- 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
806 lines
25 KiB
JSON
806 lines
25 KiB
JSON
{
|
|
"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": 0.9000799997981327,
|
|
"single_query_latency_ms_p50": 0.8559999987483025,
|
|
"single_query_latency_ms_p95": 1.2088000003132038,
|
|
"queries_per_second": 1111.0123547065562,
|
|
"samples": 150,
|
|
"batched_qps_64": 63459.326517561116,
|
|
"batched_latency_ms_64": 0.015758125005049806,
|
|
"batched_qps_256": 250582.40841407402,
|
|
"batched_latency_ms_256": 0.003990703123690764
|
|
},
|
|
"cpu": {
|
|
"single_query_latency_ms_mean": 0.3485613332304638,
|
|
"single_query_latency_ms_p50": 0.34169999707955867,
|
|
"single_query_latency_ms_p95": 0.5107999968458898,
|
|
"queries_per_second": 2868.935549253291,
|
|
"samples": 150,
|
|
"batched_qps_64": 14958.653349059248,
|
|
"batched_latency_ms_64": 0.0668509374918358,
|
|
"batched_qps_256": 17243.028361064185,
|
|
"batched_latency_ms_256": 0.057994453123910716
|
|
}
|
|
},
|
|
"metrics": {
|
|
"episodes": 300,
|
|
"answerable_episodes": 250,
|
|
"unknown_episodes": 50,
|
|
"top1": 0.184,
|
|
"recall1": 0.184,
|
|
"recall3": 0.36,
|
|
"recall5": 0.604,
|
|
"recall8": 0.708,
|
|
"evidence_count_in_top3": 0.36,
|
|
"hit3": 0.36,
|
|
"mrr": 0.3507031323442415,
|
|
"ndcg3": 0.28404553732857407,
|
|
"mean_first_positive_rank": 6.792,
|
|
"all_evidence_in_top3": 0.36,
|
|
"multi_positive_episodes": 0,
|
|
"all_evidence_in_top3_multi": null,
|
|
"hop_accuracy": 0.8333333333333334,
|
|
"hop_accuracy_answerable": 1.0,
|
|
"hop_under_prediction": 0.0,
|
|
"hop_over_prediction": 0.0,
|
|
"mean_score_margin": -18.688180733680724,
|
|
"median_score_margin": -16.592705011367798,
|
|
"negative_margin_rate": 0.816,
|
|
"hop_confusion": {
|
|
"true_0_pred_0": 0,
|
|
"true_0_pred_1": 50,
|
|
"true_1_pred_0": 0,
|
|
"true_1_pred_1": 250
|
|
},
|
|
"known_mean_need_prob": 0.9918208360671997,
|
|
"unknown_mean_need_prob": 0.9822168827056885,
|
|
"thr0.30": {
|
|
"need_precision": 0.8333333333333334,
|
|
"need_recall": 1.0,
|
|
"need_f1": 0.9090909090909091,
|
|
"specificity_unknown_refusal": 0.0,
|
|
"abstention_accuracy": 0.8333333333333334,
|
|
"known_question_refusal_rate": 0.0,
|
|
"unknown_question_read_rate": 1.0,
|
|
"counts": {
|
|
"tp": 250,
|
|
"tn": 0,
|
|
"fp": 50,
|
|
"fn": 0
|
|
}
|
|
},
|
|
"thr0.40": {
|
|
"need_precision": 0.8333333333333334,
|
|
"need_recall": 1.0,
|
|
"need_f1": 0.9090909090909091,
|
|
"specificity_unknown_refusal": 0.0,
|
|
"abstention_accuracy": 0.8333333333333334,
|
|
"known_question_refusal_rate": 0.0,
|
|
