diff --git "a/manifests/merge_step_00004510.json" "b/manifests/merge_step_00004510.json" new file mode 100644--- /dev/null +++ "b/manifests/merge_step_00004510.json" @@ -0,0 +1,4052 @@ +{ + "schema_version": 1, + "format": "comfy_scaled_fp8_merged_text_encoder", + "scale_policy": "preserve_stock_per_tensor_scale", + "base": "/workspace/Ideogram/checkpoints/text_encoders/qwen3vl_8b_fp8_scaled.safetensors", + "base_sha256": "4ba424cf62e51392e4d1a39933e803706f4e823c1065f36aaf149c6453f66bcd", + "adapter": "/workspace/Ideogram/artifacts/phase2_balanced_4000_8state_debanner_s020_v1_checkpoints/step_00004510/adapters.safetensors", + "adapter_sha256": "1f2277f2e045b144719b952e572b4f73b1df0ceb70df39881499cb4f8001d328", + "adapter_global_step": "4510", + "output": "/workspace/Ideogram/artifacts/huggingface_release/s020_v1/qwen3vl_8b_ideogram4_nl_s020_v1_step_00004510_fp8_scaled.safetensors", + "output_bytes": 10588637800, + "output_sha256": "f438aae391f0f41539cd79bceeab551770183842beba12cb106826d284e59b1f", + "output_tensors": 1254, + "merged_projections": 252, + "lora_rank": 64, + "lora_alpha": 64.0, + "elapsed_seconds": 69.229, + "projections": [ + { + "index": 1, + "total": 252, + "projection": "layers.0.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.4207, + "parameters": 50331648, + "reference_delta_l2": 3.134305206230876, + "retained_delta_l2": 2.8551292133960655, + "delta_retention_norm_ratio": 0.910928906259729, + "delta_cosine": 0.7262223187693696, + "relative_weight_rmse": 0.012330467963596634 + }, + { + "index": 2, + "total": 252, + "projection": "layers.0.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3778, + "parameters": 50331648, + "reference_delta_l2": 3.3559825783010786, + "retained_delta_l2": 3.077174281497866, + "delta_retention_norm_ratio": 0.9169220071028035, + "delta_cosine": 0.7297331310220535, + "relative_weight_rmse": 0.013610261463650438 + }, + { + "index": 3, + "total": 252, + "projection": "layers.0.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3768, + "parameters": 50331648, + "reference_delta_l2": 2.973696032745, + "retained_delta_l2": 2.6691274432529677, + "delta_retention_norm_ratio": 0.8975791116044611, + "delta_cosine": 0.7108711588641692, + "relative_weight_rmse": 0.013128998310727108 + }, + { + "index": 4, + "total": 252, + "projection": "layers.0.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0302, + "parameters": 4194304, + "reference_delta_l2": 0.6048105845388408, + "retained_delta_l2": 0.4439177039123404, + "delta_retention_norm_ratio": 0.7339780672833646, + "delta_cosine": 0.561703117976772, + "relative_weight_rmse": 0.009035763327879053 + }, + { + "index": 5, + "total": 252, + "projection": "layers.0.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1289, + "parameters": 16777216, + "reference_delta_l2": 1.9774444147559924, + "retained_delta_l2": 1.8659500759527547, + "delta_retention_norm_ratio": 0.9436169543015975, + "delta_cosine": 0.7631999662132374, + "relative_weight_rmse": 0.013642805559966107 + }, + { + "index": 6, + "total": 252, + "projection": "layers.0.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1275, + "parameters": 16777216, + "reference_delta_l2": 1.229760415708081, + "retained_delta_l2": 0.9582979496337267, + "delta_retention_norm_ratio": 0.7792558106384897, + "delta_cosine": 0.5998720966781603, + "relative_weight_rmse": 0.009831882975672031 + }, + { + "index": 7, + "total": 252, + "projection": "layers.0.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0292, + "parameters": 4194304, + "reference_delta_l2": 0.7982387638421526, + "retained_delta_l2": 0.696152310747279, + "delta_retention_norm_ratio": 0.8721103788501798, + "delta_cosine": 0.6849935283790957, + "relative_weight_rmse": 0.012011812715934966 + }, + { + "index": 8, + "total": 252, + "projection": "layers.1.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.4162, + "parameters": 50331648, + "reference_delta_l2": 2.0383382547629165, + "retained_delta_l2": 1.9029792599323412, + "delta_retention_norm_ratio": 0.9335934580463833, + "delta_cosine": 0.7566971338926844, + "relative_weight_rmse": 0.011992125686370043 + }, + { + "index": 9, + "total": 252, + "projection": "layers.1.