Files
natural-memory-nm21/zero_overlap_baseline.json
T
WpyQwq 643e22ecb9 Natural Memory NM2.1: 记忆路由器分叉、数据集缺陷修复与全轴评测证据
- 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
2026-09-19 11:11:31 +08:00

812 lines
26 KiB
JSON

{
"V2-128 v6 best": {
"checkpoint": "checkpoints\\router_v6_v2_128\\router_best.pt",
"router": {
"kind": "router_v2",
"step": 6000,
"arch_config": {
"arch": "router_v2",
"hidden_size": 2560,
"router_dim": 128,
"num_heads": 8,
"max_hops": 3
},
"parameters": 2037774,
"checkpoint_bytes": 24472973
},
"storage": {
"address_bytes_per_record": 512,
"address_mb_per_1m_records": 512,
"checkpoint_bytes": 24472973
},
"latency": {
"cuda": {
"single_query_latency_ms_mean": 0.8666539997405683,
"single_query_latency_ms_p50": 0.844600006530527,
"single_query_latency_ms_p95": 1.1691999970935285,
"queries_per_second": 1153.8630183433627,
"samples": 150,
"batched_qps_64": 61288.5926494177,
"batched_latency_ms_64": 0.016316250003001187,
"batched_qps_256": 273229.8761194092,
"batched_latency_ms_256": 0.0036599218731225847
},
"cpu": {
"single_query_latency_ms_mean": 0.33684533375587006,
"single_query_latency_ms_p50": 0.3316000002087094,
"single_query_latency_ms_p95": 0.46629999997094274,
"queries_per_second": 2968.721546027869,
"samples": 150,
"batched_qps_64": 15424.808876649771,
"batched_latency_ms_64": 0.06483062500137748,
"batched_qps_256": 17376.784343192005,
"batched_latency_ms_256": 0.057548046879674075
}
},
"metrics": {
"episodes": 300,
"answerable_episodes": 250,
"unknown_episodes": 50,
"top1": 0.076,
"recall1": 0.076,
"recall3": 0.236,
"recall5": 0.372,
"recall8": 0.488,
"evidence_count_in_top3": 0.236,
"hit3": 0.236,
"mrr": 0.22881929448111302,
"ndcg3": 0.1664743802857166,
"mean_first_positive_rank": 9.448,
"all_evidence_in_top3": 0.236,
"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": -7.889600224852562,
"median_score_margin": -7.62301778793335,
"negative_margin_rate": 0.924,
"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": 0.9972750458717347,
"unknown_mean_need_prob": 0.9953130459785462,
"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": 0.076,
"recall1": 0.076,
"recall3": 0.236,
"recall5": 0.372,
"recall8": 0.488,
"evidence_count_in_top3": 0.236,
"hit3": 0.236,
"mrr": 0.22881929448111302,
"ndcg3": 0.1664743802857166,
"mean_first_positive_rank": 9.448,
"all_evidence_in_top3": 0.236,
"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": -7.889600224852562,
"median_score_margin": -7.62301778793335,
"negative_margin_rate": 0.924,
"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": 0.9972750458717347,
"unknown_mean_need_prob": 0.9953130459785462,
"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": 0.076,
"recall1": 0.076,
"recall3": 0.236,
"recall5": 0.372,
"recall8": 0.488,
"evidence_count_in_top3": 0.236,
"hit3": 0.236,
"mrr": 0.22881929448111302,
"ndcg3": 0.1664743802857166,
"mean_first_positive_rank": 9.448,
"all_evidence_in_top3": 0.236,
"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": -7.889600224852562,
"median_score_margin": -7.62301778793335,
"negative_margin_rate": 0.924,
"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": 0.9972750458717347,
"unknown_mean_need_prob": 0.9953130459785462,
"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
}
}
}
}
},
"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": 0.8889626660675276,
"single_query_latency_ms_p50": 0.8526999954483472,