"unknown_question_read_rate": 1.0,
|
|
"counts": {
|
|
"tp": 250,
|
|
"tn": 0,
|
|
"fp": 50,
|
|
"fn": 0
|
|
}
|
|
},
|
|
"thr0.50": {
|
|
"need_precision": 0.8333333333333334,
|
|
"need_recall": 1.0,
|
|
"need_f1": 0.9090909090909091,
|
|
"specificity_unknown_refusal": 0.0,
|
|
"abstention_accuracy": 0.8333333333333334,
|
|
"known_question_refusal_rate": 0.0,
|
|
"unknown_question_read_rate": 1.0,
|
|
"counts": {
|
|
"tp": 250,
|
|
"tn": 0,
|
|
"fp": 50,
|
|
"fn": 0
|
|
}
|
|
},
|
|
"thr0.60": {
|
|
"need_precision": 0.8333333333333334,
|
|
"need_recall": 1.0,
|
|
"need_f1": 0.9090909090909091,
|
|
"specificity_unknown_refusal": 0.0,
|
|
"abstention_accuracy": 0.8333333333333334,
|
|
"known_question_refusal_rate": 0.0,
|
|
"unknown_question_read_rate": 1.0,
|
|
"counts": {
|
|
"tp": 250,
|
|
"tn": 0,
|
|
"fp": 50,
|
|
"fn": 0
|
|
}
|
|
},
|
|
"thr0.70": {
|
|
"need_precision": 0.8333333333333334,
|
|
"need_recall": 1.0,
|
|
"need_f1": 0.9090909090909091,
|
|
"specificity_unknown_refusal": 0.0,
|
|
"abstention_accuracy": 0.8333333333333334,
|
|
"known_question_refusal_rate": 0.0,
|
|
"unknown_question_read_rate": 1.0,
|
|
"counts": {
|
|
"tp": 250,
|
|
"tn": 0,
|
|
"fp": 50,
|
|
"fn": 0
|
|
}
|
|
},
|
|
"thr0.80": {
|
|
"need_precision": 0.8333333333333334,
|
|
"need_recall": 1.0,
|
|
"need_f1": 0.9090909090909091,
|
|
"specificity_unknown_refusal": 0.0,
|
|
"abstention_accuracy": 0.8333333333333334,
|
|
"known_question_refusal_rate": 0.0,
|
|
"unknown_question_read_rate": 1.0,
|
|
"counts": {
|
|
"tp": 250,
|
|
"tn": 0,
|
|
"fp": 50,
|
|
"fn": 0
|
|
}
|
|
}
|
|
},
|
|
"by_family": {
|
|
"zero_overlap_paraphrase": {
|
|
"episodes": 300,
|
|
"answerable_episodes": 250,
|
|
"unknown_episodes": 50,
|
|
"top1": 0.184,
|
|
"recall1": 0.184,
|
|
"recall3": 0.36,
|
|
"recall5": 0.604,
|
|
"recall8": 0.708,
|
|
"evidence_count_in_top3": 0.36,
|
|
"hit3": 0.36,
|
|
"mrr": 0.3507031323442415,
|
|
"ndcg3": 0.28404553732857407,
|
|
"mean_first_positive_rank": 6.792,
|
|
"all_evidence_in_top3": 0.36,
|
|
"multi_positive_episodes": 0,
|
|
"all_evidence_in_top3_multi": null,
|
|
"hop_accuracy": 0.8333333333333334,
|
|
"hop_accuracy_answerable": 1.0,
|
|
"hop_under_prediction": 0.0,
|
|
"hop_over_prediction": 0.0,
|
|
"mean_score_margin": -18.688180733680724,
|
|
"median_score_margin": -16.592705011367798,
|
|
"negative_margin_rate": 0.816,
|
|
"hop_confusion": {
|
|
"true_0_pred_0": 0,
|
|
"true_0_pred_1": 50,
|
|
"true_1_pred_0": 0,
|
|
"true_1_pred_1": 250
|
|
},
|
|
"known_mean_need_prob": 0.9918208360671997,
|
|
"unknown_mean_need_prob": 0.9822168827056885,
|
