- 引入 MemoryRouterXL 与 v5/v6 流式多线程训练/编码管线 - 修复 prepare_memory_router_dataset 候选池重建缺陷(mega 家族 3568x 加速,输出逐字节相同) - 修复 v5 被破坏的拒答与多跳标签(train 未知样本 319 -> 16319,multi_hop 平均正例 1.00 -> 2.00) - 同存储预算下 V2-128 v6 逐轴 22/22 通过:Top-1 41.12% -> 94.62%,未知拒答 0.00% -> 100.00% - 记录三条被实测推翻的显然优化(logits_to_keep=1 反而慢 55%、XL 容量未带来收益) - 记忆手术跨架构可移植性 14/14,读写关闭时与原生模型逐位相同
5251 lines
166 KiB
JSON
5251 lines
166 KiB
JSON
{
|
|
"V2-128 deployed(v3)": {
|
|
"checkpoint": "H:\\Memory\\dynamic_memory_lab\\checkpoints\\natural_memory_v2_qwen_router_entities\\memory_router_v2.pt",
|
|
"router": {
|
|
"kind": "router_v2",
|
|
"step": null,
|
|
"arch_config": {
|
|
"arch": "router_v2",
|
|
"hidden_size": 2560,
|
|
"router_dim": 128,
|
|
"num_heads": 8,
|
|
"max_hops": 3
|
|
},
|
|
"parameters": 2037774,
|
|
"checkpoint_bytes": 8157561
|
|
},
|
|
"storage": {
|
|
"address_bytes_per_record": 512,
|
|
"address_mb_per_1m_records": 512,
|
|
"checkpoint_bytes": 8157561
|
|
},
|
|
"latency": {
|
|
"cuda": {
|
|
"single_query_latency_ms_mean": 1.0995226665302955,
|
|
"single_query_latency_ms_p50": 0.9548999951221049,
|
|
"single_query_latency_ms_p95": 2.3265999989234842,
|
|
"queries_per_second": 909.4855708210602,
|
|
"samples": 150,
|
|
"batched_qps_64": 66037.93061957207,
|
|
"batched_latency_ms_64": 0.0151428124809172,
|
|
"batched_qps_256": 207287.4493168988,
|
|
"batched_latency_ms_256": 0.00482421875176442
|
|
},
|
|
"cpu": {
|
|
"single_query_latency_ms_mean": 0.4118146665374904,
|
|
"single_query_latency_ms_p50": 0.3815000018221326,
|
|
"single_query_latency_ms_p95": 0.6266999989748001,
|
|
"queries_per_second": 2428.276798415005,
|
|
"samples": 150,
|
|
"batched_qps_64": 11685.442492836237,
|
|
"batched_latency_ms_64": 0.0855765625146887,
|
|
"batched_qps_256": 12349.573987243031,
|
|
"batched_latency_ms_256": 0.08097445312955642
|
|
}
|
|
},
|
|
"metrics": {
|
|
"episodes": 21920,
|
|
"answerable_episodes": 17814,
|
|
"unknown_episodes": 4106,
|
|
"top1": 0.4112495789828225,
|
|
"recall1": 0.37027057370607386,
|
|
"recall3": 0.4673009992141013,
|
|
"recall5": 0.4891377568204783,
|
|
"recall8": 0.5160267205568654,
|
|
"evidence_count_in_top3": 0.5420455821264174,
|
|
"hit3": 0.5003368137420007,
|
|
"mrr": 0.476908110965675,
|
|
"ndcg3": 0.44220238482445295,
|
|
"mean_first_positive_rank": 14.38671831144044,
|
|
"all_evidence_in_top3": 0.43426518468620184,
|
|
"multi_positive_episodes": 2000,
|
|
"all_evidence_in_top3_multi": 0.3715,
|
|
"hop_accuracy": 0.7322992700729927,
|
|
"hop_accuracy_answerable": 0.9010890310991355,
|
|
"hop_under_prediction": 0.046817110138093636,
|
|
"hop_over_prediction": 0.05209385876277085,
|
|
"mean_score_margin": -9.29892828391031,
|
|
"median_score_margin": -9.269989609718323,
|
|
"negative_margin_rate": 0.5887504210171774,
|
|
"hop_confusion": {
|
|
"true_0_pred_0": 0,
|
|
"true_0_pred_1": 4042,
|
|
"true_0_pred_2": 64,
|
|
"true_1_pred_0": 834,
|
|
"true_1_pred_1": 14052,
|
|
"true_1_pred_2": 928,
|
|
"true_2_pred_0": 0,
|
|
"true_2_pred_1": 0,
|
|
"true_2_pred_2": 2000
|
|
},
|
|
"known_mean_need_prob": 0.9762653562039442,
|
|
"unknown_mean_need_prob": 0.998855464207621,
|
|
"thr0.30": {
|
|
"need_precision": 0.8126824817518248,
|
|
"need_recall": 1.0,
|
|
"need_f1": 0.8966628076710123,
|
|
"specificity_unknown_refusal": 0.0,
|
|
"abstention_accuracy": 0.8126824817518248,
|
|
"known_question_refusal_rate": 0.0,
|
|
"unknown_question_read_rate": 1.0,
|
|
"counts": {
|
|
"tp": 17814,
|
|
"tn": 0,
|
|
"fp": 4106,
|
|
"fn": 0
|
|
}
|
|
},
|
|
"thr0.40": {
|
|
"need_precision": 0.8126824817518248,
|
|
"need_recall": 1.0,
|
|
"need_f1": 0.8966628076710123,
|
|
"specificity_unknown_refusal": 0.0,
|
|
"abstention_accuracy": 0.8126824817518248,
|
|
"known_question_refusal_rate": 0.0,
|
|
"unknown_question_read_rate": 1.0,
|
|
"counts": {
|
|
"tp": 17814,
|
|
"tn": 0,
|
|
"fp": 4106,
|
|
"fn": 0
|
|
}
|
|
},
|
|
"thr0.50": {
|
|
"need_precision": 0.8124086257309941,
|
|
"need_recall": 0.998203660042663,
|
|
"need_f1": 0.895773512669387,
|
|
"specificity_unknown_refusal": 0.0,
|
|
"abstention_accuracy": 0.8112226277372263,
|
|
"known_question_refusal_rate": 0.001796339957336926,
|
|
"unknown_question_read_rate": 1.0,
|
|
"counts": {
|
|
"tp": 17782,
|
|
"tn": 0,
|
|
"fp": 4106,
|
|
"fn": 32
|
|
}
|
|
},
|
|
"thr0.60": {
|
|
"need_precision": 0.8115476409032495,
|
|
"need_recall": 0.9925900976759852,
|
|
"need_f1": 0.892985202767537,
|
|
"specificity_unknown_refusal": 0.0,
|
|
"abstention_accuracy": 0.8066605839416059,
|
|
"known_question_refusal_rate": 0.00740990232401482,
|
|
"unknown_question_read_rate": 1.0,
|
|
"counts": {
|
|
"tp": 17682,
|
|
"tn": 0,
|
|
"fp": 4106,
|
|
"fn": 132
|
|
}
|
|
},
|
|
"thr0.70": {
|
|
"need_precision": 0.8112270700197692,
|
|
"need_recall": 0.9905130796003143,
|
|
"need_f1": 0.8919499557689878,
|
|
"specificity_unknown_refusal": 0.0,
|
|
"abstention_accuracy": 0.8049726277372263,
|
|
"known_question_refusal_rate": 0.00948692039968564,
|
|
"unknown_question_read_rate": 1.0,
|
|
"counts": {
|
|
"tp": 17645,
|
|
"tn": 0,
|
|
"fp": 4106,
|
|
"fn": 169
|
|
}
|
|
},
|
|
"thr0.80": {
|
|
"need_precision": 0.8106525247867189,
|
|
"need_recall": 0.9868081284383069,
|
|
"need_f1": 0.8900984835059115,
|
|
"specificity_unknown_refusal": 0.0,
|
|
"abstention_accuracy": 0.8019616788321168,
|
|
"known_question_refusal_rate": 0.01319187156169305,
|
|
"unknown_question_read_rate": 1.0,
|
|
"counts": {
|
|
"tp": 17579,
|
|
"tn": 0,
|
|
"fp": 4106,
|
|
"fn": 235
|
|
}
|
|
}
|
|
},
|
|
"by_family": {
|
|
"benchmark_qa": {
|
|
"episodes": 64,
|
|
"answerable_episodes": 0,
|
|
"unknown_episodes": 64,
|
|
"top1": null,
|
|
"recall1": null,
|
|
"recall3": null,
|
|
"recall5": null,
|
|
"recall8": null,
|
|
"evidence_count_in_top3": null,
|
|
"hit3": null,
|
|
"mrr": null,
|
|
"ndcg3": null,
|
|
"mean_first_positive_rank": null,
|
|
"all_evidence_in_top3": null,
|
|
"multi_positive_episodes": 0,
|
|
"all_evidence_in_top3_multi": null,
|
|
"hop_accuracy": 0.0,
|
|
"hop_accuracy_answerable": null,
|
|
"hop_under_prediction": null,
|
|
"hop_over_prediction": null,
|
|
"mean_score_margin": null,
|
|
"median_score_margin": null,
|
|
"negative_margin_rate": null,
|
|
"hop_confusion": {
|
|
"true_0_pred_0": 0,
|
|
"true_0_pred_2": 64
|
|
},
|
|
"known_mean_need_prob": null,
|
|
"unknown_mean_need_prob": 0.9999341862276196,
|
|
"thr0.50": {
|
|
"need_precision": 0.0,
|
|
"need_recall": null,
|
|
"need_f1": null,
|
|
"specificity_unknown_refusal": 0.0,
|
|
"abstention_accuracy": 0.0,
|
|
"known_question_refusal_rate": null,
|
|
"unknown_question_read_rate": 1.0,
|
|
"counts": {
|
|
"tp": 0,
|
|
"tn": 0,
|
|
"fp": 64,
|
|
"fn": 0
|
|
}
|
|
}
|
|
},
|
|
"mega_validation": {
|
|
"episodes": 20000,
|
|
"answerable_episodes": 16000,
|
|
"unknown_episodes": 4000,
|
|
"top1": 0.442375,
|
|
"recall1": 0.39675,
|
|
"recall3": 0.49153125,
|
|
"recall5": 0.50303125,
|
|
"recall8": 0.51234375,
|
|
"evidence_count_in_top3": 0.57475,
|
|
"hit3": 0.5283125,
|
|
"mrr": 0.500402578720693,
|
|
"ndcg3": 0.4692792057097287,
|
|
"mean_first_positive_rank": 14.9525,
|
|
"all_evidence_in_top3": 0.45475,
|
|
"multi_positive_episodes": 2000,
|
|
"all_evidence_in_top3_multi": 0.3715,
|
|
"hop_accuracy": 0.74955,
|
|
"hop_accuracy_answerable": 0.9369375,
|
|
"hop_under_prediction": 0.0160625,
|
|
"hop_over_prediction": 0.047,
|
|
"mean_score_margin": -8.091850390457548,
|
|
"median_score_margin": -6.662378132343292,
|
|
"negative_margin_rate": 0.557625,
|
|
"hop_confusion": {
|
|
"true_0_pred_0": 0,
|
|
"true_0_pred_1": 4000,
|
|
"true_0_pred_2": 0,
|
|
"true_1_pred_0": 257,
|
|
"true_1_pred_1": 12991,
|
|
"true_1_pred_2": 752,
|
|
"true_2_pred_0": 0,
|
|
"true_2_pred_1": 0,
|
|
"true_2_pred_2": 2000
|
|
},
|
|
"known_mean_need_prob": 0.9821553228273988,
|
|
"unknown_mean_need_prob": 0.9988265322595835,
|
|
"thr0.50": {
|
|
"need_precision": 0.8,
|
|
"need_recall": 1.0,
|
|
"need_f1": 0.888888888888889,
|
|
"specificity_unknown_refusal": 0.0,
|
|
"abstention_accuracy": 0.8,
|
|
"known_question_refusal_rate": 0.0,
|
|
"unknown_question_read_rate": 1.0,
|
|
"counts": {
|
|
"tp": 16000,
|
|
"tn": 0,
|
|
"fp": 4000,
|
|
"fn": 0
|
|
}
|
|
}
|
|
},
|
|
"memory_policy": {
|
|
"episodes": 1728,
|
|
"answerable_episodes": 1707,
|
|
"unknown_episodes": 21,
|
|
"top1": 0.14528412419449327,
|
|
"recall1": 0.14528412419449327,
|
|
"recall3": 0.2677211482132396,
|
|
"recall5": 0.38488576449912126,
|
|
"recall8": 0.5653192735793791,
|
|
"evidence_count_in_top3": 0.2677211482132396,
|
|
"hit3": 0.2677211482132396,
|
|
"mrr": 0.2801716446629072,
|
|
"ndcg3": 0.21516992510461316,
|
|
"mean_first_positive_rank": 9.195664909197422,
|
|
"all_evidence_in_top3": 0.2677211482132396,
|
|
"multi_positive_episodes": 0,
|
|
"all_evidence_in_top3_multi": null,
|
|
"hop_accuracy": 0.5740740740740741,
|
|
"hop_accuracy_answerable": 0.5811364967779731,
|
|
"hop_under_prediction": 0.3380199179847686,
|
|
"hop_over_prediction": 0.08084358523725835,
|
|
"mean_score_margin": -19.359866383266713,
|
|
"median_score_margin": -18.159542322158813,
|
|
"negative_margin_rate": 0.8547158758055068,
|
|
"hop_confusion": {
|
|
"true_0_pred_0": 0,
|
|
"true_0_pred_1": 21,
|
|
"true_0_pred_2": 0,
|
|
"true_1_pred_0": 577,
|
|
"true_1_pred_1": 992,
|
|
"true_1_pred_2": 138
|
|
},
|
|
"known_mean_need_prob": 0.9195722371566764,
|
|
"unknown_mean_need_prob": 0.9999671209426153,
|
|
"thr0.50": {
|
|
"need_precision": 0.9876179245283019,
|
|
"need_recall": 0.981253661394259,
|
|
"need_f1": 0.9844255069056715,
|
|
"specificity_unknown_refusal": 0.0,
|
|
"abstention_accuracy": 0.9693287037037037,
|
|
"known_question_refusal_rate": 0.018746338605741066,
|
|
"unknown_question_read_rate": 1.0,
|
|
"counts": {
|
|
"tp": 1675,
|
|
"tn": 0,
|
|
"fp": 21,
|
|
"fn": 32
|
|
}
|
|
}
|
|
},
|
|
"native_memory": {
|
|
"episodes": 128,
|
|
"answerable_episodes": 107,
|
|
"unknown_episodes": 21,
|
|
"top1": 0.0,
|
|
"recall1": 0.0,
|
|
"recall3": 0.028037383177570093,
|
|
"recall5": 0.07476635514018691,
|
|
"recall8": 0.2803738317757009,
|
|
"evidence_count_in_top3": 0.028037383177570093,
|
|
"hit3": 0.028037383177570093,
|
|
"mrr": 0.1023068389893556,
|
|
"ndcg3": 0.01524233414552764,
|
|
"mean_first_positive_rank": 12.598130841121495,
|
|
"all_evidence_in_top3": 0.028037383177570093,
|
|
"multi_positive_episodes": 0,
|
|
"all_evidence_in_top3_multi": null,
|
|
"hop_accuracy": 0.5390625,
|
|
"hop_accuracy_answerable": 0.6448598130841121,
|
|
"hop_under_prediction": 0.0,
|
|
"hop_over_prediction": 0.35514018691588783,
|
|
"mean_score_margin": -29.291684916086286,
|
|
"median_score_margin": -29.07153034210205,
|
|
"negative_margin_rate": 1.0,
|
|
"hop_confusion": {
|
|
"true_0_pred_0": 0,
|
|
"true_0_pred_1": 21,
|
|
"true_0_pred_2": 0,
|
|
"true_1_pred_0": 0,
|
|
"true_1_pred_1": 69,
|
|
"true_1_pred_2": 38
|
|
},
|
|
"known_mean_need_prob": 0.9999633771236812,
|
|
"unknown_mean_need_prob": 0.9999671209426153,
|
|
"thr0.50": {
|
|
"need_precision": 0.8359375,
|
|
"need_recall": 1.0,
|
|
"need_f1": 0.9106382978723404,
|
|
"specificity_unknown_refusal": 0.0,
|
|
"abstention_accuracy": 0.8359375,
|
|
"known_question_refusal_rate": 0.0,
|
|
"unknown_question_read_rate": 1.0,
|
|
"counts": {
|
|
"tp": 107,
|
|
"tn": 0,
|
|
"fp": 21,
|
|
"fn": 0
|
|
}
|
|
}
|
|
}
|
|
},
|
|
"by_category": {
|
|
"conflict_update": {
|
|
"episodes": 2000,
|
|
"answerable_episodes": 2000,
|
|
"unknown_episodes": 0,
|
|
"top1": 0.796,
|
|
"recall1": 0.796,
|
|
"recall3": 0.9125,
|
|
"recall5": 0.919,
|
|
"recall8": 0.927,
|
|
"evidence_count_in_top3": 0.9125,
|
|
"hit3": 0.9125,
|
|
"mrr": 0.8582544951434297,
|
|
"ndcg3": 0.8678012294946459,
|
|
"mean_first_positive_rank": 2.8275,
|
|
"all_evidence_in_top3": 0.9125,
|
|
"multi_positive_episodes": 0,
|
|
"all_evidence_in_top3_multi": null,
|
|
"hop_accuracy": 0.4955,
|
|
"hop_accuracy_answerable": 0.4955,
|
|
"hop_under_prediction": 0.1285,
|
|
"hop_over_prediction": 0.376,
|
|
"mean_score_margin": 21.036133953779935,
|
|
"median_score_margin": 21.400100469589233,
|
|
"negative_margin_rate": 0.204,
|
|
"hop_confusion": {
|
|
"true_1_pred_0": 257,
|
|
"true_1_pred_1": 991,
|
|
"true_1_pred_2": 752
|
|
},
|
|
"known_mean_need_prob": 0.9764632563889026,
|
|
"unknown_mean_need_prob": null,
|
|
"thr0.50": {
|
|
"need_precision": 1.0,
|
|
"need_recall": 1.0,
|
|
"need_f1": 1.0,
|
|
"specificity_unknown_refusal": null,
|
|
"abstention_accuracy": 1.0,
|
|
"known_question_refusal_rate": 0.0,
|
|
"unknown_question_read_rate": null,
|
|
"counts": {
|
|
"tp": 2000,
|
|
"tn": 0,
|
|
"fp": 0,
|
|
"fn": 0
|
|
}
|
|
}
|
|
},
|
|
"distractor_128": {
|
|
"episodes": 2000,
|
|
"answerable_episodes": 2000,
|
|
"unknown_episodes": 0,
|
|
"top1": 0.231,
|
|
"recall1": 0.231,
|
|
"recall3": 0.2535,
|
|
"recall5": 0.255,
|
|
"recall8": 0.257,
|
|
"evidence_count_in_top3": 0.2535,
|
|
"hit3": 0.2535,
|
|
"mrr": 0.26603145346142226,
|
|
"ndcg3": 0.24486859507142916,
|
|
"mean_first_positive_rank": 23.789,
|
|
"all_evidence_in_top3": 0.2535,
|
|
"multi_positive_episodes": 0,
|
|
"all_evidence_in_top3_multi": null,
|
|
"hop_accuracy": 1.0,
|
|
"hop_accuracy_answerable": 1.0,
|
|
"hop_under_prediction": 0.0,
|
|
"hop_over_prediction": 0.0,
|
|
"mean_score_margin": -27.718347060799598,
|
|
"median_score_margin": -38.59446668624878,
|
|
"negative_margin_rate": 0.769,
|
|
"hop_confusion": {
|
|
"true_1_pred_1": 2000
|
|
},
|
|
"known_mean_need_prob": 0.9723067943453789,
|
|
"unknown_mean_need_prob": null,
|
|
"thr0.50": {
|
|
"need_precision": 1.0,
|
|
"need_recall": 1.0,
|
|
"need_f1": 1.0,
|
|
"specificity_unknown_refusal": null,
|
|
"abstention_accuracy": 1.0,
|
|
"known_question_refusal_rate": 0.0,
|
|
"unknown_question_read_rate": null,
|
|
"counts": {
|
|
"tp": 2000,
|
|
"tn": 0,
|
|
"fp": 0,
|
|
"fn": 0
|
|
}
|
|
}
|
|
},
|
|
"distractor_32": {
|
|
"episodes": 2000,
|
|
"answerable_episodes": 2000,
|
|
"unknown_episodes": 0,
|
|
"top1": 0.2255,
|
|
"recall1": 0.2255,
|
|
"recall3": 0.2455,
|
|
"recall5": 0.2475,
|
|
"recall8": 0.2485,
|
|
"evidence_count_in_top3": 0.2455,
|
|
"hit3": 0.2455,
|
|
"mrr": 0.2596338049221357,
|
|
"ndcg3": 0.23792220044107196,
|
|
"mean_first_positive_rank": 24.0695,
|
|
"all_evidence_in_top3": 0.2455,
|
|
"multi_positive_episodes": 0,
|
|
"all_evidence_in_top3_multi": null,
|
|
"hop_accuracy": 1.0,
|
|
"hop_accuracy_answerable": 1.0,
|
|
"hop_under_prediction": 0.0,
|
|
"hop_over_prediction": 0.0,
|
|
"mean_score_margin": -28.903205164782705,
|
|
"median_score_margin": -39.01692509651184,
|
|
"negative_margin_rate": 0.7745,
|
|
"hop_confusion": {
|
|
"true_1_pred_1": 2000
|
|
},
|
|
"known_mean_need_prob": 0.9724746745526791,
|
|
"unknown_mean_need_prob": null,
|
|
"thr0.50": {
|
|
"need_precision": 1.0,
|
|
"need_recall": 1.0,
|
|
"need_f1": 1.0,
|
|
"specificity_unknown_refusal": null,
|
|
"abstention_accuracy": 1.0,
|
|
"known_question_refusal_rate": 0.0,
|
|
"unknown_question_read_rate": null,
|
|
"counts": {
|
|
"tp": 2000,
|
|
"tn": 0,
|
|
"fp": 0,
|
|
"fn": 0
|
|
}
|
|
}
|
|
},
|
|
"forget_correction": {
|
|
"episodes": 2000,
|
|
"answerable_episodes": 0,
|
|
"unknown_episodes": 2000,
|
|
"top1": null,
|
|
"recall1": null,
|
|
"recall3": null,
|
|
"recall5": null,
|
|
"recall8": null,
|
|
"evidence_count_in_top3": null,
|
|
"hit3": null,
|
|
"mrr": null,
|
|
"ndcg3": null,
|
|
"mean_first_positive_rank": null,
|
|
"all_evidence_in_top3": null,
|
|
"multi_positive_episodes": 0,
|
|
"all_evidence_in_top3_multi": null,
|
|
"hop_accuracy": 0.0,
|
|
"hop_accuracy_answerable": null,
|
|
"hop_under_prediction": null,
|
|
"hop_over_prediction": null,
|
|
"mean_score_margin": null,
|
|
"median_score_margin": null,
|
|
"negative_margin_rate": null,
|
|
"hop_confusion": {
|
|
"true_0_pred_0": 0,
|
|
"true_0_pred_1": 2000
|
|
},
|
|
"known_mean_need_prob": null,
|
|
"unknown_mean_need_prob": 0.99774328109622,
|
|
"thr0.50": {
|
|
"need_precision": 0.0,
|
|
"need_recall": null,
|
|
"need_f1": null,
|
|
"specificity_unknown_refusal": 0.0,
|
|
"abstention_accuracy": 0.0,
|
|
"known_question_refusal_rate": null,
|
|
"unknown_question_read_rate": 1.0,
|
|
"counts": {
|
|
"tp": 0,
|
|
"tn": 0,
|
|
"fp": 2000,
|
|
"fn": 0
|
|
}
|
|
}
|
|
},
|
|
"long_context": {
|
|
"episodes": 2000,
|
|
"answerable_episodes": 2000,
|
|
"unknown_episodes": 0,
|
|
"top1": 0.5255,
|
|
"recall1": 0.5255,
|
|
"recall3": 0.6245,
|
|
"recall5": 0.6265,
|
|
"recall8": 0.6295,
|
|
"evidence_count_in_top3": 0.6245,
|
|
"hit3": 0.6245,
|
|
"mrr": 0.5855216777342396,
|
|
"ndcg3": 0.585867169546431,
|
|
"mean_first_positive_rank": 12.2085,
|
|
"all_evidence_in_top3": 0.6245,
|
|
"multi_positive_episodes": 0,
|
|
"all_evidence_in_top3_multi": null,
|
|
"hop_accuracy": 1.0,
|
|
"hop_accuracy_answerable": 1.0,
|
|
"hop_under_prediction": 0.0,
|
|
"hop_over_prediction": 0.0,
|
|
"mean_score_margin": -1.0039909677058458,
|
|
"median_score_margin": 2.929476022720337,
|
|
"negative_margin_rate": 0.4745,
|
|
"hop_confusion": {
|
|
"true_1_pred_1": 2000
|
|
},
|
|
"known_mean_need_prob": 0.9875293919146061,
|
|
"unknown_mean_need_prob": null,
|
|
"thr0.50": {
|
|
"need_precision": 1.0,
|
|
"need_recall": 1.0,
|
|
"need_f1": 1.0,
|
|
"specificity_unknown_refusal": null,
|
|
"abstention_accuracy": 1.0,
|
|
"known_question_refusal_rate": 0.0,
|
|
"unknown_question_read_rate": null,
|
|
"counts": {
|
|
"tp": 2000,
|
|
"tn": 0,
|
|
"fp": 0,
|
|
"fn": 0
|
|
}
|
|
}
|
|
},
|
|
"multi_hop": {
|
|
"episodes": 2000,
|
|
"answerable_episodes": 2000,
|
|
"unknown_episodes": 0,
|
|
"top1": 0.73,
|
|
"recall1": 0.365,
|
|
"recall3": 0.66575,
|
|
"recall5": 0.72725,
|
|
"recall8": 0.77375,
|
|
"evidence_count_in_top3": 1.3315,
|
|
"hit3": 0.96,
|
|
"mrr": 0.84532448592014,