"single_query_latency_ms_p95": 1.1702999981935136,
"queries_per_second": 1124.9066335076413,
"samples": 150,
"batched_qps_64": 75940.95592518139,
"batched_latency_ms_64": 0.013168124996809638,
"batched_qps_256": 296056.43592556706,
"batched_latency_ms_256": 0.0033777343730889697
},
"cpu": {
"single_query_latency_ms_mean": 0.3601646669267211,
"single_query_latency_ms_p50": 0.35260000004200265,
"single_query_latency_ms_p95": 0.5258000019239262,
"queries_per_second": 2776.5077805465558,
"samples": 150,
"batched_qps_64": 14627.50153387338,
"batched_latency_ms_64": 0.06836437498805026,
"batched_qps_256": 17832.662752528264,
"batched_latency_ms_256": 0.05607687499491476
}
},
"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
}
}
}
}
},
"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": 0.9203346665890422,
"single_query_latency_ms_p50": 0.9079000010387972,
"single_query_latency_ms_p95": 1.215000003867317,
"queries_per_second": 1086.561265486407,
"samples": 150,
"batched_qps_64": 71565.95256269597,
"batched_latency_ms_64": 0.013973124987387564,
"batched_qps_256": 257898.13019024945,
"batched_latency_ms_256": 0.003877500000726286
},
"cpu": {
"single_query_latency_ms_mean": 0.36491733257814,
"single_query_latency_ms_p50": 0.338099998771213,
"single_query_latency_ms_p95": 0.5211000025155954,
"queries_per_second": 2740.3466777941253,
"samples": 150,
"batched_qps_64": 15065.77151150895,
"batched_latency_ms_64": 0.06637562498781335,
"batched_qps_256": 15734.809454760149,
"batched_latency_ms_256": 0.06355335937655582
}
},
"metrics": {
"episodes": 300,
"answerable_episodes": 250,
"unknown_episodes": 50,
"top1": 0.116,
"recall1": 0.116,
"recall3": 0.32,
"recall5": 0.52,
"recall8": 0.64,
"evidence_count_in_top3": 0.32,
"hit3": 0.32,
"mrr": 0.2947678317950427,
"ndcg3": 0.23475900845714656,
"mean_first_positive_rank": 7.472,
"all_evidence_in_top3": 0.32,
"multi_positive_episodes": 0,
"all_evidence_in_top3_multi": null,
"hop_accuracy": 0.6233333333333333,
"hop_accuracy_answerable": 0.708,
"hop_under_prediction": 0.228,
"hop_over_prediction": 0.064,
"mean_score_margin": -22.876055246174335,
"median_score_margin": -17.429455399513245,
"negative_margin_rate": 0.884,
"hop_confusion": {
"true_0_pred_0": 10,
"true_0_pred_1": 39,
"true_0_pred_2": 1,
"true_1_pred_0": 57,
"true_1_pred_1": 177,
"true_1_pred_2": 16
},
"known_mean_need_prob": 0.7523725980222226,
"unknown_mean_need_prob": 0.8253701922297477,
"thr0.30": {
"need_precision": 0.8185328185328186,
"need_recall": 0.848,
"need_f1": 0.8330058939096268,
"specificity_unknown_refusal": 0.06,
"abstention_accuracy": 0.7166666666666667,
"known_question_refusal_rate": 0.152,
"unknown_question_read_rate": 0.94,
"counts": {
"tp": 212,
"tn": 3,
"fp": 47,
"fn": 38
}
},
"thr0.40": {
"need_precision": 0.8185328185328186,
"need_recall": 0.848,
"need_f1": 0.8330058939096268,
"specificity_unknown_refusal": 0.06,
"abstention_accuracy": 0.7166666666666667,
"known_question_refusal_rate": 0.152,
"unknown_question_read_rate": 0.94,
"counts": {
"tp": 212,
"tn": 3,
"fp": 47,
"fn": 38
}
},
"thr0.50": {
"need_precision": 0.8185328185328186,
"need_recall": 0.848,
"need_f1": 0.8330058939096268,
"specificity_unknown_refusal": 0.06,
"abstention_accuracy": 0.7166666666666667,
"known_question_refusal_rate": 0.152,
"unknown_question_read_rate": 0.94,
"counts": {
"tp": 212,
"tn": 3,
"fp": 47,
"fn": 38
}
},
"thr0.60": {