|
"thr0.50": {
|
|
"need_precision": 0.8333333333333334,
|
|
"need_recall": 1.0,
|
|
"need_f1": 0.9090909090909091,
|
|
"specificity_unknown_refusal": 0.0,
|
|
"abstention_accuracy": 0.8333333333333334,
|
|
"known_question_refusal_rate": 0.0,
|
|
"unknown_question_read_rate": 1.0,
|
|
"counts": {
|
|
"tp": 250,
|
|
"tn": 0,
|
|
"fp": 50,
|
|
"fn": 0
|
|
}
|
|
}
|
|
}
|
|
},
|
|
"by_category": {
|
|
"zero_overlap_paraphrase": {
|
|
"episodes": 300,
|
|
"answerable_episodes": 250,
|
|
"unknown_episodes": 50,
|
|
"top1": 0.184,
|
|
"recall1": 0.184,
|
|
"recall3": 0.36,
|
|
"recall5": 0.604,
|
|
"recall8": 0.708,
|
|
"evidence_count_in_top3": 0.36,
|
|
"hit3": 0.36,
|
|
"mrr": 0.3507031323442415,
|
|
"ndcg3": 0.28404553732857407,
|
|
"mean_first_positive_rank": 6.792,
|
|
"all_evidence_in_top3": 0.36,
|
|
"multi_positive_episodes": 0,
|
|
"all_evidence_in_top3_multi": null,
|
|
"hop_accuracy": 0.8333333333333334,
|
|
"hop_accuracy_answerable": 1.0,
|
|
"hop_under_prediction": 0.0,
|
|
"hop_over_prediction": 0.0,
|
|
"mean_score_margin": -18.688180733680724,
|
|
"median_score_margin": -16.592705011367798,
|
|
"negative_margin_rate": 0.816,
|
|
"hop_confusion": {
|
|
"true_0_pred_0": 0,
|
|
"true_0_pred_1": 50,
|
|
"true_1_pred_0": 0,
|
|
"true_1_pred_1": 250
|
|
},
|
|
"known_mean_need_prob": 0.9918208360671997,
|
|
"unknown_mean_need_prob": 0.9822168827056885,
|
|
"thr0.50": {
|
|
"need_precision": 0.8333333333333334,
|
|
"need_recall": 1.0,
|
|
"need_f1": 0.9090909090909091,
|
|
"specificity_unknown_refusal": 0.0,
|
|
"abstention_accuracy": 0.8333333333333334,
|
|
"known_question_refusal_rate": 0.0,
|
|
"unknown_question_read_rate": 1.0,
|
|
"counts": {
|
|
"tp": 250,
|
|
"tn": 0,
|
|
"fp": 50,
|
|
"fn": 0
|
|
}
|
|
}
|
|
}
|
|
}
|
|
},
|
|
"ZOV-128 final": {
|
|
"checkpoint": "checkpoints\\router_zov_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": 0.886220000005172,
|
|
"single_query_latency_ms_p50": 0.8515000008628704,
|
|
"single_query_latency_ms_p95": 1.2022000009892508,
|
|
"queries_per_second": 1128.3879849181512,
|
|
"samples": 150,
|
|
"batched_qps_64": 76974.88699349365,
|
|
"batched_latency_ms_64": 0.012991249991500808,
|
|
"batched_qps_256": 291924.19107650715,
|
|
"batched_latency_ms_256": 0.0034255468733590533
|
|
},
|
|
"cpu": {
|
|
"single_query_latency_ms_mean": 0.35021399982118356,
|
|
"single_query_latency_ms_p50": 0.33540000003995374,
|
|
"single_query_latency_ms_p95": 0.5363000018405728,
|
|
"queries_per_second": 2855.3969873008846,
|
|
"samples": 150,
|
|
"batched_qps_64": 15748.031493531362,
|
|