|
|
"ndcg3": 0.6658792959224611,
|
|
"mean_first_positive_rank": 1.59,
|
|
"all_evidence_in_top3": 0.3715,
|
|
"multi_positive_episodes": 2000,
|
|
"all_evidence_in_top3_multi": 0.3715,
|
|
"hop_accuracy": 1.0,
|
|
"hop_accuracy_answerable": 1.0,
|
|
"hop_under_prediction": 0.0,
|
|
"hop_over_prediction": 0.0,
|
|
"mean_score_margin": 2.786984693795443,
|
|
"median_score_margin": 5.27826452255249,
|
|
"negative_margin_rate": 0.27,
|
|
"hop_confusion": {
|
|
"true_2_pred_2": 2000
|
|
},
|
|
"known_mean_need_prob": 0.9993210740983486,
|
|
"unknown_mean_need_prob": null,
|
|
"thr0.50": {
|
|
"need_precision": 1.0,
|
|
"need_recall": 1.0,
|
|
"need_f1": 1.0,
|
|
"specificity_unknown_refusal": null,
|
|
"abstention_accuracy": 1.0,
|
|
"known_question_refusal_rate": 0.0,
|
|
"unknown_question_read_rate": null,
|
|
"counts": {
|
|
"tp": 2000,
|
|
"tn": 0,
|
|
"fp": 0,
|
|
"fn": 0
|
|
}
|
|
}
|
|
},
|
|
"paraphrase": {
|
|
"episodes": 2000,
|
|
"answerable_episodes": 2000,
|
|
"unknown_episodes": 0,
|
|
"top1": 0.443,
|
|
"recall1": 0.443,
|
|
"recall3": 0.5525,
|
|
"recall5": 0.5615,
|
|
"recall8": 0.5665,
|
|
"evidence_count_in_top3": 0.5525,
|
|
"hit3": 0.5525,
|
|
"mrr": 0.5118924660255056,
|
|
"ndcg3": 0.5092063534375025,
|
|
"mean_first_positive_rank": 13.8255,
|
|
"all_evidence_in_top3": 0.5525,
|
|
"multi_positive_episodes": 0,
|
|
"all_evidence_in_top3_multi": null,
|
|
"hop_accuracy": 1.0,
|
|
"hop_accuracy_answerable": 1.0,
|
|
"hop_under_prediction": 0.0,
|
|
"hop_over_prediction": 0.0,
|
|
"mean_score_margin": -2.995655853047967,
|
|
"median_score_margin": -5.5251545906066895,
|
|
"negative_margin_rate": 0.557,
|
|
"hop_confusion": {
|
|
"true_1_pred_1": 2000
|
|
},
|
|
"known_mean_need_prob": 0.9880377069413662,
|
|
"unknown_mean_need_prob": null,
|
|
"thr0.50": {
|
|
"need_precision": 1.0,
|
|
"need_recall": 1.0,
|
|
"need_f1": 1.0,
|
|
"specificity_unknown_refusal": null,
|
|
"abstention_accuracy": 1.0,
|
|
"known_question_refusal_rate": 0.0,
|
|
"unknown_question_read_rate": null,
|
|
"counts": {
|
|
"tp": 2000,
|
|
"tn": 0,
|
|
"fp": 0,
|
|
"fn": 0
|
|
}
|
|
}
|
|
},
|
|
"random_position": {
|
|
"episodes": 2000,
|
|
"answerable_episodes": 2000,
|
|
"unknown_episodes": 0,
|
|
"top1": 0.249,
|
|
"recall1": 0.249,
|
|
"recall3": 0.287,
|
|
"recall5": 0.293,
|
|
"recall8": 0.3005,
|
|
"evidence_count_in_top3": 0.287,
|
|
"hit3": 0.287,
|
|
"mrr": 0.2919631936667751,
|
|
"ndcg3": 0.27173149797678653,
|
|
"mean_first_positive_rank": 22.067,
|
|
"all_evidence_in_top3": 0.287,
|
|
"multi_positive_episodes": 0,
|
|
"all_evidence_in_top3_multi": null,
|
|
"hop_accuracy": 1.0,
|
|
"hop_accuracy_answerable": 1.0,
|
|
"hop_under_prediction": 0.0,
|
|
"hop_over_prediction": 0.0,
|
|
"mean_score_margin": -19.72415616109967,
|
|
"median_score_margin": -21.928287506103516,
|
|
"negative_margin_rate": 0.751,
|
|
"hop_confusion": {
|
|
"true_1_pred_1": 2000
|
|
},
|
|
"known_mean_need_prob": 0.9656237106621265,
|
|
"unknown_mean_need_prob": null,
|
|
"thr0.50": {
|
|
"need_precision": 1.0,
|
|
"need_recall": 1.0,
|
|
"need_f1": 1.0,
|
|
"specificity_unknown_refusal": null,
|
|
"abstention_accuracy": 1.0,
|
|
"known_question_refusal_rate": 0.0,
|
|
"unknown_question_read_rate": null,
|
|
"counts": {
|
|
"tp": 2000,
|
|
"tn": 0,
|
|
"fp": 0,
|
|
"fn": 0
|
|
}
|
|
}
|
|
},
|
|
"single_fact": {
|
|
"episodes": 2000,
|
|
"answerable_episodes": 2000,
|
|
"unknown_episodes": 0,
|
|
"top1": 0.339,
|
|
"recall1": 0.339,
|
|
"recall3": 0.391,
|
|
"recall5": 0.3945,
|
|
"recall8": 0.396,
|
|
"evidence_count_in_top3": 0.391,
|
|
"hit3": 0.391,
|
|
"mrr": 0.3845990528918967,
|
|
"ndcg3": 0.3709573037875013,
|
|
"mean_first_positive_rank": 19.243,
|
|
"all_evidence_in_top3": 0.391,
|
|
"multi_positive_episodes": 0,
|
|
"all_evidence_in_top3_multi": null,
|
|
"hop_accuracy": 1.0,
|
|
"hop_accuracy_answerable": 1.0,
|
|
"hop_under_prediction": 0.0,
|
|
"hop_over_prediction": 0.0,
|
|
"mean_score_margin": -8.212566563799978,
|
|
"median_score_margin": -16.937832549214363,
|
|
"negative_margin_rate": 0.661,
|
|
"hop_confusion": {
|
|
"true_1_pred_1": 2000
|
|
},
|
|
"known_mean_need_prob": 0.9954859737157822,
|
|
"unknown_mean_need_prob": null,
|
|
"thr0.50": {
|
|
"need_precision": 1.0,
|
|
"need_recall": 1.0,
|
|
"need_f1": 1.0,
|
|
"specificity_unknown_refusal": null,
|
|
"abstention_accuracy": 1.0,
|
|
"known_question_refusal_rate": 0.0,
|
|
"unknown_question_read_rate": null,
|
|
"counts": {
|
|
"tp": 2000,
|
|
"tn": 0,
|
|
"fp": 0,
|
|
"fn": 0
|
|
}
|
|
}
|
|
},
|
|
"unknown": {
|
|
"episodes": 1920,
|
|
"answerable_episodes": 1814,
|
|
"unknown_episodes": 106,
|
|
"top1": 0.13671444321940462,
|
|
"recall1": 0.13671444321940462,
|
|
"recall3": 0.2535832414553473,
|
|
"recall5": 0.36659316427783906,
|
|
"recall8": 0.5485115766262404,
|
|
"evidence_count_in_top3": 0.2535832414553473,
|
|
"hit3": 0.2535832414553473,
|
|
"mrr": 0.26968017045834825,
|
|
"ndcg3": 0.2033770627933551,
|
|
"mean_first_positive_rank": 9.396361631753033,
|
|
"all_evidence_in_top3": 0.2535832414553473,
|
|
"multi_positive_episodes": 0,
|
|
"all_evidence_in_top3_multi": null,
|
|
"hop_accuracy": 0.5526041666666667,
|
|
"hop_accuracy_answerable": 0.5848952590959207,
|
|
"hop_under_prediction": 0.31808158765159866,
|
|
"hop_over_prediction": 0.0970231532524807,
|
|
"mean_score_margin": -19.945701324287494,
|
|
"median_score_margin": -18.862584590911865,
|
|
"negative_margin_rate": 0.8632855567805954,
|
|
"hop_confusion": {
|
|
"true_0_pred_0": 0,
|
|
"true_0_pred_1": 42,
|
|
"true_0_pred_2": 64,
|
|
"true_1_pred_0": 577,
|
|
"true_1_pred_1": 1061,
|
|
"true_1_pred_2": 176
|
|
},
|
|
"known_mean_need_prob": 0.9243141621712682,
|
|
"unknown_mean_need_prob": 0.9999472358316746,
|
|
"thr0.50": {
|
|
"need_precision": 0.9438559322033898,
|
|
"need_recall": 0.9823594266813671,
|
|
"need_f1": 0.9627228525121555,
|
|
"specificity_unknown_refusal": 0.0,
|
|
"abstention_accuracy": 0.928125,
|
|
"known_question_refusal_rate": 0.017640573318632856,
|
|
"unknown_question_read_rate": 1.0,
|
|
"counts": {
|
|
"tp": 1782,
|
|
"tn": 0,
|
|
"fp": 106,
|
|
"fn": 32
|
|
}
|
|
}
|
|
},
|
|
"unknown_abstention": {
|
|
"episodes": 2000,
|
|
"answerable_episodes": 0,
|
|
"unknown_episodes": 2000,
|
|
"top1": null,
|
|
"recall1": null,
|
|
"recall3": null,
|
|
"recall5": null,
|
|
"recall8": null,
|
|
"evidence_count_in_top3": null,
|
|
"hit3": null,
|
|
"mrr": null,
|
|
"ndcg3": null,
|
|
"mean_first_positive_rank": null,
|
|
"all_evidence_in_top3": null,
|
|
"multi_positive_episodes": 0,
|
|
"all_evidence_in_top3_multi": null,
|
|
"hop_accuracy": 0.0,
|
|
"hop_accuracy_answerable": null,
|
|
"hop_under_prediction": null,
|
|
"hop_over_prediction": null,
|
|
"mean_score_margin": null,
|
|
"median_score_margin": null,
|
|
"negative_margin_rate": null,
|
|
"hop_confusion": {
|
|
"true_0_pred_0": 0,
|
|
"true_0_pred_1": 2000
|
|
},
|
|
"known_mean_need_prob": null,
|
|
"unknown_mean_need_prob": 0.9999097834229469,
|
|
"thr0.50": {
|
|
"need_precision": 0.0,
|
|
"need_recall": null,
|
|
"need_f1": null,
|
|
"specificity_unknown_refusal": 0.0,
|
|
"abstention_accuracy": 0.0,
|
|
"known_question_refusal_rate": null,
|
|
"unknown_question_read_rate": 1.0,
|
|
"counts": {
|
|
"tp": 0,
|
|
"tn": 0,
|
|
"fp": 2000,
|
|
"fn": 0
|
|
}
|
|
}
|
|
}
|
|
}
|
|
},
|
|
"V2-128 v6 final": {
|
|
"checkpoint": "checkpoints\\router_v6_v2_128\\memory_router_v2.pt",
|
|
"router": {
|
|
"kind": "router_v2",
|
|
"step": null,
|
|
"arch_config": {
|
|
"arch": "router_v2",
|
|
"hidden_size": 2560,
|
|
"router_dim": 128,
|
|
"num_heads": 8,
|
|
"max_hops": 3
|
|
},
|
|
"parameters": 2037774,
|
|
"checkpoint_bytes": 8156921
|
|
},
|
|
"storage": {
|
|
"address_bytes_per_record": 512,
|
|
"address_mb_per_1m_records": 512,
|
|
"checkpoint_bytes": 8156921
|
|
},
|
|
"latency": {
|
|
"cuda": {
|
|
"single_query_latency_ms_mean": 0.9277419995729966,
|
|
"single_query_latency_ms_p50": 0.9184999944409356,
|
|
"single_query_latency_ms_p95": 1.146100003097672,
|
|
"queries_per_second": 1077.8858782509155,
|
|
"samples": 150,
|
|
"batched_qps_64": 72856.42735601631,
|
|
"batched_latency_ms_64": 0.013725624990001961,
|
|
"batched_qps_256": 236293.15117731004,
|
|
"batched_latency_ms_256": 0.004232031250239743
|
|
},
|
|
"cpu": {
|
|
"single_query_latency_ms_mean": 0.46528599991385516,
|
|
"single_query_latency_ms_p50": 0.4394999996293336,
|
|
"single_query_latency_ms_p95": 0.6572999991476536,
|
|
"queries_per_second": 2149.215751570311,
|
|
"samples": 150,
|
|
"batched_qps_64": 10744.09578471459,
|
|
"batched_latency_ms_64": 0.093074374990465,
|
|
"batched_qps_256": 12172.553554284297,
|
|
"batched_latency_ms_256": 0.08215203125132575
|
|
}
|
|
},
|
|
"metrics": {
|
|
"episodes": 21920,
|
|
"answerable_episodes": 17814,
|
|
"unknown_episodes": 4106,
|
|
"top1": 0.9436959694622207,
|
|
"recall1": 0.8882059054676097,
|
|
"recall3": 0.9639609296059279,
|
|
"recall5": 0.973981138430448,
|
|
"recall8": 0.9864151790726395,
|
|
"evidence_count_in_top3": 1.0717974626698104,
|
|
"hit3": 0.9683956438756035,
|
|
"mrr": 0.9597054333626427,
|
|
"ndcg3": 0.9530820526370456,
|
|
"mean_first_positive_rank": 1.2573818345121814,
|
|
"all_evidence_in_top3": 0.9595262153362524,
|
|
"multi_positive_episodes": 2000,
|
|
"all_evidence_in_top3_multi": 0.921,
|
|
"hop_accuracy": 1.0,
|
|
"hop_accuracy_answerable": 1.0,
|
|
"hop_under_prediction": 0.0,
|
|
"hop_over_prediction": 0.0,
|
|
"mean_score_margin": 51.315328869764656,
|
|
"median_score_margin": 56.234654903411865,
|
|
"negative_margin_rate": 0.05630403053777928,
|
|
"hop_confusion": {
|
|
"true_0_pred_0": 4106,
|
|
"true_0_pred_1": 0,
|
|
"true_0_pred_2": 0,
|
|
"true_1_pred_0": 0,
|
|
"true_1_pred_1": 15814,
|
|
"true_1_pred_2": 0,
|
|
"true_2_pred_0": 0,
|
|
"true_2_pred_1": 0,
|
|
"true_2_pred_2": 2000
|
|
},
|
|
"known_mean_need_prob": 0.9999999984274064,
|
|
"unknown_mean_need_prob": 4.836221294362452e-06,
|
|
"thr0.30": {
|
|
"need_precision": 1.0,
|
|
"need_recall": 1.0,
|
|
"need_f1": 1.0,
|
|
"specificity_unknown_refusal": 1.0,
|
|
"abstention_accuracy": 1.0,
|
|
"known_question_refusal_rate": 0.0,
|
|
"unknown_question_read_rate": 0.0,
|
|
"counts": {
|
|
"tp": 17814,
|
|
"tn": 4106,
|
|
"fp": 0,
|
|
"fn": 0
|
|
}
|
|
},
|
|
"thr0.40": {
|
|
"need_precision": 1.0,
|
|
"need_recall": 1.0,
|
|
"need_f1": 1.0,
|
|
"specificity_unknown_refusal": 1.0,
|
|
"abstention_accuracy": 1.0,
|
|
"known_question_refusal_rate": 0.0,
|
|
"unknown_question_read_rate": 0.0,
|
|
"counts": {
|
|
"tp": 17814,
|
|
"tn": 4106,
|
|
"fp": 0,
|
|
"fn": 0
|
|
}
|
|
},
|
|
"thr0.50": {
|
|
"need_precision": 1.0,
|
|
"need_recall": 1.0,
|
|
"need_f1": 1.0,
|
|
"specificity_unknown_refusal": 1.0,
|
|
"abstention_accuracy": 1.0,
|
|
"known_question_refusal_rate": 0.0,
|
|
"unknown_question_read_rate": 0.0,
|
|
"counts": {
|
|
"tp": 17814,
|
|
"tn": 4106,
|
|
"fp": 0,
|
|
"fn": 0
|
|
}
|
|
},
|
|
"thr0.60": {
|
|
"need_precision": 1.0,
|
|
"need_recall": 1.0,
|
|
"need_f1": 1.0,
|
|
"specificity_unknown_refusal": 1.0,
|
|
"abstention_accuracy": 1.0,
|
|
"known_question_refusal_rate": 0.0,
|
|
"unknown_question_read_rate": 0.0,
|
|
"counts": {
|
|
"tp": 17814,
|
|
"tn": 4106,
|
|
"fp": 0,
|
|
"fn": 0
|
|
}
|
|
},
|
|
"thr0.70": {
|
|
"need_precision": 1.0,
|
|
"need_recall": 1.0,
|
|
"need_f1": 1.0,
|
|
"specificity_unknown_refusal": 1.0,
|
|
"abstention_accuracy": 1.0,
|
|
"known_question_refusal_rate": 0.0,
|
|
"unknown_question_read_rate": 0.0,
|
|
"counts": {
|
|
"tp": 17814,
|
|
"tn": 4106,
|
|
"fp": 0,
|
|
"fn": 0
|
|
}
|
|
},
|
|
"thr0.80": {
|
|
"need_precision": 1.0,
|
|
"need_recall": 1.0,
|
|
"need_f1": 1.0,
|
|
"specificity_unknown_refusal": 1.0,
|
|
"abstention_accuracy": 1.0,
|
|
"known_question_refusal_rate": 0.0,
|
|
"unknown_question_read_rate": 0.0,
|
|
"counts": {
|
|
"tp": 17814,
|
|
"tn": 4106,
|
|
"fp": 0,
|
|
"fn": 0
|
|
}
|
|
}
|
|
},
|
|
"by_family": {
|
|
"benchmark_qa": {
|
|
"episodes": 64,
|
|
"answerable_episodes": 0,
|
|
"unknown_episodes": 64,
|
|
"top1": null,
|
|
"recall1": null,
|
|
"recall3": null,
|
|
"recall5": null,
|
|
"recall8": null,
|
|
"evidence_count_in_top3": null,
|
|
"hit3": null,
|
|
"mrr": null,
|
|
"ndcg3": null,
|
|
"mean_first_positive_rank": null,
|
|
"all_evidence_in_top3": null,
|
|
"multi_positive_episodes": 0,
|
|
"all_evidence_in_top3_multi": null,
|
|
"hop_accuracy": 1.0,
|
|
"hop_accuracy_answerable": null,
|
|
"hop_under_prediction": null,
|
|
"hop_over_prediction": null,
|
|
"mean_score_margin": null,
|
|
"median_score_margin": null,
|
|
"negative_margin_rate": null,
|
|
"hop_confusion": {
|
|
"true_0_pred_0": 64
|
|
},
|
|
"known_mean_need_prob": null,
|
|
"unknown_mean_need_prob": 4.8406160974828055e-06,
|
|
"thr0.50": {
|
|
"need_precision": null,
|
|
"need_recall": null,
|
|
"need_f1": null,
|
|
"specificity_unknown_refusal": 1.0,
|
|
"abstention_accuracy": 1.0,
|
|
"known_question_refusal_rate": null,
|
|
"unknown_question_read_rate": 0.0,
|
|
"counts": {
|
|
"tp": 0,
|
|
"tn": 64,
|
|
"fp": 0,
|
|
"fn": 0
|
|
}
|
|
}
|
|
},
|
|
"mega_validation": {
|
|
"episodes": 20000,
|
|
"answerable_episodes": 16000,
|
|
"unknown_episodes": 4000,
|
|
"top1": 0.9838125,
|
|
"recall1": 0.92203125,
|
|
"recall3": 0.9950625,
|
|
"recall5": 0.99978125,
|
|
"recall8": 0.999875,
|
|
"evidence_count_in_top3": 1.115125,
|
|
"hit3": 1.0,
|
|
"mrr": 0.991875,
|
|
"ndcg3": 0.9876817889701721,
|
|
"mean_first_positive_rank": 1.016375,
|
|
"all_evidence_in_top3": 0.990125,
|
|
"multi_positive_episodes": 2000,
|
|
"all_evidence_in_top3_multi": 0.921,
|
|
"hop_accuracy": 1.0,
|
|
"hop_accuracy_answerable": 1.0,
|
|
"hop_under_prediction": 0.0,
|
|
"hop_over_prediction": 0.0,
|
|
"mean_score_margin": 57.110110210259094,
|
|
"median_score_margin": 59.90444374084473,
|
|
"negative_margin_rate": 0.0161875,
|
|
"hop_confusion": {
|
|
"true_0_pred_0": 4000,
|
|
"true_0_pred_1": 0,
|
|
"true_0_pred_2": 0,
|
|
"true_1_pred_0": 0,
|
|
"true_1_pred_1": 14000,
|
|
"true_1_pred_2": 0,
|
|
"true_2_pred_0": 0,
|
|
"true_2_pred_1": 0,
|
|
"true_2_pred_2": 2000
|
|
},
|
|
"known_mean_need_prob": 0.9999999999701977,
|
|
"unknown_mean_need_prob": 6.379481702447893e-11,
|
|
"thr0.50": {
|
|
"need_precision": 1.0,
|
|
"need_recall": 1.0,
|
|
"need_f1": 1.0,
|
|
"specificity_unknown_refusal": 1.0,
|
|
"abstention_accuracy": 1.0,
|
|
"known_question_refusal_rate": 0.0,
|
|
"unknown_question_read_rate": 0.0,
|
|
"counts": {
|
|
"tp": 16000,
|
|
"tn": 4000,
|
|
"fp": 0,
|
|
"fn": 0
|
|
}
|
|
}
|
|
},
|
|
"memory_policy": {
|
|
"episodes": 1728,
|
|
"answerable_episodes": 1707,
|
|
"unknown_episodes": 21,
|
|
"top1": 0.5940246045694201,
|
|
"recall1": 0.5940246045694201,
|
|
"recall3": 0.6842413591095489,
|
|
"recall5": 0.7352079671939075,
|
|
"recall8": 0.8611599297012302,
|
|
"evidence_count_in_top3": 0.6842413591095489,
|
|
"hit3": 0.6842413591095489,
|
|
"mrr": 0.6760273585431921,
|
|
"ndcg3": 0.6466497456649108,
|
|
"mean_first_positive_rank": 3.4463971880492092,
|
|
"all_evidence_in_top3": 0.6842413591095489,
|
|
"multi_positive_episodes": 0,
|
|
"all_evidence_in_top3_multi": null,
|
|
"hop_accuracy": 1.0,
|
|
"hop_accuracy_answerable": 1.0,
|
|
"hop_under_prediction": 0.0,
|
|
"hop_over_prediction": 0.0,
|
|
"mean_score_margin": 0.27154229221702947,
|
|
"median_score_margin": 7.975786209106445,
|
|
"negative_margin_rate": 0.40597539543058,
|
|
"hop_confusion": {
|
|
"true_0_pred_0": 21,
|
|
"true_0_pred_1": 0,
|
|
"true_1_pred_0": 0,
|
|
"true_1_pred_1": 1707
|
|
},
|
|
"known_mean_need_prob": 0.9999999838679872,
|
|
"unknown_mean_need_prob": 0.0004654159529796483,
|
|
"thr0.50": {
|
|
"need_precision": 1.0,
|
|
"need_recall": 1.0,
|
|
"need_f1": 1.0,
|
|
"specificity_unknown_refusal": 1.0,
|
|
"abstention_accuracy": 1.0,
|
|
"known_question_refusal_rate": 0.0,
|
|
"unknown_question_read_rate": 0.0,
|
|
"counts": {
|
|
"tp": 1707,
|
|
"tn": 21,
|
|
"fp": 0,
|
|
"fn": 0
|
|
}
|
|
}
|
|
},
|
|
"native_memory": {
|
|
"episodes": 128,
|
|
"answerable_episodes": 107,
|
|
"unknown_episodes": 21,
|
|
"top1": 0.5233644859813084,
|
|
"recall1": 0.5233644859813084,
|
|
"recall3": 0.7757009345794392,
|
|
"recall5": 0.9252336448598131,
|
|
"recall8": 0.9719626168224299,
|
|
"evidence_count_in_top3": 0.7757009345794392,
|
|
"hit3": 0.7757009345794392,
|
|
"mrr": 0.674896157840083,
|
|
"ndcg3": 0.6678873486315127,
|
|
"mean_first_positive_rank": 2.3738317757009346,
|
|
"all_evidence_in_top3": 0.7757009345794392,
|
|
"multi_positive_episodes": 0,
|
|
"all_evidence_in_top3_multi": null,
|
|
"hop_accuracy": 1.0,
|
|
"hop_accuracy_answerable": 1.0,
|
|
"hop_under_prediction": 0.0,
|
|
"hop_over_prediction": 0.0,
|
|
"mean_score_margin": -0.8786688875929217,
|
|
"median_score_margin": 0.4756507873535156,
|
|
"negative_margin_rate": 0.4766355140186916,
|
|
"hop_confusion": {
|
|
"true_0_pred_0": 21,
|
|
"true_0_pred_1": 0,
|
|
"true_1_pred_0": 0,
|
|
"true_1_pred_1": 107
|
|
},
|
|
"known_mean_need_prob": 1.0,
|
|
"unknown_mean_need_prob": 0.0004654159529796483,
|
|
"thr0.50": {
|
|
"need_precision": 1.0,
|
|
"need_recall": 1.0,
|
|
"need_f1": 1.0,
|
|
"specificity_unknown_refusal": 1.0,
|
|
"abstention_accuracy": 1.0,
|
|
"known_question_refusal_rate": 0.0,
|
|
"unknown_question_read_rate": 0.0,
|
|
"counts": {
|
|
"tp": 107,
|
|
"tn": 21,
|
|
"fp": 0,
|
|
"fn": 0
|
|
}
|
|
}
|
|
}
|
|
},
|
|
"by_category": {
|
|
"conflict_update": {
|
|
"episodes": 2000,
|
|
"answerable_episodes": 2000,
|
|
"unknown_episodes": 0,
|
|
"top1": 0.993,
|
|
"recall1": 0.993,
|
|
"recall3": 1.0,
|
|
"recall5": 1.0,
|
|
"recall8": 1.0,
|
|
"evidence_count_in_top3": 1.0,
|
|
"hit3": 1.0,
|
|
"mrr": 0.9965,
|
|
"ndcg3": 0.9974165082750002,
|