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3759, + "parameters": 50331648, + "reference_delta_l2": 3.830813255676563, + "retained_delta_l2": 3.798066137945357, + "delta_retention_norm_ratio": 0.9914516538537397, + "delta_cosine": 0.8457197498298105, + "relative_weight_rmse": 0.014365848252484843 + }, + { + "index": 10, + "total": 252, + "projection": "layers.1.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3762, + "parameters": 50331648, + "reference_delta_l2": 4.251428931776178, + "retained_delta_l2": 4.248668837106354, + "delta_retention_norm_ratio": 0.9993507842388724, + "delta_cosine": 0.9301037952968061, + "relative_weight_rmse": 0.015773663804699575 + }, + { + "index": 11, + "total": 252, + "projection": "layers.1.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0304, + "parameters": 4194304, + "reference_delta_l2": 0.6779092203442499, + "retained_delta_l2": 0.557132557038717, + "delta_retention_norm_ratio": 0.821839473957588, + "delta_cosine": 0.637988161346813, + "relative_weight_rmse": 0.01083597006662115 + }, + { + "index": 12, + "total": 252, + "projection": "layers.1.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1343, + "parameters": 16777216, + "reference_delta_l2": 1.719155842800786, + "retained_delta_l2": 1.5394124289148676, + "delta_retention_norm_ratio": 0.895446701566574, + "delta_cosine": 0.7061998273184573, + "relative_weight_rmse": 0.012686407067353635 + }, + { + "index": 13, + "total": 252, + "projection": "layers.1.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.128, + "parameters": 16777216, + "reference_delta_l2": 2.092667644901797, + "retained_delta_l2": 2.023302001419501, + "delta_retention_norm_ratio": 0.9668530052293369, + "delta_cosine": 0.7926265026900242, + "relative_weight_rmse": 0.014639248699999614 + }, + { + "index": 14, + "total": 252, + "projection": "layers.1.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.029, + "parameters": 4194304, + "reference_delta_l2": 0.9113442714515859, + "retained_delta_l2": 0.8282095652134706, + "delta_retention_norm_ratio": 0.9087779351422283, + "delta_cosine": 0.7213086683575567, + "relative_weight_rmse": 0.01240592111662902 + }, + { + "index": 15, + "total": 252, + "projection": "layers.10.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.4204, + "parameters": 50331648, + "reference_delta_l2": 2.8154059449259456, + "retained_delta_l2": 2.496875922029961, + "delta_retention_norm_ratio": 0.8868617779719994, + "delta_cosine": 0.7008178018803435, + "relative_weight_rmse": 0.011675323564227786 + }, + { + "index": 16, + "total": 252, + "projection": "layers.10.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3762, + "parameters": 50331648, + "reference_delta_l2": 2.3968528691768745, + "retained_delta_l2": 1.8596025024810308, + "delta_retention_norm_ratio": 0.7758517539375098, + "delta_cosine": 0.5966089470548691, + "relative_weight_rmse": 0.00967390290084976 + }, + { + "index": 17, + "total": 252, + "projection": "layers.10.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.377, + "parameters": 50331648, + "reference_delta_l2": 2.8577186702812782, + "retained_delta_l2": 2.5291353489987176, + "delta_retention_norm_ratio": 0.8850190101987124, + "delta_cosine": 0.6975330232746796, + "relative_weight_rmse": 0.011913550324219804 + }, + { + "index": 18, + "total": 252, + "projection": "layers.10.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0301, + "parameters": 4194304, + "reference_delta_l2": 0.9000497299947879, + "retained_delta_l2": 0.8277379544554692, + "delta_retention_norm_ratio": 0.9196580220742497, + "delta_cosine": 0.73760004795336, + "relative_weight_rmse": 0.011989591809702433 + }, + { + "index": 19, + "total": 252, + "projection": "layers.10.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1262, + "parameters": 16777216, + "reference_delta_l2": 1.7763850901409144, + "retained_delta_l2": 1.5878719671457988, + "delta_retention_norm_ratio": 0.893878233924964, + "delta_cosine": 0.7071065500058159, + "relative_weight_rmse": 0.012178572593450197 + }, + { + "index": 20, + "total": 252, + "projection": "layers.10.