"need_precision": 0.8185328185328186,
"need_recall": 0.848,
"need_f1": 0.8330058939096268,
"specificity_unknown_refusal": 0.06,
"abstention_accuracy": 0.7166666666666667,
"known_question_refusal_rate": 0.152,
"unknown_question_read_rate": 0.94,
"counts": {
"tp": 212,
"tn": 3,
"fp": 47,
"fn": 38
}
},
"thr0.70": {
"need_precision": 0.8025751072961373,
"need_recall": 0.748,
"need_f1": 0.7743271221532091,
"specificity_unknown_refusal": 0.08,
"abstention_accuracy": 0.6366666666666667,
"known_question_refusal_rate": 0.252,
"unknown_question_read_rate": 0.92,
"counts": {
"tp": 187,
"tn": 4,
"fp": 46,
"fn": 63
}
},
"thr0.80": {
"need_precision": 0.815028901734104,
"need_recall": 0.564,
"need_f1": 0.6666666666666666,
"specificity_unknown_refusal": 0.36,
"abstention_accuracy": 0.53,
"known_question_refusal_rate": 0.436,
"unknown_question_read_rate": 0.64,
"counts": {
"tp": 141,
"tn": 18,
"fp": 32,
"fn": 109
}
}
},
"by_family": {
"zero_overlap_paraphrase": {
"episodes": 300,
"answerable_episodes": 250,
"unknown_episodes": 50,
"top1": 0.116,
"recall1": 0.116,
"recall3": 0.32,
"recall5": 0.52,
"recall8": 0.64,
"evidence_count_in_top3": 0.32,
"hit3": 0.32,
"mrr": 0.2947678317950427,
"ndcg3": 0.23475900845714656,
"mean_first_positive_rank": 7.472,
"all_evidence_in_top3": 0.32,
"multi_positive_episodes": 0,
"all_evidence_in_top3_multi": null,
"hop_accuracy": 0.6233333333333333,
"hop_accuracy_answerable": 0.708,
"hop_under_prediction": 0.228,
"hop_over_prediction": 0.064,
"mean_score_margin": -22.876055246174335,
"median_score_margin": -17.429455399513245,
"negative_margin_rate": 0.884,
"hop_confusion": {
"true_0_pred_0": 10,
"true_0_pred_1": 39,
"true_0_pred_2": 1,
"true_1_pred_0": 57,
"true_1_pred_1": 177,
"true_1_pred_2": 16
},
"known_mean_need_prob": 0.7523725980222226,
"unknown_mean_need_prob": 0.8253701922297477,
"thr0.50": {
"need_precision": 0.8185328185328186,
"need_recall": 0.848,
"need_f1": 0.8330058939096268,
"specificity_unknown_refusal": 0.06,
"abstention_accuracy": 0.7166666666666667,
"known_question_refusal_rate": 0.152,
"unknown_question_read_rate": 0.94,
"counts": {
"tp": 212,
"tn": 3,
"fp": 47,
"fn": 38
}
}
}
},
"by_category": {
"zero_overlap_paraphrase": {
"episodes": 300,
"answerable_episodes": 250,
"unknown_episodes": 50,
"top1": 0.116,
"recall1": 0.116,
"recall3": 0.32,
"recall5": 0.52,
"recall8": 0.64,
"evidence_count_in_top3": 0.32,
"hit3": 0.32,
"mrr": 0.2947678317950427,
"ndcg3": 0.23475900845714656,
"mean_first_positive_rank": 7.472,
"all_evidence_in_top3": 0.32,
"multi_positive_episodes": 0,
"all_evidence_in_top3_multi": null,
"hop_accuracy": 0.6233333333333333,
"hop_accuracy_answerable": 0.708,
"hop_under_prediction": 0.228,
"hop_over_prediction": 0.064,
"mean_score_margin": -22.876055246174335,
"median_score_margin": -17.429455399513245,
"negative_margin_rate": 0.884,
"hop_confusion": {
"true_0_pred_0": 10,
"true_0_pred_1": 39,
"true_0_pred_2": 1,
"true_1_pred_0": 57,
"true_1_pred_1": 177,
"true_1_pred_2": 16
},
"known_mean_need_prob": 0.7523725980222226,
"unknown_mean_need_prob": 0.8253701922297477,
"thr0.50": {
"need_precision": 0.8185328185328186,
"need_recall": 0.848,
"need_f1": 0.8330058939096268,
"specificity_unknown_refusal": 0.06,
"abstention_accuracy": 0.7166666666666667,
"known_question_refusal_rate": 0.152,
"unknown_question_read_rate": 0.94,
"counts": {
"tp": 212,
"tn": 3,
"fp": 47,
"fn": 38
}
}
}
}
}
}