|
"mean_first_positive_rank": 1.007,
|
|
"all_evidence_in_top3": 1.0,
|
|
"multi_positive_episodes": 0,
|
|
"all_evidence_in_top3_multi": null,
|
|
"hop_accuracy": 1.0,
|
|
"hop_accuracy_answerable": 1.0,
|
|
"hop_under_prediction": 0.0,
|
|
"hop_over_prediction": 0.0,
|
|
"mean_score_margin": 54.4210530708991,
|
|
"median_score_margin": 56.92102670669556,
|
|
"negative_margin_rate": 0.007,
|
|
"hop_confusion": {
|
|
"true_1_pred_1": 2000
|
|
},
|
|
"known_mean_need_prob": 1.0,
|
|
"unknown_mean_need_prob": null,
|
|
"thr0.50": {
|
|
"need_precision": 1.0,
|
|
"need_recall": 1.0,
|
|
"need_f1": 1.0,
|
|
"specificity_unknown_refusal": null,
|
|
"abstention_accuracy": 1.0,
|
|
"known_question_refusal_rate": 0.0,
|
|
"unknown_question_read_rate": null,
|
|
"counts": {
|
|
"tp": 2000,
|
|
"tn": 0,
|
|
"fp": 0,
|
|
"fn": 0
|
|
}
|
|
}
|
|
},
|
|
"distractor_128": {
|
|
"episodes": 2000,
|
|
"answerable_episodes": 2000,
|
|
"unknown_episodes": 0,
|
|
"top1": 0.975,
|
|
"recall1": 0.975,
|
|
"recall3": 1.0,
|
|
"recall5": 1.0,
|
|
"recall8": 1.0,
|
|
"evidence_count_in_top3": 1.0,
|
|
"hit3": 1.0,
|
|
"mrr": 0.9875,
|
|
"ndcg3": 0.9907732438392864,
|
|
"mean_first_positive_rank": 1.025,
|
|
"all_evidence_in_top3": 1.0,
|
|
"multi_positive_episodes": 0,
|
|
"all_evidence_in_top3_multi": null,
|
|
"hop_accuracy": 1.0,
|
|
"hop_accuracy_answerable": 1.0,
|
|
"hop_under_prediction": 0.0,
|
|
"hop_over_prediction": 0.0,
|
|
"mean_score_margin": 64.29146421498805,
|
|
"median_score_margin": 73.40720510482788,
|
|
"negative_margin_rate": 0.025,
|
|
"hop_confusion": {
|
|
"true_1_pred_1": 2000
|
|
},
|
|
"known_mean_need_prob": 1.0,
|
|
"unknown_mean_need_prob": null,
|
|
"thr0.50": {
|
|
"need_precision": 1.0,
|
|
"need_recall": 1.0,
|
|
"need_f1": 1.0,
|
|
"specificity_unknown_refusal": null,
|
|
"abstention_accuracy": 1.0,
|
|
"known_question_refusal_rate": 0.0,
|
|
"unknown_question_read_rate": null,
|
|
"counts": {
|
|
"tp": 2000,
|
|
"tn": 0,
|
|
"fp": 0,
|
|
"fn": 0
|
|
}
|
|
}
|
|
},
|
|
"distractor_32": {
|
|
"episodes": 2000,
|
|
"answerable_episodes": 2000,
|
|
"unknown_episodes": 0,
|
|
"top1": 0.9845,
|
|
"recall1": 0.9845,
|
|
"recall3": 1.0,
|
|
"recall5": 1.0,
|
|
"recall8": 1.0,
|
|
"evidence_count_in_top3": 1.0,
|
|
"hit3": 1.0,
|
|
"mrr": 0.99225,
|
|
"ndcg3": 0.9942794111803576,
|
|
"mean_first_positive_rank": 1.0155,
|
|
"all_evidence_in_top3": 1.0,
|
|
"multi_positive_episodes": 0,
|
|
"all_evidence_in_top3_multi": null,
|
|
"hop_accuracy": 1.0,
|
|
"hop_accuracy_answerable": 1.0,
|
|
"hop_under_prediction": 0.0,
|
|
"hop_over_prediction": 0.0,
|
|
"mean_score_margin": 65.52974343589041,
|
|
"median_score_margin": 73.97303533554077,
|
|
"negative_margin_rate": 0.0155,
|
|
"hop_confusion": {
|
|
"true_1_pred_1": 2000
|
|
},
|
|
"known_mean_need_prob": 1.0,
|
|
"unknown_mean_need_prob": null,
|
|
"thr0.50": {
|
|
"need_precision": 1.0,
|
|
"need_recall": 1.0,
|
|
"need_f1": 1.0,
|
|
"specificity_unknown_refusal": null,
|
|
"abstention_accuracy": 1.0,
|
|
"known_question_refusal_rate": 0.0,
|
|
"unknown_question_read_rate": null,
|
|
"counts": {
|
|
"tp": 2000,
|
|
"tn": 0,
|
|
"fp": 0,
|
|
"fn": 0
|
|
}
|
|
}
|
|
},
|
|
"forget_correction": {
|
|
"episodes": 2000,
|
|
"answerable_episodes": 0,
|
|
"unknown_episodes": 2000,
|
|
"top1": null,
|
|
"recall1": null,
|
|
"recall3": null,
|
|
"recall5": null,
|
|
"recall8": null,
|
|
"evidence_count_in_top3": null,
|
|
"hit3": null,
|
|
"mrr": null,
|
|
"ndcg3": null,
|
|
"mean_first_positive_rank": null,
|
|
"all_evidence_in_top3": null,
|
|
"multi_positive_episodes": 0,
|
|
"all_evidence_in_top3_multi": null,
|
|
"hop_accuracy": 1.0,
|
|
"hop_accuracy_answerable": null,
|
|
"hop_under_prediction": null,
|
|
"hop_over_prediction": null,
|
|
"mean_score_margin": null,
|
|
"median_score_margin": null,
|
|
"negative_margin_rate": null,
|
|
"hop_confusion": {
|
|
"true_0_pred_0": 2000
|
|
},
|
|
"known_mean_need_prob": null,
|
|
"unknown_mean_need_prob": 1.275862421441728e-10,
|
|
"thr0.50": {
|
|
"need_precision": null,
|
|
"need_recall": null,
|
|
"need_f1": null,
|
|
"specificity_unknown_refusal": 1.0,
|
|
"abstention_accuracy": 1.0,
|
|
"known_question_refusal_rate": null,
|
|
"unknown_question_read_rate": 0.0,
|
|
"counts": {
|
|
"tp": 0,
|
|
"tn": 2000,
|
|
"fp": 0,
|
|
"fn": 0
|
|
}
|
|
}
|
|
},
|
|
"long_context": {
|
|
"episodes": 2000,
|
|
"answerable_episodes": 2000,
|
|
"unknown_episodes": 0,
|
|
"top1": 0.9795,
|
|
"recall1": 0.9795,
|
|
"recall3": 1.0,
|
|
"recall5": 1.0,
|
|
"recall8": 1.0,
|
|
"evidence_count_in_top3": 1.0,
|
|
"hit3": 1.0,
|
|
"mrr": 0.9896666666666666,
|
|
"ndcg3": 0.9923685950714292,
|
|
"mean_first_positive_rank": 1.021,
|
|
"all_evidence_in_top3": 1.0,
|
|
"multi_positive_episodes": 0,
|
|
"all_evidence_in_top3_multi": null,
|
|
"hop_accuracy": 1.0,
|
|
"hop_accuracy_answerable": 1.0,
|
|
"hop_under_prediction": 0.0,
|
|
"hop_over_prediction": 0.0,
|
|
"mean_score_margin": 60.150181606799364,
|
|
"median_score_margin": 68.01980018615723,
|
|
"negative_margin_rate": 0.0205,
|
|
"hop_confusion": {
|
|
"true_1_pred_1": 2000
|
|
},
|
|
"known_mean_need_prob": 1.0,
|
|
"unknown_mean_need_prob": null,
|
|
"thr0.50": {
|
|
"need_precision": 1.0,
|
|
"need_recall": 1.0,
|
|
"need_f1": 1.0,
|
|
"specificity_unknown_refusal": null,
|
|
"abstention_accuracy": 1.0,
|
|
"known_question_refusal_rate": 0.0,
|
|
"unknown_question_read_rate": null,
|
|
"counts": {
|
|
"tp": 2000,
|
|
"tn": 0,
|
|
"fp": 0,
|
|
"fn": 0
|
|
}
|
|
}
|
|
},
|
|
"multi_hop": {
|
|
"episodes": 2000,
|
|
"answerable_episodes": 2000,
|
|
"unknown_episodes": 0,
|
|
"top1": 0.9885,
|
|
"recall1": 0.49425,
|
|
"recall3": 0.9605,
|
|
"recall5": 0.99825,
|
|
"recall8": 0.999,
|
|
"evidence_count_in_top3": 1.921,
|
|
"hit3": 1.0,
|
|
"mrr": 0.99425,
|
|
"ndcg3": 0.9452009954703022,
|
|
"mean_first_positive_rank": 1.0115,
|
|
"all_evidence_in_top3": 0.921,
|
|
"multi_positive_episodes": 2000,
|
|
"all_evidence_in_top3_multi": 0.921,
|
|
"hop_accuracy": 1.0,
|
|
"hop_accuracy_answerable": 1.0,
|
|
"hop_under_prediction": 0.0,
|
|
"hop_over_prediction": 0.0,
|
|
"mean_score_margin": 55.6439196386747,
|
|
"median_score_margin": 62.50904202461243,
|
|
"negative_margin_rate": 0.0115,
|
|
"hop_confusion": {
|
|
"true_2_pred_2": 2000
|
|
},
|
|
"known_mean_need_prob": 0.9999999997615814,
|
|
"unknown_mean_need_prob": null,
|
|
"thr0.50": {
|
|
"need_precision": 1.0,
|
|
"need_recall": 1.0,
|
|
"need_f1": 1.0,
|
|
"specificity_unknown_refusal": null,
|
|
"abstention_accuracy": 1.0,
|
|
"known_question_refusal_rate": 0.0,
|
|
"unknown_question_read_rate": null,
|
|
"counts": {
|
|
"tp": 2000,
|
|
"tn": 0,
|
|
"fp": 0,
|
|
"fn": 0
|
|
}
|
|
}
|
|
},
|
|
"paraphrase": {
|
|
"episodes": 2000,
|
|
"answerable_episodes": 2000,
|
|
"unknown_episodes": 0,
|
|
"top1": 0.995,
|
|
"recall1": 0.995,
|
|
"recall3": 1.0,
|
|
"recall5": 1.0,
|
|
"recall8": 1.0,
|
|
"evidence_count_in_top3": 1.0,
|
|
"hit3": 1.0,
|
|
"mrr": 0.9975,
|
|
"ndcg3": 0.9981546487678573,
|
|
"mean_first_positive_rank": 1.005,
|
|
"all_evidence_in_top3": 1.0,
|
|
"multi_positive_episodes": 0,
|
|
"all_evidence_in_top3_multi": null,
|
|
"hop_accuracy": 1.0,
|
|
"hop_accuracy_answerable": 1.0,
|
|
"hop_under_prediction": 0.0,
|
|
"hop_over_prediction": 0.0,
|
|
"mean_score_margin": 48.13904330630228,
|
|
"median_score_margin": 49.25835180282593,
|
|
"negative_margin_rate": 0.005,
|
|
"hop_confusion": {
|
|
"true_1_pred_1": 2000
|
|
},
|
|
"known_mean_need_prob": 1.0,
|
|
"unknown_mean_need_prob": null,
|
|
"thr0.50": {
|
|
"need_precision": 1.0,
|
|
"need_recall": 1.0,
|
|
"need_f1": 1.0,
|
|
"specificity_unknown_refusal": null,
|
|
"abstention_accuracy": 1.0,
|
|
"known_question_refusal_rate": 0.0,
|
|
"unknown_question_read_rate": null,
|
|
"counts": {
|
|
"tp": 2000,
|
|
"tn": 0,
|
|
"fp": 0,
|
|
"fn": 0
|
|
}
|
|
}
|
|
},
|
|
"random_position": {
|
|
"episodes": 2000,
|
|
"answerable_episodes": 2000,
|
|
"unknown_episodes": 0,
|
|
"top1": 0.9805,
|
|
"recall1": 0.9805,
|
|
"recall3": 1.0,
|
|
"recall5": 1.0,
|
|
"recall8": 1.0,
|
|
"evidence_count_in_top3": 1.0,
|
|
"hit3": 1.0,
|
|
"mrr": 0.9901666666666666,
|
|
"ndcg3": 0.9927376653178577,
|
|
"mean_first_positive_rank": 1.02,
|
|
"all_evidence_in_top3": 1.0,
|
|
"multi_positive_episodes": 0,
|
|
"all_evidence_in_top3_multi": null,
|
|
"hop_accuracy": 1.0,
|
|
"hop_accuracy_answerable": 1.0,
|
|
"hop_under_prediction": 0.0,
|
|
"hop_over_prediction": 0.0,
|
|
"mean_score_margin": 50.85701001610607,
|
|
"median_score_margin": 54.9089093208313,
|
|
"negative_margin_rate": 0.0195,
|
|
"hop_confusion": {
|
|
"true_1_pred_1": 2000
|
|
},
|
|
"known_mean_need_prob": 1.0,
|
|
"unknown_mean_need_prob": null,
|
|
"thr0.50": {
|
|
"need_precision": 1.0,
|
|
"need_recall": 1.0,
|
|
"need_f1": 1.0,
|
|
"specificity_unknown_refusal": null,
|
|
"abstention_accuracy": 1.0,
|
|
"known_question_refusal_rate": 0.0,
|
|
"unknown_question_read_rate": null,
|
|
"counts": {
|
|
"tp": 2000,
|
|
"tn": 0,
|
|
"fp": 0,
|
|
"fn": 0
|
|
}
|
|
}
|
|
},
|
|
"single_fact": {
|
|
"episodes": 2000,
|
|
"answerable_episodes": 2000,
|
|
"unknown_episodes": 0,
|
|
"top1": 0.9745,
|
|
"recall1": 0.9745,
|
|
"recall3": 1.0,
|
|
"recall5": 1.0,
|
|
"recall8": 1.0,
|
|
"evidence_count_in_top3": 1.0,
|
|
"hit3": 1.0,
|
|
"mrr": 0.9871666666666666,
|
|
"ndcg3": 0.9905232438392865,
|
|
"mean_first_positive_rank": 1.026,
|
|
"all_evidence_in_top3": 1.0,
|
|
"multi_positive_episodes": 0,
|
|
"all_evidence_in_top3_multi": null,
|
|
"hop_accuracy": 1.0,
|
|
"hop_accuracy_answerable": 1.0,
|
|
"hop_under_prediction": 0.0,
|
|
"hop_over_prediction": 0.0,
|
|
"mean_score_margin": 57.84846639241278,
|
|
"median_score_margin": 63.093711376190186,
|
|
"negative_margin_rate": 0.0255,
|
|
"hop_confusion": {
|
|
"true_1_pred_1": 2000
|
|
},
|
|
"known_mean_need_prob": 1.0,
|
|
"unknown_mean_need_prob": null,
|
|
"thr0.50": {
|
|
"need_precision": 1.0,
|
|
"need_recall": 1.0,
|
|
"need_f1": 1.0,
|
|
"specificity_unknown_refusal": null,
|
|
"abstention_accuracy": 1.0,
|
|
"known_question_refusal_rate": 0.0,
|
|
"unknown_question_read_rate": null,
|
|
"counts": {
|
|
"tp": 2000,
|
|
"tn": 0,
|
|
"fp": 0,
|
|
"fn": 0
|
|
}
|
|
}
|
|
},
|
|
"unknown": {
|
|
"episodes": 1920,
|
|
"answerable_episodes": 1814,
|
|
"unknown_episodes": 106,
|
|
"top1": 0.5898566703417861,
|
|
"recall1": 0.5898566703417861,
|
|
"recall3": 0.6896361631753032,
|
|
"recall5": 0.7464167585446527,
|
|
"recall8": 0.8676957001102535,
|
|
"evidence_count_in_top3": 0.6896361631753032,
|
|
"hit3": 0.6896361631753032,
|
|
"mrr": 0.6759606339151697,
|
|
"ndcg3": 0.6479024598421029,
|
|
"mean_first_positive_rank": 3.383131201764057,
|
|
"all_evidence_in_top3": 0.6896361631753032,
|
|
"multi_positive_episodes": 0,
|
|
"all_evidence_in_top3_multi": null,
|
|
"hop_accuracy": 1.0,
|
|
"hop_accuracy_answerable": 1.0,
|
|
"hop_under_prediction": 0.0,
|
|
"hop_over_prediction": 0.0,
|
|
"mean_score_margin": 0.20369631854576997,
|
|
"median_score_margin": 6.827788352966309,
|
|
"negative_margin_rate": 0.4101433296582139,
|
|
"hop_confusion": {
|
|
"true_0_pred_0": 106,
|
|
"true_0_pred_1": 0,
|
|
"true_1_pred_0": 0,
|
|
"true_1_pred_1": 1814
|
|
},
|
|
"known_mean_need_prob": 0.9999999848195447,
|
|
"unknown_mean_need_prob": 0.00018733273071117103,
|
|
"thr0.50": {
|
|
"need_precision": 1.0,
|
|
"need_recall": 1.0,
|
|
"need_f1": 1.0,
|
|
"specificity_unknown_refusal": 1.0,
|
|
"abstention_accuracy": 1.0,
|
|
"known_question_refusal_rate": 0.0,
|
|
"unknown_question_read_rate": 0.0,
|
|
"counts": {
|
|
"tp": 1814,
|
|
"tn": 106,
|
|
"fp": 0,
|
|
"fn": 0
|
|
}
|
|
}
|
|
},
|
|
"unknown_abstention": {
|
|
"episodes": 2000,
|
|
"answerable_episodes": 0,
|
|
"unknown_episodes": 2000,
|
|
"top1": null,
|
|
"recall1": null,
|
|
"recall3": null,
|
|
"recall5": null,
|
|
"recall8": null,
|
|
"evidence_count_in_top3": null,
|
|
"hit3": null,
|
|
"mrr": null,
|
|
"ndcg3": null,
|
|
"mean_first_positive_rank": null,
|
|
"all_evidence_in_top3": null,
|
|
"multi_positive_episodes": 0,
|
|
"all_evidence_in_top3_multi": null,
|
|
"hop_accuracy": 1.0,
|
|
"hop_accuracy_answerable": null,
|
|
"hop_under_prediction": null,
|
|
"hop_over_prediction": null,
|
|
"mean_score_margin": null,
|
|
"median_score_margin": null,
|
|
"negative_margin_rate": null,
|
|
"hop_confusion": {
|
|
"true_0_pred_0": 2000
|
|
},
|
|
"known_mean_need_prob": null,
|
|
"unknown_mean_need_prob": 3.3919047850488803e-15,
|
|
"thr0.50": {
|
|
"need_precision": null,
|
|
"need_recall": null,
|
|
"need_f1": null,
|
|
"specificity_unknown_refusal": 1.0,
|
|
"abstention_accuracy": 1.0,
|
|
"known_question_refusal_rate": null,
|
|
"unknown_question_read_rate": 0.0,
|
|
"counts": {
|
|
"tp": 0,
|
|
"tn": 2000,
|
|
"fp": 0,
|
|
"fn": 0
|
|
}
|
|
}
|
|
}
|
|
}
|
|
},
|
|
"REPLAY-128 v7 final": {
|
|
"checkpoint": "checkpoints\\router_replay_v7_v2_128\\memory_router_v2.pt",
|
|
"router": {
|
|
"kind": "router_v2",
|
|
"step": null,
|
|
"arch_config": {
|
|
"arch": "router_v2",
|
|
"hidden_size": 2560,
|
|
"router_dim": 128,
|
|
"num_heads": 8,
|
|
"max_hops": 3
|
|
},
|
|
"parameters": 2037774,
|
|
"checkpoint_bytes": 8156921
|
|
},
|
|
"storage": {
|
|
"address_bytes_per_record": 512,
|
|
"address_mb_per_1m_records": 512,
|
|
"checkpoint_bytes": 8156921
|
|
},
|
|
"latency": {
|
|
"cuda": {
|
|
"single_query_latency_ms_mean": 0.9663366669459114,
|
|
"single_query_latency_ms_p50": 0.9168999968096614,
|
|
"single_query_latency_ms_p95": 1.2540000025182962,
|
|
"queries_per_second": 1034.8360299319709,
|
|
"samples": 150,
|
|
"batched_qps_64": 69378.19790068052,
|
|
"batched_latency_ms_64": 0.014413750000130676,
|
|
"batched_qps_256": 242759.87633474605,
|
|
"batched_latency_ms_256": 0.004119296875160217
|
|
},
|
|
"cpu": {
|
|
"single_query_latency_ms_mean": 0.3868726668103288,
|
|
"single_query_latency_ms_p50": 0.36950000503566116,
|
|
"single_query_latency_ms_p95": 0.5363999953260645,
|
|
"queries_per_second": 2584.829805229605,
|
|
"samples": 150,
|
|
"batched_qps_64": 13315.579227047761,
|
|
"batched_latency_ms_64": 0.07510000000365835,
|
|
"batched_qps_256": 12736.711281807726,
|
|
"batched_latency_ms_256": 0.07851320312397547
|
|
}
|
|
},
|
|
"metrics": {
|
|
"episodes": 21920,
|
|
"answerable_episodes": 17814,
|
|
"unknown_episodes": 4106,
|
|
"top1": 0.9413944088918828,
|
|
"recall1": 0.8857920736499383,
|
|
"recall3": 0.9647748961490962,
|
|
"recall5": 0.9714550353654429,
|
|
"recall8": 0.9804086673402942,
|
|
"evidence_count_in_top3": 1.0744919726058157,
|
|
"hit3": 0.9673290670259347,
|
|
"mrr": 0.957689277532011,
|
|
"ndcg3": 0.9536787031890833,
|
|
"mean_first_positive_rank": 1.34001347254968,
|
|
"all_evidence_in_top3": 0.9622207252722578,
|
|
"multi_positive_episodes": 2000,
|
|
"all_evidence_in_top3_multi": 0.9545,
|
|
"hop_accuracy": 1.0,
|
|
"hop_accuracy_answerable": 1.0,
|
|
"hop_under_prediction": 0.0,
|
|
"hop_over_prediction": 0.0,
|
|
"mean_score_margin": 40.14557690149532,
|
|
"median_score_margin": 45.34479379653931,
|
|
"negative_margin_rate": 0.05860559110811721,
|
|
"hop_confusion": {
|
|
"true_0_pred_0": 4106,
|
|
"true_0_pred_1": 0,
|
|
"true_0_pred_2": 0,
|
|
"true_1_pred_0": 0,
|
|
"true_1_pred_1": 15814,
|
|
"true_1_pred_2": 0,
|
|
"true_2_pred_0": 0,
|
|
"true_2_pred_1": 0,
|
|
"true_2_pred_2": 2000
|
|
},
|
|
"known_mean_need_prob": 0.9999999984274064,
|
|
"unknown_mean_need_prob": 4.836221294362452e-06,
|
|
"thr0.30": {
|
|
"need_precision": 1.0,
|
|
"need_recall": 1.0,
|
|
"need_f1": 1.0,
|
|
"specificity_unknown_refusal": 1.0,
|
|
"abstention_accuracy": 1.0,
|
|
"known_question_refusal_rate": 0.0,
|
|
"unknown_question_read_rate": 0.0,
|
|
"counts": {
|
|
"tp": 17814,
|
|
"tn": 4106,
|
|
"fp": 0,
|
|
"fn": 0
|
|
}
|
|
},
|
|
"thr0.40": {
|
|
"need_precision": 1.0,
|
|
"need_recall": 1.0,
|
|
"need_f1": 1.0,
|
|
"specificity_unknown_refusal": 1.0,
|
|
"abstention_accuracy": 1.0,
|
|
"known_question_refusal_rate": 0.0,
|
|
"unknown_question_read_rate": 0.0,
|
|
"counts": {
|
|
"tp": 17814,
|
|
"tn": 4106,
|
|
"fp": 0,
|
|
"fn": 0
|
|
}
|
|
},
|
|
"thr0.50": {
|
|
"need_precision": 1.0,
|
|
"need_recall": 1.0,
|
|
"need_f1": 1.0,
|
|
"specificity_unknown_refusal": 1.0,
|
|
"abstention_accuracy": 1.0,
|
|
"known_question_refusal_rate": 0.0,
|
|
"unknown_question_read_rate": 0.0,
|
|
"counts": {
|
|
"tp": 17814,
|
|
"tn": 4106,
|
|
"fp": 0,
|
|
"fn": 0
|
|
}
|
|
},
|
|
"thr0.60": {
|
|
"need_precision": 1.0,
|
|
"need_recall": 1.0,
|
|
"need_f1": 1.0,
|
|
"specificity_unknown_refusal": 1.0,
|
|
"abstention_accuracy": 1.0,
|
|
"known_question_refusal_rate": 0.0,
|
|
"unknown_question_read_rate": 0.0,
|
|
"counts": {
|
|
"tp": 17814,
|
|
"tn": 4106,
|
|
"fp": 0,
|
|
"fn": 0
|
|
}
|
|
},
|
|
"thr0.70": {
|
|
"need_precision": 1.0,
|
|
"need_recall": 1.0,
|
|
"need_f1": 1.0,
|
|
"specificity_unknown_refusal": 1.0,
|
|
"abstention_accuracy": 1.0,
|
|
"known_question_refusal_rate": 0.0,
|
|
"unknown_question_read_rate": 0.0,
|
|
"counts": {
|
|
"tp": 17814,
|
|
"tn": 4106,
|
|
"fp": 0,
|
|
"fn": 0
|
|
}
|
|
},
|
|
"thr0.80": {
|
|
"need_precision": 1.0,
|
|
"need_recall": 1.0,
|
|
"need_f1": 1.0,
|
|
"specificity_unknown_refusal": 1.0,
|
|
"abstention_accuracy": 1.0,
|
|
"known_question_refusal_rate": 0.0,
|
|
"unknown_question_read_rate": 0.0,
|
|
"counts": {
|
|
"tp": 17814,
|
|
"tn": 4106,
|
|
"fp": 0,
|
|
"fn": 0
|
|
}
|
|
}
|
|
},
|
|
"by_family": {
|
|
"benchmark_qa": {
|
|
"episodes": 64,
|
|
"answerable_episodes": 0,
|
|
"unknown_episodes": 64,
|
|
"top1": null,
|
|
"recall1": null,
|
|
"recall3": null,
|
|
"recall5": null,
|
|
"recall8": null,
|
|
"evidence_count_in_top3": null,
|
|
"hit3": null,
|
|
"mrr": null,