"batched_latency_ms_64": 0.06350000001020817,
|
|
"batched_qps_256": 15840.858772079004,
|
|
"batched_latency_ms_256": 0.06312789062690172
|
|
}
|
|
},
|
|
"metrics": {
|
|
"episodes": 300,
|
|
"answerable_episodes": 250,
|
|
"unknown_episodes": 50,
|
|
"top1": 0.668,
|
|
"recall1": 0.668,
|
|
"recall3": 0.844,
|
|
"recall5": 0.904,
|
|
"recall8": 0.964,
|
|
"evidence_count_in_top3": 0.844,
|
|
"hit3": 0.844,
|
|
"mrr": 0.7695227994227993,
|
|
"ndcg3": 0.7696166943714315,
|
|
"mean_first_positive_rank": 2.148,
|
|
"all_evidence_in_top3": 0.844,
|
|
"multi_positive_episodes": 0,
|
|
"all_evidence_in_top3_multi": null,
|
|
"hop_accuracy": 0.5833333333333334,
|
|
"hop_accuracy_answerable": 0.62,
|
|
"hop_under_prediction": 0.38,
|
|
"hop_over_prediction": 0.0,
|
|
"mean_score_margin": 3.126375099360943,
|
|
"median_score_margin": 4.672530174255371,
|
|
"negative_margin_rate": 0.332,
|
|
"hop_confusion": {
|
|
"true_0_pred_0": 20,
|
|
"true_0_pred_1": 30,
|
|
"true_1_pred_0": 95,
|
|
"true_1_pred_1": 155
|
|
},
|
|
"known_mean_need_prob": 0.6688127560317516,
|
|
"unknown_mean_need_prob": 0.6399642534554004,
|
|
"thr0.30": {
|
|
"need_precision": 0.844,
|
|
"need_recall": 0.844,
|
|
"need_f1": 0.844,
|
|
"specificity_unknown_refusal": 0.22,
|
|
"abstention_accuracy": 0.74,
|
|
"known_question_refusal_rate": 0.156,
|
|
"unknown_question_read_rate": 0.78,
|
|
"counts": {
|
|
"tp": 211,
|
|
"tn": 11,
|
|
"fp": 39,
|
|
"fn": 39
|
|
}
|
|
},
|
|
"thr0.40": {
|
|
"need_precision": 0.8421052631578947,
|
|
"need_recall": 0.768,
|
|
"need_f1": 0.803347280334728,
|
|
"specificity_unknown_refusal": 0.28,
|
|
"abstention_accuracy": 0.6866666666666666,
|
|
"known_question_refusal_rate": 0.232,
|
|
"unknown_question_read_rate": 0.72,
|
|
"counts": {
|
|
"tp": 192,
|
|
"tn": 14,
|
|
"fp": 36,
|
|
"fn": 58
|
|
}
|
|
},
|
|
"thr0.50": {
|
|
"need_precision": 0.8378378378378378,
|
|
"need_recall": 0.62,
|
|
"need_f1": 0.7126436781609196,
|
|
"specificity_unknown_refusal": 0.4,
|
|
"abstention_accuracy": 0.5833333333333334,
|
|
"known_question_refusal_rate": 0.38,
|
|
"unknown_question_read_rate": 0.6,
|
|
"counts": {
|
|
"tp": 155,
|
|
"tn": 20,
|
|
"fp": 30,
|
|
"fn": 95
|
|
}
|
|
},
|
|
"thr0.60": {
|
|
"need_precision": 0.8378378378378378,
|
|
"need_recall": 0.62,
|
|
"need_f1": 0.7126436781609196,
|
|
"specificity_unknown_refusal": 0.4,
|
|
"abstention_accuracy": 0.5833333333333334,
|
|
"known_question_refusal_rate": 0.38,
|
|
"unknown_question_read_rate": 0.6,
|
|
"counts": {
|
|
"tp": 155,
|
|
"tn": 20,
|
|
"fp": 30,
|
|
"fn": 95
|
|
}
|
|
},
|
|
"thr0.70": {
|
|