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1253, + "parameters": 16777216, + "reference_delta_l2": 2.123153151203661, + "retained_delta_l2": 2.0261615701124023, + "delta_retention_norm_ratio": 0.9543171998514228, + "delta_cosine": 0.7804211120035812, + "relative_weight_rmse": 0.013466313266177627 + }, + { + "index": 21, + "total": 252, + "projection": "layers.10.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0289, + "parameters": 4194304, + "reference_delta_l2": 0.9135031519131191, + "retained_delta_l2": 0.816820385165288, + "delta_retention_norm_ratio": 0.8941626347480557, + "delta_cosine": 0.7122999270297641, + "relative_weight_rmse": 0.0112335137878804 + }, + { + "index": 22, + "total": 252, + "projection": "layers.11.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.4232, + "parameters": 50331648, + "reference_delta_l2": 2.6109010771483048, + "retained_delta_l2": 2.235147165148276, + "delta_retention_norm_ratio": 0.8560826699683172, + "delta_cosine": 0.6706252596501824, + "relative_weight_rmse": 0.011055882773351417 + }, + { + "index": 23, + "total": 252, + "projection": "layers.11.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.5072, + "parameters": 50331648, + "reference_delta_l2": 2.7837904557374036, + "retained_delta_l2": 2.352648766663104, + "delta_retention_norm_ratio": 0.8451242304586126, + "delta_cosine": 0.660464602486499, + "relative_weight_rmse": 0.010840530665069759 + }, + { + "index": 24, + "total": 252, + "projection": "layers.11.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.5614, + "parameters": 50331648, + "reference_delta_l2": 2.678626283286802, + "retained_delta_l2": 2.2772349899339264, + "delta_retention_norm_ratio": 0.8501503192672518, + "delta_cosine": 0.6636888571816134, + "relative_weight_rmse": 0.011386402464295113 + }, + { + "index": 25, + "total": 252, + "projection": "layers.11.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0427, + "parameters": 4194304, + "reference_delta_l2": 1.1352019041397279, + "retained_delta_l2": 1.0932414195471012, + "delta_retention_norm_ratio": 0.9630369853683209, + "delta_cosine": 0.7884213686708266, + "relative_weight_rmse": 0.013700759185643972 + }, + { + "index": 26, + "total": 252, + "projection": "layers.11.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1276, + "parameters": 16777216, + "reference_delta_l2": 1.6714994547029598, + "retained_delta_l2": 1.4680214426427542, + "delta_retention_norm_ratio": 0.878266181010292, + "delta_cosine": 0.6931568524501354, + "relative_weight_rmse": 0.011735150801091823 + }, + { + "index": 27, + "total": 252, + "projection": "layers.11.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1268, + "parameters": 16777216, + "reference_delta_l2": 1.4130009406082322, + "retained_delta_l2": 1.1770579340057459, + "delta_retention_norm_ratio": 0.8330199224772464, + "delta_cosine": 0.6476754056028551, + "relative_weight_rmse": 0.010787635442405654 + }, + { + "index": 28, + "total": 252, + "projection": "layers.11.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0294, + "parameters": 4194304, + "reference_delta_l2": 1.0009756815806787, + "retained_delta_l2": 0.911410842151049, + "delta_retention_norm_ratio": 0.9105224621559292, + "delta_cosine": 0.7279115063868035, + "relative_weight_rmse": 0.011910522316250768 + }, + { + "index": 29, + "total": 252, + "projection": "layers.12.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.421, + "parameters": 50331648, + "reference_delta_l2": 2.5701887321570265, + "retained_delta_l2": 2.1904728701434952, + "delta_retention_norm_ratio": 0.852261486768384, + "delta_cosine": 0.6668780159488222, + "relative_weight_rmse": 0.010843526137798286 + }, + { + "index": 30, + "total": 252, + "projection": "layers.12.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3764, + "parameters": 50331648, + "reference_delta_l2": 3.227729167637915, + "retained_delta_l2": 2.8970759705000475, + "delta_retention_norm_ratio": 0.897558568279803, + "delta_cosine": 0.7136826757993329, + "relative_weight_rmse": 0.012018947370414802 + }, + { + "index": 31, + "total": 252, + "projection": "layers.12.