|
|
"ndcg3": null,
|
|
"mean_first_positive_rank": null,
|
|
"all_evidence_in_top3": null,
|
|
"multi_positive_episodes": 0,
|
|
"all_evidence_in_top3_multi": null,
|
|
"hop_accuracy": 1.0,
|
|
"hop_accuracy_answerable": null,
|
|
"hop_under_prediction": null,
|
|
"hop_over_prediction": null,
|
|
"mean_score_margin": null,
|
|
"median_score_margin": null,
|
|
"negative_margin_rate": null,
|
|
"hop_confusion": {
|
|
"true_0_pred_0": 64
|
|
},
|
|
"known_mean_need_prob": null,
|
|
"unknown_mean_need_prob": 4.8406160974828055e-06,
|
|
"thr0.50": {
|
|
"need_precision": null,
|
|
"need_recall": null,
|
|
"need_f1": null,
|
|
"specificity_unknown_refusal": 1.0,
|
|
"abstention_accuracy": 1.0,
|
|
"known_question_refusal_rate": null,
|
|
"unknown_question_read_rate": 0.0,
|
|
"counts": {
|
|
"tp": 0,
|
|
"tn": 64,
|
|
"fp": 0,
|
|
"fn": 0
|
|
}
|
|
}
|
|
},
|
|
"mega_validation": {
|
|
"episodes": 20000,
|
|
"answerable_episodes": 16000,
|
|
"unknown_episodes": 4000,
|
|
"top1": 0.98225,
|
|
"recall1": 0.92034375,
|
|
"recall3": 0.99715625,
|
|
"recall5": 0.99959375,
|
|
"recall8": 0.9996875,
|
|
"evidence_count_in_top3": 1.1193125,
|
|
"hit3": 1.0,
|
|
"mrr": 0.9911041666666667,
|
|
"ndcg3": 0.9893498205729669,
|
|
"mean_first_positive_rank": 1.017875,
|
|
"all_evidence_in_top3": 0.9943125,
|
|
"multi_positive_episodes": 2000,
|
|
"all_evidence_in_top3_multi": 0.9545,
|
|
"hop_accuracy": 1.0,
|
|
"hop_accuracy_answerable": 1.0,
|
|
"hop_under_prediction": 0.0,
|
|
"hop_over_prediction": 0.0,
|
|
"mean_score_margin": 44.73340179644042,
|
|
"median_score_margin": 47.92497134208679,
|
|
"negative_margin_rate": 0.01775,
|
|
"hop_confusion": {
|
|
"true_0_pred_0": 4000,
|
|
"true_0_pred_1": 0,
|
|
"true_0_pred_2": 0,
|
|
"true_1_pred_0": 0,
|
|
"true_1_pred_1": 14000,
|
|
"true_1_pred_2": 0,
|
|
"true_2_pred_0": 0,
|
|
"true_2_pred_1": 0,
|
|
"true_2_pred_2": 2000
|
|
},
|
|
"known_mean_need_prob": 0.9999999999701977,
|
|
"unknown_mean_need_prob": 6.379481702447893e-11,
|
|
"thr0.50": {
|
|
"need_precision": 1.0,
|
|
"need_recall": 1.0,
|
|
"need_f1": 1.0,
|
|
"specificity_unknown_refusal": 1.0,
|
|
"abstention_accuracy": 1.0,
|
|
"known_question_refusal_rate": 0.0,
|
|
"unknown_question_read_rate": 0.0,
|
|
"counts": {
|
|
"tp": 16000,
|
|
"tn": 4000,
|
|
"fp": 0,
|
|
"fn": 0
|
|
}
|
|
}
|
|
},
|
|
"memory_policy": {
|
|
"episodes": 1728,
|
|
"answerable_episodes": 1707,
|
|
"unknown_episodes": 21,
|
|
"top1": 0.5875805506736965,
|
|
"recall1": 0.5875805506736965,
|
|
"recall3": 0.6731107205623902,
|
|
"recall5": 0.7106033977738723,
|
|
"recall8": 0.7996485061511424,
|
|
"evidence_count_in_top3": 0.6731107205623902,
|
|
"hit3": 0.6731107205623902,
|
|
"mrr": 0.6633131807141506,
|
|
"ndcg3": 0.6380925044585338,
|
|
"mean_first_positive_rank": 4.295254833040421,
|
|
"all_evidence_in_top3": 0.6731107205623902,
|
|
"multi_positive_episodes": 0,
|
|
"all_evidence_in_top3_multi": null,
|
|
"hop_accuracy": 1.0,
|
|
"hop_accuracy_answerable": 1.0,
|
|
"hop_under_prediction": 0.0,
|
|
"hop_over_prediction": 0.0,
|
|
"mean_score_margin": -0.2721888081208264,
|
|
"median_score_margin": 8.391286849975586,
|
|
"negative_margin_rate": 0.4124194493263035,
|
|
"hop_confusion": {
|
|
"true_0_pred_0": 21,
|
|
"true_0_pred_1": 0,
|
|
"true_1_pred_0": 0,
|
|
"true_1_pred_1": 1707
|
|
},
|
|
"known_mean_need_prob": 0.9999999838679872,
|
|
"unknown_mean_need_prob": 0.0004654159529796483,
|
|
"thr0.50": {
|
|
"need_precision": 1.0,
|
|
"need_recall": 1.0,
|
|
"need_f1": 1.0,
|
|
"specificity_unknown_refusal": 1.0,
|
|
"abstention_accuracy": 1.0,
|
|
"known_question_refusal_rate": 0.0,
|
|
"unknown_question_read_rate": 0.0,
|
|
"counts": {
|
|
"tp": 1707,
|
|
"tn": 21,
|
|
"fp": 0,
|
|
"fn": 0
|
|
}
|
|
}
|
|
},
|
|
"native_memory": {
|
|
"episodes": 128,
|
|
"answerable_episodes": 107,
|
|
"unknown_episodes": 21,
|
|
"top1": 0.4766355140186916,
|
|
"recall1": 0.4766355140186916,
|
|
"recall3": 0.7757009345794392,
|
|
"recall5": 0.9252336448598131,
|
|
"recall8": 0.9813084112149533,
|
|
"evidence_count_in_top3": 0.7757009345794392,
|
|
"hit3": 0.7757009345794392,
|
|
"mrr": 0.6573319982198488,
|
|
"ndcg3": 0.6543120031041451,
|
|
"mean_first_positive_rank": 2.364485981308411,
|
|
"all_evidence_in_top3": 0.7757009345794392,
|
|
"multi_positive_episodes": 0,
|
|
"all_evidence_in_top3_multi": null,
|
|
"hop_accuracy": 1.0,
|
|
"hop_accuracy_answerable": 1.0,
|
|
"hop_under_prediction": 0.0,
|
|
"hop_over_prediction": 0.0,
|
|
"mean_score_margin": -1.0887432181946586,
|
|
"median_score_margin": -0.6999244689941406,
|
|
"negative_margin_rate": 0.5233644859813084,
|
|
"hop_confusion": {
|
|
"true_0_pred_0": 21,
|
|
"true_0_pred_1": 0,
|
|
"true_1_pred_0": 0,
|
|
"true_1_pred_1": 107
|
|
},
|
|
"known_mean_need_prob": 1.0,
|
|
"unknown_mean_need_prob": 0.0004654159529796483,
|
|
"thr0.50": {
|
|
"need_precision": 1.0,
|
|
"need_recall": 1.0,
|
|
"need_f1": 1.0,
|
|
"specificity_unknown_refusal": 1.0,
|
|
"abstention_accuracy": 1.0,
|
|
"known_question_refusal_rate": 0.0,
|
|
"unknown_question_read_rate": 0.0,
|
|
"counts": {
|
|
"tp": 107,
|
|
"tn": 21,
|
|
"fp": 0,
|
|
"fn": 0
|
|
}
|
|
}
|
|
}
|
|
},
|
|
"by_category": {
|
|
"conflict_update": {
|
|
"episodes": 2000,
|
|
"answerable_episodes": 2000,
|
|
"unknown_episodes": 0,
|
|
"top1": 0.991,
|
|
"recall1": 0.991,
|
|
"recall3": 1.0,
|
|
"recall5": 1.0,
|
|
"recall8": 1.0,
|
|
"evidence_count_in_top3": 1.0,
|
|
"hit3": 1.0,
|
|
"mrr": 0.9955,
|
|
"ndcg3": 0.9966783677821431,
|
|
"mean_first_positive_rank": 1.009,
|
|
"all_evidence_in_top3": 1.0,
|
|
"multi_positive_episodes": 0,
|
|
"all_evidence_in_top3_multi": null,
|
|
"hop_accuracy": 1.0,
|
|
"hop_accuracy_answerable": 1.0,
|
|
"hop_under_prediction": 0.0,
|
|
"hop_over_prediction": 0.0,
|
|
"mean_score_margin": 42.24227243242413,
|
|
"median_score_margin": 42.77137851715088,
|
|
"negative_margin_rate": 0.009,
|
|
"hop_confusion": {
|
|
"true_1_pred_1": 2000
|
|
},
|
|
"known_mean_need_prob": 1.0,
|
|
"unknown_mean_need_prob": null,
|
|
"thr0.50": {
|
|
"need_precision": 1.0,
|
|
"need_recall": 1.0,
|
|
"need_f1": 1.0,
|
|
"specificity_unknown_refusal": null,
|
|
"abstention_accuracy": 1.0,
|
|
"known_question_refusal_rate": 0.0,
|
|
"unknown_question_read_rate": null,
|
|
"counts": {
|
|
"tp": 2000,
|
|
"tn": 0,
|
|
"fp": 0,
|
|
"fn": 0
|
|
}
|
|
}
|
|
},
|
|
"distractor_128": {
|
|
"episodes": 2000,
|
|
"answerable_episodes": 2000,
|
|
"unknown_episodes": 0,
|
|
"top1": 0.973,
|
|
"recall1": 0.973,
|
|
"recall3": 1.0,
|
|
"recall5": 1.0,
|
|
"recall8": 1.0,
|
|
"evidence_count_in_top3": 1.0,
|
|
"hit3": 1.0,
|
|
"mrr": 0.9865,
|
|
"ndcg3": 0.9900351033464293,
|
|
"mean_first_positive_rank": 1.027,
|
|
"all_evidence_in_top3": 1.0,
|
|
"multi_positive_episodes": 0,
|
|
"all_evidence_in_top3_multi": null,
|
|
"hop_accuracy": 1.0,
|
|
"hop_accuracy_answerable": 1.0,
|
|
"hop_under_prediction": 0.0,
|
|
"hop_over_prediction": 0.0,
|
|
"mean_score_margin": 50.82861385926604,
|
|
"median_score_margin": 54.98523235321045,
|
|
"negative_margin_rate": 0.027,
|
|
"hop_confusion": {
|
|
"true_1_pred_1": 2000
|
|
},
|
|
"known_mean_need_prob": 1.0,
|
|
"unknown_mean_need_prob": null,
|
|
"thr0.50": {
|
|
"need_precision": 1.0,
|
|
"need_recall": 1.0,
|
|
"need_f1": 1.0,
|
|
"specificity_unknown_refusal": null,
|
|
"abstention_accuracy": 1.0,
|
|
"known_question_refusal_rate": 0.0,
|
|
"unknown_question_read_rate": null,
|
|
"counts": {
|
|
"tp": 2000,
|
|
"tn": 0,
|
|
"fp": 0,
|
|
"fn": 0
|
|
}
|
|
}
|
|
},
|
|
"distractor_32": {
|
|
"episodes": 2000,
|
|
"answerable_episodes": 2000,
|
|
"unknown_episodes": 0,
|
|
"top1": 0.983,
|
|
"recall1": 0.983,
|
|
"recall3": 1.0,
|
|
"recall5": 1.0,
|
|
"recall8": 1.0,
|
|
"evidence_count_in_top3": 1.0,
|
|
"hit3": 1.0,
|
|
"mrr": 0.9915,
|
|
"ndcg3": 0.9937258058107148,
|
|
"mean_first_positive_rank": 1.017,
|
|
"all_evidence_in_top3": 1.0,
|
|
"multi_positive_episodes": 0,
|
|
"all_evidence_in_top3_multi": null,
|
|
"hop_accuracy": 1.0,
|
|
"hop_accuracy_answerable": 1.0,
|
|
"hop_under_prediction": 0.0,
|
|
"hop_over_prediction": 0.0,
|
|
"mean_score_margin": 51.79758185644448,
|
|
"median_score_margin": 55.65491724014282,
|
|
"negative_margin_rate": 0.017,
|
|
"hop_confusion": {
|
|
"true_1_pred_1": 2000
|
|
},
|
|
"known_mean_need_prob": 1.0,
|
|
"unknown_mean_need_prob": null,
|
|
"thr0.50": {
|
|
"need_precision": 1.0,
|
|
"need_recall": 1.0,
|
|
"need_f1": 1.0,
|
|
"specificity_unknown_refusal": null,
|
|
"abstention_accuracy": 1.0,
|
|
"known_question_refusal_rate": 0.0,
|
|
"unknown_question_read_rate": null,
|
|
"counts": {
|
|
"tp": 2000,
|
|
"tn": 0,
|
|
"fp": 0,
|
|
"fn": 0
|
|
}
|
|
}
|
|
},
|
|
"forget_correction": {
|
|
"episodes": 2000,
|
|
"answerable_episodes": 0,
|
|
"unknown_episodes": 2000,
|
|
"top1": null,
|
|
"recall1": null,
|
|
"recall3": null,
|
|
"recall5": null,
|
|
"recall8": null,
|
|
"evidence_count_in_top3": null,
|
|
"hit3": null,
|
|
"mrr": null,
|
|
"ndcg3": null,
|
|
"mean_first_positive_rank": null,
|
|
"all_evidence_in_top3": null,
|
|
"multi_positive_episodes": 0,
|
|
"all_evidence_in_top3_multi": null,
|
|
"hop_accuracy": 1.0,
|
|
"hop_accuracy_answerable": null,
|
|
"hop_under_prediction": null,
|
|
"hop_over_prediction": null,
|
|
"mean_score_margin": null,
|
|
"median_score_margin": null,
|
|
"negative_margin_rate": null,
|
|
"hop_confusion": {
|
|
"true_0_pred_0": 2000
|
|
},
|
|
"known_mean_need_prob": null,
|
|
"unknown_mean_need_prob": 1.275862421441728e-10,
|
|
"thr0.50": {
|
|
"need_precision": null,
|
|
"need_recall": null,
|
|
"need_f1": null,
|
|
"specificity_unknown_refusal": 1.0,
|
|
"abstention_accuracy": 1.0,
|
|
"known_question_refusal_rate": null,
|
|
"unknown_question_read_rate": 0.0,
|
|
"counts": {
|
|
"tp": 0,
|
|
"tn": 2000,
|
|
"fp": 0,
|
|
"fn": 0
|
|
}
|
|
}
|
|
},
|
|
"long_context": {
|
|
"episodes": 2000,
|
|
"answerable_episodes": 2000,
|
|
"unknown_episodes": 0,
|
|
"top1": 0.978,
|
|
"recall1": 0.978,
|
|
"recall3": 1.0,
|
|
"recall5": 1.0,
|
|
"recall8": 1.0,
|
|
"evidence_count_in_top3": 1.0,
|
|
"hit3": 1.0,
|
|
"mrr": 0.9889166666666667,
|
|
"ndcg3": 0.9918149897017863,
|
|
"mean_first_positive_rank": 1.0225,
|
|
"all_evidence_in_top3": 1.0,
|
|
"multi_positive_episodes": 0,
|
|
"all_evidence_in_top3_multi": null,
|
|
"hop_accuracy": 1.0,
|
|
"hop_accuracy_answerable": 1.0,
|
|
"hop_under_prediction": 0.0,
|
|
"hop_over_prediction": 0.0,
|
|
"mean_score_margin": 47.034401270009575,
|
|
"median_score_margin": 50.28937745094299,
|
|
"negative_margin_rate": 0.022,
|
|
"hop_confusion": {
|
|
"true_1_pred_1": 2000
|
|
},
|
|
"known_mean_need_prob": 1.0,
|
|
"unknown_mean_need_prob": null,
|
|
"thr0.50": {
|
|
"need_precision": 1.0,
|
|
"need_recall": 1.0,
|
|
"need_f1": 1.0,
|
|
"specificity_unknown_refusal": null,
|
|
"abstention_accuracy": 1.0,
|
|
"known_question_refusal_rate": 0.0,
|
|
"unknown_question_read_rate": null,
|
|
"counts": {
|
|
"tp": 2000,
|
|
"tn": 0,
|
|
"fp": 0,
|
|
"fn": 0
|
|
}
|
|
}
|
|
},
|
|
"multi_hop": {
|
|
"episodes": 2000,
|
|
"answerable_episodes": 2000,
|
|
"unknown_episodes": 0,
|
|
"top1": 0.9905,
|
|
"recall1": 0.49525,
|
|
"recall3": 0.97725,
|
|
"recall5": 0.99675,
|
|
"recall8": 0.9975,
|
|
"evidence_count_in_top3": 1.9545,
|
|
"hit3": 1.0,
|
|
"mrr": 0.99525,
|
|
"ndcg3": 0.963831301989089,
|
|
"mean_first_positive_rank": 1.0095,
|
|
"all_evidence_in_top3": 0.9545,
|
|
"multi_positive_episodes": 2000,
|
|
"all_evidence_in_top3_multi": 0.9545,
|
|
"hop_accuracy": 1.0,
|
|
"hop_accuracy_answerable": 1.0,
|
|
"hop_under_prediction": 0.0,
|
|
"hop_over_prediction": 0.0,
|
|
"mean_score_margin": 46.103771095309405,
|
|
"median_score_margin": 52.25391101837158,
|
|
"negative_margin_rate": 0.0095,
|
|
"hop_confusion": {
|
|
"true_2_pred_2": 2000
|
|
},
|
|
"known_mean_need_prob": 0.9999999997615814,
|
|
"unknown_mean_need_prob": null,
|
|
"thr0.50": {
|
|
"need_precision": 1.0,
|
|
"need_recall": 1.0,
|
|
"need_f1": 1.0,
|
|
"specificity_unknown_refusal": null,
|
|
"abstention_accuracy": 1.0,
|
|
"known_question_refusal_rate": 0.0,
|
|
"unknown_question_read_rate": null,
|
|
"counts": {
|
|
"tp": 2000,
|
|
"tn": 0,
|
|
"fp": 0,
|
|
"fn": 0
|
|
}
|
|
}
|
|
},
|
|
"paraphrase": {
|
|
"episodes": 2000,
|
|
"answerable_episodes": 2000,
|
|
"unknown_episodes": 0,
|
|
"top1": 0.994,
|
|
"recall1": 0.994,
|
|
"recall3": 1.0,
|
|
"recall5": 1.0,
|
|
"recall8": 1.0,
|
|
"evidence_count_in_top3": 1.0,
|
|
"hit3": 1.0,
|
|
"mrr": 0.997,
|
|
"ndcg3": 0.9977855785214288,
|
|
"mean_first_positive_rank": 1.006,
|
|
"all_evidence_in_top3": 1.0,
|
|
"multi_positive_episodes": 0,
|
|
"all_evidence_in_top3_multi": null,
|
|
"hop_accuracy": 1.0,
|
|
"hop_accuracy_answerable": 1.0,
|
|
"hop_under_prediction": 0.0,
|
|
"hop_over_prediction": 0.0,
|
|
"mean_score_margin": 36.59333083114028,
|
|
"median_score_margin": 35.444119691848755,
|
|
"negative_margin_rate": 0.006,
|
|
"hop_confusion": {
|
|
"true_1_pred_1": 2000
|
|
},
|
|
"known_mean_need_prob": 1.0,
|
|
"unknown_mean_need_prob": null,
|
|
"thr0.50": {
|
|
"need_precision": 1.0,
|
|
"need_recall": 1.0,
|
|
"need_f1": 1.0,
|
|
"specificity_unknown_refusal": null,
|
|
"abstention_accuracy": 1.0,
|
|
"known_question_refusal_rate": 0.0,
|
|
"unknown_question_read_rate": null,
|
|
"counts": {
|
|
"tp": 2000,
|
|
"tn": 0,
|
|
"fp": 0,
|
|
"fn": 0
|
|
}
|
|
}
|
|
},
|
|
"random_position": {
|
|
"episodes": 2000,
|
|
"answerable_episodes": 2000,
|
|
"unknown_episodes": 0,
|
|
"top1": 0.978,
|
|
"recall1": 0.978,
|
|
"recall3": 1.0,
|
|
"recall5": 1.0,
|
|
"recall8": 1.0,
|
|
"evidence_count_in_top3": 1.0,
|
|
"hit3": 1.0,
|
|
"mrr": 0.9889166666666667,
|
|
"ndcg3": 0.9918149897017863,
|
|
"mean_first_positive_rank": 1.0225,
|
|
"all_evidence_in_top3": 1.0,
|
|
"multi_positive_episodes": 0,
|
|
"all_evidence_in_top3_multi": null,
|
|
"hop_accuracy": 1.0,
|
|
"hop_accuracy_answerable": 1.0,
|
|
"hop_under_prediction": 0.0,
|
|
"hop_over_prediction": 0.0,
|
|
"mean_score_margin": 35.780669214252384,
|
|
"median_score_margin": 35.064393758773804,
|
|
"negative_margin_rate": 0.022,
|
|
"hop_confusion": {
|
|
"true_1_pred_1": 2000
|
|
},
|
|
"known_mean_need_prob": 1.0,
|
|
"unknown_mean_need_prob": null,
|
|
"thr0.50": {
|
|
"need_precision": 1.0,
|
|
"need_recall": 1.0,
|
|
"need_f1": 1.0,
|
|
"specificity_unknown_refusal": null,
|
|
"abstention_accuracy": 1.0,
|
|
"known_question_refusal_rate": 0.0,
|
|
"unknown_question_read_rate": null,
|
|
"counts": {
|
|
"tp": 2000,
|
|
"tn": 0,
|
|
"fp": 0,
|
|
"fn": 0
|
|
}
|
|
}
|
|
},
|
|
"single_fact": {
|
|
"episodes": 2000,
|
|
"answerable_episodes": 2000,
|
|
"unknown_episodes": 0,
|
|
"top1": 0.9705,
|
|
"recall1": 0.9705,
|
|
"recall3": 1.0,
|
|
"recall5": 1.0,
|
|
"recall8": 1.0,
|
|
"evidence_count_in_top3": 1.0,
|
|
"hit3": 1.0,
|
|
"mrr": 0.98525,
|
|
"ndcg3": 0.989112427730358,
|
|
"mean_first_positive_rank": 1.0295,
|
|
"all_evidence_in_top3": 1.0,
|
|
"multi_positive_episodes": 0,
|
|
"all_evidence_in_top3_multi": null,
|
|
"hop_accuracy": 1.0,
|
|
"hop_accuracy_answerable": 1.0,
|
|
"hop_under_prediction": 0.0,
|
|
"hop_over_prediction": 0.0,
|
|
"mean_score_margin": 47.48657381267706,
|
|
"median_score_margin": 51.48151856660843,
|
|
"negative_margin_rate": 0.0295,
|
|
"hop_confusion": {
|
|
"true_1_pred_1": 2000
|
|
},
|
|
"known_mean_need_prob": 1.0,
|
|
"unknown_mean_need_prob": null,
|
|
"thr0.50": {
|
|
"need_precision": 1.0,
|
|
"need_recall": 1.0,
|
|
"need_f1": 1.0,
|
|
"specificity_unknown_refusal": null,
|
|
"abstention_accuracy": 1.0,
|
|
"known_question_refusal_rate": 0.0,
|
|
"unknown_question_read_rate": null,
|
|
"counts": {
|
|
"tp": 2000,
|
|
"tn": 0,
|
|
"fp": 0,
|
|
"fn": 0
|
|
}
|
|
}
|
|
},
|
|
"unknown": {
|
|
"episodes": 1920,
|
|
"answerable_episodes": 1814,
|
|
"unknown_episodes": 106,
|
|
"top1": 0.5810363836824697,
|
|
"recall1": 0.5810363836824697,
|
|
"recall3": 0.679162072767365,
|
|
"recall5": 0.7232635060639471,
|
|
"recall8": 0.8103638368246968,
|
|
"evidence_count_in_top3": 0.679162072767365,
|
|
"hit3": 0.679162072767365,
|
|
"mrr": 0.662960376675071,
|
|
"ndcg3": 0.6390492224051051,
|
|
"mean_first_positive_rank": 4.181367144432194,
|
|
"all_evidence_in_top3": 0.679162072767365,
|
|
"multi_positive_episodes": 0,
|
|
"all_evidence_in_top3_multi": null,
|
|
"hop_accuracy": 1.0,
|
|
"hop_accuracy_answerable": 1.0,
|
|
"hop_under_prediction": 0.0,
|
|
"hop_over_prediction": 0.0,
|
|
"mean_score_margin": -0.32035381466873164,
|
|
"median_score_margin": 7.441610336303711,
|
|
"negative_margin_rate": 0.4189636163175303,
|
|
"hop_confusion": {
|
|
"true_0_pred_0": 106,
|
|
"true_0_pred_1": 0,
|
|
"true_1_pred_0": 0,
|
|
"true_1_pred_1": 1814
|
|
},
|
|
"known_mean_need_prob": 0.9999999848195447,
|
|
"unknown_mean_need_prob": 0.00018733273071117103,
|
|
"thr0.50": {
|
|
"need_precision": 1.0,
|
|
"need_recall": 1.0,
|
|
"need_f1": 1.0,
|
|
"specificity_unknown_refusal": 1.0,
|
|
"abstention_accuracy": 1.0,
|
|
"known_question_refusal_rate": 0.0,
|
|
"unknown_question_read_rate": 0.0,
|
|
"counts": {
|
|
"tp": 1814,
|
|
"tn": 106,
|
|
"fp": 0,
|
|
"fn": 0
|
|
}
|
|
}
|
|
},
|
|
"unknown_abstention": {
|
|
"episodes": 2000,
|
|
"answerable_episodes": 0,
|
|
"unknown_episodes": 2000,
|
|
"top1": null,
|
|
"recall1": null,
|
|
"recall3": null,
|
|
"recall5": null,
|
|
"recall8": null,
|
|
"evidence_count_in_top3": null,
|
|
"hit3": null,
|
|
"mrr": null,
|
|
"ndcg3": null,
|
|
"mean_first_positive_rank": null,
|
|
"all_evidence_in_top3": null,
|
|
"multi_positive_episodes": 0,
|
|
"all_evidence_in_top3_multi": null,
|
|
"hop_accuracy": 1.0,
|
|
"hop_accuracy_answerable": null,
|
|
"hop_under_prediction": null,
|
|
