{ "n": 512, "steps": { "2": { "vs_teacher_l2": 0.4342995882034302, "vs_32step_l2": 0.30502474308013916, "vs_1step_l2": 4.2227020263671875 }, "3": { "vs_teacher_l2": 0.43872418999671936, "vs_32step_l2": 0.29536527395248413, "vs_1step_l2": 4.216776371002197 }, "4": { "vs_teacher_l2": 0.42935511469841003, "vs_32step_l2": 0.29940515756607056, "vs_1step_l2": 4.232732772827148 }, "8": { "vs_teacher_l2": 0.4444255232810974, "vs_32step_l2": 0.2452448457479477, "vs_1step_l2": 4.195967197418213 }, "16": { "vs_teacher_l2": 0.5166380405426025, "vs_32step_l2": 0.12673738598823547, "vs_1step_l2": 4.085569381713867 }, "32": { "vs_teacher_l2": 0.6008902192115784, "vs_32step_l2": 0.0, "vs_1step_l2": 3.970090866088867 } }, "vs_teacher": { "a1_1step": { "l2_mean": 4.325636386871338, "l2_p95": 4.9564337730407715, "per_axis_norm": [ 0.3975203335285187, 0.36649173498153687, 0.3425973951816559, 0.3447648882865906, 0.3368167281150818, 0.32962310314178467 ] }, "aU_32step_rneq_t": { "l2_mean": 0.6008902192115784, "l2_p95": 1.9733507633209229, "per_axis_norm": [ 0.05318881571292877, 0.04853545129299164, 0.04775547981262207, 0.04531646892428398, 0.0381026417016983, 0.034751877188682556 ] }, "aF_flow_map_r_eq_t": { "l2_mean": 0.6651337146759033, "l2_p95": 1.9794291257858276, "per_axis_norm": [ 0.06039007008075714, 0.05440700799226761, 0.05297258496284485, 0.05065540224313736, 0.04315900802612305, 0.03988185524940491 ] } }, "mutual": [ { "l2_mean": 3.970090866088867, "pair": [ "a1", "aU_32" ] }, { "l2_mean": 3.891127109527588, "pair": [ "a1", "aF_flow" ] }, { "l2_mean": 0.08254596590995789, "pair": [ "aU_32", "aF_flow" ] } ], "note": "动作块是 ±1 归一化的 16×6;l2_mean 无单位权重,越小越贴老师" }