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3806, + "parameters": 50331648, + "reference_delta_l2": 2.691843635194388, + "retained_delta_l2": 2.2957754331565856, + "delta_retention_norm_ratio": 0.8528635924986777, + "delta_cosine": 0.6665708676441654, + "relative_weight_rmse": 0.011236440043849646 + }, + { + "index": 32, + "total": 252, + "projection": "layers.12.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0302, + "parameters": 4194304, + "reference_delta_l2": 0.9324759911703842, + "retained_delta_l2": 0.865664776687525, + "delta_retention_norm_ratio": 0.9283507402705329, + "delta_cosine": 0.7477112390956875, + "relative_weight_rmse": 0.012291956052749723 + }, + { + "index": 33, + "total": 252, + "projection": "layers.12.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1257, + "parameters": 16777216, + "reference_delta_l2": 1.7752423438997837, + "retained_delta_l2": 1.6082069434909523, + "delta_retention_norm_ratio": 0.9059083955591695, + "delta_cosine": 0.7220035227224644, + "relative_weight_rmse": 0.012041195982630933 + }, + { + "index": 34, + "total": 252, + "projection": "layers.12.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1258, + "parameters": 16777216, + "reference_delta_l2": 1.808164638639432, + "retained_delta_l2": 1.6533763990488697, + "delta_retention_norm_ratio": 0.9143948309336288, + "delta_cosine": 0.7314398087130457, + "relative_weight_rmse": 0.012442322897772152 + }, + { + "index": 35, + "total": 252, + "projection": "layers.12.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0294, + "parameters": 4194304, + "reference_delta_l2": 1.088169586668509, + "retained_delta_l2": 1.02521247620021, + "delta_retention_norm_ratio": 0.9421440267770708, + "delta_cosine": 0.7662794220487094, + "relative_weight_rmse": 0.012197573996043104 + }, + { + "index": 36, + "total": 252, + "projection": "layers.13.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.4177, + "parameters": 50331648, + "reference_delta_l2": 2.348918943037561, + "retained_delta_l2": 1.9246704916857462, + "delta_retention_norm_ratio": 0.8193856571299187, + "delta_cosine": 0.6362000951544078, + "relative_weight_rmse": 0.010204729829850571 + }, + { + "index": 37, + "total": 252, + "projection": "layers.13.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3795, + "parameters": 50331648, + "reference_delta_l2": 3.0046582979771133, + "retained_delta_l2": 2.632487583998917, + "delta_retention_norm_ratio": 0.8761354280356071, + "delta_cosine": 0.6889241535168036, + "relative_weight_rmse": 0.011927461191097697 + }, + { + "index": 38, + "total": 252, + "projection": "layers.13.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3785, + "parameters": 50331648, + "reference_delta_l2": 2.40636297798598, + "retained_delta_l2": 1.9470638035278636, + "delta_retention_norm_ratio": 0.8091313826467986, + "delta_cosine": 0.6256307252856597, + "relative_weight_rmse": 0.010364549956162825 + }, + { + "index": 39, + "total": 252, + "projection": "layers.13.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0306, + "parameters": 4194304, + "reference_delta_l2": 1.3441852025176408, + "retained_delta_l2": 1.3300534995723514, + "delta_retention_norm_ratio": 0.9894867887856369, + "delta_cosine": 0.834868233635624, + "relative_weight_rmse": 0.014792573377830402 + }, + { + "index": 40, + "total": 252, + "projection": "layers.13.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1273, + "parameters": 16777216, + "reference_delta_l2": 1.7339808147340978, + "retained_delta_l2": 1.5549879861344538, + "delta_retention_norm_ratio": 0.8967734665350999, + "delta_cosine": 0.7125385544579776, + "relative_weight_rmse": 0.012279755473908797 + }, + { + "index": 41, + "total": 252, + "projection": "layers.13.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.127, + "parameters": 16777216, + "reference_delta_l2": 1.4129089644993296, + "retained_delta_l2": 1.1771269073550152, + "delta_retention_norm_ratio": 0.833122966115609, + "delta_cosine": 0.647700392964135, + "relative_weight_rmse": 0.010767695495616146 + }, + { + "index": 42, + "total": 252, + "projection": "layers.13.