"hop_over_prediction": null,
|
|
"mean_score_margin": null,
|
|
"median_score_margin": null,
|
|
"negative_margin_rate": null,
|
|
"hop_confusion": {
|
|
"true_0_pred_0": 2000
|
|
},
|
|
"known_mean_need_prob": null,
|
|
"unknown_mean_need_prob": 3.3919047850488803e-15,
|
|
"thr0.50": {
|
|
"need_precision": null,
|
|
"need_recall": null,
|
|
"need_f1": null,
|
|
"specificity_unknown_refusal": 1.0,
|
|
"abstention_accuracy": 1.0,
|
|
"known_question_refusal_rate": null,
|
|
"unknown_question_read_rate": 0.0,
|
|
"counts": {
|
|
"tp": 0,
|
|
"tn": 2000,
|
|
"fp": 0,
|
|
"fn": 0
|
|
}
|
|
}
|
|
}
|
|
}
|
|
},
|
|
"V2-512 v6 final": {
|
|
"checkpoint": "checkpoints\\router_v6_v2_512\\memory_router_v2.pt",
|
|
"router": {
|
|
"kind": "router_v2",
|
|
"step": null,
|
|
"arch_config": {
|
|
"arch": "router_v2",
|
|
"hidden_size": 2560,
|
|
"router_dim": 512,
|
|
"num_heads": 8,
|
|
"max_hops": 3
|
|
},
|
|
"parameters": 4741902,
|
|
"checkpoint_bytes": 18973433
|
|
},
|
|
"storage": {
|
|
"address_bytes_per_record": 2048,
|
|
"address_mb_per_1m_records": 2048,
|
|
"checkpoint_bytes": 18973433
|
|
},
|
|
"latency": {
|
|
"cuda": {
|
|
"single_query_latency_ms_mean": 0.9488446664181538,
|
|
"single_query_latency_ms_p50": 0.8728999964660034,
|
|
"single_query_latency_ms_p95": 1.1364000019966625,
|
|
"queries_per_second": 1053.9132856961248,
|
|
"samples": 150,
|
|
"batched_qps_64": 71586.76538686478,
|
|
"batched_latency_ms_64": 0.01396906250192842,
|
|
"batched_qps_256": 177958.2077420389,
|
|
"batched_latency_ms_256": 0.005619296871373081
|
|
},
|
|
"cpu": {
|
|
"single_query_latency_ms_mean": 0.8265033330826554,
|
|
"single_query_latency_ms_p50": 0.8115000018733554,
|
|
"single_query_latency_ms_p95": 1.039499999023974,
|
|
"queries_per_second": 1209.9164757996127,
|
|
"samples": 150,
|
|
"batched_qps_64": 3171.5826196153826,
|
|
"batched_latency_ms_64": 0.3153000000111206,
|
|
"batched_qps_256": 3134.2351278810493,
|
|
"batched_latency_ms_256": 0.3190571093739436
|
|
}
|
|
},
|
|
"metrics": {
|
|
"episodes": 21920,
|
|
"answerable_episodes": 17814,
|
|
"unknown_episodes": 4106,
|
|
"top1": 0.941843493881217,
|
|
"recall1": 0.8864657011339396,
|
|
"recall3": 0.9666273717300999,
|
|
"recall5": 0.9764230380599529,
|
|
"recall8": 0.984506567867969,
|
|
"evidence_count_in_top3": 1.07668126192882,
|
|
"hit3": 0.9688447288649377,
|
|
"mrr": 0.958724954839147,
|
|
"ndcg3": 0.9548399733910118,
|
|
"mean_first_positive_rank": 1.2825867295385651,
|
|
"all_evidence_in_top3": 0.9644100145952621,
|
|
"multi_positive_episodes": 2000,
|
|
"all_evidence_in_top3_multi": 0.9605,
|
|
"hop_accuracy": 1.0,
|
|
"hop_accuracy_answerable": 1.0,
|
|
"hop_under_prediction": 0.0,
|
|
"hop_over_prediction": 0.0,
|
|
"mean_score_margin": 73.24820388849786,
|
|
"median_score_margin": 76.59281730651855,
|
|
"negative_margin_rate": 0.05815650611878298,
|
|
"hop_confusion": {
|
|
"true_0_pred_0": 4106,
|
|
"true_0_pred_1": 0,
|
|
"true_0_pred_2": 0,
|
|
"true_1_pred_0": 0,
|
|
"true_1_pred_1": 15814,
|
|
"true_1_pred_2": 0,
|
|
"true_2_pred_0": 0,
|
|
"true_2_pred_1": 0,
|
|
"true_2_pred_2": 2000
|
|
},
|
|
"known_mean_need_prob": 0.9999999984274064,
|
|
"unknown_mean_need_prob": 4.522551017238816e-06,
|
|
"thr0.30": {
|
|
"need_precision": 1.0,
|
|
"need_recall": 1.0,
|
|
"need_f1": 1.0,
|
|
"specificity_unknown_refusal": 1.0,
|
|
"abstention_accuracy": 1.0,
|
|
"known_question_refusal_rate": 0.0,
|
|
"unknown_question_read_rate": 0.0,
|
|
"counts": {
|
|
"tp": 17814,
|
|
"tn": 4106,
|
|
"fp": 0,
|
|
"fn": 0
|
|
}
|
|
},
|
|
"thr0.40": {
|
|
"need_precision": 1.0,
|
|
"need_recall": 1.0,
|
|
"need_f1": 1.0,
|
|
"specificity_unknown_refusal": 1.0,
|
|
"abstention_accuracy": 1.0,
|
|
"known_question_refusal_rate": 0.0,
|
|
"unknown_question_read_rate": 0.0,
|
|
"counts": {
|
|
"tp": 17814,
|
|
"tn": 4106,
|
|
"fp": 0,
|
|
"fn": 0
|
|
}
|
|
},
|
|
"thr0.50": {
|
|
"need_precision": 1.0,
|
|
"need_recall": 1.0,
|
|
"need_f1": 1.0,
|
|
"specificity_unknown_refusal": 1.0,
|
|
"abstention_accuracy": 1.0,
|
|
"known_question_refusal_rate": 0.0,
|
|
"unknown_question_read_rate": 0.0,
|
|
"counts": {
|
|
"tp": 17814,
|
|
"tn": 4106,
|
|
"fp": 0,
|
|
"fn": 0
|
|
}
|
|
},
|
|
"thr0.60": {
|
|
"need_precision": 1.0,
|
|
"need_recall": 1.0,
|
|
"need_f1": 1.0,
|
|
"specificity_unknown_refusal": 1.0,
|
|
"abstention_accuracy": 1.0,
|
|
"known_question_refusal_rate": 0.0,
|
|
"unknown_question_read_rate": 0.0,
|
|
"counts": {
|
|
"tp": 17814,
|
|
"tn": 4106,
|
|
"fp": 0,
|
|
"fn": 0
|
|
}
|
|
},
|
|
"thr0.70": {
|
|
"need_precision": 1.0,
|
|
"need_recall": 1.0,
|
|
"need_f1": 1.0,
|
|
"specificity_unknown_refusal": 1.0,
|
|
"abstention_accuracy": 1.0,
|
|
"known_question_refusal_rate": 0.0,
|
|
"unknown_question_read_rate": 0.0,
|
|
"counts": {
|
|
"tp": 17814,
|
|
"tn": 4106,
|
|
"fp": 0,
|
|
"fn": 0
|
|
}
|
|
},
|
|
"thr0.80": {
|
|
"need_precision": 1.0,
|
|
"need_recall": 1.0,
|
|
"need_f1": 1.0,
|
|
"specificity_unknown_refusal": 1.0,
|
|
"abstention_accuracy": 1.0,
|
|
"known_question_refusal_rate": 0.0,
|
|
"unknown_question_read_rate": 0.0,
|
|
"counts": {
|
|
"tp": 17814,
|
|
"tn": 4106,
|
|
"fp": 0,
|
|
"fn": 0
|
|
}
|
|
}
|
|
},
|
|
"by_family": {
|
|
"benchmark_qa": {
|
|
"episodes": 64,
|
|
"answerable_episodes": 0,
|
|
"unknown_episodes": 64,
|
|
"top1": null,
|
|
"recall1": null,
|
|
"recall3": null,
|
|
"recall5": null,
|
|
"recall8": null,
|
|
"evidence_count_in_top3": null,
|
|
"hit3": null,
|
|
"mrr": null,
|
|
"ndcg3": null,
|
|
"mean_first_positive_rank": null,
|
|
"all_evidence_in_top3": null,
|
|
"multi_positive_episodes": 0,
|
|
"all_evidence_in_top3_multi": null,
|
|
"hop_accuracy": 1.0,
|
|
"hop_accuracy_answerable": null,
|
|
"hop_under_prediction": null,
|
|
"hop_over_prediction": null,
|
|
"mean_score_margin": null,
|
|
"median_score_margin": null,
|
|
"negative_margin_rate": null,
|
|
"hop_confusion": {
|
|
"true_0_pred_0": 64
|
|
},
|
|
"known_mean_need_prob": null,
|
|
"unknown_mean_need_prob": 4.432710963581804e-06,
|
|
"thr0.50": {
|
|
"need_precision": null,
|
|
"need_recall": null,
|
|
"need_f1": null,
|
|
"specificity_unknown_refusal": 1.0,
|
|
"abstention_accuracy": 1.0,
|
|
"known_question_refusal_rate": null,
|
|
"unknown_question_read_rate": 0.0,
|
|
"counts": {
|
|
"tp": 0,
|
|
"tn": 64,
|
|
"fp": 0,
|
|
"fn": 0
|
|
}
|
|
}
|
|
},
|
|
"mega_validation": {
|
|
"episodes": 20000,
|
|
"answerable_episodes": 16000,
|
|
"unknown_episodes": 4000,
|
|
"top1": 0.98375,
|
|
"recall1": 0.92209375,
|
|
"recall3": 0.99753125,
|
|
"recall5": 0.9999375,
|
|
"recall8": 1.0,
|
|
"evidence_count_in_top3": 1.1200625,
|
|
"hit3": 1.0,
|
|
"mrr": 0.99184375,
|
|
"ndcg3": 0.9902417180168516,
|
|
"mean_first_positive_rank": 1.0164375,
|
|
"all_evidence_in_top3": 0.9950625,
|
|
"multi_positive_episodes": 2000,
|
|
"all_evidence_in_top3_multi": 0.9605,
|
|
"hop_accuracy": 1.0,
|
|
"hop_accuracy_answerable": 1.0,
|
|
"hop_under_prediction": 0.0,
|
|
"hop_over_prediction": 0.0,
|
|
"mean_score_margin": 81.85253369444608,
|
|
"median_score_margin": 83.23931503295898,
|
|
"negative_margin_rate": 0.01625,
|
|
"hop_confusion": {
|
|
"true_0_pred_0": 4000,
|
|
"true_0_pred_1": 0,
|
|
"true_0_pred_2": 0,
|
|
"true_1_pred_0": 0,
|
|
"true_1_pred_1": 14000,
|
|
"true_1_pred_2": 0,
|
|
"true_2_pred_0": 0,
|
|
"true_2_pred_1": 0,
|
|
"true_2_pred_2": 2000
|
|
},
|
|
"known_mean_need_prob": 0.9999999999701977,
|
|
"unknown_mean_need_prob": 6.058075942262249e-11,
|
|
"thr0.50": {
|
|
"need_precision": 1.0,
|
|
"need_recall": 1.0,
|
|
"need_f1": 1.0,
|
|
"specificity_unknown_refusal": 1.0,
|
|
"abstention_accuracy": 1.0,
|
|
"known_question_refusal_rate": 0.0,
|
|
"unknown_question_read_rate": 0.0,
|
|
"counts": {
|
|
"tp": 16000,
|
|
"tn": 4000,
|
|
"fp": 0,
|
|
"fn": 0
|
|
}
|
|
}
|
|
},
|
|
"memory_policy": {
|
|
"episodes": 1728,
|
|
"answerable_episodes": 1707,
|
|
"unknown_episodes": 21,
|
|
"top1": 0.5770357352079672,
|
|
"recall1": 0.5770357352079672,
|
|
"recall3": 0.6859988283538372,
|
|
"recall5": 0.7562975981253661,
|
|
"recall8": 0.8388986526069128,
|
|
"evidence_count_in_top3": 0.6859988283538372,
|
|
"hit3": 0.6859988283538372,
|
|
"mrr": 0.6663638286240626,
|
|
"ndcg3": 0.6401845706816489,
|
|
"mean_first_positive_rank": 3.719976567076743,
|
|
"all_evidence_in_top3": 0.6859988283538372,
|
|
"multi_positive_episodes": 0,
|
|
"all_evidence_in_top3_multi": null,
|
|
"hop_accuracy": 1.0,
|
|
"hop_accuracy_answerable": 1.0,
|
|
"hop_under_prediction": 0.0,
|
|
"hop_over_prediction": 0.0,
|
|
"mean_score_margin": -2.648404171182946,
|
|
"median_score_margin": 5.544002532958984,
|
|
"negative_margin_rate": 0.4229642647920328,
|
|
"hop_confusion": {
|
|
"true_0_pred_0": 21,
|
|
"true_0_pred_1": 0,
|
|
"true_1_pred_0": 0,
|
|
"true_1_pred_1": 1707
|
|
},
|
|
"known_mean_need_prob": 0.9999999838679872,
|
|
"unknown_mean_need_prob": 0.00043537282504942034,
|
|
"thr0.50": {
|
|
"need_precision": 1.0,
|
|
"need_recall": 1.0,
|
|
"need_f1": 1.0,
|
|
"specificity_unknown_refusal": 1.0,
|
|
"abstention_accuracy": 1.0,
|
|
"known_question_refusal_rate": 0.0,
|
|
"unknown_question_read_rate": 0.0,
|
|
"counts": {
|
|
"tp": 1707,
|
|
"tn": 21,
|
|
"fp": 0,
|
|
"fn": 0
|
|
}
|
|
}
|
|
},
|
|
"native_memory": {
|
|
"episodes": 128,
|
|
"answerable_episodes": 107,
|
|
"unknown_episodes": 21,
|
|
"top1": 0.4953271028037383,
|
|
"recall1": 0.4953271028037383,
|
|
"recall3": 0.822429906542056,
|
|
"recall5": 0.9719626168224299,
|
|
"recall8": 0.9906542056074766,
|
|
"evidence_count_in_top3": 0.822429906542056,
|
|
"hit3": 0.822429906542056,
|
|
"mrr": 0.6704980377877574,
|
|
"ndcg3": 0.680904070694264,
|
|
"mean_first_positive_rank": 2.196261682242991,
|
|
"all_evidence_in_top3": 0.822429906542056,
|
|
"multi_positive_episodes": 0,
|
|
"all_evidence_in_top3_multi": null,
|
|
"hop_accuracy": 1.0,
|
|
"hop_accuracy_answerable": 1.0,
|
|
"hop_under_prediction": 0.0,
|
|
"hop_over_prediction": 0.0,
|
|
"mean_score_margin": -2.581393656329574,
|
|
"median_score_margin": -0.056890010833740234,
|
|
"negative_margin_rate": 0.5046728971962616,
|
|
"hop_confusion": {
|
|
"true_0_pred_0": 21,
|
|
"true_0_pred_1": 0,
|
|
"true_1_pred_0": 0,
|
|
"true_1_pred_1": 107
|
|
},
|
|
"known_mean_need_prob": 1.0,
|
|
"unknown_mean_need_prob": 0.00043537282504942034,
|
|
"thr0.50": {
|
|
"need_precision": 1.0,
|
|
"need_recall": 1.0,
|
|
"need_f1": 1.0,
|
|
"specificity_unknown_refusal": 1.0,
|
|
"abstention_accuracy": 1.0,
|
|
"known_question_refusal_rate": 0.0,
|
|
"unknown_question_read_rate": 0.0,
|
|
"counts": {
|
|
"tp": 107,
|
|
"tn": 21,
|
|
"fp": 0,
|
|
"fn": 0
|
|
}
|
|
}
|
|
}
|
|
},
|
|
"by_category": {
|
|
"conflict_update": {
|
|
"episodes": 2000,
|
|
"answerable_episodes": 2000,
|
|
"unknown_episodes": 0,
|
|
"top1": 0.9935,
|
|
"recall1": 0.9935,
|
|
"recall3": 1.0,
|
|
"recall5": 1.0,
|
|
"recall8": 1.0,
|
|
"evidence_count_in_top3": 1.0,
|
|
"hit3": 1.0,
|
|
"mrr": 0.99675,
|
|
"ndcg3": 0.9976010433982145,
|
|
"mean_first_positive_rank": 1.0065,
|
|
"all_evidence_in_top3": 1.0,
|
|
"multi_positive_episodes": 0,
|
|
"all_evidence_in_top3_multi": null,
|
|
"hop_accuracy": 1.0,
|
|
"hop_accuracy_answerable": 1.0,
|
|
"hop_under_prediction": 0.0,
|
|
"hop_over_prediction": 0.0,
|
|
"mean_score_margin": 64.64610423742235,
|
|
"median_score_margin": 60.52951920032501,
|
|
"negative_margin_rate": 0.0065,
|
|
"hop_confusion": {
|
|
"true_1_pred_1": 2000
|
|
},
|
|
"known_mean_need_prob": 1.0,
|
|
"unknown_mean_need_prob": null,
|
|
"thr0.50": {
|
|
"need_precision": 1.0,
|
|
"need_recall": 1.0,
|
|
"need_f1": 1.0,
|
|
"specificity_unknown_refusal": null,
|
|
"abstention_accuracy": 1.0,
|
|
"known_question_refusal_rate": 0.0,
|
|
"unknown_question_read_rate": null,
|
|
"counts": {
|
|
"tp": 2000,
|
|
"tn": 0,
|
|
"fp": 0,
|
|
"fn": 0
|
|
}
|
|
}
|
|
},
|
|
"distractor_128": {
|
|
"episodes": 2000,
|
|
"answerable_episodes": 2000,
|
|
"unknown_episodes": 0,
|
|
"top1": 0.98,
|
|
"recall1": 0.98,
|
|
"recall3": 1.0,
|
|
"recall5": 1.0,
|
|
"recall8": 1.0,
|
|
"evidence_count_in_top3": 1.0,
|
|
"hit3": 1.0,
|
|
"mrr": 0.99,
|
|
"ndcg3": 0.9926185950714291,
|
|
"mean_first_positive_rank": 1.02,
|
|
"all_evidence_in_top3": 1.0,
|
|
"multi_positive_episodes": 0,
|
|
"all_evidence_in_top3_multi": null,
|
|
"hop_accuracy": 1.0,
|
|
"hop_accuracy_answerable": 1.0,
|
|
"hop_under_prediction": 0.0,
|
|
"hop_over_prediction": 0.0,
|
|
"mean_score_margin": 88.94486605204642,
|
|
"median_score_margin": 95.25923618674278,
|
|
"negative_margin_rate": 0.02,
|
|
"hop_confusion": {
|
|
"true_1_pred_1": 2000
|
|
},
|
|
"known_mean_need_prob": 1.0,
|
|
"unknown_mean_need_prob": null,
|
|
"thr0.50": {
|
|
"need_precision": 1.0,
|
|
"need_recall": 1.0,
|
|
"need_f1": 1.0,
|
|
"specificity_unknown_refusal": null,
|
|
"abstention_accuracy": 1.0,
|
|
"known_question_refusal_rate": 0.0,
|
|
"unknown_question_read_rate": null,
|
|
"counts": {
|
|
"tp": 2000,
|
|
"tn": 0,
|
|
"fp": 0,
|
|
"fn": 0
|
|
}
|
|
}
|
|
},
|
|
"distractor_32": {
|
|
"episodes": 2000,
|
|
"answerable_episodes": 2000,
|
|
"unknown_episodes": 0,
|
|
"top1": 0.9805,
|
|
"recall1": 0.9805,
|
|
"recall3": 1.0,
|
|
"recall5": 1.0,
|
|
"recall8": 1.0,
|
|
"evidence_count_in_top3": 1.0,
|
|
"hit3": 1.0,
|
|
"mrr": 0.9901666666666666,
|
|
"ndcg3": 0.9927376653178577,
|
|
"mean_first_positive_rank": 1.02,
|
|
"all_evidence_in_top3": 1.0,
|
|
"multi_positive_episodes": 0,
|
|
"all_evidence_in_top3_multi": null,
|
|
"hop_accuracy": 1.0,
|
|
"hop_accuracy_answerable": 1.0,
|
|
"hop_under_prediction": 0.0,
|
|
"hop_over_prediction": 0.0,
|
|
"mean_score_margin": 90.02319332621246,
|
|
"median_score_margin": 96.11362898349762,
|
|
"negative_margin_rate": 0.0195,
|
|
"hop_confusion": {
|
|
"true_1_pred_1": 2000
|
|
},
|
|
"known_mean_need_prob": 1.0,
|
|
"unknown_mean_need_prob": null,
|
|
"thr0.50": {
|
|
"need_precision": 1.0,
|
|
"need_recall": 1.0,
|
|
"need_f1": 1.0,
|
|
"specificity_unknown_refusal": null,
|
|
"abstention_accuracy": 1.0,
|
|
"known_question_refusal_rate": 0.0,
|
|
"unknown_question_read_rate": null,
|
|
"counts": {
|
|
"tp": 2000,
|
|
"tn": 0,
|
|
"fp": 0,
|
|
"fn": 0
|
|
}
|
|
}
|
|
},
|
|
"forget_correction": {
|
|
"episodes": 2000,
|
|
"answerable_episodes": 0,
|
|
"unknown_episodes": 2000,
|
|
"top1": null,
|
|
"recall1": null,
|
|
"recall3": null,
|
|
"recall5": null,
|
|
"recall8": null,
|
|
"evidence_count_in_top3": null,
|
|
"hit3": null,
|
|
"mrr": null,
|
|
"ndcg3": null,
|
|
"mean_first_positive_rank": null,
|
|
"all_evidence_in_top3": null,
|
|
"multi_positive_episodes": 0,
|
|
"all_evidence_in_top3_multi": null,
|
|
"hop_accuracy": 1.0,
|
|
"hop_accuracy_answerable": null,
|
|
"hop_under_prediction": null,
|
|
"hop_over_prediction": null,
|
|
"mean_score_margin": null,
|
|
"median_score_margin": null,
|
|
"negative_margin_rate": null,
|
|
"hop_confusion": {
|
|
"true_0_pred_0": 2000
|
|
},
|
|
"known_mean_need_prob": null,
|
|
"unknown_mean_need_prob": 1.2115745555928435e-10,
|
|
"thr0.50": {
|
|
"need_precision": null,
|
|
"need_recall": null,
|
|
"need_f1": null,
|
|
"specificity_unknown_refusal": 1.0,
|
|
"abstention_accuracy": 1.0,
|
|
"known_question_refusal_rate": null,
|
|
"unknown_question_read_rate": 0.0,
|
|
"counts": {
|
|
"tp": 0,
|
|
"tn": 2000,
|
|
"fp": 0,
|
|
"fn": 0
|
|
}
|
|
}
|
|
},
|
|
"long_context": {
|
|
"episodes": 2000,
|
|
"answerable_episodes": 2000,
|
|
"unknown_episodes": 0,
|
|
"top1": 0.977,
|
|
"recall1": 0.977,
|
|
"recall3": 1.0,
|
|
"recall5": 1.0,
|
|
"recall8": 1.0,
|
|
"evidence_count_in_top3": 1.0,
|
|
"hit3": 1.0,
|
|
"mrr": 0.9884166666666666,
|
|
"ndcg3": 0.9914459194553578,
|
|
"mean_first_positive_rank": 1.0235,
|
|
"all_evidence_in_top3": 1.0,
|
|
"multi_positive_episodes": 0,
|
|
"all_evidence_in_top3_multi": null,
|
|
"hop_accuracy": 1.0,
|
|
"hop_accuracy_answerable": 1.0,
|
|
"hop_under_prediction": 0.0,
|
|
"hop_over_prediction": 0.0,
|
|
"mean_score_margin": 78.39938361007721,
|
|
"median_score_margin": 81.58803009986877,
|
|
"negative_margin_rate": 0.023,
|
|
"hop_confusion": {
|
|
"true_1_pred_1": 2000
|
|
},
|
|
"known_mean_need_prob": 1.0,
|
|
"unknown_mean_need_prob": null,
|
|
"thr0.50": {
|
|
"need_precision": 1.0,
|
|
"need_recall": 1.0,
|
|
"need_f1": 1.0,
|
|
"specificity_unknown_refusal": null,
|
|
"abstention_accuracy": 1.0,
|
|
"known_question_refusal_rate": 0.0,
|
|
"unknown_question_read_rate": null,
|
|
"counts": {
|
|
"tp": 2000,
|
|
"tn": 0,
|
|
"fp": 0,
|
|
"fn": 0
|
|
}
|
|
}
|
|
},
|
|
"multi_hop": {
|
|
"episodes": 2000,
|
|
"answerable_episodes": 2000,
|
|
"unknown_episodes": 0,
|
|
"top1": 0.9865,
|
|
"recall1": 0.49325,
|
|
"recall3": 0.98025,
|
|
"recall5": 0.9995,
|
|
"recall8": 1.0,
|
|
"evidence_count_in_top3": 1.9605,
|
|
"hit3": 1.0,
|
|
"mrr": 0.99325,
|
|
"ndcg3": 0.9651268224740949,
|
|
"mean_first_positive_rank": 1.0135,
|
|
"all_evidence_in_top3": 0.9605,
|
|
"multi_positive_episodes": 2000,
|
|
"all_evidence_in_top3_multi": 0.9605,
|
|
"hop_accuracy": 1.0,
|
|
"hop_accuracy_answerable": 1.0,
|
|
"hop_under_prediction": 0.0,
|
|
"hop_over_prediction": 0.0,
|
|
"mean_score_margin": 104.36254599650204,
|
|
"median_score_margin": 118.10550880432129,
|
|
"negative_margin_rate": 0.0135,
|
|
"hop_confusion": {
|
|
"true_2_pred_2": 2000
|
|
},
|
|
"known_mean_need_prob": 0.9999999997615814,
|
|
"unknown_mean_need_prob": null,
|
|
"thr0.50": {
|
|
"need_precision": 1.0,
|
|
"need_recall": 1.0,
|
|
"need_f1": 1.0,
|
|
"specificity_unknown_refusal": null,
|
|
"abstention_accuracy": 1.0,
|
|
"known_question_refusal_rate": 0.0,
|
|
"unknown_question_read_rate": null,
|
|
"counts": {
|
|
"tp": 2000,
|
|
"tn": 0,
|
|
"fp": 0,
|
|
"fn": 0
|
|
}
|
|
}
|
|
},
|
