "action_delta": { "distinct_n": 11, "max": 11.900000000000006, "min": -34.5, "n": 12 }, "action_signs": { "consistency": 0.5, "negative": 6, "positive": 6, "zero": 0 }, "effect_to_repeat_median": 0.44408945686900986, "effect_to_repeat_median_is_infinite": false, "gate_feature": true, "qualifies": false, "repeat_abs_median": 15.650000000000006, "repeat_abs_p95": 50.385, "repeat_delta": { "distinct_n": 24, "max": 65.0, "min": -46.900000000000006, "n": 24 } } }, "sanity": { "action_pairs": 12, "invariants": { "efficacy_label_balance": true, "expected_action_pair_count": true, "expected_repeat_pair_count": true, "finite_deltas": true, "matched_action_request_hashes": true, "probabilities_bounded": true }, "red_flags": [], "repeat_pairs": 24, "trials": 36 } }, "5": { "actions": [ { "delta_outcome": { "admitted_fraction": 0.0, "admitted_input_tokens_mean_over_limit": 0.0, "completed_fail_fraction_of_total": 0.0, "completed_over_admitted": 0.0, "completed_pass_rate": 0.0, "outstanding_over_admitted": 0.0, "tpot_max_over_slo": -0.0008452736944928041, "tpot_mean_over_slo": -0.0021708008398531042, "ttft_max_over_slo_max": -0.00014104516594670713, "ttft_mean_over_slo_max": -0.00020825385150534392 }, "delta_state": { "batch_size.mean": -0.06880455035794819, "batch_tokens.mean": -1.0156712758654507, "decode_batch_size.mean": -0.067941551436697, "graph_full_share": 0.0008629989212514122, "graph_none_share": -0.006688241639697952, "graph_padding_fraction": 0.00030316259031496623, "kv_usage_end_minus_start": 2.700440393244019e-05, "kv_usage_max": 0.00012663114366773254, "kv_usage_mean": -6.934750080297807e-05, "preemptions": 0.0, "prefill_token_fraction": -0.000948468871704522, "queue_running_mean": -0.028601565200000234, "queue_waiting_mean": 0.0, "scheduler_steps_per_s": 4.0 }, "full_action_efficacy": 0, "full_feasibility_transition": "false->false", "group": { "level": "high", "offered_rate_per_gpu": 2.9833333333333334, "replicate": 1, "request_hash": "3f0ab9be34dbee884f5dbe2135b983d31a47578a43f606a5f390b0528dbedbe3", "tp": 1 }, "kind": "matched_mns_increase", "source": { "cell": "tp1_mns8", "early_stopped": true, "full_feasible": false, "full_pass_rate": 0.20670391061452514, "mns": 8, "requests_sha256": "5f5bfa4a16846190f488d3369eed89ae895804145cea59197e6b544b703223fc", "result_sha256": "f65e3dc91d17f5caab1f4aecf97d2c464b99f107231af2cfd96f5484880136ee", "trial_id": "cells/tp1_mns8/high-rep1/result.json" }, "target": { "cell": "tp1_mns64", "early_stopped": true, "full_feasible": false, "full_pass_rate": 0.24581005586592178, "mns": 64, "requests_sha256": "e970dcac20144cb1e13a35997377f6695a9c57d788a88bad7f18cf6c64e35cec", "result_sha256": "3233686b60b920dcd6cc2f45193a81f3af2d46f781ef42c4d117630612084feb", "trial_id": "cells/tp1_mns64/high-rep1/result.json" } }, { "delta_outcome": { "admitted_fraction": 0.0, "admitted_input_tokens_mean_over_limit": 0.0, "completed_fail_fraction_of_total": 0.0, "completed_over_admitted": -0.375, "completed_pass_rate": -1.0, "outstanding_over_admitted": 0.375, "tpot_max_over_slo": -0.562845061415821, "tpot_mean_over_slo": -0.49685783692809143, "ttft_max_over_slo_max": -0.1160576971694051, "ttft_mean_over_slo_max": -0.06709235994559196 }, "delta_state": { "batch_size.mean": 1.9650793650793652, "batch_tokens.mean": 127.72380952380956, "decode_batch_size.mean": 1.9158730158730162, "graph_full_share": -0.041269841269841345, "graph_none_share": 0.033333333333333326, "graph_padding_fraction": 0.0032253509426556993, "kv_usage_end_minus_start": 0.08606331121732713, "kv_usage_max": 0.04171319199456147, "kv_usage_mean": 0.01377304273482051, "preemptions": 0.0, "prefill_token_fraction": 0.0018893720284961502, "queue_running_mean": 1.9109891312, "queue_waiting_mean": -1.558876379, "scheduler_steps_per_s": -10.8 }, "full_action_efficacy": 0, "full_feasibility_transition": "false->false", "group": { "level": "high", "offered_rate_per_gpu": 2.9833333333333334, "replicate": 2, "request_hash": "b19c91876a8417fe06cb2775f65a1111b202f88ce10ec33576af3d84c059f02a", "tp": 1 }, "kind": "matched_mns_increase", "source": { "cell": "tp1_mns8", "early_stopped": true, "full_feasible": false, "full_pass_rate": 0.7988826815642458, "mns": 8, "requests_sha256": "e802d09057de349e5164c18314fcad09dc88c9673834fd6e5b100a57dd12fbf4", "result_sha256": "4835a985dd6bb8dc50b53737baa980ac95a49e22f0b8a7326e1fb416c9b4e798", "trial_id": "cells/tp1_mns8/high-rep2/result.json" }, "target": { "cell": "tp1_mns64", "early_stopped": true, "full_feasible": false, "full_pass_rate": 0.88268156424581, "mns": 64, "requests_sha256": "ec06878e2aefdc4853d972197cedc29ab94df7147abb2701e0d88b5dabf29728", "result_sha256": "8215477a014f061c55e94d879a25bbf1676ee2ef1c5e81efadd23d20d337e07a", "trial_id": "cells/tp1_mns64/high-rep2/result.json" } }, { "delta_outcome": { "admitted_fraction": 0.0, "admitted_input_tokens_mean_over_limit": 0.0, "completed_fail_fraction_of_total": 0.0, "completed_over_admitted": 0.0, "completed_pass_rate": 0.0, "outstanding_over_admitted": 0.0, "tpot_max_over_slo": -0.003354239532372044, "tpot_mean_over_slo": -0.002150681978709712, "ttft_max_over_slo_max": -0.0004135148386315776, "ttft_mean_over_slo_max": -0.00028004554942959864 }, "delta_state": { "batch_size.mean": -0.03143779072203845, "batch_tokens.mean": -11.489561245442736, "decode_batch_size.mean": -0.02922096970204935, "graph_full_share": 0.0022168210199891014, "graph_none_share": -0.00629845367305033, "graph_padding_fraction": 0.001725218661472923, "kv_usage_end_minus_start": -0.01687018923627459, "kv_usage_max": -0.01682167078536223, "kv_usage_mean": -0.00010651327558790225, "preemptions": 0.0, "prefill_token_fraction": -0.009907841139181417, "queue_running_mean": -0.027900191600000124, "queue_waiting_mean": 0.0, "scheduler_steps_per_s": 0.5999999999999943 }, "full_action_efficacy": 0, "full_feasibility_transition": "false->false", "group": { "level": "high", "offered_rate_per_gpu": 2.9833333333333334, "replicate": 3, "request_hash": "a85849f1f057a67c82523cff57a52af22d2eb9514a953e076990aa6c25c1ef7e", "tp": 1 }, "kind": "matched_mns_increase", "source": { "cell": "tp1_mns8", "early_stopped": true, "full_feasible": false, "full_pass_rate": 0.6424581005586593, "mns": 8, "requests_sha256": "cf1226185845cd8aa4e7ead0cb5c95c77eda62b84851a2ab579f984bbb88e8ac", "result_sha256": "95d5103f2678eeb99c258870324566932e2f2853c3b6e0ee94bc6c538f6ed865", "trial_id": "cells/tp1_mns8/high-rep3/result.json" }, "target": { "cell": "tp1_mns64", "early_stopped": true, "full_feasible": false, "full_pass_rate": 0.9273743016759777, "mns": 64, "requests_sha256": "18d72d1fe16136359e28a1909e395ffd553dbe3edd2c0fd93a9cc270a5ec6d60", "result_sha256": "480a3b50acaf9b9c1902e2669cf3aa7723a6703a00778da269b5b9757990ceef", "trial_id": "cells/tp1_mns64/high-rep3/result.json" } }, { "delta_outcome": { "admitted_fraction": 0.0, "admitted_input_tokens_mean_over_limit": 0.0, "completed_fail_fraction_of_total": 0.0, "completed_over_admitted": 0.0, "completed_pass_rate": 0.0, "outstanding_over_admitted": 0.0, "tpot_max_over_slo": -0.019319413383821316, "tpot_mean_over_slo": -0.01717012501436968, "ttft_max_over_slo_max": -0.0006420235052549544, "ttft_mean_over_slo_max": -0.0016948439976355693 }, "delta_state": { "batch_size.mean": -0.06047484495981781, "batch_tokens.mean": -1.7159186818832382, "decode_batch_size.mean": -0.059649187185791375, "graph_full_share": 0.0008256577740266557, "graph_none_share": -0.007128178782430002, "graph_padding_fraction": 0.0006142370124512038, "kv_usage_end_minus_start": 9.20038674978585e-05, "kv_usage_max": 0.00016629145416469093, "kv_usage_mean": -5.445550568086016e-05, "preemptions": 0.0, "prefill_token_fraction": -0.00036424705933824075, "queue_running_mean": -0.16474551560000017, "queue_waiting_mean": 0.0, "scheduler_steps_per_s": 3.6000000000000085 }, "full_action_efficacy": 0, "full_feasibility_transition": "true->true", "group": { "level": "low", "offered_rate_per_gpu": 2.033333333333333, "replicate": 1, "request_hash": "c8ea9278f180ebb23cee595e167627cbed711032e70c9ca5ddc7205320cd75c0", "tp": 1 }, "kind": "matched_mns_increase", "source": { "cell": "tp1_mns8", "early_stopped": false, "full_feasible": true, "full_pass_rate": 1.0, "mns": 8, "requests_sha256": "0edc6095b16e5361a88b2765cc5d0f267ed16ff1d883e02fc3c58f57e59c1018", "result_sha256": "8bc433109d148c21b7d158df95ec5d1fa4fa619243114563068e1e98c4241a29", "trial_id": "cells/tp1_mns8/low-rep1/result.json" }, "target": { "cell": "tp1_mns64", "early_stopped": false, "full_feasible": true, "full_pass_rate": 1.0, "mns": 64, "requests_sha256": "44d729c1ace9b09fe2ea4c4e110bd0f95fee69f83edc78ee0f65256922dcd2c2", "result_sha256": "d2c08e71c5d4e0dd99662bab8457bd1ad0a4aba6bda57794dbfce62e36720f62", "trial_id": "cells/tp1_mns64/low-rep1/result.json" } }, { "delta_outcome": { "admitted_fraction": 0.0, "admitted_input_tokens_mean_over_limit": 0.0, "completed_fail_fraction_of_total": 0.0, "completed_over_admitted": -0.15384615384615385, "completed_pass_rate": 0.0, "outstanding_over_admitted": 0.15384615384615385, "tpot_max_over_slo": 0.040732697633234394, "tpot_mean_over_slo": 0.014527851366132327, "ttft_max_over_slo_max": -0.03623925517119157, "ttft_mean_over_slo_max": -0.015724865515949205 }, "delta_state": { "batch_size.mean": 0.05866351271436976, "batch_tokens.mean": 16.67403903015969, "decode_batch_size.mean": 0.053696037847427824, "graph_full_share": 0.0010053222945002327, "graph_none_share": -0.012241277350680077, "graph_padding_fraction": 0.0038721626639551332, "kv_usage_end_minus_start": 0.02762750849475526, "kv_usage_max": 0.01654350181309283, "kv_usage_mean": 0.004906340576442361, "preemptions": 0.0, "prefill_token_fraction": 0.0010152672534560203, "queue_running_mean": 0.6418094667999998, "queue_waiting_mean": -0.3949698864, "scheduler_steps_per_s": -2.3999999999999986 }, "full_action_efficacy": 0, "full_feasibility_transition": "true->true", "group": { "level": "low", "offered_rate_per_gpu": 2.033333333333333, "replicate": 2, "request_hash": "1f27fe31ca32fd67f2f70a54eee83146329a7838ff2a33be48ce4372640a858c", "tp": 1 }, "kind": "matched_mns_increase", "source": { "cell": "tp1_mns8", "early_stopped": false, "full_feasible": true, "full_pass_rate": 1.0, "mns": 8, "requests_sha256": "ca58a878401ac44cca0a9a595eab41659dafec2e923f7f2474e4f82ea30b2f3e", "result_sha256": "42a56af136d1d3bdfc16d9cea092c2db0eaa4db2c85dceb21958c5e418122900", "trial_id": "cells/tp1_mns8/low-rep2/result.json" }, "target": { "cell": "tp1_mns64", "early_stopped": false, "full_feasible": true, "full_pass_rate": 1.0, "mns": 64, "requests_sha256": "97de3cced64fd88a5c2dcd087ca6c29d8b09a0a43d7bf0e4e844ef246462465b", "result_sha256": "f194db2f7abb995171cd24626e4e515c000481f81283c929b9a5b63f110f1506", "trial_id": "cells/tp1_mns64/low-rep2/result.json" } }, { "delta_outcome": { "admitted_fraction": 0.0, "admitted_input_tokens_mean_over_limit": 0.0, "completed_fail_fraction_of_total": 0.0, "completed_over_admitted": 0.0, "completed_pass_rate": 0.0, "outstanding_over_admitted": 