"need_precision": 0.8378378378378378,
|
|
"need_recall": 0.62,
|
|
"need_f1": 0.7126436781609196,
|
|
"specificity_unknown_refusal": 0.4,
|
|
"abstention_accuracy": 0.5833333333333334,
|
|
"known_question_refusal_rate": 0.38,
|
|
"unknown_question_read_rate": 0.6,
|
|
"counts": {
|
|
"tp": 155,
|
|
"tn": 20,
|
|
"fp": 30,
|
|
"fn": 95
|
|
}
|
|
},
|
|
"thr0.80": {
|
|
"need_precision": 0.8417721518987342,
|
|
"need_recall": 0.532,
|
|
"need_f1": 0.6519607843137255,
|
|
"specificity_unknown_refusal": 0.5,
|
|
"abstention_accuracy": 0.5266666666666666,
|
|
"known_question_refusal_rate": 0.468,
|
|
"unknown_question_read_rate": 0.5,
|
|
"counts": {
|
|
"tp": 133,
|
|
"tn": 25,
|
|
"fp": 25,
|
|
"fn": 117
|
|
}
|
|
}
|
|
},
|
|
"by_family": {
|
|
"zero_overlap_paraphrase": {
|
|
"episodes": 300,
|
|
"answerable_episodes": 250,
|
|
"unknown_episodes": 50,
|
|
"top1": 0.668,
|
|
"recall1": 0.668,
|
|
"recall3": 0.844,
|
|
"recall5": 0.904,
|
|
"recall8": 0.964,
|
|
"evidence_count_in_top3": 0.844,
|
|
"hit3": 0.844,
|
|
"mrr": 0.7695227994227993,
|
|
"ndcg3": 0.7696166943714315,
|
|
"mean_first_positive_rank": 2.148,
|
|
"all_evidence_in_top3": 0.844,
|
|
"multi_positive_episodes": 0,
|
|
"all_evidence_in_top3_multi": null,
|
|
"hop_accuracy": 0.5833333333333334,
|
|
"hop_accuracy_answerable": 0.62,
|
|
"hop_under_prediction": 0.38,
|
|
"hop_over_prediction": 0.0,
|
|
"mean_score_margin": 3.126375099360943,
|
|
"median_score_margin": 4.672530174255371,
|
|
"negative_margin_rate": 0.332,
|
|
"hop_confusion": {
|
|
"true_0_pred_0": 20,
|
|
"true_0_pred_1": 30,
|
|
"true_1_pred_0": 95,
|
|
"true_1_pred_1": 155
|
|
},
|
|
"known_mean_need_prob": 0.6688127560317516,
|
|
"unknown_mean_need_prob": 0.6399642534554004,
|
|
"thr0.50": {
|
|
"need_precision": 0.8378378378378378,
|
|
"need_recall": 0.62,
|
|
"need_f1": 0.7126436781609196,
|
|
"specificity_unknown_refusal": 0.4,
|
|
"abstention_accuracy": 0.5833333333333334,
|
|
"known_question_refusal_rate": 0.38,
|
|
"unknown_question_read_rate": 0.6,
|
|
"counts": {
|
|
"tp": 155,
|
|
"tn": 20,
|
|
"fp": 30,
|
|
"fn": 95
|
|
}
|
|
}
|
|
}
|
|
},
|
|
"by_category": {
|
|
"zero_overlap_paraphrase": {
|
|
"episodes": 300,
|
|
"answerable_episodes": 250,
|
|
"unknown_episodes": 50,
|
|
"top1": 0.668,
|
|
"recall1": 0.668,
|
|
"recall3": 0.844,
|
|
"recall5": 0.904,
|
|
"recall8": 0.964,
|
|
"evidence_count_in_top3": 0.844,
|
|
"hit3": 0.844,
|
|
"mrr": 0.7695227994227993,
|
|
"ndcg3": 0.7696166943714315,
|
|
"mean_first_positive_rank": 2.148,
|
|
"all_evidence_in_top3": 0.844,
|
|
"multi_positive_episodes": 0,
|
|