|
"paraphrase": {
|
|
"episodes": 2000,
|
|
"answerable_episodes": 2000,
|
|
"unknown_episodes": 0,
|
|
"top1": 0.9955,
|
|
"recall1": 0.9955,
|
|
"recall3": 1.0,
|
|
"recall5": 1.0,
|
|
"recall8": 1.0,
|
|
"evidence_count_in_top3": 1.0,
|
|
"hit3": 1.0,
|
|
"mrr": 0.99775,
|
|
"ndcg3": 0.9983391838910716,
|
|
"mean_first_positive_rank": 1.0045,
|
|
"all_evidence_in_top3": 1.0,
|
|
"multi_positive_episodes": 0,
|
|
"all_evidence_in_top3_multi": null,
|
|
"hop_accuracy": 1.0,
|
|
"hop_accuracy_answerable": 1.0,
|
|
"hop_under_prediction": 0.0,
|
|
"hop_over_prediction": 0.0,
|
|
"mean_score_margin": 57.61693591082841,
|
|
"median_score_margin": 57.403422355651855,
|
|
"negative_margin_rate": 0.0045,
|
|
"hop_confusion": {
|
|
"true_1_pred_1": 2000
|
|
},
|
|
"known_mean_need_prob": 1.0,
|
|
"unknown_mean_need_prob": null,
|
|
"thr0.50": {
|
|
"need_precision": 1.0,
|
|
"need_recall": 1.0,
|
|
"need_f1": 1.0,
|
|
"specificity_unknown_refusal": null,
|
|
"abstention_accuracy": 1.0,
|
|
"known_question_refusal_rate": 0.0,
|
|
"unknown_question_read_rate": null,
|
|
"counts": {
|
|
"tp": 2000,
|
|
"tn": 0,
|
|
"fp": 0,
|
|
"fn": 0
|
|
}
|
|
}
|
|
},
|
|
"random_position": {
|
|
"episodes": 2000,
|
|
"answerable_episodes": 2000,
|
|
"unknown_episodes": 0,
|
|
"top1": 0.9825,
|
|
"recall1": 0.9825,
|
|
"recall3": 1.0,
|
|
"recall5": 1.0,
|
|
"recall8": 1.0,
|
|
"evidence_count_in_top3": 1.0,
|
|
"hit3": 1.0,
|
|
"mrr": 0.99125,
|
|
"ndcg3": 0.9935412706875005,
|
|
"mean_first_positive_rank": 1.0175,
|
|
"all_evidence_in_top3": 1.0,
|
|
"multi_positive_episodes": 0,
|
|
"all_evidence_in_top3_multi": null,
|
|
"hop_accuracy": 1.0,
|
|
"hop_accuracy_answerable": 1.0,
|
|
"hop_under_prediction": 0.0,
|
|
"hop_over_prediction": 0.0,
|
|
"mean_score_margin": 84.6615379076004,
|
|
"median_score_margin": 90.02162456512451,
|
|
"negative_margin_rate": 0.0175,
|
|
"hop_confusion": {
|
|
"true_1_pred_1": 2000
|
|
},
|
|
"known_mean_need_prob": 1.0,
|
|
"unknown_mean_need_prob": null,
|
|
"thr0.50": {
|
|
"need_precision": 1.0,
|
|
"need_recall": 1.0,
|
|
"need_f1": 1.0,
|
|
"specificity_unknown_refusal": null,
|
|
"abstention_accuracy": 1.0,
|
|
"known_question_refusal_rate": 0.0,
|
|
"unknown_question_read_rate": null,
|
|
"counts": {
|
|
"tp": 2000,
|
|
"tn": 0,
|
|
"fp": 0,
|
|
"fn": 0
|
|
}
|
|
}
|
|
},
|
|
"single_fact": {
|
|
"episodes": 2000,
|
|
"answerable_episodes": 2000,
|
|
"unknown_episodes": 0,
|
|
"top1": 0.9745,
|
|
"recall1": 0.9745,
|
|
"recall3": 1.0,
|
|
"recall5": 1.0,
|
|
"recall8": 1.0,
|
|
"evidence_count_in_top3": 1.0,
|
|
"hit3": 1.0,
|
|
"mrr": 0.9871666666666666,
|
|
"ndcg3": 0.9905232438392865,
|
|
"mean_first_positive_rank": 1.026,
|
|
"all_evidence_in_top3": 1.0,
|
|
"multi_positive_episodes": 0,
|
|
"all_evidence_in_top3_multi": null,
|
|
"hop_accuracy": 1.0,
|
|
"hop_accuracy_answerable": 1.0,
|
|
"hop_under_prediction": 0.0,
|
|
"hop_over_prediction": 0.0,
|
|
"mean_score_margin": 86.1657025148794,
|
|
"median_score_margin": 91.7922420501709,
|
|
"negative_margin_rate": 0.0255,
|
|
"hop_confusion": {
|
|
"true_1_pred_1": 2000
|
|
},
|
|
"known_mean_need_prob": 1.0,
|
|
"unknown_mean_need_prob": null,
|
|
"thr0.50": {
|
|
"need_precision": 1.0,
|
|
"need_recall": 1.0,
|
|
"need_f1": 1.0,
|
|
"specificity_unknown_refusal": null,
|
|
"abstention_accuracy": 1.0,
|
|
"known_question_refusal_rate": 0.0,
|
|
"unknown_question_read_rate": null,
|
|
"counts": {
|
|
"tp": 2000,
|
|
"tn": 0,
|
|
"fp": 0,
|
|
"fn": 0
|
|
}
|
|
}
|
|
},
|
|
"unknown": {
|
|
"episodes": 1920,
|
|
"answerable_episodes": 1814,
|
|
"unknown_episodes": 106,
|
|
"top1": 0.5722160970231532,
|
|
"recall1": 0.5722160970231532,
|
|
"recall3": 0.6940463065049615,
|
|
"recall5": 0.7690187431091511,
|
|
"recall8": 0.8478500551267916,
|
|
"evidence_count_in_top3": 0.6940463065049615,
|
|
"hit3": 0.6940463065049615,
|
|
"mrr": 0.666607687709242,
|
|
"ndcg3": 0.6425864375511913,
|
|
"mean_first_positive_rank": 3.630099228224917,
|
|
"all_evidence_in_top3": 0.6940463065049615,
|
|
"multi_positive_episodes": 0,
|
|
"all_evidence_in_top3_multi": null,
|
|
"hop_accuracy": 1.0,
|
|
"hop_accuracy_answerable": 1.0,
|
|
"hop_under_prediction": 0.0,
|
|
"hop_over_prediction": 0.0,
|
|
"mean_score_margin": -2.644451511266016,
|
|
"median_score_margin": 4.475192725658417,
|
|
"negative_margin_rate": 0.42778390297684676,
|
|
"hop_confusion": {
|
|
"true_0_pred_0": 106,
|
|
"true_0_pred_1": 0,
|
|
"true_1_pred_0": 0,
|
|
"true_1_pred_1": 1814
|
|
},
|
|
"known_mean_need_prob": 0.9999999848195447,
|
|
"unknown_mean_need_prob": 0.00017518256748815932,
|
|
"thr0.50": {
|
|
"need_precision": 1.0,
|
|
"need_recall": 1.0,
|
|
"need_f1": 1.0,
|
|
"specificity_unknown_refusal": 1.0,
|
|
"abstention_accuracy": 1.0,
|
|
"known_question_refusal_rate": 0.0,
|
|
"unknown_question_read_rate": 0.0,
|
|
"counts": {
|
|
"tp": 1814,
|
|
"tn": 106,
|
|
"fp": 0,
|
|
"fn": 0
|
|
}
|
|
}
|
|
},
|
|
"unknown_abstention": {
|
|
"episodes": 2000,
|
|
"answerable_episodes": 0,
|
|
"unknown_episodes": 2000,
|
|
"top1": null,
|
|
"recall1": null,
|
|
"recall3": null,
|
|
"recall5": null,
|
|
"recall8": null,
|
|
"evidence_count_in_top3": null,
|
|
"hit3": null,
|
|
"mrr": null,
|
|
"ndcg3": null,
|
|
"mean_first_positive_rank": null,
|
|
"all_evidence_in_top3": null,
|
|
"multi_positive_episodes": 0,
|
|
"all_evidence_in_top3_multi": null,
|
|
"hop_accuracy": 1.0,
|
|
"hop_accuracy_answerable": null,
|
|
"hop_under_prediction": null,
|
|
"hop_over_prediction": null,
|
|
"mean_score_margin": null,
|
|
"median_score_margin": null,
|
|
"negative_margin_rate": null,
|
|
"hop_confusion": {
|
|
"true_0_pred_0": 2000
|
|
},
|
|
"known_mean_need_prob": null,
|
|
"unknown_mean_need_prob": 4.063285960620704e-15,
|
|
"thr0.50": {
|
|
"need_precision": null,
|
|
"need_recall": null,
|
|
"need_f1": null,
|
|
"specificity_unknown_refusal": 1.0,
|
|
"abstention_accuracy": 1.0,
|
|
"known_question_refusal_rate": null,
|
|
"unknown_question_read_rate": 0.0,
|
|
"counts": {
|
|
"tp": 0,
|
|
"tn": 2000,
|
|
"fp": 0,
|
|
"fn": 0
|
|
}
|
|
}
|
|
}
|
|
}
|
|
},
|
|
"XL-512 v6 final": {
|
|
"checkpoint": "checkpoints\\router_v6_xl512\\memory_router_xl.pt",
|
|
"router": {
|
|
"kind": "router_xl",
|
|
"step": null,
|
|
"arch_config": {
|
|
"arch": "router_xl",
|
|
"arch_version": 1,
|
|
"hidden_size": 2560,
|
|
"router_dim": 512,
|
|
"num_heads": 8,
|
|
"max_hops": 3,
|
|
"encoder_layers": 2,
|
|
"encoder_hidden": 512,
|
|
"pair_blocks": 1,
|
|
"pair_hidden": 512,
|
|
"pair_expansion": 2,
|
|
"pair_dropout": 0.05,
|
|
"use_interaction": true,
|
|
"policy_layers": 2,
|
|
"policy_hidden": 512,
|
|
"policy_dropout": 0.0,
|
|
"learnable_cosine_scale": true
|
|
},
|
|
"parameters": 7898127,
|
|
"checkpoint_bytes": 35802155
|
|
},
|
|
"storage": {
|
|
"address_bytes_per_record": 2048,
|
|
"address_mb_per_1m_records": 2048,
|
|
"checkpoint_bytes": 35802155
|
|
},
|
|
"latency": {
|
|
"cuda": {
|
|
"single_query_latency_ms_mean": 1.450062666942055,
|
|
"single_query_latency_ms_p50": 1.4245999991544522,
|
|
"single_query_latency_ms_p95": 1.9877999948221259,
|
|
"queries_per_second": 689.6253677841637,
|
|
"samples": 150,
|
|
"batched_qps_64": 46891.9433921584,
|
|
"batched_latency_ms_64": 0.021325624993551173,
|
|
"batched_qps_256": 85734.57110724591,
|
|
"batched_latency_ms_256": 0.011663906252579181
|
|
},
|
|
"cpu": {
|
|
"single_query_latency_ms_mean": 2.099922667063462,
|
|
"single_query_latency_ms_p50": 1.9889000031980686,
|
|
"single_query_latency_ms_p95": 2.6856999975279905,
|
|
"queries_per_second": 476.2080126495338,
|
|
"samples": 150,
|
|
"batched_qps_64": 1630.631411069136,
|
|
"batched_latency_ms_64": 0.613259374995323,
|
|
"batched_qps_256": 1668.1377991342565,
|
|
"batched_latency_ms_256": 0.5994708593732412
|
|
}
|
|
},
|
|
"metrics": {
|
|
"episodes": 21920,
|
|
"answerable_episodes": 17814,
|
|
"unknown_episodes": 4106,
|
|
"top1": 0.9502077018075671,
|
|
"recall1": 0.8948299090602897,
|
|
"recall3": 0.965336252385764,
|
|
"recall5": 0.9760862243179522,
|
|
"recall8": 0.9854608734703043,
|
|
"evidence_count_in_top3": 1.0657348153137982,
|
|
"hit3": 0.9772089367912877,
|
|
"mrr": 0.9664213223997663,
|
|
"ndcg3": 0.9560053718886551,
|
|
"mean_first_positive_rank": 1.1716066015493432,
|
|
"all_evidence_in_top3": 0.9534635679802402,
|
|
"multi_positive_episodes": 2000,
|
|
"all_evidence_in_top3_multi": 0.7885,
|
|
"hop_accuracy": 1.0,
|
|
"hop_accuracy_answerable": 1.0,
|
|
"hop_under_prediction": 0.0,
|
|
"hop_over_prediction": 0.0,
|
|
"mean_score_margin": 215.75331382084113,
|
|
"median_score_margin": 239.68688774108887,
|
|
"negative_margin_rate": 0.049792298192432916,
|
|
"hop_confusion": {
|
|
"true_0_pred_0": 4106,
|
|
"true_0_pred_1": 0,
|
|
"true_0_pred_2": 0,
|
|
"true_1_pred_0": 0,
|
|
"true_1_pred_1": 15814,
|
|
"true_1_pred_2": 0,
|
|
"true_2_pred_0": 0,
|
|
"true_2_pred_1": 0,
|
|
"true_2_pred_2": 2000
|
|
},
|
|
"known_mean_need_prob": 0.9999999979790499,
|
|
"unknown_mean_need_prob": 4.1244088529706584e-06,
|
|
"thr0.30": {
|
|
"need_precision": 1.0,
|
|
"need_recall": 1.0,
|
|
"need_f1": 1.0,
|
|
"specificity_unknown_refusal": 1.0,
|
|
"abstention_accuracy": 1.0,
|
|
"known_question_refusal_rate": 0.0,
|
|
"unknown_question_read_rate": 0.0,
|
|
"counts": {
|
|
"tp": 17814,
|
|
"tn": 4106,
|
|
"fp": 0,
|
|
"fn": 0
|
|
}
|
|
},
|
|
"thr0.40": {
|
|
"need_precision": 1.0,
|
|
"need_recall": 1.0,
|
|
"need_f1": 1.0,
|
|
"specificity_unknown_refusal": 1.0,
|
|
"abstention_accuracy": 1.0,
|
|
"known_question_refusal_rate": 0.0,
|
|
"unknown_question_read_rate": 0.0,
|
|
"counts": {
|
|
"tp": 17814,
|
|
"tn": 4106,
|
|
"fp": 0,
|
|
"fn": 0
|
|
}
|
|
},
|
|
"thr0.50": {
|
|
"need_precision": 1.0,
|
|
"need_recall": 1.0,
|
|
"need_f1": 1.0,
|
|
"specificity_unknown_refusal": 1.0,
|
|
"abstention_accuracy": 1.0,
|
|
"known_question_refusal_rate": 0.0,
|
|
"unknown_question_read_rate": 0.0,
|
|
"counts": {
|
|
"tp": 17814,
|
|
"tn": 4106,
|
|
"fp": 0,
|
|
"fn": 0
|
|
}
|
|
},
|
|
"thr0.60": {
|
|
"need_precision": 1.0,
|
|
"need_recall": 1.0,
|
|
"need_f1": 1.0,
|
|
"specificity_unknown_refusal": 1.0,
|
|
"abstention_accuracy": 1.0,
|
|
"known_question_refusal_rate": 0.0,
|
|
"unknown_question_read_rate": 0.0,
|
|
"counts": {
|
|
"tp": 17814,
|
|
"tn": 4106,
|
|
"fp": 0,
|
|
"fn": 0
|
|
}
|
|
},
|
|
"thr0.70": {
|
|
"need_precision": 1.0,
|
|
"need_recall": 1.0,
|
|
"need_f1": 1.0,
|
|
"specificity_unknown_refusal": 1.0,
|
|
"abstention_accuracy": 1.0,
|
|
"known_question_refusal_rate": 0.0,
|
|
"unknown_question_read_rate": 0.0,
|
|
"counts": {
|
|
"tp": 17814,
|
|
"tn": 4106,
|
|
"fp": 0,
|
|
"fn": 0
|
|
}
|
|
},
|
|
"thr0.80": {
|
|
"need_precision": 1.0,
|
|
"need_recall": 1.0,
|
|
"need_f1": 1.0,
|
|
"specificity_unknown_refusal": 1.0,
|
|
"abstention_accuracy": 1.0,
|
|
"known_question_refusal_rate": 0.0,
|
|
"unknown_question_read_rate": 0.0,
|
|
"counts": {
|
|
"tp": 17814,
|
|
"tn": 4106,
|
|
"fp": 0,
|
|
"fn": 0
|
|
}
|
|
}
|
|
},
|
|
"by_family": {
|
|
"benchmark_qa": {
|
|
"episodes": 64,
|
|
"answerable_episodes": 0,
|
|
"unknown_episodes": 64,
|
|
"top1": null,
|
|
"recall1": null,
|
|
"recall3": null,
|
|
"recall5": null,
|
|
"recall8": null,
|
|
"evidence_count_in_top3": null,
|
|
"hit3": null,
|
|
"mrr": null,
|
|
"ndcg3": null,
|
|
"mean_first_positive_rank": null,
|
|
"all_evidence_in_top3": null,
|
|
"multi_positive_episodes": 0,
|
|
"all_evidence_in_top3_multi": null,
|
|
"hop_accuracy": 1.0,
|
|
"hop_accuracy_answerable": null,
|
|
"hop_under_prediction": null,
|
|
"hop_over_prediction": null,
|
|
"mean_score_margin": null,
|
|
"median_score_margin": null,
|
|
"negative_margin_rate": null,
|
|
"hop_confusion": {
|
|
"true_0_pred_0": 64
|
|
},
|
|
"known_mean_need_prob": null,
|
|
"unknown_mean_need_prob": 4.264552800687521e-06,
|
|
"thr0.50": {
|
|
"need_precision": null,
|
|
"need_recall": null,
|
|
"need_f1": null,
|
|
"specificity_unknown_refusal": 1.0,
|
|
"abstention_accuracy": 1.0,
|
|
"known_question_refusal_rate": null,
|
|
"unknown_question_read_rate": 0.0,
|
|
"counts": {
|
|
"tp": 0,
|
|
"tn": 64,
|
|
"fp": 0,
|
|
"fn": 0
|
|
}
|
|
}
|
|
},
|
|
"mega_validation": {
|
|
"episodes": 20000,
|
|
"answerable_episodes": 16000,
|
|
"unknown_episodes": 4000,
|
|
"top1": 0.984,
|
|
"recall1": 0.92234375,
|
|
"recall3": 0.98678125,
|
|
"recall5": 0.990875,
|
|
"recall8": 0.9933125,
|
|
"evidence_count_in_top3": 1.0985625,
|
|
"hit3": 1.0,
|
|
"mrr": 0.99196875,
|
|
"ndcg3": 0.9823026449115756,
|
|
"mean_first_positive_rank": 1.0161875,
|
|
"all_evidence_in_top3": 0.9735625,
|
|
"multi_positive_episodes": 2000,
|
|
"all_evidence_in_top3_multi": 0.7885,
|
|
"hop_accuracy": 1.0,
|
|
"hop_accuracy_answerable": 1.0,
|
|
"hop_under_prediction": 0.0,
|
|
"hop_over_prediction": 0.0,
|
|
"mean_score_margin": 235.8910635419197,
|
|
"median_score_margin": 250.7837634086609,
|
|
"negative_margin_rate": 0.016,
|
|
"hop_confusion": {
|
|
"true_0_pred_0": 4000,
|
|
"true_0_pred_1": 0,
|
|
"true_0_pred_2": 0,
|
|
"true_1_pred_0": 0,
|
|
"true_1_pred_1": 14000,
|
|
"true_1_pred_2": 0,
|
|
"true_2_pred_0": 0,
|
|
"true_2_pred_1": 0,
|
|
"true_2_pred_2": 2000
|
|
},
|
|
"known_mean_need_prob": 0.9999999999627471,
|
|
"unknown_mean_need_prob": 7.011441629130955e-11,
|
|
"thr0.50": {
|
|
"need_precision": 1.0,
|
|
"need_recall": 1.0,
|
|
"need_f1": 1.0,
|
|
"specificity_unknown_refusal": 1.0,
|
|
"abstention_accuracy": 1.0,
|
|
"known_question_refusal_rate": 0.0,
|
|
"unknown_question_read_rate": 0.0,
|
|
"counts": {
|
|
"tp": 16000,
|
|
"tn": 4000,
|
|
"fp": 0,
|
|
"fn": 0
|
|
}
|
|
}
|
|
},
|
|
"memory_policy": {
|
|
"episodes": 1728,
|
|
"answerable_episodes": 1707,
|
|
"unknown_episodes": 21,
|
|
"top1": 0.6690099589923842,
|
|
"recall1": 0.6690099589923842,
|
|
"recall3": 0.7709431751611013,
|
|
"recall5": 0.8388986526069128,
|
|
"recall8": 0.9121265377855887,
|
|
"evidence_count_in_top3": 0.7709431751611013,
|
|
"hit3": 0.7709431751611013,
|
|
"mrr": 0.7488416468644941,
|
|
"ndcg3": 0.7281836459473614,
|
|
"mean_first_positive_rank": 2.558289396602226,
|
|
"all_evidence_in_top3": 0.7709431751611013,
|
|
"multi_positive_episodes": 0,
|
|
"all_evidence_in_top3_multi": null,
|
|
"hop_accuracy": 1.0,
|
|
"hop_accuracy_answerable": 1.0,
|
|
"hop_under_prediction": 0.0,
|
|
"hop_over_prediction": 0.0,
|
|
"mean_score_margin": 40.85260479970152,
|
|
"median_score_margin": 62.14678978919983,
|
|
"negative_margin_rate": 0.3309900410076157,
|
|
"hop_confusion": {
|
|
"true_0_pred_0": 21,
|
|
"true_0_pred_1": 0,
|
|
"true_1_pred_0": 0,
|
|
"true_1_pred_1": 1707
|
|
},
|
|
"known_mean_need_prob": 0.9999999792588407,
|
|
"unknown_mean_need_prob": 0.00039670502174734185,
|
|
"thr0.50": {
|
|
"need_precision": 1.0,
|
|
"need_recall": 1.0,
|
|
"need_f1": 1.0,
|
|
"specificity_unknown_refusal": 1.0,
|
|
"abstention_accuracy": 1.0,
|
|
"known_question_refusal_rate": 0.0,
|
|
"unknown_question_read_rate": 0.0,
|
|
"counts": {
|
|
"tp": 1707,
|
|
"tn": 21,
|
|
"fp": 0,
|
|
"fn": 0
|
|
}
|
|
}
|
|
},
|
|
"native_memory": {
|
|
"episodes": 128,
|
|
"answerable_episodes": 107,
|
|
"unknown_episodes": 21,
|
|
"top1": 0.38317757009345793,
|
|
"recall1": 0.38317757009345793,
|
|
"recall3": 0.8598130841121495,
|
|
"recall5": 0.9532710280373832,
|
|
"recall8": 0.9813084112149533,
|
|
"evidence_count_in_top3": 0.8598130841121495,
|
|
"hit3": 0.8598130841121495,
|
|
"mrr": 0.617352766651832,
|
|
"ndcg3": 0.6582046038050816,
|
|
"mean_first_positive_rank": 2.289719626168224,
|
|
"all_evidence_in_top3": 0.8598130841121495,
|
|
"multi_positive_episodes": 0,
|
|
"all_evidence_in_top3_multi": null,
|
|
"hop_accuracy": 1.0,
|
|
"hop_accuracy_answerable": 1.0,
|
|
"hop_under_prediction": 0.0,
|
|
"hop_over_prediction": 0.0,
|
|
"mean_score_margin": -5.260566909736562,
|
|
"median_score_margin": -7.525552272796631,
|
|
"negative_margin_rate": 0.616822429906542,
|
|
"hop_confusion": {
|
|
"true_0_pred_0": 21,
|
|
"true_0_pred_1": 0,
|
|
"true_1_pred_0": 0,
|
|
"true_1_pred_1": 107
|
|
},
|
|
"known_mean_need_prob": 1.0,
|
|
"unknown_mean_need_prob": 0.00039670502174734185,
|
|
"thr0.50": {
|
|
"need_precision": 1.0,
|
|
"need_recall": 1.0,
|
|
"need_f1": 1.0,
|
|
"specificity_unknown_refusal": 1.0,
|
|
"abstention_accuracy": 1.0,
|
|
"known_question_refusal_rate": 0.0,
|
|
"unknown_question_read_rate": 0.0,
|
|
"counts": {
|
|
"tp": 107,
|
|
"tn": 21,
|
|
"fp": 0,
|
|
"fn": 0
|
|
}
|
|
}
|
|
}
|
|
},
|
|
"by_category": {
|
|
"conflict_update": {
|
|
"episodes": 2000,
|
|
"answerable_episodes": 2000,
|
|
"unknown_episodes": 0,
|
|
"top1": 0.9925,
|
|
"recall1": 0.9925,
|
|
"recall3": 1.0,
|
|
"recall5": 1.0,
|
|
"recall8": 1.0,
|
|
"evidence_count_in_top3": 1.0,
|
|
"hit3": 1.0,
|
|
"mrr": 0.99625,
|
|
"ndcg3": 0.9972319731517859,
|
|
"mean_first_positive_rank": 1.0075,
|
|
"all_evidence_in_top3": 1.0,
|
|
"multi_positive_episodes": 0,
|
|
"all_evidence_in_top3_multi": null,
|
|
"hop_accuracy": 1.0,
|
|
"hop_accuracy_answerable": 1.0,
|
|
"hop_under_prediction": 0.0,
|
|
"hop_over_prediction": 0.0,
|
|
"mean_score_margin": 272.5490972074904,
|
|
"median_score_margin": 288.63626432418823,
|
|
"negative_margin_rate": 0.0075,
|
|
"hop_confusion": {
|
|
"true_1_pred_1": 2000
|
|
},
|
|
"known_mean_need_prob": 1.0,
|
|
"unknown_mean_need_prob": null,
|
|
"thr0.50": {
|
|
"need_precision": 1.0,
|
|
"need_recall": 1.0,
|
|
"need_f1": 1.0,
|
|
"specificity_unknown_refusal": null,
|
|
"abstention_accuracy": 1.0,
|
|
"known_question_refusal_rate": 0.0,
|
|
"unknown_question_read_rate": null,
|
|
"counts": {
|
|
"tp": 2000,
|
|
"tn": 0,
|
|
"fp": 0,
|
|
"fn": 0
|
|
}
|
|
}
|
|
},
|
|
"distractor_128": {
|
|
"episodes": 2000,
|
|
"answerable_episodes": 2000,