0.0, "tpot_max_over_slo": -0.0007078415662871984, "tpot_mean_over_slo": -0.000578866876769335, "ttft_max_over_slo_max": -5.364099827906421e-05, "ttft_mean_over_slo_max": 4.2679616146623633e-05 }, "delta_state": { "batch_size.mean": -0.001802258455126493, "batch_tokens.mean": -0.2360817774774091, "decode_batch_size.mean": -0.00173185773422313, "graph_full_share": 7.040072090336302e-05, "graph_none_share": -7.040072090338037e-05, "graph_padding_fraction": 0.0, "kv_usage_end_minus_start": 0.00011716903913294718, "kv_usage_max": 0.00013871042605262396, "kv_usage_mean": 8.130641993444807e-06, "preemptions": 0.0, "prefill_token_fraction": -3.2355679021600636e-05, "queue_running_mean": -0.0025477285999999655, "queue_waiting_mean": 0.0, "scheduler_steps_per_s": 0.19999999999999574 }, "full_action_efficacy": 0, "full_feasibility_transition": "true->true", "group": { "level": "low", "offered_rate_per_gpu": 2.033333333333333, "replicate": 3, "request_hash": "a46c62ebd8b7df5534831feb70aed45d23b1a5e7f8bf3d4cd82a79f182fbc69e", "tp": 1 }, "kind": "matched_mns_increase", "source": { "cell": "tp1_mns8", "early_stopped": false, "full_feasible": true, "full_pass_rate": 1.0, "mns": 8, "requests_sha256": "8a72240f84bbeff9eabd63e7d1f7a4c19910660fc5e4335013a1deaf89408127", "result_sha256": "0214f32cabbcdc1bad6c331c3945c505cad136fcc9a8336aeb6314b27474e279", "trial_id": "cells/tp1_mns8/low-rep3/result.json" }, "target": { "cell": "tp1_mns64", "early_stopped": false, "full_feasible": true, "full_pass_rate": 1.0, "mns": 64, "requests_sha256": "30f4623100c58580cfa4efe9a1998c63581c8e44511c556ef39cf5bd777883bf", "result_sha256": "0be457aa2fc9ce95bdb92a6eb25544f5f61a9739932ee5afa40535d26456e348", "trial_id": "cells/tp1_mns64/low-rep3/result.json" } }, { "delta_outcome": { "admitted_fraction": 0.0, "admitted_input_tokens_mean_over_limit": 0.0, "completed_fail_fraction_of_total": 0.0, "completed_over_admitted": -0.27586206896551724, "completed_pass_rate": 0.0, "outstanding_over_admitted": 0.27586206896551724, "tpot_max_over_slo": 0.207529695590914, "tpot_mean_over_slo": 0.05943365186246716, "ttft_max_over_slo_max": -0.10603234449323887, "ttft_mean_over_slo_max": -0.017237129418515705 }, "delta_state": { "batch_size.mean": 4.237397287896355, "batch_tokens.mean": 201.35936633563546, "decode_batch_size.mean": 4.139436406714726, "graph_full_share": -0.05746836719261983, "graph_none_share": 0.044500176437969455, "graph_padding_fraction": 0.0022441507829442276, "kv_usage_end_minus_start": 0.02343349054119659, "kv_usage_max": 0.0207004044749316, "kv_usage_mean": 0.007304490614720681, "preemptions": 0.0, "prefill_token_fraction": 0.016050535408914768, "queue_running_mean": 9.929849956000002, "queue_waiting_mean": -5.506802823999999, "scheduler_steps_per_s": -29.19999999999999 }, "full_action_efficacy": 1, "full_feasibility_transition": "false->true", "group": { "level": "high", "offered_rate_per_gpu": 3.125, "replicate": 1, "request_hash": "5240182174ec5e25247be01c017cbefe2f1e59426bb24a57746dc40be4a82208", "tp": 4 }, "kind": "matched_mns_increase", "source": { "cell": "tp4_mns16", "early_stopped": true, "full_feasible": false, "full_pass_rate": 0.10666666666666667, "mns": 16, "requests_sha256": "8695422a0f1e0e1c097a08a7f4c082f1ba2e3bb26b58caadae0afc7438aeca90", "result_sha256": "c54a2bb5c1b78bdef3fa09c39f5974215370d6fc7c2b6d403a3d376d53a1d9eb", "trial_id": "cells/tp4_mns16/high-rep1/result.json" }, "target": { "cell": "tp4_mns64", "early_stopped": false, "full_feasible": true, "full_pass_rate": 1.0, "mns": 64, "requests_sha256": "b4aeaf5b3175bd85fdac08a2ef657fb2278c6c4b36ea32ea417275ab6d55812d", "result_sha256": "111a1042b614d05772888665a7cfbc86925b2fbeb52ffe19edb62a5bcfe2f513", "trial_id": "cells/tp4_mns64/high-rep1/result.json" } }, { "delta_outcome": { "admitted_fraction": 0.0, "admitted_input_tokens_mean_over_limit": 0.0, "completed_fail_fraction_of_total": 0.0, "completed_over_admitted": -0.2631578947368421, "completed_pass_rate": 0.0, "outstanding_over_admitted": 0.26315789473684204, "tpot_max_over_slo": 0.2331512886664177, "tpot_mean_over_slo": 0.22120977902069205, "ttft_max_over_slo_max": -0.17202288067589203, "ttft_mean_over_slo_max": -0.05936201432352671 }, "delta_state": { "batch_size.mean": 10.325762195121952, "batch_tokens.mean": 461.67439024390245, "decode_batch_size.mean": 10.164329268292683, "graph_full_share": -0.08612804878048774, "graph_none_share": 0.07111280487804877, "graph_padding_fraction": 0.0007778735353099922, "kv_usage_end_minus_start": 0.03093665701006798, "kv_usage_max": 0.026463529328836888, "kv_usage_mean": 0.010597923626744828, "preemptions": 0.0, "prefill_token_fraction": 0.009329314654311416, "queue_running_mean": 12.835752482200002, "queue_waiting_mean": -6.7273728942, "scheduler_steps_per_s": -23.0 }, "full_action_efficacy": 1, "full_feasibility_transition": "false->true", "group": { "level": "high", "offered_rate_per_gpu": 3.125, "replicate": 2, "request_hash": "51be592ba9a59c8d66eb1bcc70c3d2b011c87c66128aff36574973b40389ad96", "tp": 4 }, "kind": "matched_mns_increase", "source": { "cell": "tp4_mns16", "early_stopped": true, "full_feasible": false, "full_pass_rate": 0.092, "mns": 16, "requests_sha256": "aa0482b10414d911af227e2df172c19ecca0add2e0024dc88f85ffbd063854bb", "result_sha256": "91e2491de4ca0868b667e7d126b755f03bdd78378681b8070a2bbc0c38fb9055", "trial_id": "cells/tp4_mns16/high-rep2/result.json" }, "target": { "cell": "tp4_mns64", "early_stopped": false, "full_feasible": true, "full_pass_rate": 1.0, "mns": 64, "requests_sha256": "5d82b6885f5c654ae1ae126696df4bafe04f5be1c1109db24c7856b0456d9443", "result_sha256": "6ed574c18723e265ab1327c5e43d3c76d37b3fc5711d06c77a51718e4ba16601", "trial_id": "cells/tp4_mns64/high-rep2/result.json" } }, { "delta_outcome": { "admitted_fraction": 0.0, "admitted_input_tokens_mean_over_limit": 0.0, "completed_fail_fraction_of_total": 0.0, "completed_over_admitted": -0.07843137254901955, "completed_pass_rate": 0.0, "outstanding_over_admitted": 0.0784313725490196, "tpot_max_over_slo": 0.05021820802747551, "tpot_mean_over_slo": -0.012818677862446504, "ttft_max_over_slo_max": -0.08022667251255673, "ttft_mean_over_slo_max": -0.02098792773025101 }, "delta_state": { "batch_size.mean": 0.32844164919636576, "batch_tokens.mean": 20.103174603174637, "decode_batch_size.mean": 0.3230008984725963, "graph_full_share": 0.004542278127183774, "graph_none_share": -0.017769791354697015, "graph_padding_fraction": 0.006616409980351494, "kv_usage_end_minus_start": 0.007401906100479305, "kv_usage_max": 0.004352176770658023, "kv_usage_mean": 0.0009581054950093242, "preemptions": 0.0, "prefill_token_fraction": 0.0019433062474734086, "queue_running_mean": 0.822417270599999, "queue_waiting_mean": -1.9255238218, "scheduler_steps_per_s": 1.3999999999999915 }, "full_action_efficacy": 1, "full_feasibility_transition": "false->true", "group": { "level": "high", "offered_rate_per_gpu": 3.125, "replicate": 3, "request_hash": "604e338cf576d187157293922dada2bceb311567268ddae74eb7a089debf0242", "tp": 4 }, "kind": "matched_mns_increase", "source": { "cell": "tp4_mns16", "early_stopped": true, "full_feasible": false, "full_pass_rate": 0.14666666666666667, "mns": 16, "requests_sha256": "3a23ffbe0e724648226152ef5433cc5825ad640c37cf47997c2f091210eceeec", "result_sha256": "136dde0451f806841ff3996b081fe0bffac0a76f60c3d78b7516f2689e106008", "trial_id": "cells/tp4_mns16/high-rep3/result.json" }, "target": { "cell": "tp4_mns64", "early_stopped": false, "full_feasible": true, "full_pass_rate": 1.0, "mns": 64, "requests_sha256": "5c5861d716181f06dc3f951ae0d6ee479b1b61fc21ebf24541c5b45053fcc167", "result_sha256": "35b8f0465f6b0c197f71b5ecaf4e771e6613f86998ae44f8701ab9e4e1e440b4", "trial_id": "cells/tp4_mns64/high-rep3/result.json" } }, { "delta_outcome": { "admitted_fraction": 0.0, "admitted_input_tokens_mean_over_limit": 0.0, "completed_fail_fraction_of_total": 0.0, "completed_over_admitted": 0.0, "completed_pass_rate": 0.0, "outstanding_over_admitted": 0.0, "tpot_max_over_slo": 0.03209297007177109, "tpot_mean_over_slo": 0.00165350785497842, "ttft_max_over_slo_max": -0.04185720401195188, "ttft_mean_over_slo_max": -0.008807509172654063 }, "delta_state": { "batch_size.mean": -1.1226839723743751, "batch_tokens.mean": -34.76613479399856, "decode_batch_size.mean": -1.1063943796141942, "graph_full_share": 0.012598237675637058, "graph_none_share": -0.01938556799237913, "graph_padding_fraction": -9.478587342680138e-05, "kv_usage_end_minus_start": 3.947117882996398e-05, "kv_usage_max": 0.004831301652910902, "kv_usage_mean": 1.4883144002681564e-05, "preemptions": 0.0, "prefill_token_fraction": -0.00096100288388834, "queue_running_mean": -0.07484160059999923, "queue_waiting_mean": -0.1708327272, "scheduler_steps_per_s": 12.400000000000006 }, "full_action_efficacy": 1, "full_feasibility_transition": "false->true", "group": { "level": "low", "offered_rate_per_gpu": 2.125, "replicate": 1, "request_hash": "6fb170f366b68a4e4e0fc976d370cb1f5b7b915d4dc22ebe5db3c6b9615afe68", "tp": 4 }, "kind": "matched_mns_increase", "source": { "cell": "tp4_mns16", "early_stopped": true, "full_feasible": false, "full_pass_rate": 0.6196078431372549, "mns": 16, "requests_sha256": "d7eaaefd35cc906f759dcaa7ac3c8543249a1a02f28572bec953e41e4379a29f", "result_sha256": "11d9ec9d2a3358aed988f7b4b1014af087d61c4b5a72eb9076ae19bb5cc481cd", "trial_id": "cells/tp4_mns16/low-rep1/result.json" }, "target": { "cell": "tp4_mns64", "early_stopped": false, "full_feasible": true, "full_pass_rate": 1.0, "mns": 64, "requests_sha256": "931d372ded1e3aaa9c67f06653fac8f79b4ee4f369028c1148163580386681ce", "result_sha256": "325ce57f33133670a6761b5dffb831ed3b2f464755ff743b56976b4a974c1215", "trial_id": "cells/tp4_mns64/low-rep1/result.json" } }, { "delta_outcome": { "admitted_fraction": 0.0, "admitted_input_tokens_mean_over_limit": 0.0, "completed_fail_fraction_of_total": 0.0, "completed_over_admitted": -0.10869565217391303, "completed_pass_rate": 0.0, "outstanding_over_admitted": 0.10869565217391308, "tpot_max_over_slo": 0.031286743469536304, "tpot_mean_over_slo": -0.024426390991776098, "ttft_max_over_slo_max": -0.014295891169846676, "ttft_mean_over_slo_max": -0.003572963410477855 }, "delta_state": { "batch_size.mean": 0.7171517639433311, "batch_tokens.mean": 19.4169881786475, "decode_batch_size.mean": 0.701919812339888, "graph_full_share": 0.004382851322571657, "graph_none_share": -0.007268742862434022, "graph_padding_fraction": 0.002283307241934776, "kv_usage_end_minus_start": 0.008891646331802172, "kv_usage_max": 0.0030741209471074615, "kv_usage_mean": 0.0003701503920676315, "preemptions": 0.0, "prefill_token_fraction": 0.00017846155000900144, "queue_running_mean": 0.8948316157999994, "queue_waiting_mean": -1.1446055668, "scheduler_steps_per_s": -0.8000000000000114 }, "full_action_efficacy": 1, "full_feasibility_transition": "false->true", "group": { "level": "low", "offered_rate_per_gpu": 2.125, "replicate": 2, "request_hash": "eb4c0445556d23d73d27d728b45e63fec44e17e84a4f2f341787108b7f7a4901", "tp": 4 }, "kind": "matched_mns_increase", "source": { "cell": "tp4_mns16", "early_stopped": true, "full_feasible": false, "full_pass_rate": 0.6803921568627451, "mns": 16, "requests_sha256": "7fdb90f679bba1747fbacc92fac2045c820e09965c755b20f6dafd4fd9e26a7a", "result_sha256": "800e2aa6e9f9efef4be12ee778be6817364914d63ae0624775d02729089768d6", "trial_id": "cells/tp4_mns16/low-rep2/result.json" }, "target": { "cell": "tp4_mns64", "early_stopped": false, "full_feasible": true, "full_pass_rate": 1.0, "mns": 64, "requests_sha256": "72c8f70a240f79edcaa4eb0a2fe9bade8ac07093339c5b2d09c49ddd1cf27d16", "result_sha256": "ff8b2283d02d5967893ca6225b561ae11011da6149d1f6cff7b4275c00170e1d", "trial_id": "cells/tp4_mns64/low-rep2/result.json" } }, { "delta_outcome": { "admitted_fraction": 0.0, "admitted_input_tokens_mean_over_limit": 0.0, "completed_fail_fraction_of_total": 0.0, "completed_over_admitted": 0.06666666666666676, "completed_pass_rate": 0.0, "outstanding_over_admitted": -0.06666666666666665, "tpot_max_over_slo": -0.0004955744884145297, "tpot_mean_over_slo": -0.0060310517702504285, "ttft_max_over_slo_max": 0.0002604483355147097, "ttft_mean_over_slo_max": -0.003228990856401895 }, "delta_state": { "batch_size.mean": -0.09294265910522981, "batch_tokens.mean": -6.586326402016368, "decode_batch_size.mean": -0.0904221802142402, "graph_full_share": 0.0001575299306868505, "graph_none_share": -0.0036231884057971037, "graph_padding_fraction": 0.0004465429838699502, "kv_usage_end_minus_start": -0.002635620484154888, "kv_usage_max": 0.0003123116904728551, "kv_usage_mean": -0.00022787683290566817, "preemptions": 0.0, "prefill_token_fraction": -0.0009035229748303175, "queue_running_mean": -0.21642387019999987, "queue_waiting_mean": -0.0184107002, "scheduler_steps_per_s": 4.6000000000000085 }, "full_action_efficacy": 1, "full_feasibility_transition": "false->true", "group": { "level": "low", "offered_rate_per_gpu": 2.125, "replicate": 3, "request_hash": "a660c44d1f8b9c0ada5a19285fa579871bcfcae7c6fa71649546efda9aa3baef", "tp": 4 }, "kind": "matched_mns_increase", "source": { "cell": "tp4_mns16", "early_stopped": true, "full_feasible": false, "full_pass_rate": 0.6294117647058823, "mns": 16, "requests_sha256": "b592fe254476957265fc9139aefdea77be4b5540f3c1b8a429c10ea6ccab9a6a", "result_sha256": "cc1b1f7f5280c2892d0c6deac9df9f34ce126b0284e6ad893ffbf0ddbf27969f", "trial_id": "cells/tp4_mns16/low-rep3/result.json" }, "target": { "cell": "tp4_mns64", "early_stopped": false, "full_feasible": true, "full_pass_rate": 1.0, "mns": 64, "requests_sha256": "8b2c779d2097c5e4038e94e1ac9b986e29f55d9acfd67eeb366c4e0e9270f6e0", "result_sha256": "f026b69356701894fab184e0f2f7504a30b537982e823373899c9ab123a33891", "trial_id": "cells/tp4_mns64/low-rep3/result.json" } } ], "efficacy": { "feasibility_transitions": { "false->false": 3, "false->true": 6, "true->true": 3 }, "minimum_balanced_accuracy": 0.75, "minimum_delta_over_best_outcome": 0.15, "outcome_delta": { "best_accuracy": 0.8333333333333334, "best_balanced_accuracy": 0.8333333333333334, "best_feature": "tpot_max_over_slo", "features": { "admitted_fraction": { "accuracy": 0.5, "balanced_accuracy": 0.5, "folds": [ { "direction": 1, "held_out_replicate": 1, "test_labels": [ 0, 0, 1, 1 ], "test_predictions": [ 1, 1, 1, 1 ], "threshold": 0.0, "train_balanced_accuracy": 0.5 }, { "direction": 1, "held_out_replicate": 2, "test_labels": [ 0, 0, 1, 1 ], "test_predictions": [ 1, 1, 1, 1 ], "threshold": 0.0, "train_balanced_accuracy": 0.5 }, { "direction": 1, "held_out_replicate": 3, "test_labels": [ 0, 0, 1, 1 ], "test_predictions": [ 1, 1, 1, 1 ], "threshold": 0.0, "train_balanced_accuracy": 0.5 } ] }, "admitted_input_tokens_mean_over_limit": { "accuracy": 0.5, "balanced_accuracy": 0.5, "folds": [ { "direction": 1, "held_out_replicate": 1, "test_labels": [ 0, 0, 1, 1 ], "test_predictions": [ 1, 1, 1, 1 ], "threshold": 0.0, "train_balanced_accuracy": 0.5 }, { "direction": 1, "held_out_replicate": 2, "test_labels": [ 0, 0, 1, 1 ], "test_predictions": [ 1, 1, 1, 1 ], "threshold": 0.0, "train_balanced_accuracy": 0.5 }, { "direction": 1, "held_out_replicate": 3, "test_labels": [ 0, 0, 1, 1 ], "test_predictions": [ 1, 1, 1, 1 ], "threshold": 0.0, "train_balanced_accuracy": 0.5 } ] }, "completed_fail_fraction_of_total": { "accuracy": 0.5, "balanced_accuracy": 0.5, "folds": [ { "direction": 1, "held_out_replicate": 1, "test_labels": [ 0, 0, 1, 1 ], "test_predictions": [ 1, 1, 1, 1 ], "threshold": 0.0, "train_balanced_accuracy": 0.5 }, { "direction": 1, "held_out_replicate": 2, "test_labels": [ 0, 0, 1, 1 ], "test_predictions": [ 1, 1, 1, 1 ], "threshold": 0.0, "train_balanced_accuracy": 0.5 }, { "direction": 1, "held_out_replicate": 3, "test_labels": [ 0, 0, 1, 1 ], "test_predictions": [ 1, 1, 1, 1 ], "threshold": 0.0, "train_balanced_accuracy": 0.5 } ] }, "completed_over_admitted": { "accuracy": 0.5833333333333334, "balanced_accuracy": 0.5833333333333333, "folds": [ { "direction": 1, "held_out_replicate": 1, "test_labels": [ 0, 0, 1, 1 ], "test_predictions": [ 0, 0, 0, 0 ], "threshold": 0.03333333333333338, "train_balanced_accuracy": 0.625 }, { "direction": -1, "held_out_replicate": 2, "test_labels": [ 0, 0, 1, 1 ], "test_predictions": [ 1, 1, 1, 1 ], "threshold": -0.039215686274509776, "train_balanced_accuracy": 0.75 }, { "direction": -1, "held_out_replicate": 3, "test_labels": [ 0, 0, 1, 1 ], "test_predictions": [ 0, 0, 1, 0 ], "threshold": -0.05434782608695651, "train_balanced_accuracy": 0.625 } ] }, "completed_pass_rate": { "accuracy": 0.5833333333333334, "balanced_accuracy": 0.5833333333333334, "folds": [ { "direction": 1, "held_out_replicate": 1, "test_labels": [ 0, 0, 1, 1 ], "test_predictions": [ 1, 1, 1, 1 ], "threshold": -0.5, "train_balanced_accuracy": 0.625 }, { "direction": 1, "held_out_replicate": 2, "test_labels": [ 0, 0, 1, 1 ], "test_predictions": [ 0, 1, 1, 1 ], "threshold": 0.0, "train_balanced_accuracy": 0.5 }, { "direction": 1, "held_out_replicate": 3, "test_labels": [ 0, 0, 1, 1 ], "test_predictions": [ 1, 1, 1, 1 ], "threshold": -0.5, "train_balanced_accuracy": 0.625 } ] }, "outstanding_over_admitted": { "accuracy": 0.5833333333333334, "balanced_accuracy": 0.5833333333333333, "folds": [ { "direction": -1, "held_out_replicate": 1, "test_labels": [ 0, 0, 1, 1 ], "test_predictions": [ 0, 0, 0, 0 ], "threshold": -0.033333333333333326, "train_balanced_accuracy": 0.625 }, { "direction": 1, "held_out_replicate": 2, "test_labels": [ 0, 0, 1, 1 ], "test_predictions": [ 1, 1, 1, 1 ], "threshold": 0.0392156862745098, "train_balanced_accuracy": 0.75 }, { "direction": 1, "held_out_replicate": 3, "test_labels": [ 0, 0, 1, 1 ], "test_predictions": [ 0, 0, 1, 0 ], "threshold": 0.05434782608695654, "train_balanced_accuracy": 0.625 } ] }, "tpot_max_over_slo": { "accuracy": 0.8333333333333334, "balanced_accuracy": 0.8333333333333334, "folds": [ { "direction": 1, "held_out_replicate": 1, "test_labels": [ 0, 0, 1, 1 ], "test_predictions": [ 0, 0, 1, 1 ], "threshold": -0.000601708027350864, "train_balanced_accuracy": 0.875 }, { "direction": 1, "held_out_replicate": 2, "test_labels": [ 0, 0, 1, 1 ], "test_predictions": [ 0, 1, 1, 1 ], "threshold": -0.000601708027350864, "train_balanced_accuracy": 1.0 }, { "direction": 1, "held_out_replicate": 3, "test_labels": [ 0, 0, 1, 1 ], "test_predictions": [ 0, 0, 1, 0 ], "threshold": 0.01522073488752175, "train_balanced_accuracy": 0.875 } ] }, "tpot_mean_over_slo": { "accuracy": 0.4166666666666667, "balanced_accuracy": 0.41666666666666663, "folds": [ { "direction": -1, "held_out_replicate": 1, "test_labels": [ 0, 0, 1, 1 ], "test_predictions": [ 0, 1, 0, 0 ], "threshold": -0.00409086687448007, "train_balanced_accuracy": 0.75 }, { "direction": 1, "held_out_replicate": 2, "test_labels": [ 0, 0, 1, 1 ], "test_predictions": [ 0, 1, 1, 0 ], "threshold": 0.0005373204891045424, "train_balanced_accuracy": 0.75 }, { "direction": 1, "held_out_replicate": 3, "test_labels": [ 0, 0, 1, 1 ], "test_predictions": [ 0, 0, 0, 0 ], "threshold": -0.00025864649243734217, "train_balanced_accuracy": 0.75 } ] }, "ttft_max_over_slo_max": { "accuracy": 0.5, "balanced_accuracy": 0.5, "folds": [ { "direction": 1, "held_out_replicate": 1, "test_labels": [ 0, 0, 1, 1 ], "test_predictions": [ 0, 0, 0, 0 ], "threshold": 0.00010340366861782274, "train_balanced_accuracy": 0.625 }, { "direction": -1, "held_out_replicate": 2, "test_labels": [ 0, 0, 1, 1 ], "test_predictions": [ 1, 1, 1, 0 ], "threshold": -0.021249613758603417, "train_balanced_accuracy": 0.875 }, { "direction": -1, "held_out_replicate": 3, "test_labels": [ 0, 0, 1, 1 ], "test_predictions": [ 0, 0, 1, 0 ], "threshold": -0.0074689573375508155, "train_balanced_accuracy": 0.75 } ] }, "ttft_mean_over_slo_max": { "accuracy": 0.8333333333333334, "balanced_accuracy": 0.8333333333333333, "folds": [ { "direction": -1, "held_out_replicate": 1, "test_labels": [ 0, 0, 1, 1 ], "test_predictions": [ 0, 0, 1, 1 ], "threshold": -0.0017545182029157468, "train_balanced_accuracy": 0.75 }, { "direction": -1, "held_out_replicate": 2, "test_labels": [ 0, 0, 1, 1 ], "test_predictions": [ 1, 1, 1, 1 ], "threshold": -0.002461917427018732, "train_balanced_accuracy": 1.0 }, { "direction": -1, "held_out_replicate": 3, "test_labels": [ 0, 0, 1, 1 ], "test_predictions": [ 0, 0, 1, 1 ], "threshold": -0.002633903704056712, "train_balanced_accuracy": 0.75 } ] } }, "labels": { "distinct_n": 2, "max": 1.0, "min": 0.0, "n": 12 }, "negative": 6, "positive": 6 }, "telemetry_delta": { "best_accuracy": 0.75, "best_balanced_accuracy": 0.75, "best_feature": "queue_waiting_mean", "features": { "decode_batch_size.mean": { "accuracy": 0.5833333333333334, "balanced_accuracy": 0.5833333333333334, "folds": [ { "direction": 1, "held_out_replicate": 1, "test_labels": [ 0, 0, 1, 1 ], "test_predictions": [ 0, 0, 1, 0 ], "threshold": 0.18834846816001205, "train_balanced_accuracy": 0.75 }, { "direction": -1, "held_out_replicate": 2, "test_labels": [ 0, 0, 1, 1 ], "test_predictions": [ 0, 0, 0, 0 ], "threshold": -0.0791818658254686, "train_balanced_accuracy": 0.75 }, { "direction": 1, "held_out_replicate": 3, "test_labels": [ 0, 0, 1, 1 ], "test_predictions": [ 0, 0, 0, 0 ], "threshold": 0.3778079250936579, "train_balanced_accuracy": 0.75 } ] }, "graph_padding_fraction": { "accuracy": 0.25, "balanced_accuracy": 0.25, "folds": [ { "direction": 1, "held_out_replicate": 1, "test_labels": [ 0, 0, 1, 1 ], "test_predictions": [ 1, 1, 1, 0 ], "threshold": 0.0002232714919349751, "train_balanced_accuracy": 0.625 }, { "direction": 1, "held_out_replicate": 2, "test_labels": [ 0, 0, 1, 1 ], "test_predictions": [ 1, 1, 0, 1 ], "threshold": 0.0019846847222085755, "train_balanced_accuracy": 0.75 }, { "direction": -1, "held_out_replicate": 3, "test_labels": [ 0, 0, 1, 1 ], "test_predictions": [ 1, 1, 0, 1 ], "threshold": 0.0027543290922952377, "train_balanced_accuracy": 0.75 } ] }, "kv_usage_mean": { "accuracy": 0.5, "balanced_accuracy": 0.5, "folds": [ { "direction": -1, "held_out_replicate": 1, "test_labels": [ 0, 0, 1, 1 ], "test_predictions": [ 0, 0, 0, 0 ], "threshold": -0.0001671950542467852, "train_balanced_accuracy": 0.625 }, { "direction": 1, "held_out_replicate": 2, "test_labels": [ 0, 0, 1, 1 ], "test_predictions": [ 1, 1, 1, 1 ], "threshold": 1.1506892998063185e-05, "train_balanced_accuracy": 0.875 }, { "direction": 1, "held_out_replicate": 3, "test_labels": [ 0, 0, 1, 1 ], "test_predictions": [ 0, 1, 1, 0 ], "threshold": -1.97861808390893e-05, "train_balanced_accuracy": 0.75 } ] }, "prefill_token_fraction": { "accuracy": 0.5, "balanced_accuracy": 0.5, "folds": [ { "direction": 1, "held_out_replicate": 1, "test_labels": [ 0, 0, 1, 1 ], "test_predictions": [ 0, 0, 1, 0 ], "threshold": 0.0019163391379847794, "train_balanced_accuracy": 0.75 }, { "direction": 1, "held_out_replicate": 2, "test_labels": [ 0, 0, 1, 1 ], "test_predictions": [ 1, 1, 1, 0 ], "threshold": 0.000955475284225904, "train_balanced_accuracy": 0.75 }, { "direction": 1, "held_out_replicate": 3, "test_labels": [ 0, 0, 1, 1 ], "test_predictions": [ 0, 0, 0, 0 ], "threshold": 0.005609343341403783, "train_balanced_accuracy": 0.75 } ] }, "queue_running_mean": { "accuracy": 0.6666666666666666, "balanced_accuracy": 0.6666666666666666, "folds": [ { "direction": 1, "held_out_replicate": 1, "test_labels": [ 0, 0, 1, 1 ], "test_predictions": [ 0, 0, 1, 0 ], "threshold": 0.7321133686999994, "train_balanced_accuracy": 0.75 }, { "direction": 1, "held_out_replicate": 2, "test_labels": [ 0, 0, 1, 1 ], "test_predictions": [ 1, 1, 1, 1 ], "threshold": 0.40993477099999953, "train_balanced_accuracy": 0.75 }, { "direction": 1, "held_out_replicate": 3, "test_labels": [ 0, 0, 1, 1 ], "test_predictions": [ 0, 0, 1, 0 ], "threshold": 0.7683205412999996, "train_balanced_accuracy": 0.75 } ] }, "queue_waiting_mean": { "accuracy": 0.75, "balanced_accuracy": 0.75, "folds": [ { "direction": -1, "held_out_replicate": 1, "test_labels": [ 0, 0, 1, 1 ], "test_predictions": [ 0, 0, 1, 1 ], "threshold": -0.0092053501, "train_balanced_accuracy": 0.75 }, { "direction": -1, "held_out_replicate": 2, "test_labels": [ 0, 0, 1, 1 ], "test_predictions": [ 1, 1, 1, 1 ], "threshold": -0.0092053501, "train_balanced_accuracy": 1.0 }, { "direction": -1, "held_out_replicate": 3, "test_labels": [ 0, 0, 1, 1 ], "test_predictions": [ 0, 0, 1, 0 ], "threshold": -0.0854163636, "train_balanced_accuracy": 0.75 } ] }, "scheduler_steps_per_s": { "accuracy": 0.4166666666666667, "balanced_accuracy": 0.41666666666666663, "folds": [ { "direction": 1, "held_out_replicate": 1, "test_labels": [ 0, 0, 1, 1 ], "test_predictions": [ 1, 1, 0, 1 ], "threshold": 0.9999999999999929, "train_balanced_accuracy": 0.75 }, { "direction": 1, "held_out_replicate": 2, "test_labels": [ 0, 0, 1, 1 ], "test_predictions": [ 0, 0, 0, 0 ], "threshold": 4.300000000000004, "train_balanced_accuracy": 0.75 }, { "direction": -1, "held_out_replicate": 3, "test_labels": [ 0, 0, 1, 1 ], "test_predictions": [ 0, 0, 0, 0 ], "threshold": -16.9, "train_balanced_accuracy": 0.75 } ] } }, "labels": { "distinct_n": 2, "max": 1.0, "min": 0.0, "n": 12 }, "negative": 6, "positive": 6 }, "telemetry_qualifying_features": [] }, "horizon_s": 5.0, "qualifying_response_features": [], "repeats": [ { "delta_outcome": { "admitted_fraction": 0.027932960893854754, "admitted_input_tokens_mean_over_limit": 0.25888824462890625, "completed_fail_fraction_of_total": 0.0, "completed_over_admitted": -0.8181818181818182, "completed_pass_rate": -1.0, "outstanding_over_admitted": 0.8181818181818181, "tpot_max_over_slo": -0.22724595748614082, "tpot_mean_over_slo": -0.16274999088627282, "ttft_max_over_slo_max": -0.034610742829196774, "ttft_mean_over_slo_max": -0.0228428806456375 }, "delta_state": { "batch_size.mean": 5.528833410386808, "batch_tokens.mean": 384.9526121128063, "decode_batch_size.mean": 5.423208506703652, "graph_full_share": -0.09768839574664823, "graph_none_share": 0.09557713052858684, "graph_padding_fraction": -0.01691520035196146, "kv_usage_end_minus_start": 0.16350284743730636, "kv_usage_max": 0.1339294634828655, "kv_usage_mean": 0.07938311082464784, "preemptions": 0.0, "prefill_token_fraction": 0.07360971724683552, "queue_running_mean": 5.2693423148, "queue_waiting_mean": 0.0, "scheduler_steps_per_s": -77.8 }, "group": { "cell": "tp1_mns64", "level": "high", "mns": 64, "source_replicate": 1, "target_replicate": 2, "tp": 1 }, "kind": "same_config_workload_repeat", "source": { "requests_sha256": "e970dcac20144cb1e13a35997377f6695a9c57d788a88bad7f18cf6c64e35cec", "result_sha256": "3233686b60b920dcd6cc2f45193a81f3af2d46f781ef42c4d117630612084feb", "trial_id": "cells/tp1_mns64/high-rep1/result.json" }, "target": { "requests_sha256": "ec06878e2aefdc4853d972197cedc29ab94df7147abb2701e0d88b5dabf29728", "result_sha256": "8215477a014f061c55e94d879a25bbf1676ee2ef1c5e81efadd23d20d337e07a", "trial_id": "cells/tp1_mns64/high-rep2/result.json" } }, { "delta_outcome": { "admitted_fraction": -0.011173184357541902, "admitted_input_tokens_mean_over_limit": -0.18295179094587052, "completed_fail_fraction_of_total": 0.0, "completed_over_admitted": 0.5, "completed_pass_rate": 1.0, "outstanding_over_admitted": -0.5, "tpot_max_over_slo": 0.16581994141305348, "tpot_mean_over_slo": 0.14577003201512728, "ttft_max_over_slo_max": 0.04351074216538109, "ttft_mean_over_slo_max": 0.02181711295270361 }, "delta_state": { "batch_size.mean": -5.595011337868481, "batch_tokens.mean": -365.5517006802721, "decode_batch_size.mean": -5.494557823129252, "graph_full_share": 0.09251700680272112, "graph_none_share": -0.08866213151927438, "graph_padding_fraction": 0.004841272439424705, "kv_usage_end_minus_start": -0.1340793286042561, "kv_usage_max": -0.12953341992207013, "kv_usage_mean": -0.0768057629061545, "preemptions": 0.0, "prefill_token_fraction": -0.033037405519596885, "queue_running_mean": -4.789778502000001, "queue_waiting_mean": 0.0, "scheduler_steps_per_s": 72.8 }, "group": { "cell": "tp1_mns64", "level": "high", "mns": 64, "source_replicate": 2, "target_replicate": 3, "tp": 1 }, "kind": "same_config_workload_repeat", "source": { "requests_sha256": "ec06878e2aefdc4853d972197cedc29ab94df7147abb2701e0d88b5dabf29728", "result_sha256": "8215477a014f061c55e94d879a25bbf1676ee2ef1c5e81efadd23d20d337e07a", "trial_id": "cells/tp1_mns64/high-rep2/result.json" }, "target": { "requests_sha256": "18d72d1fe16136359e28a1909e395ffd553dbe3edd2c0fd93a9cc270a5ec6d60", "result_sha256": "480a3b50acaf9b9c1902e2669cf3aa7723a6703a00778da269b5b9757990ceef", "trial_id": "cells/tp1_mns64/high-rep3/result.json" } }, { "delta_outcome": { "admitted_fraction": 0.02459016393442623, "admitted_input_tokens_mean_over_limit": 0.20505746694711535, "completed_fail_fraction_of_total": 0.0, "completed_over_admitted": -0.4692307692307692, "completed_pass_rate": 0.0, "outstanding_over_admitted": 0.4692307692307693, "tpot_max_over_slo": 0.38761381102051196, "tpot_mean_over_slo": 0.33823327695869493, "ttft_max_over_slo_max": -0.043946471996605396, "ttft_mean_over_slo_max": -0.0057849611233298966 }, "delta_state": { "batch_size.mean": 3.1123123406666036, "batch_tokens.mean": 225.4005759607214, "decode_batch_size.mean": 3.0546690586346896, "graph_full_share": -0.04640732697573413, "graph_none_share": 0.041473892927957695, "graph_padding_fraction": 0.005316401000291102, "kv_usage_end_minus_start": 0.11649515436107505, "kv_usage_max": 0.10080927165551001, "kv_usage_mean": 0.05787529741227895, "preemptions": 0.0, "prefill_token_fraction": 0.025873876159429043, "queue_running_mean": 4.3325417076, "queue_waiting_mean": 0.0, "scheduler_steps_per_s": -59.6 }, "group": { "cell": "tp1_mns64", "level": "low", "mns": 64, "source_replicate": 1, "target_replicate": 2, "tp": 1 }, "kind": "same_config_workload_repeat", "source": { "requests_sha256": "44d729c1ace9b09fe2ea4c4e110bd0f95fee69f83edc78ee0f65256922dcd2c2", "result_sha256": "d2c08e71c5d4e0dd99662bab8457bd1ad0a4aba6bda57794dbfce62e36720f62", "trial_id": "cells/tp1_mns64/low-rep1/result.json" }, "target": { "requests_sha256": "97de3cced64fd88a5c2dcd087ca6c29d8b09a0a43d7bf0e4e844ef246462465b", "result_sha256": "f194db2f7abb995171cd24626e4e515c000481f81283c929b9a5b63f110f1506", "trial_id": "cells/tp1_mns64/low-rep2/result.json" } }, { "delta_outcome": { "admitted_fraction": -0.06557377049180327, "admitted_input_tokens_mean_over_limit": 0.058956204927884615, "completed_fail_fraction_of_total": 0.0, "completed_over_admitted": 0.3692307692307692, "completed_pass_rate": 0.0, "outstanding_over_admitted": -0.36923076923076925, "tpot_max_over_slo": -0.41323108928586083, "tpot_mean_over_slo": -0.3718732634627717, "ttft_max_over_slo_max": 0.05646172333702755, "ttft_mean_over_slo_max": 0.02716241966764857 }, "delta_state": { "batch_size.mean": -3.6161048689138573, "batch_tokens.mean": -189.70786516853934, "decode_batch_size.mean": -3.556179775280899, "graph_full_share": 0.048689138576779034, "graph_none_share": -0.037453183520599245, "graph_padding_fraction": -0.008781536045622663, "kv_usage_end_minus_start": -0.11025077430312713, "kv_usage_max": -0.10765311219902085, "kv_usage_mean": -0.06249167924800681, "preemptions": 0.0, "prefill_token_fraction": 0.00031032253091034523, "queue_running_mean": -5.3659050614, "queue_waiting_mean": 0.0, "scheduler_steps_per_s": 17.799999999999997 }, "group": { "cell": "tp1_mns64", "level": "low", "mns": 64, "source_replicate": 2, "target_replicate": 3, "tp": 1 }, "kind": "same_config_workload_repeat", "source": { "requests_sha256": "97de3cced64fd88a5c2dcd087ca6c29d8b09a0a43d7bf0e4e844ef246462465b", "result_sha256": "f194db2f7abb995171cd24626e4e515c000481f81283c929b9a5b63f110f1506", "trial_id": "cells/tp1_mns64/low-rep2/result.json" }, "target": { "requests_sha256": "30f4623100c58580cfa4efe9a1998c63581c8e44511c556ef39cf5bd777883bf", "result_sha256": "0be457aa2fc9ce95bdb92a6eb25544f5f61a9739932ee5afa40535d26456e348", "trial_id": "cells/tp1_mns64/low-rep3/result.json" } }, { "delta_outcome": { "admitted_fraction": 0.027932960893854754, "admitted_input_tokens_mean_over_limit": 0.25888824462890625, "completed_fail_fraction_of_total": 0.0, "completed_over_admitted": -0.44318181818181823, "completed_pass_rate": 0.0, "outstanding_over_admitted": 0.4431818181818182, "tpot_max_over_slo": 0.3347538302351874, "tpot_mean_over_slo": 0.33193704520196554, "ttft_max_over_slo_max": 0.08130590917426161, "ttft_mean_over_slo_max": 0.044041225448449114 }, "delta_state": { "batch_size.mean": 3.494949494949495, "batch_tokens.mean": 256.21313131313127, "decode_batch_size.mean": 3.4393939393939394, "graph_full_share": -0.05555555555555547, "graph_none_share": 0.05555555555555555, "graph_padding_fraction": -0.01983738870430219, "kv_usage_end_minus_start": 0.07746654062391167, "kv_usage_max": 0.09234290263197176, "kv_usage_mean": 0.06554072058902435, "preemptions": 0.0, "prefill_token_fraction": 0.07077187634663484, "queue_running_mean": 3.3297516184, "queue_waiting_mean": 1.558876379, "scheduler_steps_per_s": -63.0 }, "group": { "cell": "tp1_mns8", "level": "high", "mns": 8, "source_replicate": 1, "target_replicate": 2, "tp": 1 }, "kind": "same_config_workload_repeat", "source": { "requests_sha256": "5f5bfa4a16846190f488d3369eed89ae895804145cea59197e6b544b703223fc", "result_sha256": "f65e3dc91d17f5caab1f4aecf97d2c464b99f107231af2cfd96f5484880136ee", "trial_id": "cells/tp1_mns8/high-rep1/result.json" }, "target": { "requests_sha256": "e802d09057de349e5164c18314fcad09dc88c9673834fd6e5b100a57dd12fbf4", "result_sha256": "4835a985dd6bb8dc50b53737baa980ac95a49e22f0b8a7326e1fb416c9b4e798", "trial_id": "cells/tp1_mns8/high-rep2/result.json" } }, { "delta_outcome": { "admitted_fraction": -0.011173184357541902, "admitted_input_tokens_mean_over_limit": -0.18295179094587052, "completed_fail_fraction_of_total": 0.0, "completed_over_admitted": 0.125, "completed_pass_rate": 0.0, "outstanding_over_admitted": -0.125, "tpot_max_over_slo": -0.3936708804703955, "tpot_mean_over_slo": -0.34893712293425444, "ttft_max_over_slo_max": -0.07213344016539244, "ttft_mean_over_slo_max": -0.04499520144345875 }, "delta_state": { "batch_size.mean": -3.598494182067077, "batch_tokens.mean": -226.33832991101983, "decode_batch_size.mean": -3.5494638375541867, "graph_full_share": 0.04903034451289068, "graph_none_share": -0.04903034451289072, "graph_padding_fraction": 0.006341404720607481, "kv_usage_end_minus_start": -0.031145828150654364, "kv_usage_max": -0.07099855714214642, "kv_usage_mean": -0.06292620689574609, "preemptions": 0.0, "prefill_token_fraction": -0.021240192351919318, "queue_running_mean": -2.8508891792, "queue_waiting_mean": -1.558876379, "scheduler_steps_per_s": 61.400000000000006 }, "group": { "cell": "tp1_mns8", "level": "high", "mns": 8, "source_replicate": 2, "target_replicate": 3, "tp": 1 }, "kind": "same_config_workload_repeat", "source": { "requests_sha256": "e802d09057de349e5164c18314fcad09dc88c9673834fd6e5b100a57dd12fbf4", "result_sha256": "4835a985dd6bb8dc50b53737baa980ac95a49e22f0b8a7326e1fb416c9b4e798", "trial_id": "cells/tp1_mns8/high-rep2/result.json" }, "target": { "requests_sha256": "cf1226185845cd8aa4e7ead0cb5c95c77eda62b84851a2ab579f984bbb88e8ac", "result_sha256": "95d5103f2678eeb99c258870324566932e2f2853c3b6e0ee94bc6c538f6ed865", "trial_id": "cells/tp1_mns8/high-rep3/result.json" } }, { "delta_outcome": { "admitted_fraction": 0.02459016393442623, "admitted_input_tokens_mean_over_limit": 0.20505746694711535, "completed_fail_fraction_of_total": 0.0, "completed_over_admitted": -0.3153846153846153, "completed_pass_rate": 0.0, "outstanding_over_admitted": 0.31538461538461543, "tpot_max_over_slo": 0.3275617000034563, "tpot_mean_over_slo": 0.3065353005781929, "ttft_max_over_slo_max": -0.008349240330668778, "ttft_mean_over_slo_max": 0.008245060394983739 }, "delta_state": { "batch_size.mean": 2.993173982992416, "batch_tokens.mean": 207.01061824867847, "decode_batch_size.mean": 2.9413238336014706, "graph_full_share": -0.04658699149620771, "graph_none_share": 0.04658699149620778, "graph_padding_fraction": 0.0020584753487871725, "kv_usage_end_minus_start": 0.08895964973381765, "kv_usage_max": 0.08443206129658187, "kv_usage_mean": 0.05291450133015573, "preemptions": 0.0, "prefill_token_fraction": 0.024494361846634782, "queue_running_mean": 3.5259867252, "queue_waiting_mean": 0.3949698864, "scheduler_steps_per_s": -53.599999999999994 }, "group": { "cell": "tp1_mns8", "level": "low", "mns": 8, "source_replicate": 1, "target_replicate": 2, "tp": 1 }, "kind": "same_config_workload_repeat", "source": { "requests_sha256": "0edc6095b16e5361a88b2765cc5d0f267ed16ff1d883e02fc3c58f57e59c1018", "result_sha256": "8bc433109d148c21b7d158df95ec5d1fa4fa619243114563068e1e98c4241a29", "trial_id": "cells/tp1_mns8/low-rep1/result.json" }, "target": { "requests_sha256": "ca58a878401ac44cca0a9a595eab41659dafec2e923f7f2474e4f82ea30b2f3e", "result_sha256": "42a56af136d1d3bdfc16d9cea092c2db0eaa4db2c85dceb21958c5e418122900", "trial_id": "cells/tp1_mns8/low-rep2/result.json" } }, { "delta_outcome": { "admitted_fraction": -0.06557377049180327, "admitted_input_tokens_mean_over_limit": 0.058956204927884615, "completed_fail_fraction_of_total": 0.0, "completed_over_admitted": 0.21538461538461534, "completed_pass_rate": 0.0, "outstanding_over_admitted": -0.2153846153846154, "tpot_max_over_slo": -0.3717905500863392, "tpot_mean_over_slo": -0.35676654521987006, "ttft_max_over_slo_max": 0.02027610916411504, "ttft_mean_over_slo_max": 0.01139487453555274 }, "delta_state": { "batch_size.mean": -3.5556390977443613, "batch_tokens.mean": -172.79774436090224, "decode_batch_size.mean": -3.500751879699248, "graph_full_share": 0.049624060150375904, "graph_none_share": -0.049624060150375945, "graph_padding_fraction": -0.00490937338166753, "kv_usage_end_minus_start": -0.08274043484750482, "kv_usage_max": -0.09124832081198064, "kv_usage_mean": -0.057593469313557896, "preemptions": 0.0, "prefill_token_fraction": 0.0013579454633879662, "queue_running_mean": -4.721547866, "queue_waiting_mean": -0.3949698864, "scheduler_steps_per_s": 15.200000000000003 }, "group": { "cell": "tp1_mns8", "level": "low", "mns": 8, "source_replicate": 2, "target_replicate": 3, "tp": 1 }, "kind": "same_config_workload_repeat", "source": { "requests_sha256": "ca58a878401ac44cca0a9a595eab41659dafec2e923f7f2474e4f82ea30b2f3e", "result_sha256": "42a56af136d1d3bdfc16d9cea092c2db0eaa4db2c85dceb21958c5e418122900", "trial_id": "cells/tp1_mns8/low-rep2/result.json" }, "target": { "requests_sha256": "8a72240f84bbeff9eabd63e7d1f7a4c19910660fc5e4335013a1deaf89408127", "result_sha256": "0214f32cabbcdc1bad6c331c3945c505cad136fcc9a8336aeb6314b27474e279", "trial_id": "cells/tp1_mns8/low-rep3/result.json" } }, { "delta_outcome": { "admitted_fraction": 0.005797101449275366, "admitted_input_tokens_mean_over_limit": 0.08894061053240737, "completed_fail_fraction_of_total": 0.0, "completed_over_admitted": -0.4148148148148148, "completed_pass_rate": 0.0, "outstanding_over_admitted": 0.41481481481481475, "tpot_max_over_slo": 0.18239212881473055, "tpot_mean_over_slo": 0.22534509859074725, "ttft_max_over_slo_max": 0.009043970329609387, "ttft_mean_over_slo_max": 0.0034825192906686835 }, "delta_state": { "batch_size.mean": 4.136376094514087, "batch_tokens.mean": 294.2141424923907, "decode_batch_size.mean": 4.035904389596526, "graph_full_share": -0.08650808533964194, "graph_none_share": 0.08037736393404786, "graph_padding_fraction": -0.007242852466043042, "kv_usage_end_minus_start": 0.03774188185038829, "kv_usage_max": 0.02683241888245158, "kv_usage_mean": 0.01589560565810534, "preemptions": 0.0, "prefill_token_fraction": 0.01670564225806337, "queue_running_mean": 3.658735814, "queue_waiting_mean": 0.0, "scheduler_steps_per_s": -40.39999999999999 }, "group": { "cell": "tp2_mns64", "level": "high", "mns": 64, "source_replicate": 1, "target_replicate": 2, "tp": 2 }, "kind": "same_config_workload_repeat", "source": { "requests_sha256": "19537ec05f1e40543d850964db12234561954c209d55cd8cd0a1da716ab105fc", "result_sha256": "49523b3269db649aa70072058666da3309ecabb93242cd504b6413f5de0f5d54", "trial_id": "cells/tp2_mns64/high-rep1/result.json" }, "target": { "requests_sha256": "d95e0dcc3d3f25af11b4d3da0ca133f647fd602753b804337a7131cc56e03059", "result_sha256": "f37a42d9ea734adcb24892f86cbb19e2c51cee0baf8a6a63c5d51c4820bd28f2", "trial_id": "cells/tp2_mns64/high-rep2/result.json" } }, { "delta_outcome": { "admitted_fraction": 0.0, "admitted_input_tokens_mean_over_limit": -0.07515914351851849, "completed_fail_fraction_of_total": 0.0, "completed_over_admitted": 0.3333333333333333, "completed_pass_rate": 0.0, "outstanding_over_admitted": -0.3333333333333333, "tpot_max_over_slo": -0.2056584360561972, "tpot_mean_over_slo": -0.22610848720556842, "ttft_max_over_slo_max": 0.0024637060026483013, "ttft_mean_over_slo_max": -0.0009699875146560837 }, "delta_state": { "batch_size.mean": -3.8698393491171377, "batch_tokens.mean": -263.8179979938542, "decode_batch_size.mean": -3.7790214466540357, "graph_full_share": 0.0771251612081455, "graph_none_share": -0.06919610871399684, "graph_padding_fraction": 0.008047836044516822, "kv_usage_end_minus_start": -0.028942690148230588, "kv_usage_max": -0.029901904359948328, "kv_usage_mean": -0.01598686261764442, "preemptions": 0.0, "prefill_token_fraction": -0.011907711608659222, "queue_running_mean": -3.4762310214000003, "queue_waiting_mean": 0.0, "scheduler_steps_per_s": 33.2 }, "group": { "cell": "tp2_mns64", "level": "high", "mns": 64, "source_replicate": 2, "target_replicate": 3, "tp": 2 }, "kind": "same_config_workload_repeat", "source": { "requests_sha256": "d95e0dcc3d3f25af11b4d3da0ca133f647fd602753b804337a7131cc56e03059", "result_sha256": "f37a42d9ea734adcb24892f86cbb19e2c51cee0baf8a6a63c5d51c4820bd28f2", "trial_id": "cells/tp2_mns64/high-rep2/result.json" }, "target": { "requests_sha256": "ccba3d0a18a7db8501b040855bda473ee55c519619d4b9fa2a105bda06300829", "result_sha256": "55447419925f30397d19863e149890241c58863d0adaafd511a67757905cb58e", "trial_id": "cells/tp2_mns64/high-rep3/result.json" } }, { "delta_outcome": { "admitted_fraction": 0.017021276595744678, "admitted_input_tokens_mean_over_limit": 0.045296223958333326, "completed_fail_fraction_of_total": 0.0, "completed_over_admitted": 0.33333333333333337, "completed_pass_rate": 0.0, "outstanding_over_admitted": -0.33333333333333337, "tpot_max_over_slo": 0.22320900677050662, "tpot_mean_over_slo": 0.15232562887487847, "ttft_max_over_slo_max": 0.0061178085064360205, "ttft_mean_over_slo_max": 0.018187332690771045 }, "delta_state": { "batch_size.mean": 4.4320356115196375, "batch_tokens.mean": 99.22318191879197, "decode_batch_size.mean": 4.402188829591008, "graph_full_share": -0.026162662010280058, "graph_none_share": 0.02238508644647205, "graph_padding_fraction": 0.010726240511491474, "kv_usage_end_minus_start": -0.01923544232564689, "kv_usage_max": 0.02004118226348972, "kv_usage_mean": 0.014893345374172863, "preemptions": 0.0, "prefill_token_fraction": -0.00708953487763786, "queue_running_mean": 4.799239215399999, "queue_waiting_mean": 0.0, "scheduler_steps_per_s": -38.400000000000006 }, "group": { "cell": "tp2_mns64", "level": "low", "mns": 64, "source_replicate": 1, "target_replicate": 2, "tp": 2 }, "kind": "same_config_workload_repeat", "source": { "requests_sha256": "8685f475baf247f7ce52a6a2df1183cd77a6f580cd57a3a87c407b4e96c0895f", "result_sha256": "7a8f4e462d496bd56b73b0a686653cbe499a238df62a8c9a50db687d051f1c68", "trial_id": "cells/tp2_mns64/low-rep1/result.json" }, "target": { "requests_sha256": "016a76ab3dc015700f9ecb9f60a20fcc3b456c45270472fdd1d04e84ebd7021a", "result_sha256": "368869326f44a0032985eff9d3abfa2306b93c25370caa05cee255a8c0208c0e", "trial_id": "cells/tp2_mns64/low-rep2/result.json" } }, { "delta_outcome": { "admitted_fraction": -0.042553191489361694, "admitted_input_tokens_mean_over_limit": -0.026056198846726164, "completed_fail_fraction_of_total": 0.0, "completed_over_admitted": -0.11904761904761907, "completed_pass_rate": 0.0, "outstanding_over_admitted": 0.11904761904761904, "tpot_max_over_slo": -0.2355341601559496, "tpot_mean_over_slo": -0.1643432670810964, "ttft_max_over_slo_max": -0.026516160830700145, "ttft_mean_over_slo_max": -0.01345709933763525 }, "delta_state": { "batch_size.mean": -5.32536868824603, "batch_tokens.mean": -133.89198850405924, "decode_batch_size.mean": -5.279523379974407, "graph_full_share": 0.04036585621682853, "graph_none_share": -0.034001132811680546, "graph_padding_fraction": -0.008571843774165616, "kv_usage_end_minus_start": 0.0019567969919042083, "kv_usage_max": -0.028865953011293133, "kv_usage_mean": -0.018137793999508895, "preemptions": 0.0, "prefill_token_fraction": 0.0041833878419771375, "queue_running_mean": -6.512575867199999, "queue_waiting_mean": 0.0, "scheduler_steps_per_s": 57.599999999999994 }, "group": { "cell": "tp2_mns64", "level": "low", "mns": 64, "source_replicate": 2, "target_replicate": 3, "tp": 2 }, "kind": "same_config_workload_repeat", "source": { "requests_sha256": "016a76ab3dc015700f9ecb9f60a20fcc3b456c45270472fdd1d04e84ebd7021a", "result_sha256": "368869326f44a0032985eff9d3abfa2306b93c25370caa05cee255a8c0208c0e", "trial_id": "cells/tp2_mns64/low-rep2/result.json" }, "target": { "requests_sha256": "5455d1cdeabd6e2937130ed8fc8d86c1aaa6f04cbf26c3cb69ad064be5f67d73", "result_sha256": "2de5c4524f57b487b94b4fdba35defc2a566236fc3e31f7260b7a9f85bb7007c", "trial_id": "cells/tp2_mns64/low-rep3/result.json" } }, { "delta_outcome": { "admitted_fraction": 0.005952380952380959, "admitted_input_tokens_mean_over_limit": 0.11506300706129807, "completed_fail_fraction_of_total": 0.0, "completed_over_admitted": -0.16666666666666663, "completed_pass_rate": 0.0, "outstanding_over_admitted": 0.16666666666666669, "tpot_max_over_slo": 0.0634281492909402, "tpot_mean_over_slo": 0.06688242088001703, "ttft_max_over_slo_max": 0.13068378849614723, "ttft_mean_over_slo_max": 0.05409936663613152 }, "delta_state": { "batch_size.mean": 1.552585603284773, "batch_tokens.mean": 89.12552429250253, "decode_batch_size.mean": 1.5358014452458475, "graph_full_share": -0.018898322943787682, "graph_none_share": 0.01889832294378769, "graph_padding_fraction": -0.003394579943621764, "kv_usage_end_minus_start": 0.000535967229432166, "kv_usage_max": 0.01052792772099076, "kv_usage_mean": 0.006505251073294799, "preemptions": 0.0, "prefill_token_fraction": 0.010329730802724923, "queue_running_mean": 0.5542123245999999, "queue_waiting_mean": 2.0910913376, "scheduler_steps_per_s": -32.39999999999999 }, "group": { "cell": "tp2_mns8", "level": "high", "mns": 8, "source_replicate": 1, "target_replicate": 2, "tp": 2 }, "kind": "same_config_workload_repeat", "source": { "requests_sha256": "5d41a316f329e2efcc0f728cdd078377eb72fb9258738636db704bcfcca290aa", "result_sha256": "466b3991190a232c984ba1b54fb6793cac1893dce196ee96f7238d70d49fcb74", "trial_id": "cells/tp2_mns8/high-rep1/result.json" }, "target": { "requests_sha256": "1b7d2cdcd639c00c443ae1b98287a7cc57323eb80c7f0e1bfc72f1ab5bf48760", "result_sha256": "e9156d6871cc8cb3a20dce629cc2443df7473f3836b46eb0944d491ea5c4d44c", "trial_id": "cells/tp2_mns8/high-rep2/result.json" } }, { "delta_outcome": { "admitted_fraction": -0.0029761904761904795, "admitted_input_tokens_mean_over_limit": -0.09617018479567307, "completed_fail_fraction_of_total": 0.0, "completed_over_admitted": 0.14, "completed_pass_rate": 0.0, "outstanding_over_admitted": -0.14, "tpot_max_over_slo": -0.07476021338613548, "tpot_mean_over_slo": -0.06040260059613095, "ttft_max_over_slo_max": -0.0990663813330078, "ttft_mean_over_slo_max": -0.03931008928012512 }, "delta_state": { "batch_size.mean": -1.5731600373589236, "batch_tokens.mean": -75.68624207727024, "decode_batch_size.mean": -1.5588509992086177, "graph_full_share": 0.01874362795074891, "graph_none_share": -0.018743627950748964, "graph_padding_fraction": 0.004413545938114254, "kv_usage_end_minus_start": 0.003471025866799371, "kv_usage_max": -0.012199635031839073, "kv_usage_mean": -0.007236177184589677, "preemptions": 0.0, "prefill_token_fraction": -0.006096255385559024, "queue_running_mean": -0.43295287099999946, "queue_waiting_mean": -1.9712983662, "scheduler_steps_per_s": 28.0 }, "group": { "cell": "tp2_mns8", "level": "high", "mns": 8, "source_replicate": 2, "target_replicate": 3, "tp": 2 }, "kind": "same_config_workload_repeat", "source": { "requests_sha256": "1b7d2cdcd639c00c443ae1b98287a7cc57323eb80c7f0e1bfc72f1ab5bf48760", "result_sha256": "e9156d6871cc8cb3a20dce629cc2443df7473f3836b46eb0944d491ea5c4d44c", "trial_id": "cells/tp2_mns8/high-rep2/result.json" }, "target": { "requests_sha256": "5a68fda7c29477e5b3e66d70ad2b05b77be6c200cc1beea89d663a5729f31a2b", "result_sha256": "f2116797b35057949acbf1f57898befac9ff18b94c808462f0969bdcf2446dd6", "trial_id": "cells/tp2_mns8/high-rep3/result.json" } }, { "delta_outcome": { "admitted_fraction": 0.017467248908296942, "admitted_input_tokens_mean_over_limit": 0.045296223958333326, "completed_fail_fraction_of_total": 0.0, "completed_over_admitted": 0.16666666666666663, "completed_pass_rate": 0.0, "outstanding_over_admitted": -0.16666666666666669, "tpot_max_over_slo": 0.06375630676753702, "tpot_mean_over_slo": 0.0900300343637355, "ttft_max_over_slo_max": 0.051512272330001, "ttft_mean_over_slo_max": 0.0365448741568495 }, "delta_state": { "batch_size.mean": 3.2164733311475753, "batch_tokens.mean": 83.75307177093255, "decode_batch_size.mean": 3.188651751468882, "graph_full_share": -0.027821579678693587, "graph_none_share": 0.027821579678693545, "graph_padding_fraction": -0.0015661122787500655, "kv_usage_end_minus_start": -0.011063894950423037, "kv_usage_max": 0.003075431006980267, "kv_usage_mean": 0.009516064845812437, "preemptions": 0.0, "prefill_token_fraction": -0.003183579698054051, "queue_running_mean": 2.5618635706, "queue_waiting_mean": 1.0030269042, "scheduler_steps_per_s": -25.60000000000001 }, "group": { "cell": "tp2_mns8", "level": "low", "mns": 8, "source_replicate": 1, "target_replicate": 2, "tp": 2 }, "kind": "same_config_workload_repeat", "source": { "requests_sha256": "455a1c21388700f0e3234c49d15ba3b0426b0d3b551c096b297793afa0d4602f", "result_sha256": "13de25a73cc8c752245f169cfca1d64220f697c15b835a646ba969a1005047fb", "trial_id": "cells/tp2_mns8/low-rep1/result.json" }, "target": { "requests_sha256": "cae4b764ebb8714a6f5d2009bb79497420b388986a992c51410f1681810c14c6", "result_sha256": "92f92a576e990946c4767a2107d2ce0e6ba1b066337911399360c199fae3f5e4", "trial_id": "cells/tp2_mns8/low-rep2/result.json" } }, { "delta_outcome": { "admitted_fraction": -0.04803493449781659, "admitted_input_tokens_mean_over_limit": -0.009297688802083315, "completed_fail_fraction_of_total": 0.0, "completed_over_admitted": 0.10256410256410264, "completed_pass_rate": 0.0, "outstanding_over_admitted": -0.10256410256410253, "tpot_max_over_slo": -0.07959600126436789, "tpot_mean_over_slo": -0.121404858913335, "ttft_max_over_slo_max": -0.07178450832725503, "ttft_mean_over_slo_max": -0.03313614673934353 }, "delta_state": { "batch_size.mean": -4.304456872572413, "batch_tokens.mean": -123.40500688035755, "decode_batch_size.mean": -4.261865986626437, "graph_full_share": 0.04259088594597549, "graph_none_share": -0.04259088594597552, "graph_padding_fraction": 0.005238254196308781, "kv_usage_end_minus_start": -0.0053979556678535046, "kv_usage_max": -0.011102178323953882, "kv_usage_mean": -0.01280541269135944, "preemptions": 0.0, "prefill_token_fraction": 0.000758348233463435, "queue_running_mean": -4.5151126308, "queue_waiting_mean": -1.010089547, "scheduler_steps_per_s": 49.60000000000001 }, "group": { "cell": "tp2_mns8", "level": "low", "mns": 8, "source_replicate": 2, "target_replicate": 3, "tp": 2 }, "kind": "same_config_workload_repeat", "source": { "requests_sha256": "cae4b764ebb8714a6f5d2009bb79497420b388986a992c51410f1681810c14c6", "result_sha256": "92f92a576e990946c4767a2107d2ce0e6ba1b066337911399360c199fae3f5e4", "trial_id": "cells/tp2_mns8/low-rep2/result.json" }, "target": { "requests_sha256": "cc8c01e4560411cd5b84cdc9bebdda9a86b88615ee1c39a996affe669d26c167", "result_sha256": "c9a7d026bd89cc027c45b8f1899d56e21d53d4234e95db365a9492f670a78d36", "trial_id": "cells/tp2_mns8/low-rep3/result.json" } }, { "delta_outcome": { "admitted_fraction": -0.0013333333333333391, "admitted_input_tokens_mean_over_limit": 0.0868109341325431, "completed_fail_fraction_of_total": 0.0, "completed_over_admitted": 0.025105868118572294, "completed_pass_rate": 0.0, "outstanding_over_admitted": -0.025105868118572294, "tpot_max_over_slo": 0.029982028815044692, "tpot_mean_over_slo": 0.05498080777948092, "ttft_max_over_slo_max": 0.05707128983340226, "ttft_mean_over_slo_max": 0.04147020173474 }, "delta_state": { "batch_size.mean": 3.8799424686192463, "batch_tokens.mean": 164.1232217573222, "decode_batch_size.mean": 3.8344665271966534, "graph_full_share": -0.04971234309623429, "graph_none_share": 0.042429393305439336, "graph_padding_fraction": -0.000239797608965738, "kv_usage_end_minus_start": 0.003295296296107475, "kv_usage_max": 0.004723598096277648, "kv_usage_mean": 0.003950784826693938, "preemptions": 0.0, "prefill_token_fraction": 0.013157938495005683, "queue_running_mean": 2.4192883438, "queue_waiting_mean": 1.2621521842000005, "scheduler_steps_per_s": -31.599999999999994 }, "group": { "cell": "tp4_mns16", "level": "high", "mns": 16, "source_replicate": 1, "target_replicate": 2, "tp": 4 }, "kind": "same_config_workload_repeat", "source": { "requests_sha256": "8695422a0f1e0e1c097a08a7f4c082f1ba2e3bb26b58caadae0afc7438aeca90", "result_sha256": "c54a2bb5c1b78bdef3fa09c39f5974215370d6fc7c2b6d403a3d376d53a1d9eb", "trial_id": "cells/tp4_mns16/high-rep1/result.json" }, "target": { "requests_sha256": "aa0482b10414d911af227e2df172c19ecca0add2e0024dc88f85ffbd063854bb", "result_sha256": "91e2491de4ca0868b667e7d126b755f03bdd78378681b8070a2bbc0c38fb9055", "trial_id": "cells/tp4_mns16/high-rep2/result.json" } }, { "delta_outcome": { "admitted_fraction": -0.007999999999999993, "admitted_input_tokens_mean_over_limit": -0.08157169117647056, "completed_fail_fraction_of_total": 0.0, "completed_over_admitted": 0.1713106295149639, "completed_pass_rate": 0.0, "outstanding_over_admitted": -0.1713106295149639, "tpot_max_over_slo": -0.04532225479511176, "tpot_mean_over_slo": -0.04295654146532207, "ttft_max_over_slo_max": -0.031098171496220545, "ttft_mean_over_slo_max": -0.018361275960999665 }, "delta_state": { "batch_size.mean": 0.28468665768194157, "batch_tokens.mean": -55.978571428571456, "decode_batch_size.mean": 0.2811657681940698, "graph_full_share": 0.016324123989218342, "graph_none_share": -0.006949123989218334, "graph_padding_fraction": 0.0015473261462516127, "kv_usage_end_minus_start": -0.0032850941403919087, "kv_usage_max": -0.00460117222769163, "kv_usage_mean": -0.0019998572533929705, "preemptions": 0.0, "prefill_token_fraction": -0.008209306549717632, "queue_running_mean": 0.27542344140000097, "queue_waiting_mean": -4.8449581018000005, "scheduler_steps_per_s": 10.200000000000003 }, "group": { "cell": "tp4_mns16", "level": "high", "mns": 16, "source_replicate": 2, "target_replicate": 3, "tp": 4 }, "kind": "same_config_workload_repeat", "source": { "requests_sha256": "aa0482b10414d911af227e2df172c19ecca0add2e0024dc88f85ffbd063854bb", "result_sha256": "91e2491de4ca0868b667e7d126b755f03bdd78378681b8070a2bbc0c38fb9055", "trial_id": "cells/tp4_mns16/high-rep2/result.json" }, "target": { "requests_sha256": "3a23ffbe0e724648226152ef5433cc5825ad640c37cf47997c2f091210eceeec", "result_sha256": "136dde0451f806841ff3996b081fe0bffac0a76f60c3d78b7516f2689e106008", "trial_id": "cells/tp4_mns16/high-rep3/result.json" } }, { "delta_outcome": { "admitted_fraction": 0.0196078431372549, "admitted_input_tokens_mean_over_limit": -0.009788771182442646, "completed_fail_fraction_of_total": 0.0, "completed_over_admitted": -0.20048309178743962, "completed_pass_rate": 0.0, "outstanding_over_admitted": 0.20048309178743962, "tpot_max_over_slo": 0.042659409131532156, "tpot_mean_over_slo": 0.04424057361688544, "ttft_max_over_slo_max": -0.048664552169308685, "ttft_mean_over_slo_max": -0.015179490139230319 }, "delta_state": { "batch_size.mean": 2.2298197150334396, "batch_tokens.mean": 48.29412619947661, "decode_batch_size.mean": 2.216560046525153, "graph_full_share": -0.03405059610351846, "graph_none_share": 0.02576330328583891, "graph_padding_fraction": -0.0005586075805253821, "kv_usage_end_minus_start": 0.005978463249284638, "kv_usage_max": 0.005988665405000093, "kv_usage_mean": 0.0033715964225947276, "preemptions": 0.0, "prefill_token_fraction": -0.0013328880320482295, "queue_running_mean": 2.578025910000001, "queue_waiting_mean": 0.9737728395999999, "scheduler_steps_per_s": -3.5999999999999943 }, "group": { "cell": "tp4_mns16", "level": "low", "mns": 16, "source_replicate": 1, "target_replicate": 2, "tp": 4 }, "kind": "same_config_workload_repeat", "source": { "requests_sha256": "d7eaaefd35cc906f759dcaa7ac3c8543249a1a02f28572bec953e41e4379a29f", "result_sha256": "11d9ec9d2a3358aed988f7b4b1014af087d61c4b5a72eb9076ae19bb5cc481cd", "trial_id": "cells/tp4_mns16/low-rep1/result.json" }, "target": { "requests_sha256": "7fdb90f679bba1747fbacc92fac2045c820e09965c755b20f6dafd4fd9e26a7a", "result_sha256": "800e2aa6e9f9efef4be12ee778be6817364914d63ae0624775d02729089768d6", "trial_id": "cells/tp4_mns16/low-rep2/result.json" } }, { "delta_outcome": { "admitted_fraction": -0.03137254901960784, "admitted_input_tokens_mean_over_limit": 0.04353239639945655, "completed_fail_fraction_of_total": 0.0, "completed_over_admitted": 0.17826086956521736, "completed_pass_rate": 0.0, "outstanding_over_admitted": -0.17826086956521742, "tpot_max_over_slo": -0.06113145905583517, "tpot_mean_over_slo": -0.07880983479690457, "ttft_max_over_slo_max": 0.028715976329597957, "ttft_mean_over_slo_max": 0.012690293574927458 }, "delta_state": { "batch_size.mean": -5.802285141359179, "batch_tokens.mean": -131.81382573186144, "decode_batch_size.mean": -5.749516966234634, "graph_full_share": 0.0604758274237851, "graph_none_share": -0.055969252942589476, "graph_padding_fraction": 0.00012825203851504416, "kv_usage_end_minus_start": -0.0044583420476747015, "kv_usage_max": -0.003932931028326259, "kv_usage_mean": -0.00443657185283492, "preemptions": 0.0, "prefill_token_fraction": 0.003659185530759501, "queue_running_mean": -4.786444050200001, "queue_waiting_mean": -1.1252851525999998, "scheduler_steps_per_s": 33.39999999999999 }, "group": { "cell": "tp4_mns16", "level": "low", "mns": 16, "source_replicate": 2, "target_replicate": 3, "tp": 4 }, "kind": "same_config_workload_repeat", "source": { "requests_sha256": "7fdb90f679bba1747fbacc92fac2045c820e09965c755b20f6dafd4fd9e26a7a", "result_sha256": "800e2aa6e9f9efef4be12ee778be6817364914d63ae0624775d02729089768d6", "trial_id": "cells/tp4_mns16/low-rep2/result.json" }, "target": { "requests_sha256": "b592fe254476957265fc9139aefdea77be4b5540f3c1b8a429c10ea6ccab9a6a", "result_sha256": "cc1b1f7f5280c2892d0c6deac9df9f34ce126b0284e6ad893ffbf0ddbf27969f", "trial_id": "cells/tp4_mns16/low-rep3/result.json" } }, { "delta_outcome": { "admitted_fraction": -0.0013333333333333391, "admitted_input_tokens_mean_over_limit": 0.0868109341325431, "completed_fail_fraction_of_total": 0.0, "completed_over_admitted": 0.03781004234724741, "completed_pass_rate": 0.0, "outstanding_over_admitted": -0.0378100423472475, "tpot_max_over_slo": 0.055603621890548405, "tpot_mean_over_slo": 0.21675693493770581, "ttft_max_over_slo_max": -0.008919246349250898, "ttft_mean_over_slo_max": -0.0006546831702710018 }, "delta_state": { "batch_size.mean": 9.968307375844843, "batch_tokens.mean": 424.4382456655892, "decode_batch_size.mean": 9.85935938877461, "graph_full_share": -0.0783720246841022, "graph_none_share": 0.06904202174551866, "graph_padding_fraction": -0.0017060748565999734, "kv_usage_end_minus_start": 0.010798462764978867, "kv_usage_max": 0.010486722950182936, "kv_usage_mean": 0.007244217838718085, "preemptions": 0.0, "prefill_token_fraction": 0.006436717740402331, "queue_running_mean": 5.32519087, "queue_waiting_mean": 0.041582113999999996, "scheduler_steps_per_s": -25.400000000000006 }, "group": { "cell": "tp4_mns64", "level": "high", "mns": 64, "source_replicate": 1, "target_replicate": 2, "tp": 4 }, "kind": "same_config_workload_repeat", "source": { "requests_sha256": "b4aeaf5b3175bd85fdac08a2ef657fb2278c6c4b36ea32ea417275ab6d55812d", "result_sha256": "111a1042b614d05772888665a7cfbc86925b2fbeb52ffe19edb62a5bcfe2f513", "trial_id": "cells/tp4_mns64/high-rep1/result.json" }, "target": { "requests_sha256": "5d82b6885f5c654ae1ae126696df4bafe04f5be1c1109db24c7856b0456d9443", "result_sha256": "6ed574c18723e265ab1327c5e43d3c76d37b3fc5711d06c77a51718e4ba16601", "trial_id": "cells/tp4_mns64/high-rep2/result.json" } }, { "delta_outcome": { "admitted_fraction": -0.007999999999999993, "admitted_input_tokens_mean_over_limit": -0.08157169117647056, "completed_fail_fraction_of_total": 0.0, "completed_over_admitted": 0.35603715170278644, "completed_pass_rate": 0.0, "outstanding_over_admitted": -0.3560371517027863, "tpot_max_over_slo": -0.22825533543405396, "tpot_mean_over_slo": -0.27698499834846063, "ttft_max_over_slo_max": 0.06069803666711475, "ttft_mean_over_slo_max": 0.020012810632276036 }, "delta_state": { "batch_size.mean": -9.712633888243644, "batch_tokens.mean": -497.54978706929927, "decode_batch_size.mean": -9.560162601626017, "graph_full_share": 0.10699445089688986, "graph_none_share": -0.09583172022196412, "graph_padding_fraction": 0.007385862591293114, "kv_usage_end_minus_start": -0.026819845049980584, "kv_usage_max": -0.026712524785870495, "kv_usage_mean": -0.011639675385128475, "preemptions": 0.0, "prefill_token_fraction": -0.01559531495655564, "queue_running_mean": -11.737911770200002, "queue_waiting_mean": -0.0431090294, "scheduler_steps_per_s": 34.599999999999994 }, "group": { "cell": "tp4_mns64", "level": "high", "mns": 64, "source_replicate": 2, "target_replicate": 3, "tp": 4 }, "kind": "same_config_workload_repeat", "source": { "requests_sha256": "5d82b6885f5c654ae1ae126696df4bafe04f5be1c1109db24c7856b0456d9443", "result_sha256": "6ed574c18723e265ab1327c5e43d3c76d37b3fc5711d06c77a51718e4ba16601", "trial_id": "cells/tp4_mns64/high-rep2/result.json" }, "target": { "requests_sha256": "5c5861d716181f06dc3f951ae0d6ee479b1b61fc21ebf24541c5b45053fcc167", "result_sha256": "35b8f0465f6b0c197f71b5ecaf4e771e6613f86998ae44f8701ab9e4e1e440b4", "trial_id": "cells/tp4_mns64/high-rep3/result.json" } }, { "delta_outcome": { "admitted_fraction": 0.0196078431372549, "admitted_input_tokens_mean_over_limit": -0.009788771182442646, "completed_fail_fraction_of_total": 0.0, "completed_over_admitted": -0.30917874396135264, "completed_pass_rate": 0.0, "outstanding_over_admitted": 0.3091787439613527, "tpot_max_over_slo": 0.04185318252929737, "tpot_mean_over_slo": 0.01816067477013092, "ttft_max_over_slo_max": -0.021103239327203482, "ttft_mean_over_slo_max": -0.00994494437705411 }, "delta_state": { "batch_size.mean": 4.069655451351146, "batch_tokens.mean": 102.47724917212267, "decode_batch_size.mean": 4.024874238479235, "graph_full_share": -0.04226598245658386, "graph_none_share": 0.03788012841578402, "graph_padding_fraction": 0.0018194855348361953, "kv_usage_end_minus_start": 0.014830638402256846, "kv_usage_max": 0.004231484699196653, "kv_usage_mean": 0.0037268636706596775, "preemptions": 0.0, "prefill_token_fraction": -0.00019342359815088805, "queue_running_mean": 3.5476991263999995, "queue_waiting_mean": 0.0, "scheduler_steps_per_s": -16.80000000000001 }, "group": { "cell": "tp4_mns64", "level": "low", "mns": 64, "source_replicate": 1, "target_replicate": 2, "tp": 4 }, "kind": "same_config_workload_repeat", "source": { "requests_sha256": "931d372ded1e3aaa9c67f06653fac8f79b4ee4f369028c1148163580386681ce", "result_sha256": "325ce57f33133670a6761b5dffb831ed3b2f464755ff743b56976b4a974c1215", "trial_id": "cells/tp4_mns64/low-rep1/result.json" }, "target": { "requests_sha256": "72c8f70a240f79edcaa4eb0a2fe9bade8ac07093339c5b2d09c49ddd1cf27d16", "result_sha256": "ff8b2283d02d5967893ca6225b561ae11011da6149d1f6cff7b4275c00170e1d", "trial_id": "cells/tp4_mns64/low-rep2/result.json" } }, { "delta_outcome": { "admitted_fraction": -0.03137254901960784, "admitted_input_tokens_mean_over_limit": 0.04353239639945655, "completed_fail_fraction_of_total": 0.0, "completed_over_admitted": 0.35362318840579715, "completed_pass_rate": 0.0, "outstanding_over_admitted": -0.35362318840579715, "tpot_max_over_slo": -0.092913777013786, "tpot_mean_over_slo": -0.060414495575378896, "ttft_max_over_slo_max": 0.04327231583495934, "ttft_mean_over_slo_max": 0.013034266129003418 }, "delta_state": { "batch_size.mean": -6.61237956440774, "batch_tokens.mean": -157.8171403125253, "decode_batch_size.mean": -6.541858958788763, "graph_full_share": 0.056250506031900294, "graph_none_share": -0.05232369848595256, "graph_padding_fraction": -0.0017085122195497817, "kv_usage_end_minus_start": -0.01598560886363176, "kv_usage_max": -0.006694740284960865, "kv_usage_mean": -0.00503459907780822, "preemptions": 0.0, "prefill_token_fraction": 0.002577201005920182, "queue_running_mean": -5.8976995362, "queue_waiting_mean": 0.000909714, "scheduler_steps_per_s": 38.80000000000001 }, "group": { "cell": "tp4_mns64", "level": "low", "mns": 64, "source_replicate": 2, "target_replicate": 3, "tp": 4 }, "kind": "same_config_workload_repeat", "source": { "requests_sha256": "72c8f70a240f79edcaa4eb0a2fe9bade8ac07093339c5b2d09c49ddd1cf27d16", "result_sha256": "ff8b2283d02d5967893ca6225b561ae11011da6149d1f6cff7b4275c00170e1d", "trial_id": "cells/tp4_mns64/low-rep2/result.json" }, "target": { "requests_sha256": "8b2c779d2097c5e4038e94e1ac9b986e29f55d9acfd67eeb366c4e0e9270f6e0", "result_sha256": "f026b69356701894fab184e0f2f7504a30b537982e823373899c9ab123a33891", "trial_id": "cells/tp4_mns64/low-rep3/result.json" } } ], "response_statistics": { "batch_size.mean": { "action_above_repeat_p95_fraction": 0.08333333333333333, "action_abs_median": 0.21069215415079778, "action_delta": { "distinct_n": 12, "max": 10.325762195121952, "min": -1.1226839723743751, "n": 12 }, "action_signs": { "consistency": 0.5, "negative": 6, "positive": 6, "zero": 0 }, "effect_to_repeat_median": 0.054373699571412804, "effect_to_repeat_median_is_infinite": false, "gate_feature": false, "qualifies": false, "repeat_abs_median": 