"all_evidence_in_top3_multi": null,
|
|
"hop_accuracy": 0.5833333333333334,
|
|
"hop_accuracy_answerable": 0.62,
|
|
"hop_under_prediction": 0.38,
|
|
"hop_over_prediction": 0.0,
|
|
"mean_score_margin": 3.126375099360943,
|
|
"median_score_margin": 4.672530174255371,
|
|
"negative_margin_rate": 0.332,
|
|
"hop_confusion": {
|
|
"true_0_pred_0": 20,
|
|
"true_0_pred_1": 30,
|
|
"true_1_pred_0": 95,
|
|
"true_1_pred_1": 155
|
|
},
|
|
"known_mean_need_prob": 0.6688127560317516,
|
|
"unknown_mean_need_prob": 0.6399642534554004,
|
|
"thr0.50": {
|
|
"need_precision": 0.8378378378378378,
|
|
"need_recall": 0.62,
|
|
"need_f1": 0.7126436781609196,
|
|
"specificity_unknown_refusal": 0.4,
|
|
"abstention_accuracy": 0.5833333333333334,
|
|
"known_question_refusal_rate": 0.38,
|
|
"unknown_question_read_rate": 0.6,
|
|
"counts": {
|
|
"tp": 155,
|
|
"tn": 20,
|
|
"fp": 30,
|
|
"fn": 95
|
|
}
|
|
}
|
|
}
|
|
}
|
|
},
|
|
"REPLAY-128 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": 0.8708980005758349,
|
|
"single_query_latency_ms_p50": 0.8460000026389025,
|
|
"single_query_latency_ms_p95": 1.0751000008895062,
|
|
"queries_per_second": 1148.2400916511501,
|
|
"samples": 150,
|
|
"batched_qps_64": 72823.26707561837,
|
|
"batched_latency_ms_64": 0.01373187499211781,
|
|
"batched_qps_256": 234140.6306666912,
|
|
"batched_latency_ms_256": 0.004270937500905347
|
|
},
|
|
"cpu": {
|
|
"single_query_latency_ms_mean": 0.3600686669233255,
|
|
"single_query_latency_ms_p50": 0.3471999953035265,
|
|
"single_query_latency_ms_p95": 0.4761000018334016,
|
|
"queries_per_second": 2777.2480414491165,
|
|
"samples": 150,
|
|
"batched_qps_64": 14568.101321990402,
|
|
"batched_latency_ms_64": 0.0686431249960151,
|
|
"batched_qps_256": 16973.44848648407,
|
|
"batched_latency_ms_256": 0.05891554687877942
|
|
}
|
|
},
|
|
"metrics": {
|
|
"episodes": 300,
|
|
"answerable_episodes": 250,
|
|
"unknown_episodes": 50,
|
|
"top1": 0.596,
|
|
"recall1": 0.596,
|
|
"recall3": 0.832,
|
|
"recall5": 0.908,
|
|
"recall8": 0.968,
|
|
"evidence_count_in_top3": 0.832,
|
|
"hit3": 0.832,
|
|
"mrr": 0.7306051948051948,
|
|
"ndcg3": 0.7344250415571473,
|
|
"mean_first_positive_rank": 2.2,
|
|
"all_evidence_in_top3": 0.832,
|
|
"multi_positive_episodes": 0,
|
|
"all_evidence_in_top3_multi": null,
|
|
"hop_accuracy": 0.8333333333333334,
|
|
"hop_accuracy_answerable": 1.0,
|
|
"hop_under_prediction": 0.0,
|
|
"hop_over_prediction": 0.0,
|
|
"mean_score_margin": 1.5204011157155037,
|
|
"median_score_margin": 2.312741279602051,
|
|
"negative_margin_rate": 0.404,
|
|
"hop_confusion": {
|
|