|
|
"unknown_episodes": 0,
|
|
"top1": 0.98,
|
|
"recall1": 0.98,
|
|
"recall3": 1.0,
|
|
"recall5": 1.0,
|
|
"recall8": 1.0,
|
|
"evidence_count_in_top3": 1.0,
|
|
"hit3": 1.0,
|
|
"mrr": 0.99,
|
|
"ndcg3": 0.9926185950714291,
|
|
"mean_first_positive_rank": 1.02,
|
|
"all_evidence_in_top3": 1.0,
|
|
"multi_positive_episodes": 0,
|
|
"all_evidence_in_top3_multi": null,
|
|
"hop_accuracy": 1.0,
|
|
"hop_accuracy_answerable": 1.0,
|
|
"hop_under_prediction": 0.0,
|
|
"hop_over_prediction": 0.0,
|
|
"mean_score_margin": 215.3791910160221,
|
|
"median_score_margin": 239.17540168762207,
|
|
"negative_margin_rate": 0.02,
|
|
"hop_confusion": {
|
|
"true_1_pred_1": 2000
|
|
},
|
|
"known_mean_need_prob": 1.0,
|
|
"unknown_mean_need_prob": null,
|
|
"thr0.50": {
|
|
"need_precision": 1.0,
|
|
"need_recall": 1.0,
|
|
"need_f1": 1.0,
|
|
"specificity_unknown_refusal": null,
|
|
"abstention_accuracy": 1.0,
|
|
"known_question_refusal_rate": 0.0,
|
|
"unknown_question_read_rate": null,
|
|
"counts": {
|
|
"tp": 2000,
|
|
"tn": 0,
|
|
"fp": 0,
|
|
"fn": 0
|
|
}
|
|
}
|
|
},
|
|
"distractor_32": {
|
|
"episodes": 2000,
|
|
"answerable_episodes": 2000,
|
|
"unknown_episodes": 0,
|
|
"top1": 0.982,
|
|
"recall1": 0.982,
|
|
"recall3": 1.0,
|
|
"recall5": 1.0,
|
|
"recall8": 1.0,
|
|
"evidence_count_in_top3": 1.0,
|
|
"hit3": 1.0,
|
|
"mrr": 0.9909166666666667,
|
|
"ndcg3": 0.9932912706875006,
|
|
"mean_first_positive_rank": 1.0185,
|
|
"all_evidence_in_top3": 1.0,
|
|
"multi_positive_episodes": 0,
|
|
"all_evidence_in_top3_multi": null,
|
|
"hop_accuracy": 1.0,
|
|
"hop_accuracy_answerable": 1.0,
|
|
"hop_under_prediction": 0.0,
|
|
"hop_over_prediction": 0.0,
|
|
"mean_score_margin": 219.1182684260085,
|
|
"median_score_margin": 239.48388767242432,
|
|
"negative_margin_rate": 0.018,
|
|
"hop_confusion": {
|
|
"true_1_pred_1": 2000
|
|
},
|
|
"known_mean_need_prob": 1.0,
|
|
"unknown_mean_need_prob": null,
|
|
"thr0.50": {
|
|
"need_precision": 1.0,
|
|
"need_recall": 1.0,
|
|
"need_f1": 1.0,
|
|
"specificity_unknown_refusal": null,
|
|
"abstention_accuracy": 1.0,
|
|
"known_question_refusal_rate": 0.0,
|
|
"unknown_question_read_rate": null,
|
|
"counts": {
|
|
"tp": 2000,
|
|
"tn": 0,
|
|
"fp": 0,
|
|
"fn": 0
|
|
}
|
|
}
|
|
},
|
|
"forget_correction": {
|
|
"episodes": 2000,
|
|
"answerable_episodes": 0,
|
|
"unknown_episodes": 2000,
|
|
"top1": null,
|
|
"recall1": null,
|
|
"recall3": null,
|
|
"recall5": null,
|
|
"recall8": null,
|
|
"evidence_count_in_top3": null,
|
|
"hit3": null,
|
|
"mrr": null,
|
|
"ndcg3": null,
|
|
"mean_first_positive_rank": null,
|
|
"all_evidence_in_top3": null,
|
|
"multi_positive_episodes": 0,
|
|
"all_evidence_in_top3_multi": null,
|
|
"hop_accuracy": 1.0,
|
|
"hop_accuracy_answerable": null,
|
|
"hop_under_prediction": null,
|
|
"hop_over_prediction": null,
|
|
"mean_score_margin": null,
|
|
"median_score_margin": null,
|
|
"negative_margin_rate": null,
|
|
"hop_confusion": {
|
|
"true_0_pred_0": 2000
|
|
},
|
|
"known_mean_need_prob": null,
|
|
"unknown_mean_need_prob": 1.4022628460683615e-10,
|
|
"thr0.50": {
|
|
"need_precision": null,
|
|
"need_recall": null,
|
|
"need_f1": null,
|
|
"specificity_unknown_refusal": 1.0,
|
|
"abstention_accuracy": 1.0,
|
|
"known_question_refusal_rate": null,
|
|
"unknown_question_read_rate": 0.0,
|
|
"counts": {
|
|
"tp": 0,
|
|
"tn": 2000,
|
|
"fp": 0,
|
|
"fn": 0
|
|
}
|
|
}
|
|
},
|
|
"long_context": {
|
|
"episodes": 2000,
|
|
"answerable_episodes": 2000,
|
|
"unknown_episodes": 0,
|
|
"top1": 0.983,
|
|
"recall1": 0.983,
|
|
"recall3": 1.0,
|
|
"recall5": 1.0,
|
|
"recall8": 1.0,
|
|
"evidence_count_in_top3": 1.0,
|
|
"hit3": 1.0,
|
|
"mrr": 0.9914166666666666,
|
|
"ndcg3": 0.9936603409339291,
|
|
"mean_first_positive_rank": 1.0175,
|
|
"all_evidence_in_top3": 1.0,
|
|
"multi_positive_episodes": 0,
|
|
"all_evidence_in_top3_multi": null,
|
|
"hop_accuracy": 1.0,
|
|
"hop_accuracy_answerable": 1.0,
|
|
"hop_under_prediction": 0.0,
|
|
"hop_over_prediction": 0.0,
|
|
"mean_score_margin": 211.44927242165804,
|
|
"median_score_margin": 233.2896122932434,
|
|
"negative_margin_rate": 0.017,
|
|
"hop_confusion": {
|
|
"true_1_pred_1": 2000
|
|
},
|
|
"known_mean_need_prob": 1.0,
|
|
"unknown_mean_need_prob": null,
|
|
"thr0.50": {
|
|
"need_precision": 1.0,
|
|
"need_recall": 1.0,
|
|
"need_f1": 1.0,
|
|
"specificity_unknown_refusal": null,
|
|
"abstention_accuracy": 1.0,
|
|
"known_question_refusal_rate": 0.0,
|
|
"unknown_question_read_rate": null,
|
|
"counts": {
|
|
"tp": 2000,
|
|
"tn": 0,
|
|
"fp": 0,
|
|
"fn": 0
|
|
}
|
|
}
|
|
},
|
|
"multi_hop": {
|
|
"episodes": 2000,
|
|
"answerable_episodes": 2000,
|
|
"unknown_episodes": 0,
|
|
"top1": 0.9865,
|
|
"recall1": 0.49325,
|
|
"recall3": 0.89425,
|
|
"recall5": 0.927,
|
|
"recall8": 0.9465,
|
|
"evidence_count_in_top3": 1.7885,
|
|
"hit3": 1.0,
|
|
"mrr": 0.99325,
|
|
"ndcg3": 0.9008760971390303,
|
|
"mean_first_positive_rank": 1.0135,
|
|
"all_evidence_in_top3": 0.7885,
|
|
"multi_positive_episodes": 2000,
|
|
"all_evidence_in_top3_multi": 0.7885,
|
|
"hop_accuracy": 1.0,
|
|
"hop_accuracy_answerable": 1.0,
|
|
"hop_under_prediction": 0.0,
|
|
"hop_over_prediction": 0.0,
|
|
"mean_score_margin": 274.43398497897016,
|
|
"median_score_margin": 301.56740951538086,
|
|
"negative_margin_rate": 0.0135,
|
|
"hop_confusion": {
|
|
"true_2_pred_2": 2000
|
|
},
|
|
"known_mean_need_prob": 0.9999999997019767,
|
|
"unknown_mean_need_prob": null,
|
|
"thr0.50": {
|
|
"need_precision": 1.0,
|
|
"need_recall": 1.0,
|
|
"need_f1": 1.0,
|
|
"specificity_unknown_refusal": null,
|
|
"abstention_accuracy": 1.0,
|
|
"known_question_refusal_rate": 0.0,
|
|
"unknown_question_read_rate": null,
|
|
"counts": {
|
|
"tp": 2000,
|
|
"tn": 0,
|
|
"fp": 0,
|
|
"fn": 0
|
|
}
|
|
}
|
|
},
|
|
"paraphrase": {
|
|
"episodes": 2000,
|
|
"answerable_episodes": 2000,
|
|
"unknown_episodes": 0,
|
|
"top1": 0.995,
|
|
"recall1": 0.995,
|
|
"recall3": 1.0,
|
|
"recall5": 1.0,
|
|
"recall8": 1.0,
|
|
"evidence_count_in_top3": 1.0,
|
|
"hit3": 1.0,
|
|
"mrr": 0.9975,
|
|
"ndcg3": 0.9981546487678573,
|
|
"mean_first_positive_rank": 1.005,
|
|
"all_evidence_in_top3": 1.0,
|
|
"multi_positive_episodes": 0,
|
|
"all_evidence_in_top3_multi": null,
|
|
"hop_accuracy": 1.0,
|
|
"hop_accuracy_answerable": 1.0,
|
|
"hop_under_prediction": 0.0,
|
|
"hop_over_prediction": 0.0,
|
|
"mean_score_margin": 223.85322645658067,
|
|
"median_score_margin": 222.2765736579895,
|
|
"negative_margin_rate": 0.005,
|
|
"hop_confusion": {
|
|
"true_1_pred_1": 2000
|
|
},
|
|
"known_mean_need_prob": 1.0,
|
|
"unknown_mean_need_prob": null,
|
|
"thr0.50": {
|
|
"need_precision": 1.0,
|
|
"need_recall": 1.0,
|
|
"need_f1": 1.0,
|
|
"specificity_unknown_refusal": null,
|
|
"abstention_accuracy": 1.0,
|
|
"known_question_refusal_rate": 0.0,
|
|
"unknown_question_read_rate": null,
|
|
"counts": {
|
|
"tp": 2000,
|
|
"tn": 0,
|
|
"fp": 0,
|
|
"fn": 0
|
|
}
|
|
}
|
|
},
|
|
"random_position": {
|
|
"episodes": 2000,
|
|
"answerable_episodes": 2000,
|
|
"unknown_episodes": 0,
|
|
"top1": 0.9805,
|
|
"recall1": 0.9805,
|
|
"recall3": 1.0,
|
|
"recall5": 1.0,
|
|
"recall8": 1.0,
|
|
"evidence_count_in_top3": 1.0,
|
|
"hit3": 1.0,
|
|
"mrr": 0.99025,
|
|
"ndcg3": 0.9928031301946435,
|
|
"mean_first_positive_rank": 1.0195,
|
|
"all_evidence_in_top3": 1.0,
|
|
"multi_positive_episodes": 0,
|
|
"all_evidence_in_top3_multi": null,
|
|
"hop_accuracy": 1.0,
|
|
"hop_accuracy_answerable": 1.0,
|
|
"hop_under_prediction": 0.0,
|
|
"hop_over_prediction": 0.0,
|
|
"mean_score_margin": 225.64811652239038,
|
|
"median_score_margin": 241.96742022037506,
|
|
"negative_margin_rate": 0.0195,
|
|
"hop_confusion": {
|
|
"true_1_pred_1": 2000
|
|
},
|
|
"known_mean_need_prob": 1.0,
|
|
"unknown_mean_need_prob": null,
|
|
"thr0.50": {
|
|
"need_precision": 1.0,
|
|
"need_recall": 1.0,
|
|
"need_f1": 1.0,
|
|
"specificity_unknown_refusal": null,
|
|
"abstention_accuracy": 1.0,
|
|
"known_question_refusal_rate": 0.0,
|
|
"unknown_question_read_rate": null,
|
|
"counts": {
|
|
"tp": 2000,
|
|
"tn": 0,
|
|
"fp": 0,
|
|
"fn": 0
|
|
}
|
|
}
|
|
},
|
|
"single_fact": {
|
|
"episodes": 2000,
|
|
"answerable_episodes": 2000,
|
|
"unknown_episodes": 0,
|
|
"top1": 0.9725,
|
|
"recall1": 0.9725,
|
|
"recall3": 1.0,
|
|
"recall5": 1.0,
|
|
"recall8": 1.0,
|
|
"evidence_count_in_top3": 1.0,
|
|
"hit3": 1.0,
|
|
"mrr": 0.9861666666666666,
|
|
"ndcg3": 0.9897851033464293,
|
|
"mean_first_positive_rank": 1.028,
|
|
"all_evidence_in_top3": 1.0,
|
|
"multi_positive_episodes": 0,
|
|
"all_evidence_in_top3_multi": null,
|
|
"hop_accuracy": 1.0,
|
|
"hop_accuracy_answerable": 1.0,
|
|
"hop_under_prediction": 0.0,
|
|
"hop_over_prediction": 0.0,
|
|
"mean_score_margin": 244.6973513062373,
|
|
"median_score_margin": 263.6290616989136,
|
|
"negative_margin_rate": 0.0275,
|
|
"hop_confusion": {
|
|
"true_1_pred_1": 2000
|
|
},
|
|
"known_mean_need_prob": 1.0,
|
|
"unknown_mean_need_prob": null,
|
|
"thr0.50": {
|
|
"need_precision": 1.0,
|
|
"need_recall": 1.0,
|
|
"need_f1": 1.0,
|
|
"specificity_unknown_refusal": null,
|
|
"abstention_accuracy": 1.0,
|
|
"known_question_refusal_rate": 0.0,
|
|
"unknown_question_read_rate": null,
|
|
"counts": {
|
|
"tp": 2000,
|
|
"tn": 0,
|
|
"fp": 0,
|
|
"fn": 0
|
|
}
|
|
}
|
|
},
|
|
"unknown": {
|
|
"episodes": 1920,
|
|
"answerable_episodes": 1814,
|
|
"unknown_episodes": 106,
|
|
"top1": 0.6521499448732084,
|
|
"recall1": 0.6521499448732084,
|
|
"recall3": 0.7761852260198456,
|
|
"recall5": 0.8456449834619625,
|
|
"recall8": 0.9162072767364939,
|
|
"evidence_count_in_top3": 0.7761852260198456,
|
|
"hit3": 0.7761852260198456,
|
|
"mrr": 0.7410856875575729,
|
|
"ndcg3": 0.7240558854681862,
|
|
"mean_first_positive_rank": 2.54244762954796,
|
|
"all_evidence_in_top3": 0.7761852260198456,
|
|
"multi_positive_episodes": 0,
|
|
"all_evidence_in_top3_multi": null,
|
|
"hop_accuracy": 1.0,
|
|
"hop_accuracy_answerable": 1.0,
|
|
"hop_under_prediction": 0.0,
|
|
"hop_over_prediction": 0.0,
|
|
"mean_score_margin": 38.1325886073587,
|
|
"median_score_margin": 54.42159175872803,
|
|
"negative_margin_rate": 0.3478500551267916,
|
|
"hop_confusion": {
|
|
"true_0_pred_0": 106,
|
|
"true_0_pred_1": 0,
|
|
"true_1_pred_0": 0,
|
|
"true_1_pred_1": 1814
|
|
},
|
|
"known_mean_need_prob": 0.9999999804822718,
|
|
"unknown_mean_need_prob": 0.0001597598329493619,
|
|
"thr0.50": {
|
|
"need_precision": 1.0,
|
|
"need_recall": 1.0,
|
|
"need_f1": 1.0,
|
|
"specificity_unknown_refusal": 1.0,
|
|
"abstention_accuracy": 1.0,
|
|
"known_question_refusal_rate": 0.0,
|
|
"unknown_question_read_rate": 0.0,
|
|
"counts": {
|
|
"tp": 1814,
|
|
"tn": 106,
|
|
"fp": 0,
|
|
"fn": 0
|
|
}
|
|
}
|
|
},
|
|
"unknown_abstention": {
|
|
"episodes": 2000,
|
|
"answerable_episodes": 0,
|
|
"unknown_episodes": 2000,
|
|
"top1": null,
|
|
"recall1": null,
|
|
"recall3": null,
|
|
"recall5": null,
|
|
"recall8": null,
|
|
"evidence_count_in_top3": null,
|
|
"hit3": null,
|
|
"mrr": null,
|
|
"ndcg3": null,
|
|
"mean_first_positive_rank": null,
|
|
"all_evidence_in_top3": null,
|
|
"multi_positive_episodes": 0,
|
|
"all_evidence_in_top3_multi": null,
|
|
"hop_accuracy": 1.0,
|
|
"hop_accuracy_answerable": null,
|
|
"hop_under_prediction": null,
|
|
"hop_over_prediction": null,
|
|
"mean_score_margin": null,
|
|
"median_score_margin": null,
|
|
"negative_margin_rate": null,
|
|
"hop_confusion": {
|
|
"true_0_pred_0": 2000
|
|
},
|
|
"known_mean_need_prob": null,
|
|
"unknown_mean_need_prob": 2.5479757829358922e-15,
|
|
"thr0.50": {
|
|
"need_precision": null,
|
|
"need_recall": null,
|
|
"need_f1": null,
|
|
"specificity_unknown_refusal": 1.0,
|
|
"abstention_accuracy": 1.0,
|
|
"known_question_refusal_rate": null,
|
|
"unknown_question_read_rate": 0.0,
|
|
"counts": {
|
|
"tp": 0,
|
|
"tn": 2000,
|
|
"fp": 0,
|
|
"fn": 0
|
|
}
|
|
}
|
|
}
|
|
}
|
|
},
|
|
"XL-128 v6 final": {
|
|
"checkpoint": "checkpoints\\router_v6_xl128\\memory_router_xl.pt",
|
|
"router": {
|
|
"kind": "router_xl",
|
|
"step": null,
|
|
"arch_config": {
|
|
"arch": "router_xl",
|
|
"arch_version": 1,
|
|
"hidden_size": 2560,
|
|
"router_dim": 128,
|
|
"num_heads": 8,
|
|
"max_hops": 3,
|
|
"encoder_layers": 2,
|
|
"encoder_hidden": 512,
|
|
"pair_blocks": 1,
|
|
"pair_hidden": 512,
|
|
"pair_expansion": 2,
|
|
"pair_dropout": 0.05,
|
|
"use_interaction": true,
|
|
"policy_layers": 2,
|
|
"policy_hidden": 512,
|
|
"policy_dropout": 0.0,
|
|
"learnable_cosine_scale": true
|
|
},
|
|
"parameters": 6714639,
|
|
"checkpoint_bytes": 27922475
|
|
},
|
|
"storage": {
|
|
"address_bytes_per_record": 512,
|
|
"address_mb_per_1m_records": 512,
|
|
"checkpoint_bytes": 27922475
|
|
},
|
|
"latency": {
|
|
"cuda": {
|
|
"single_query_latency_ms_mean": 1.5358359999663662,
|
|
"single_query_latency_ms_p50": 1.4220000011846423,
|
|
"single_query_latency_ms_p95": 2.4550000016461127,
|
|
"queries_per_second": 651.1111863648849,
|
|
"samples": 150,
|
|
"batched_qps_64": 33111.55489890671,
|
|
"batched_latency_ms_64": 0.03020093749910302,
|
|
"batched_qps_256": 131545.14154150538,
|
|
"batched_latency_ms_256": 0.007601953126368244
|
|
},
|
|
"cpu": {
|
|
"single_query_latency_ms_mean": 1.8319966668301884,
|
|
"single_query_latency_ms_p50": 1.8206000022473745,
|
|
"single_query_latency_ms_p95": 2.2160000007716008,
|
|
"queries_per_second": 545.8525215170013,
|
|
"samples": 150,
|
|
"batched_qps_64": 2173.5896629680874,
|
|
"batched_latency_ms_64": 0.4600684374963748,
|
|
"batched_qps_256": 2384.8277260004847,
|
|
"batched_latency_ms_256": 0.41931750000117063
|
|
}
|
|
},
|
|
"metrics": {
|
|
"episodes": 21920,
|
|
"answerable_episodes": 17814,
|
|
"unknown_episodes": 4106,
|
|
"top1": 0.9518917705175705,
|
|
"recall1": 0.8963736387111261,
|
|
"recall3": 0.9745424946671157,
|
|
"recall5": 0.9820366004266308,
|
|
"recall8": 0.9912709105198159,
|
|
"evidence_count_in_top3": 1.0851577411025037,
|
|
"hit3": 0.9761984955652857,
|
|
"mrr": 0.9670023550641874,
|
|
"ndcg3": 0.9642642772784563,
|
|
"mean_first_positive_rank": 1.1905243067250477,
|
|
"all_evidence_in_top3": 0.9728864937689458,
|
|
"multi_positive_episodes": 2000,
|
|
"all_evidence_in_top3_multi": 0.9705,
|
|
"hop_accuracy": 1.0,
|
|
"hop_accuracy_answerable": 1.0,
|
|
"hop_under_prediction": 0.0,
|
|
"hop_over_prediction": 0.0,
|
|
"mean_score_margin": 200.33457717921095,
|
|
"median_score_margin": 219.4529800415039,
|
|
"negative_margin_rate": 0.04810822948242955,
|
|
"hop_confusion": {
|
|
"true_0_pred_0": 4106,
|
|
"true_0_pred_1": 0,
|
|
"true_0_pred_2": 0,
|
|
"true_1_pred_0": 0,
|
|
"true_1_pred_1": 15814,
|
|
"true_1_pred_2": 0,
|
|
"true_2_pred_0": 0,
|
|
"true_2_pred_1": 0,
|
|
"true_2_pred_2": 2000
|
|
},
|
|
"known_mean_need_prob": 0.9999999979790499,
|
|
"unknown_mean_need_prob": 3.8975788264237145e-06,
|
|
"thr0.30": {
|
|
"need_precision": 1.0,
|
|
"need_recall": 1.0,
|
|
"need_f1": 1.0,
|
|
"specificity_unknown_refusal": 1.0,
|
|
"abstention_accuracy": 1.0,
|
|
"known_question_refusal_rate": 0.0,
|
|
"unknown_question_read_rate": 0.0,
|
|
"counts": {
|
|
"tp": 17814,
|
|
"tn": 4106,
|
|
"fp": 0,
|
|
"fn": 0
|
|
}
|
|
},
|
|
"thr0.40": {
|
|
"need_precision": 1.0,
|
|
"need_recall": 1.0,
|
|
"need_f1": 1.0,
|
|
"specificity_unknown_refusal": 1.0,
|
|
"abstention_accuracy": 1.0,
|
|
"known_question_refusal_rate": 0.0,
|
|
"unknown_question_read_rate": 0.0,
|
|
"counts": {
|
|
"tp": 17814,
|
|
"tn": 4106,
|
|
"fp": 0,
|
|
"fn": 0
|
|
}
|
|
},
|
|
"thr0.50": {
|
|
"need_precision": 1.0,
|
|
"need_recall": 1.0,
|
|
"need_f1": 1.0,
|
|
"specificity_unknown_refusal": 1.0,
|
|
"abstention_accuracy": 1.0,
|
|
"known_question_refusal_rate": 0.0,
|
|
"unknown_question_read_rate": 0.0,
|
|
"counts": {
|
|
"tp": 17814,
|
|
"tn": 4106,
|
|
"fp": 0,
|
|
"fn": 0
|
|
}
|
|
},
|
|
"thr0.60": {
|
|
"need_precision": 1.0,
|
|
"need_recall": 1.0,
|
|
"need_f1": 1.0,
|
|
"specificity_unknown_refusal": 1.0,
|
|
"abstention_accuracy": 1.0,
|
|
"known_question_refusal_rate": 0.0,
|
|
"unknown_question_read_rate": 0.0,
|
|
"counts": {
|
|
"tp": 17814,
|
|
"tn": 4106,
|
|
"fp": 0,
|
|
"fn": 0
|
|
}
|
|
},
|
|
"thr0.70": {
|
|
"need_precision": 1.0,
|
|
"need_recall": 1.0,
|
|
"need_f1": 1.0,
|
|
"specificity_unknown_refusal": 1.0,
|
|
"abstention_accuracy": 1.0,
|
|
"known_question_refusal_rate": 0.0,
|
|
"unknown_question_read_rate": 0.0,
|
|
"counts": {
|
|
"tp": 17814,
|
|
"tn": 4106,
|
|
"fp": 0,
|
|
"fn": 0
|
|
}
|
|
},
|
|
"thr0.80": {
|
|
"need_precision": 1.0,
|
|
"need_recall": 1.0,
|
|
"need_f1": 1.0,
|
|
"specificity_unknown_refusal": 1.0,
|
|
"abstention_accuracy": 1.0,
|
|
"known_question_refusal_rate": 0.0,
|
|
"unknown_question_read_rate": 0.0,
|
|
"counts": {
|
|
"tp": 17814,
|
|
"tn": 4106,
|
|
"fp": 0,
|
|
"fn": 0
|
|
}
|
|
}
|
|
},
|
|
"by_family": {
|
|
"benchmark_qa": {
|
|
"episodes": 64,
|
|
"answerable_episodes": 0,
|
|
"unknown_episodes": 64,
|
|
"top1": null,
|
|
"recall1": null,
|
|
"recall3": null,
|
|
"recall5": null,
|
|
"recall8": null,
|
|
"evidence_count_in_top3": null,
|
|
"hit3": null,
|
|
"mrr": null,
|
|
"ndcg3": null,
|
|
"mean_first_positive_rank": null,
|
|
"all_evidence_in_top3": null,
|
|
"multi_positive_episodes": 0,
|
|
"all_evidence_in_top3_multi": null,
|
|
"hop_accuracy": 1.0,
|
|
"hop_accuracy_answerable": null,
|
|
"hop_under_prediction": null,
|
|
"hop_over_prediction": null,
|
|
"mean_score_margin": null,
|
|
"median_score_margin": null,
|
|
"negative_margin_rate": null,
|
|
"hop_confusion": {
|
|
"true_0_pred_0": 64
|
|
},
|
|
"known_mean_need_prob": null,
|
|
"unknown_mean_need_prob": 3.895038815020929e-06,
|
|
"thr0.50": {
|
|
"need_precision": null,
|
|
"need_recall": null,
|
|