3.874890908868192, "repeat_abs_p95": 9.247595739668252, "repeat_delta": { "distinct_n": 24, "max": 9.968307375844843, "min": -9.712633888243644, "n": 24 } }, "batch_tokens.mean": { "action_above_repeat_p95_fraction": 0.08333333333333333, "action_abs_median": 18.045513604403595, "action_delta": { "distinct_n": 12, "max": 461.67439024390245, "min": -34.76613479399856, "n": 12 }, "action_signs": { "consistency": 0.5, "negative": 6, "positive": 6, "zero": 0 }, "effect_to_repeat_median": 0.10712015825142496, "effect_to_repeat_median_is_infinite": false, "gate_feature": false, "qualifies": false, "repeat_abs_median": 168.46048305911222, "repeat_abs_p95": 418.5154006326717, "repeat_delta": { "distinct_n": 24, "max": 424.4382456655892, "min": -497.54978706929927, "n": 24 } }, "decode_batch_size.mean": { "action_above_repeat_p95_fraction": 0.08333333333333333, "action_abs_median": 0.20671153934341824, "action_delta": { "distinct_n": 12, "max": 10.164329268292683, "min": -1.1063943796141942, "n": 12 }, "action_signs": { "consistency": 0.5, "negative": 6, "positive": 6, "zero": 0 }, "effect_to_repeat_median": 0.05430140299778245, "effect_to_repeat_median_is_infinite": false, "gate_feature": true, "qualifies": false, "repeat_abs_median": 3.8067439869253445, "repeat_abs_p95": 9.107417055200422, "repeat_delta": { "distinct_n": 24, "max": 9.85935938877461, "min": -9.560162601626017, "n": 24 } }, "graph_full_share": { "action_above_repeat_p95_fraction": 0.0, "action_abs_median": 0.003299836171280379, "action_delta": { "distinct_n": 12, "max": 0.012598237675637058, "min": -0.08612804878048774, "n": 12 }, "action_signs": { "consistency": 0.75, "negative": 3, "positive": 9, "zero": 0 }, "effect_to_repeat_median": 0.06753691417406232, "effect_to_repeat_median_is_infinite": false, "gate_feature": false, "qualifies": false, "repeat_abs_median": 0.048859741544834856, "repeat_abs_p95": 0.09691268740505915, "repeat_delta": { "distinct_n": 24, "max": 0.10699445089688986, "min": -0.09768839574664823, "n": 24 } }, "graph_none_share": { "action_above_repeat_p95_fraction": 0.0, "action_abs_median": 0.00975501010655705, "action_delta": { "distinct_n": 12, "max": 0.07111280487804877, "min": -0.01938556799237913, "n": 12 }, "action_signs": { "consistency": 0.75, "negative": 9, "positive": 3, "zero": 0 }, "effect_to_repeat_median": 0.21877645860951367, "effect_to_repeat_median_is_infinite": false, "gate_feature": false, "qualifies": false, "repeat_abs_median": 0.04458893872109165, "repeat_abs_p95": 0.09453988067718995, "repeat_delta": { "distinct_n": 24, "max": 0.09557713052858684, "min": -0.09583172022196412, "n": 24 } }, "graph_padding_fraction": { "action_above_repeat_p95_fraction": 0.0, "action_abs_median": 0.0012515460983914576, "action_delta": { "distinct_n": 12, "max": 0.006616409980351494, "min": -9.478587342680138e-05, "n": 12 }, "action_signs": { "consistency": 0.9090909090909091, "negative": 1, "positive": 10, "zero": 1 }, "effect_to_repeat_median": 0.2567104007991265, "effect_to_repeat_median_is_infinite": false, "gate_feature": true, "qualifies": false, "repeat_abs_median": 0.004875322910546117, "repeat_abs_p95": 0.015986856375890948, "repeat_delta": { "distinct_n": 24, "max": 0.010726240511491474, "min": -0.01983738870430219, "n": 24 } }, "kv_usage_end_minus_start": { "action_above_repeat_p95_fraction": 0.0, "action_abs_median": 0.008146776216140739, "action_delta": { "distinct_n": 12, "max": 0.08606331121732713, "min": -0.01687018923627459, "n": 12 }, "action_signs": { "consistency": 0.8333333333333334, "negative": 2, "positive": 10, "zero": 0 }, "effect_to_repeat_median": 0.4626083515997178, "effect_to_repeat_median_is_infinite": false, "gate_feature": false, "qualifies": false, "repeat_abs_median": 0.017610525594639326, "repeat_abs_p95": 0.1314417024677789, "repeat_delta": { "distinct_n": 24, "max": 0.16350284743730636, "min": -0.1340793286042561, "n": 24 } }, "kv_usage_max": { "action_above_repeat_p95_fraction": 0.0, "action_abs_median": 0.004591739211784462, "action_delta": { "distinct_n": 12, "max": 0.04171319199456147, "min": -0.01682167078536223, "n": 12 }, "action_signs": { "consistency": 0.9166666666666666, "negative": 1, "positive": 11, "zero": 0 }, "effect_to_repeat_median": 0.19642246579235886, "effect_to_repeat_median_is_infinite": false, "gate_feature": false, "qualifies": false, "repeat_abs_median": 0.023376853524680108, "repeat_abs_p95": 0.1262513737636127, "repeat_delta": { "distinct_n": 24, "max": 0.1339294634828655, "min": -0.12953341992207013, "n": 24 } }, "kv_usage_mean": { "action_above_repeat_p95_fraction": 0.0, "action_abs_median": 0.0002990136124866498, "action_delta": { "distinct_n": 12, "max": 0.01377304273482051, "min": -0.00022787683290566817, "n": 12 }, "action_signs": { "consistency": 0.6666666666666666, "negative": 4, "positive": 8, "zero": 0 }, "effect_to_repeat_median": 0.021590398513840623, "effect_to_repeat_median_is_infinite": false, "gate_feature": true, "qualifies": false, "repeat_abs_median": 0.013849379032766153, "repeat_abs_p95": 0.07511600655858494, "repeat_delta": { "distinct_n": 24, "max": 0.07938311082464784, "min": -0.0768057629061545, "n": 24 } }, "preemptions": { "action_above_repeat_p95_fraction": 0.0, "action_abs_median": 0.0, "action_delta": { "distinct_n": 1, "max": 0.0, "min": 0.0, "n": 12 }, "action_signs": { "consistency": 0.0, "negative": 0, "positive": 0, "zero": 12 }, "effect_to_repeat_median": 0.0, "effect_to_repeat_median_is_infinite": false, "gate_feature": false, "qualifies": false, "repeat_abs_median": 0.0, "repeat_abs_p95": 0.0, "repeat_delta": { "distinct_n": 1, "max": 0.0, "min": 0.0, "n": 24 } }, "prefill_token_fraction": { "action_above_repeat_p95_fraction": 0.0, "action_abs_median": 0.0009881350686721802, "action_delta": { "distinct_n": 12, "max": 0.016050535408914768, "min": -0.009907841139181417, "n": 12 }, "action_signs": { "consistency": 0.5, "negative": 6, "positive": 6, "zero": 0 }, "effect_to_repeat_median": 0.12917776465155323, "effect_to_repeat_median_is_infinite": false, "gate_feature": true, "qualifies": false, "repeat_abs_median": 0.007649420713677746, "repeat_abs_p95": 0.06511170572257907, "repeat_delta": { "distinct_n": 24, "max": 0.07360971724683552, "min": -0.033037405519596885, "n": 24 } }, "queue_running_mean": { "action_above_repeat_p95_fraction": 0.16666666666666666, "action_abs_median": 0.42911666849999985, "action_delta": { "distinct_n": 12, "max": 12.835752482200002, "min": -0.21642387019999987, "n": 12 }, "action_signs": { "consistency": 0.5, "negative": 6, "positive": 6, "zero": 0 }, "effect_to_repeat_median": 0.10739626232229334, "effect_to_repeat_median_is_infinite": false, "gate_feature": true, "qualifies": false, "repeat_abs_median": 3.9956387608, "repeat_abs_p95": 6.420344417549997, "repeat_delta": { "distinct_n": 24, "max": 5.32519087, "min": -11.737911770200002, "n": 24 } }, "queue_waiting_mean": { "action_above_repeat_p95_fraction": 0.16666666666666666, "action_abs_median": 0.28290130680000003, "action_delta": { "distinct_n": 9, "max": 0.0, "min": -6.7273728942, "n": 12 }, "action_signs": { "consistency": 1.0, "negative": 8, "positive": 0, "zero": 4 }, "effect_to_repeat_median": 1.2915540857901293, "effect_to_repeat_median_is_infinite": false, "gate_feature": true, "qualifies": false, "repeat_abs_median": 0.2190394579, "repeat_abs_p95": 2.0731223918899997, "repeat_delta": { "distinct_n": 16, "max": 2.0910913376, "min": -4.8449581018000005, "n": 24 } }, "scheduler_steps_per_s": { "action_above_repeat_p95_fraction": 0.0, "action_abs_median": 3.8000000000000043, "action_delta": { "distinct_n": 12, "max": 12.400000000000006, "min": -29.19999999999999, "n": 12 }, "action_signs": { "consistency": 0.5833333333333334, "negative": 5, "positive": 7, "zero": 0 }, "effect_to_repeat_median": 0.11176470588235309, "effect_to_repeat_median_is_infinite": false, "gate_feature": true, "qualifies": false, "repeat_abs_median": 33.99999999999999, "repeat_abs_p95": 71.32999999999997, "repeat_delta": { "distinct_n": 24, "max": 72.8, "min": -77.8, "n": 24 } } }, "sanity": { "action_pairs": 12, "invariants": { "efficacy_label_balance": true, "expected_action_pair_count": true, "expected_repeat_pair_count": true, "finite_deltas": true, "matched_action_request_hashes": true, "probabilities_bounded": true }, "red_flags": [], "repeat_pairs": 24, "trials": 36 } } }, "provenance": { "analysis_script": "/home/gahow/phd/aituner/runs/intervention-response-v0/analyze_p1.py", "analysis_script_sha256": "902adcba58bcb73f97c455fdea3710f6cb28fa8d8f4ea5bc4915143ed106b17a", "manifest": "/home/gahow/phd/replayserve/runs/fidelity_p1_frontier_committed_20260714/real/p1b/pilot-manifest.json", "manifest_sha256": "86b249a087ab1bcd51cd8de3d1da6bf34afd41af977eb58c93229b941dbf5c8a", "manifest_validation": { "expected_trials": 36, "matched_trials": 36, "schema": "fidelity-prefix-pilot-manifest-v1" }, "phase6_v0_script_sha256": "f1b40c350f22e797b08b129df7b55386b407bd5ae7f466d36987e94d768ded2f", "run_root": "/home/gahow/phd/replayserve/runs/fidelity_p1_frontier_committed_20260714/real/p1b", "streams": [ { "bytes": 24501686, "path": "/home/gahow/phd/replayserve/runs/fidelity_p1_frontier_committed_20260714/real/p1b/cells/tp1_mns64/opprof/opprof-v1-dp0-pid234511-1784005631788937266.jsonl", "sha256": "a813076afe677e5c13010b2ec44f46998794ca44cdf30ee81be836e3623f0068" }, { "bytes": 25524027, "path": "/home/gahow/phd/replayserve/runs/fidelity_p1_frontier_committed_20260714/real/p1b/cells/tp1_mns8/opprof/opprof-v1-dp0-pid227734-1784005177936445360.jsonl", "sha256": "1708759d2e77c42bff1940d43214fb2d927c3eebeda62d53a3fb7d04817e9b22" }, { "bytes": 29431988, "path": "/home/gahow/phd/replayserve/runs/fidelity_p1_frontier_committed_20260714/real/p1b/cells/tp2_mns64/opprof/opprof-v1-dp0-pid247791-1784006495467408663.jsonl", "sha256": "65355de883f2eb5d31be161d82b26dddc53c41a0a35a0fdb807777bc5b21c5ec" }, { "bytes": 26036993, "path": "/home/gahow/phd/replayserve/runs/fidelity_p1_frontier_committed_20260714/real/p1b/cells/tp2_mns8/opprof/opprof-v1-dp0-pid241594-1784006096655191299.jsonl", "sha256": "485e203d37d45fdd3c4de768b95da2050478f111f4d739b129d4f4ebd4737b70" }, { "bytes": 17449143, "path": "/home/gahow/phd/replayserve/runs/fidelity_p1_frontier_committed_20260714/real/p1b/cells/tp4_mns16/opprof/opprof-v1-dp0-pid256090-1784007028237607411.jsonl", "sha256": "59e98014d01bfc68d9f97112ecd782cb03571ad437382ed67416c7b85ae48c56" }, { "bytes": 22674678, "path": "/home/gahow/phd/replayserve/runs/fidelity_p1_frontier_committed_20260714/real/p1b/cells/tp4_mns64/opprof/opprof-v1-dp0-pid262293-1784007376383569515.jsonl", "sha256": "03e4f3cca3023a78e859b663703b9468dedf89e58320f0292e2973070e7616c8" } ] }, "sanity": { "red_flags": [], "stream_bytes": { "distinct_n": 6, "max": 29431988.0, "min": 17449143.0, "n": 6 }, "stream_count": 6 }, "schema": "intervention-response-p1-confirmation-v1", "stable_incremental_efficacy_features": [], "stable_response_features": [], "status": "COMPLETE" }