"true_0_pred_0": 0,
|
|
"true_0_pred_1": 50,
|
|
"true_1_pred_0": 0,
|
|
"true_1_pred_1": 250
|
|
},
|
|
"known_mean_need_prob": 0.9918208360671997,
|
|
"unknown_mean_need_prob": 0.9822168827056885,
|
|
"thr0.30": {
|
|
"need_precision": 0.8333333333333334,
|
|
"need_recall": 1.0,
|
|
"need_f1": 0.9090909090909091,
|
|
"specificity_unknown_refusal": 0.0,
|
|
"abstention_accuracy": 0.8333333333333334,
|
|
"known_question_refusal_rate": 0.0,
|
|
"unknown_question_read_rate": 1.0,
|
|
"counts": {
|
|
"tp": 250,
|
|
"tn": 0,
|
|
"fp": 50,
|
|
"fn": 0
|
|
}
|
|
},
|
|
"thr0.40": {
|
|
"need_precision": 0.8333333333333334,
|
|
"need_recall": 1.0,
|
|
"need_f1": 0.9090909090909091,
|
|
"specificity_unknown_refusal": 0.0,
|
|
"abstention_accuracy": 0.8333333333333334,
|
|
"known_question_refusal_rate": 0.0,
|
|
"unknown_question_read_rate": 1.0,
|
|
"counts": {
|
|
"tp": 250,
|
|
"tn": 0,
|
|
"fp": 50,
|
|
"fn": 0
|
|
}
|
|
},
|
|
"thr0.50": {
|
|
"need_precision": 0.8333333333333334,
|
|
"need_recall": 1.0,
|
|
"need_f1": 0.9090909090909091,
|
|
"specificity_unknown_refusal": 0.0,
|
|
"abstention_accuracy": 0.8333333333333334,
|
|
"known_question_refusal_rate": 0.0,
|
|
"unknown_question_read_rate": 1.0,
|
|
"counts": {
|
|
"tp": 250,
|
|
"tn": 0,
|
|
"fp": 50,
|
|
"fn": 0
|
|
}
|
|
},
|
|
"thr0.60": {
|
|
"need_precision": 0.8333333333333334,
|
|
"need_recall": 1.0,
|
|
"need_f1": 0.9090909090909091,
|
|
"specificity_unknown_refusal": 0.0,
|
|
"abstention_accuracy": 0.8333333333333334,
|
|
"known_question_refusal_rate": 0.0,
|
|
"unknown_question_read_rate": 1.0,
|
|
"counts": {
|
|
"tp": 250,
|
|
"tn": 0,
|
|
"fp": 50,
|
|
"fn": 0
|
|
}
|
|
},
|
|
"thr0.70": {
|
|
"need_precision": 0.8333333333333334,
|
|
"need_recall": 1.0,
|
|
"need_f1": 0.9090909090909091,
|
|
"specificity_unknown_refusal": 0.0,
|
|
"abstention_accuracy": 0.8333333333333334,
|
|
"known_question_refusal_rate": 0.0,
|
|
"unknown_question_read_rate": 1.0,
|
|
"counts": {
|
|
"tp": 250,
|
|
"tn": 0,
|
|
"fp": 50,
|
|
"fn": 0
|
|
}
|
|
},
|
|
"thr0.80": {
|
|
"need_precision": 0.8333333333333334,
|
|
"need_recall": 1.0,
|
|
"need_f1": 0.9090909090909091,
|
|
"specificity_unknown_refusal": 0.0,
|
|
"abstention_accuracy": 0.8333333333333334,
|
|
"known_question_refusal_rate": 0.0,
|
|
"unknown_question_read_rate": 1.0,
|
|
"counts": {
|
|
"tp": 250,
|
|
"tn": 0,
|
|
"fp": 50,
|
|
"fn": 0
|
|
}
|
|
}
|
|
},
|
|
"by_family": {
|
|
"zero_overlap_paraphrase": {
|
|
"episodes": 300,
|
|
"answerable_episodes": 250,
|
|
"unknown_episodes": 50,
|
|
"top1": 0.596,
|
|
"recall1": 0.596,
|
|