"need_f1": null,
|
|
"specificity_unknown_refusal": 1.0,
|
|
"abstention_accuracy": 1.0,
|
|
"known_question_refusal_rate": null,
|
|
"unknown_question_read_rate": 0.0,
|
|
"counts": {
|
|
"tp": 0,
|
|
"tn": 64,
|
|
"fp": 0,
|
|
"fn": 0
|
|
}
|
|
}
|
|
},
|
|
"mega_validation": {
|
|
"episodes": 20000,
|
|
"answerable_episodes": 16000,
|
|
"unknown_episodes": 4000,
|
|
"top1": 0.984375,
|
|
"recall1": 0.9225625,
|
|
"recall3": 0.99815625,
|
|
"recall5": 0.9995,
|
|
"recall8": 0.99959375,
|
|
"evidence_count_in_top3": 1.1213125,
|
|
"hit3": 1.0,
|
|
"mrr": 0.9921354166666666,
|
|
"ndcg3": 0.9914167841247224,
|
|
"mean_first_positive_rank": 1.0159375,
|
|
"all_evidence_in_top3": 0.9963125,
|
|
"multi_positive_episodes": 2000,
|
|
"all_evidence_in_top3_multi": 0.9705,
|
|
"hop_accuracy": 1.0,
|
|
"hop_accuracy_answerable": 1.0,
|
|
"hop_under_prediction": 0.0,
|
|
"hop_over_prediction": 0.0,
|
|
"mean_score_margin": 217.4091286659348,
|
|
"median_score_margin": 228.10742282867432,
|
|
"negative_margin_rate": 0.015625,
|
|
"hop_confusion": {
|
|
"true_0_pred_0": 4000,
|
|
"true_0_pred_1": 0,
|
|
"true_0_pred_2": 0,
|
|
"true_1_pred_0": 0,
|
|
"true_1_pred_1": 14000,
|
|
"true_1_pred_2": 0,
|
|
"true_2_pred_0": 0,
|
|
"true_2_pred_1": 0,
|
|
"true_2_pred_2": 2000
|
|
},
|
|
"known_mean_need_prob": 0.9999999999627471,
|
|
"unknown_mean_need_prob": 6.477803048274694e-11,
|
|
"thr0.50": {
|
|
"need_precision": 1.0,
|
|
"need_recall": 1.0,
|
|
"need_f1": 1.0,
|
|
"specificity_unknown_refusal": 1.0,
|
|
"abstention_accuracy": 1.0,
|
|
"known_question_refusal_rate": 0.0,
|
|
"unknown_question_read_rate": 0.0,
|
|
"counts": {
|
|
"tp": 16000,
|
|
"tn": 4000,
|
|
"fp": 0,
|
|
"fn": 0
|
|
}
|
|
}
|
|
},
|
|
"memory_policy": {
|
|
"episodes": 1728,
|
|
"answerable_episodes": 1707,
|
|
"unknown_episodes": 21,
|
|
"top1": 0.6754540128881078,
|
|
"recall1": 0.6754540128881078,
|
|
"recall3": 0.7592267135325131,
|
|
"recall5": 0.8189806678383128,
|
|
"recall8": 0.9127123608670181,
|
|
"evidence_count_in_top3": 0.7592267135325131,
|
|
"hit3": 0.7592267135325131,
|
|
"mrr": 0.7481470978152184,
|
|
"ndcg3": 0.7247037236923608,
|
|
"mean_first_positive_rank": 2.7809021675454013,
|
|
"all_evidence_in_top3": 0.7592267135325131,
|
|
"multi_positive_episodes": 0,
|
|
"all_evidence_in_top3_multi": null,
|
|
"hop_accuracy": 1.0,
|
|
"hop_accuracy_answerable": 1.0,
|
|
"hop_under_prediction": 0.0,
|
|
"hop_over_prediction": 0.0,
|
|
"mean_score_margin": 52.43232965865677,
|
|
"median_score_margin": 72.44926834106445,
|
|
"negative_margin_rate": 0.3245459871118922,
|
|
"hop_confusion": {
|
|
"true_0_pred_0": 21,
|
|
"true_0_pred_1": 0,
|
|
"true_1_pred_0": 0,
|
|
"true_1_pred_1": 1707
|
|
},
|
|
"known_mean_need_prob": 0.9999999792588407,
|
|
"unknown_mean_need_prob": 0.0003750932634526786,
|
|
"thr0.50": {
|
|
"need_precision": 1.0,
|
|
"need_recall": 1.0,
|
|
"need_f1": 1.0,
|
|
"specificity_unknown_refusal": 1.0,
|
|
"abstention_accuracy": 1.0,
|
|
"known_question_refusal_rate": 0.0,
|
|
"unknown_question_read_rate": 0.0,
|
|
"counts": {
|
|
"tp": 1707,
|
|
"tn": 21,
|
|
"fp": 0,
|
|
"fn": 0
|
|
}
|
|
}
|
|
},
|
|
"native_memory": {
|
|
"episodes": 128,
|
|
"answerable_episodes": 107,
|
|
"unknown_episodes": 21,
|
|
"top1": 0.5046728971962616,
|
|
"recall1": 0.5046728971962616,
|
|
"recall3": 0.8785046728971962,
|
|
"recall5": 0.9719626168224299,
|
|
"recall8": 1.0,
|
|
"evidence_count_in_top3": 0.8785046728971962,
|
|
"hit3": 0.8785046728971962,
|
|
"mrr": 0.7002447708055184,
|
|
"ndcg3": 0.7258507766355217,
|
|
"mean_first_positive_rank": 1.925233644859813,
|
|
"all_evidence_in_top3": 0.8785046728971962,
|
|
"multi_positive_episodes": 0,
|
|
"all_evidence_in_top3_multi": null,
|
|
"hop_accuracy": 1.0,
|
|
"hop_accuracy_answerable": 1.0,
|
|
"hop_under_prediction": 0.0,
|
|
"hop_over_prediction": 0.0,
|
|
"mean_score_margin": 6.655256898882233,
|
|
"median_score_margin": 1.441606044769287,
|
|
"negative_margin_rate": 0.4953271028037383,
|
|
"hop_confusion": {
|
|
"true_0_pred_0": 21,
|
|
"true_0_pred_1": 0,
|
|
"true_1_pred_0": 0,
|
|
"true_1_pred_1": 107
|
|
},
|
|
"known_mean_need_prob": 1.0,
|
|
"unknown_mean_need_prob": 0.0003750932634526786,
|
|
"thr0.50": {
|
|
"need_precision": 1.0,
|
|
"need_recall": 1.0,
|
|
"need_f1": 1.0,
|
|
"specificity_unknown_refusal": 1.0,
|
|
"abstention_accuracy": 1.0,
|
|
"known_question_refusal_rate": 0.0,
|
|
"unknown_question_read_rate": 0.0,
|
|
"counts": {
|
|
"tp": 107,
|
|
"tn": 21,
|
|
"fp": 0,
|
|
"fn": 0
|
|
}
|
|
}
|
|
}
|
|
},
|
|
"by_category": {
|
|
"conflict_update": {
|
|
"episodes": 2000,
|
|
"answerable_episodes": 2000,
|
|
"unknown_episodes": 0,
|
|
"top1": 0.9945,
|
|
"recall1": 0.9945,
|
|
"recall3": 1.0,
|
|
"recall5": 1.0,
|
|
"recall8": 1.0,
|
|
"evidence_count_in_top3": 1.0,
|
|
"hit3": 1.0,
|
|
"mrr": 0.99725,
|
|
"ndcg3": 0.9979701136446429,
|
|
"mean_first_positive_rank": 1.0055,
|
|
"all_evidence_in_top3": 1.0,
|
|
"multi_positive_episodes": 0,
|
|
"all_evidence_in_top3_multi": null,
|
|
"hop_accuracy": 1.0,
|
|
"hop_accuracy_answerable": 1.0,
|
|
"hop_under_prediction": 0.0,
|
|
"hop_over_prediction": 0.0,
|
|
"mean_score_margin": 243.0819873433709,
|
|
"median_score_margin": 235.45626258850098,
|
|
"negative_margin_rate": 0.0055,
|
|
"hop_confusion": {
|
|
"true_1_pred_1": 2000
|
|
},
|
|
"known_mean_need_prob": 1.0,
|
|
"unknown_mean_need_prob": null,
|
|
"thr0.50": {
|
|
"need_precision": 1.0,
|
|
"need_recall": 1.0,
|
|
"need_f1": 1.0,
|
|
"specificity_unknown_refusal": null,
|
|
"abstention_accuracy": 1.0,
|
|
"known_question_refusal_rate": 0.0,
|
|
"unknown_question_read_rate": null,
|
|
"counts": {
|
|
"tp": 2000,
|
|
"tn": 0,
|
|
"fp": 0,
|
|
"fn": 0
|
|
}
|
|
}
|
|
},
|
|
"distractor_128": {
|
|
"episodes": 2000,
|
|
"answerable_episodes": 2000,
|
|
"unknown_episodes": 0,
|
|
"top1": 0.9785,
|
|
"recall1": 0.9785,
|
|
"recall3": 1.0,
|
|
"recall5": 1.0,
|
|
"recall8": 1.0,
|
|
"evidence_count_in_top3": 1.0,
|
|
"hit3": 1.0,
|
|
"mrr": 0.98925,
|
|
"ndcg3": 0.9920649897017864,
|
|
"mean_first_positive_rank": 1.0215,
|
|
"all_evidence_in_top3": 1.0,
|
|
"multi_positive_episodes": 0,
|
|
"all_evidence_in_top3_multi": null,
|
|
"hop_accuracy": 1.0,
|
|
"hop_accuracy_answerable": 1.0,
|
|
"hop_under_prediction": 0.0,
|
|
"hop_over_prediction": 0.0,
|
|
"mean_score_margin": 229.82924858646467,
|
|
"median_score_margin": 248.98048400878906,
|
|
"negative_margin_rate": 0.0215,
|
|
"hop_confusion": {
|
|
"true_1_pred_1": 2000
|
|
},
|
|
"known_mean_need_prob": 1.0,
|
|
"unknown_mean_need_prob": null,
|
|
"thr0.50": {
|
|
"need_precision": 1.0,
|
|
"need_recall": 1.0,
|
|
"need_f1": 1.0,
|
|
"specificity_unknown_refusal": null,
|
|
"abstention_accuracy": 1.0,
|
|
"known_question_refusal_rate": 0.0,
|
|
"unknown_question_read_rate": null,
|
|
"counts": {
|
|
"tp": 2000,
|
|
"tn": 0,
|
|
"fp": 0,
|
|
"fn": 0
|
|
}
|
|
}
|
|
},
|
|
"distractor_32": {
|
|
"episodes": 2000,
|
|
"answerable_episodes": 2000,
|
|
"unknown_episodes": 0,
|
|
"top1": 0.9845,
|
|
"recall1": 0.9845,
|
|
"recall3": 1.0,
|
|
"recall5": 1.0,
|
|
"recall8": 1.0,
|
|
"evidence_count_in_top3": 1.0,
|
|
"hit3": 1.0,
|
|
"mrr": 0.9921666666666666,
|
|
"ndcg3": 0.9942139463035718,
|
|
"mean_first_positive_rank": 1.016,
|
|
"all_evidence_in_top3": 1.0,
|
|
"multi_positive_episodes": 0,
|
|
"all_evidence_in_top3_multi": null,
|
|
"hop_accuracy": 1.0,
|
|
"hop_accuracy_answerable": 1.0,
|
|
"hop_under_prediction": 0.0,
|
|
"hop_over_prediction": 0.0,
|
|
"mean_score_margin": 232.02638172297924,
|
|
"median_score_margin": 249.2184543609619,
|
|
"negative_margin_rate": 0.0155,
|
|
"hop_confusion": {
|
|
"true_1_pred_1": 2000
|
|
},
|
|
"known_mean_need_prob": 1.0,
|
|
"unknown_mean_need_prob": null,
|
|
"thr0.50": {
|
|
"need_precision": 1.0,
|
|
"need_recall": 1.0,
|
|
"need_f1": 1.0,
|
|
"specificity_unknown_refusal": null,
|
|
"abstention_accuracy": 1.0,
|
|
"known_question_refusal_rate": 0.0,
|
|
"unknown_question_read_rate": null,
|
|
"counts": {
|
|
"tp": 2000,
|
|
"tn": 0,
|
|
"fp": 0,
|
|
"fn": 0
|
|
}
|
|
}
|
|
},
|
|
"forget_correction": {
|
|
"episodes": 2000,
|
|
"answerable_episodes": 0,
|
|
"unknown_episodes": 2000,
|
|
"top1": null,
|
|
"recall1": null,
|
|
"recall3": null,
|
|
"recall5": null,
|
|
"recall8": null,
|
|
"evidence_count_in_top3": null,
|
|
"hit3": null,
|
|
"mrr": null,
|
|
"ndcg3": null,
|
|
"mean_first_positive_rank": null,
|
|
"all_evidence_in_top3": null,
|
|
"multi_positive_episodes": 0,
|
|
"all_evidence_in_top3_multi": null,
|
|
"hop_accuracy": 1.0,
|
|
"hop_accuracy_answerable": null,
|
|
"hop_under_prediction": null,
|
|
"hop_over_prediction": null,
|
|
"mean_score_margin": null,
|
|
"median_score_margin": null,
|
|
"negative_margin_rate": null,
|
|
"hop_confusion": {
|
|
"true_0_pred_0": 2000
|
|
},
|
|
"known_mean_need_prob": null,
|
|
"unknown_mean_need_prob": 1.2955399044485864e-10,
|
|
"thr0.50": {
|
|
"need_precision": null,
|
|
"need_recall": null,
|
|
"need_f1": null,
|
|
"specificity_unknown_refusal": 1.0,
|
|
"abstention_accuracy": 1.0,
|
|
"known_question_refusal_rate": null,
|
|
"unknown_question_read_rate": 0.0,
|
|
"counts": {
|
|
"tp": 0,
|
|
"tn": 2000,
|
|
"fp": 0,
|
|
"fn": 0
|
|
}
|
|
}
|
|
},
|
|
"long_context": {
|
|
"episodes": 2000,
|
|
"answerable_episodes": 2000,
|
|
"unknown_episodes": 0,
|
|
"top1": 0.9815,
|
|
"recall1": 0.9815,
|
|
"recall3": 1.0,
|
|
"recall5": 1.0,
|
|
"recall8": 1.0,
|
|
"evidence_count_in_top3": 1.0,
|
|
"hit3": 1.0,
|
|
"mrr": 0.9906666666666666,
|
|
"ndcg3": 0.9931067355642863,
|
|
"mean_first_positive_rank": 1.019,
|
|
"all_evidence_in_top3": 1.0,
|
|
"multi_positive_episodes": 0,
|
|
"all_evidence_in_top3_multi": null,
|
|
"hop_accuracy": 1.0,
|
|
"hop_accuracy_answerable": 1.0,
|
|
"hop_under_prediction": 0.0,
|
|
"hop_over_prediction": 0.0,
|
|
"mean_score_margin": 211.99539810842276,
|
|
"median_score_margin": 225.90464401245117,
|
|
"negative_margin_rate": 0.0185,
|
|
"hop_confusion": {
|
|
"true_1_pred_1": 2000
|
|
},
|
|
"known_mean_need_prob": 1.0,
|
|
"unknown_mean_need_prob": null,
|
|
"thr0.50": {
|
|
"need_precision": 1.0,
|
|
"need_recall": 1.0,
|
|
"need_f1": 1.0,
|
|
"specificity_unknown_refusal": null,
|
|
"abstention_accuracy": 1.0,
|
|
"known_question_refusal_rate": 0.0,
|
|
"unknown_question_read_rate": null,
|
|
"counts": {
|
|
"tp": 2000,
|
|
"tn": 0,
|
|
"fp": 0,
|
|
"fn": 0
|
|
}
|
|
}
|
|
},
|
|
"multi_hop": {
|
|
"episodes": 2000,
|
|
"answerable_episodes": 2000,
|
|
"unknown_episodes": 0,
|
|
"top1": 0.989,
|
|
"recall1": 0.4945,
|
|
"recall3": 0.98525,
|
|
"recall5": 0.996,
|
|
"recall8": 0.99675,
|
|
"evidence_count_in_top3": 1.9705,
|
|
"hit3": 1.0,
|
|
"mrr": 0.9945,
|
|
"ndcg3": 0.9737356054745618,
|
|
"mean_first_positive_rank": 1.011,
|
|
"all_evidence_in_top3": 0.9705,
|
|
"multi_positive_episodes": 2000,
|
|
"all_evidence_in_top3_multi": 0.9705,
|
|
"hop_accuracy": 1.0,
|
|
"hop_accuracy_answerable": 1.0,
|
|
"hop_under_prediction": 0.0,
|
|
"hop_over_prediction": 0.0,
|
|
"mean_score_margin": 234.29880876231567,
|
|
"median_score_margin": 254.67290210723877,
|
|
"negative_margin_rate": 0.011,
|
|
"hop_confusion": {
|
|
"true_2_pred_2": 2000
|
|
},
|
|
"known_mean_need_prob": 0.9999999997019767,
|
|
"unknown_mean_need_prob": null,
|
|
"thr0.50": {
|
|
"need_precision": 1.0,
|
|
"need_recall": 1.0,
|
|
"need_f1": 1.0,
|
|
"specificity_unknown_refusal": null,
|
|
"abstention_accuracy": 1.0,
|
|
"known_question_refusal_rate": 0.0,
|
|
"unknown_question_read_rate": null,
|
|
"counts": {
|
|
"tp": 2000,
|
|
"tn": 0,
|
|
"fp": 0,
|
|
"fn": 0
|
|
}
|
|
}
|
|
},
|
|
"paraphrase": {
|
|
"episodes": 2000,
|
|
"answerable_episodes": 2000,
|
|
"unknown_episodes": 0,
|
|
"top1": 0.997,
|
|
"recall1": 0.997,
|
|
"recall3": 1.0,
|
|
"recall5": 1.0,
|
|
"recall8": 1.0,
|
|
"evidence_count_in_top3": 1.0,
|
|
"hit3": 1.0,
|
|
"mrr": 0.9985,
|
|
"ndcg3": 0.9988927892607143,
|
|
"mean_first_positive_rank": 1.003,
|
|
"all_evidence_in_top3": 1.0,
|
|
"multi_positive_episodes": 0,
|
|
"all_evidence_in_top3_multi": null,
|
|
"hop_accuracy": 1.0,
|
|
"hop_accuracy_answerable": 1.0,
|
|
"hop_under_prediction": 0.0,
|
|
"hop_over_prediction": 0.0,
|
|
"mean_score_margin": 155.67033711057155,
|
|
"median_score_margin": 155.824444770813,
|
|
"negative_margin_rate": 0.003,
|
|
"hop_confusion": {
|
|
"true_1_pred_1": 2000
|
|
},
|
|
"known_mean_need_prob": 1.0,
|
|
"unknown_mean_need_prob": null,
|
|
"thr0.50": {
|
|
"need_precision": 1.0,
|
|
"need_recall": 1.0,
|
|
"need_f1": 1.0,
|
|
"specificity_unknown_refusal": null,
|
|
"abstention_accuracy": 1.0,
|
|
"known_question_refusal_rate": 0.0,
|
|
"unknown_question_read_rate": null,
|
|
"counts": {
|
|
"tp": 2000,
|
|
"tn": 0,
|
|
"fp": 0,
|
|
"fn": 0
|
|
}
|
|
}
|
|
},
|
|
"random_position": {
|
|
"episodes": 2000,
|
|
"answerable_episodes": 2000,
|
|
"unknown_episodes": 0,
|
|
"top1": 0.981,
|
|
"recall1": 0.981,
|
|
"recall3": 1.0,
|
|
"recall5": 1.0,
|
|
"recall8": 1.0,
|
|
"evidence_count_in_top3": 1.0,
|
|
"hit3": 1.0,
|
|
"mrr": 0.9905,
|
|
"ndcg3": 0.9929876653178578,
|
|
"mean_first_positive_rank": 1.019,
|
|
"all_evidence_in_top3": 1.0,
|
|
"multi_positive_episodes": 0,
|
|
"all_evidence_in_top3_multi": null,
|
|
"hop_accuracy": 1.0,
|
|
"hop_accuracy_answerable": 1.0,
|
|
"hop_under_prediction": 0.0,
|
|
"hop_over_prediction": 0.0,
|
|
"mean_score_margin": 217.68418576198818,
|
|
"median_score_margin": 236.54215621948242,
|
|
"negative_margin_rate": 0.019,
|
|
"hop_confusion": {
|
|
"true_1_pred_1": 2000
|
|
},
|
|
"known_mean_need_prob": 1.0,
|
|
"unknown_mean_need_prob": null,
|
|
"thr0.50": {
|
|
"need_precision": 1.0,
|
|
"need_recall": 1.0,
|
|
"need_f1": 1.0,
|
|
"specificity_unknown_refusal": null,
|
|
"abstention_accuracy": 1.0,
|
|
"known_question_refusal_rate": 0.0,
|
|
"unknown_question_read_rate": null,
|
|
"counts": {
|
|
"tp": 2000,
|
|
"tn": 0,
|
|
"fp": 0,
|
|
"fn": 0
|
|
}
|
|
}
|
|
},
|
|
"single_fact": {
|
|
"episodes": 2000,
|
|
"answerable_episodes": 2000,
|
|
"unknown_episodes": 0,
|
|
"top1": 0.969,
|
|
"recall1": 0.969,
|
|
"recall3": 1.0,
|
|
"recall5": 1.0,
|
|
"recall8": 1.0,
|
|
"evidence_count_in_top3": 1.0,
|
|
"hit3": 1.0,
|
|
"mrr": 0.98425,
|
|
"ndcg3": 0.988362427730358,
|
|
"mean_first_positive_rank": 1.0325,
|
|
"all_evidence_in_top3": 1.0,
|
|
"multi_positive_episodes": 0,
|
|
"all_evidence_in_top3_multi": null,
|
|
"hop_accuracy": 1.0,
|
|
"hop_accuracy_answerable": 1.0,
|
|
"hop_under_prediction": 0.0,
|
|
"hop_over_prediction": 0.0,
|
|
"mean_score_margin": 214.68668193136529,
|
|
"median_score_margin": 221.31105995178223,
|
|
"negative_margin_rate": 0.031,
|
|
"hop_confusion": {
|
|
"true_1_pred_1": 2000
|
|
},
|
|
"known_mean_need_prob": 1.0,
|
|
"unknown_mean_need_prob": null,
|
|
"thr0.50": {
|
|
"need_precision": 1.0,
|
|
"need_recall": 1.0,
|
|
"need_f1": 1.0,
|
|
"specificity_unknown_refusal": null,
|
|
"abstention_accuracy": 1.0,
|
|
"known_question_refusal_rate": 0.0,
|
|
"unknown_question_read_rate": null,
|
|
"counts": {
|
|
"tp": 2000,
|
|
"tn": 0,
|
|
"fp": 0,
|
|
"fn": 0
|
|
}
|
|
}
|
|
},
|
|
"unknown": {
|
|
"episodes": 1920,
|
|
"answerable_episodes": 1814,
|
|
"unknown_episodes": 106,
|
|
"top1": 0.665380374862183,
|
|
"recall1": 0.665380374862183,
|
|
"recall3": 0.7662624035281147,
|
|
"recall5": 0.8280044101433297,
|
|
"recall8": 0.9178610804851157,
|
|
"evidence_count_in_top3": 0.7662624035281147,
|
|
"hit3": 0.7662624035281147,
|
|
"mrr": 0.745321547104062,
|
|
"ndcg3": 0.7247713833753366,
|
|
"mean_first_positive_rank": 2.7304299889746417,
|
|
"all_evidence_in_top3": 0.7662624035281147,
|
|
"multi_positive_episodes": 0,
|
|
"all_evidence_in_top3_multi": null,
|
|
"hop_accuracy": 1.0,
|
|
"hop_accuracy_answerable": 1.0,
|
|
"hop_under_prediction": 0.0,
|
|
"hop_over_prediction": 0.0,
|
|
"mean_score_margin": 49.73213848704934,
|
|
"median_score_margin": 64.13029479980469,
|
|
"negative_margin_rate": 0.33461962513781696,
|
|
"hop_confusion": {
|
|
"true_0_pred_0": 106,
|
|
"true_0_pred_1": 0,
|
|
"true_1_pred_0": 0,
|
|
"true_1_pred_1": 1814
|
|
},
|
|
"known_mean_need_prob": 0.9999999804822718,
|
|
"unknown_mean_need_prob": 0.0001509735806525834,
|
|
"thr0.50": {
|
|
"need_precision": 1.0,
|
|
"need_recall": 1.0,
|
|
"need_f1": 1.0,
|
|
"specificity_unknown_refusal": 1.0,
|
|
"abstention_accuracy": 1.0,
|
|
"known_question_refusal_rate": 0.0,
|
|
"unknown_question_read_rate": 0.0,
|
|
"counts": {
|
|
"tp": 1814,
|
|
"tn": 106,
|
|
"fp": 0,
|
|
"fn": 0
|
|
}
|
|
}
|
|
},
|
|
"unknown_abstention": {
|
|
"episodes": 2000,
|
|
"answerable_episodes": 0,
|
|
"unknown_episodes": 2000,
|
|
"top1": null,
|
|
"recall1": null,
|
|
"recall3": null,
|
|
"recall5": null,
|
|
"recall8": null,
|
|
"evidence_count_in_top3": null,
|
|
"hit3": null,
|
|
"mrr": null,
|
|
"ndcg3": null,
|
|
"mean_first_positive_rank": null,
|
|
"all_evidence_in_top3": null,
|
|
"multi_positive_episodes": 0,
|
|
"all_evidence_in_top3_multi": null,
|
|
"hop_accuracy": 1.0,
|
|
"hop_accuracy_answerable": null,
|
|
"hop_under_prediction": null,
|
|
"hop_over_prediction": null,
|
|
"mean_score_margin": null,
|
|
"median_score_margin": null,
|
|
"negative_margin_rate": null,
|
|
"hop_confusion": {
|
|
"true_0_pred_0": 2000
|
|
},
|
|
"known_mean_need_prob": null,
|
|
"unknown_mean_need_prob": 2.0705206352511507e-15,
|
|
"thr0.50": {
|
|
"need_precision": null,
|
|
"need_recall": null,
|
|
"need_f1": null,
|
|
"specificity_unknown_refusal": 1.0,
|
|
"abstention_accuracy": 1.0,
|
|
"known_question_refusal_rate": null,
|
|
"unknown_question_read_rate": 0.0,
|
|
"counts": {
|
|
"tp": 0,
|
|
"tn": 2000,
|
|
"fp": 0,
|
|
"fn": 0
|
|
}
|
|
}
|
|
}
|
|
}
|
|
}
|
|
} |