"recall3": 0.832,
|
|
"recall5": 0.908,
|
|
"recall8": 0.968,
|
|
"evidence_count_in_top3": 0.832,
|
|
"hit3": 0.832,
|
|
"mrr": 0.7306051948051948,
|
|
"ndcg3": 0.7344250415571473,
|
|
"mean_first_positive_rank": 2.2,
|
|
"all_evidence_in_top3": 0.832,
|
|
"multi_positive_episodes": 0,
|
|
"all_evidence_in_top3_multi": null,
|
|
"hop_accuracy": 0.8333333333333334,
|
|
"hop_accuracy_answerable": 1.0,
|
|
"hop_under_prediction": 0.0,
|
|
"hop_over_prediction": 0.0,
|
|
"mean_score_margin": 1.5204011157155037,
|
|
"median_score_margin": 2.312741279602051,
|
|
"negative_margin_rate": 0.404,
|
|
"hop_confusion": {
|
|
"true_0_pred_0": 0,
|
|
"true_0_pred_1": 50,
|
|
"true_1_pred_0": 0,
|
|
"true_1_pred_1": 250
|
|
},
|
|
"known_mean_need_prob": 0.9918208360671997,
|
|
"unknown_mean_need_prob": 0.9822168827056885,
|
|
"thr0.50": {
|
|
"need_precision": 0.8333333333333334,
|
|
"need_recall": 1.0,
|
|
"need_f1": 0.9090909090909091,
|
|
"specificity_unknown_refusal": 0.0,
|
|
"abstention_accuracy": 0.8333333333333334,
|
|
"known_question_refusal_rate": 0.0,
|
|
"unknown_question_read_rate": 1.0,
|
|
"counts": {
|
|
"tp": 250,
|
|
"tn": 0,
|
|
"fp": 50,
|
|
"fn": 0
|
|
}
|
|
}
|
|
}
|
|
},
|
|
"by_category": {
|
|
"zero_overlap_paraphrase": {
|
|
"episodes": 300,
|
|
"answerable_episodes": 250,
|
|
"unknown_episodes": 50,
|
|
"top1": 0.596,
|
|
"recall1": 0.596,
|
|
"recall3": 0.832,
|
|
"recall5": 0.908,
|
|
"recall8": 0.968,
|
|
"evidence_count_in_top3": 0.832,
|
|
"hit3": 0.832,
|
|
"mrr": 0.7306051948051948,
|
|
"ndcg3": 0.7344250415571473,
|
|
"mean_first_positive_rank": 2.2,
|
|
"all_evidence_in_top3": 0.832,
|
|
"multi_positive_episodes": 0,
|
|
"all_evidence_in_top3_multi": null,
|
|
"hop_accuracy": 0.8333333333333334,
|
|
"hop_accuracy_answerable": 1.0,
|
|
"hop_under_prediction": 0.0,
|
|
"hop_over_prediction": 0.0,
|
|
"mean_score_margin": 1.5204011157155037,
|
|
"median_score_margin": 2.312741279602051,
|
|
"negative_margin_rate": 0.404,
|
|
"hop_confusion": {
|
|
"true_0_pred_0": 0,
|
|
"true_0_pred_1": 50,
|
|
"true_1_pred_0": 0,
|
|
"true_1_pred_1": 250
|
|
},
|
|
"known_mean_need_prob": 0.9918208360671997,
|
|
"unknown_mean_need_prob": 0.9822168827056885,
|
|
"thr0.50": {
|
|
"need_precision": 0.8333333333333334,
|
|
"need_recall": 1.0,
|
|
"need_f1": 0.9090909090909091,
|
|
"specificity_unknown_refusal": 0.0,
|
|
"abstention_accuracy": 0.8333333333333334,
|
|
"known_question_refusal_rate": 0.0,
|
|
"unknown_question_read_rate": 1.0,
|
|
"counts": {
|
|
"tp": 250,
|
|
"tn": 0,
|
|
"fp": 50,
|
|
"fn": 0
|
|
}
|
|
}
|
|
}
|
|
}
|
|
}
|
|
} |