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.029, + "parameters": 4194304, + "reference_delta_l2": 1.0336547858771412, + "retained_delta_l2": 0.953239391689645, + "delta_retention_norm_ratio": 0.9222028521647515, + "delta_cosine": 0.7397786048648096, + "relative_weight_rmse": 0.012355256936773435 + }, + { + "index": 43, + "total": 252, + "projection": "layers.14.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.4221, + "parameters": 50331648, + "reference_delta_l2": 2.470164573793053, + "retained_delta_l2": 2.089345235869997, + "delta_retention_norm_ratio": 0.845832402438559, + "delta_cosine": 0.6614611720767581, + "relative_weight_rmse": 0.010527219099625394 + }, + { + "index": 44, + "total": 252, + "projection": "layers.14.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3787, + "parameters": 50331648, + "reference_delta_l2": 2.745108894397341, + "retained_delta_l2": 2.331755421786474, + "delta_retention_norm_ratio": 0.8494218304219168, + "delta_cosine": 0.6626063395532109, + "relative_weight_rmse": 0.011396063131030322 + }, + { + "index": 45, + "total": 252, + "projection": "layers.14.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3793, + "parameters": 50331648, + "reference_delta_l2": 2.950458176756739, + "retained_delta_l2": 2.616798122966152, + "delta_retention_norm_ratio": 0.8869124611156634, + "delta_cosine": 0.7013218772256531, + "relative_weight_rmse": 0.011800066616802376 + }, + { + "index": 46, + "total": 252, + "projection": "layers.14.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0308, + "parameters": 4194304, + "reference_delta_l2": 0.7298318333268707, + "retained_delta_l2": 0.619879028570166, + "delta_retention_norm_ratio": 0.8493450138294254, + "delta_cosine": 0.6663762705060717, + "relative_weight_rmse": 0.010729798361705333 + }, + { + "index": 47, + "total": 252, + "projection": "layers.14.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1273, + "parameters": 16777216, + "reference_delta_l2": 1.5592192910587424, + "retained_delta_l2": 1.3582477852635342, + "delta_retention_norm_ratio": 0.8711076068981006, + "delta_cosine": 0.6853565213261322, + "relative_weight_rmse": 0.01146121264696887 + }, + { + "index": 48, + "total": 252, + "projection": "layers.14.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.126, + "parameters": 16777216, + "reference_delta_l2": 2.3184811672777266, + "retained_delta_l2": 2.260027350123298, + "delta_retention_norm_ratio": 0.974787883559536, + "delta_cosine": 0.8055590776751921, + "relative_weight_rmse": 0.013913195491072323 + }, + { + "index": 49, + "total": 252, + "projection": "layers.14.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0292, + "parameters": 4194304, + "reference_delta_l2": 1.0189491596777331, + "retained_delta_l2": 0.9465701129781681, + "delta_retention_norm_ratio": 0.9289669695370701, + "delta_cosine": 0.7463598755487639, + "relative_weight_rmse": 0.012390833025527957 + }, + { + "index": 50, + "total": 252, + "projection": "layers.15.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.4183, + "parameters": 50331648, + "reference_delta_l2": 2.370632470259436, + "retained_delta_l2": 1.972016566615466, + "delta_retention_norm_ratio": 0.8318525082884963, + "delta_cosine": 0.6479575030195999, + "relative_weight_rmse": 0.01029952986547441 + }, + { + "index": 51, + "total": 252, + "projection": "layers.15.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3795, + "parameters": 50331648, + "reference_delta_l2": 2.8947211752330935, + "retained_delta_l2": 2.5376154944413374, + "delta_retention_norm_ratio": 0.8766355516907425, + "delta_cosine": 0.6884670381775119, + "relative_weight_rmse": 0.012080161860821805 + }, + { + "index": 52, + "total": 252, + "projection": "layers.15.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3792, + "parameters": 50331648, + "reference_delta_l2": 2.788330378720977, + "retained_delta_l2": 2.439307077906408, + "delta_retention_norm_ratio": 0.8748271354506175, + "delta_cosine": 0.6891756444416203, + "relative_weight_rmse": 0.011425579547482719 + }, + { + "index": 53, + "total": 252, + "projection": "layers.15.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0302, + "parameters": 4194304, + "reference_delta_l2": 0.8166695591874434, + "retained_delta_l2": 0.7227354016638761, + "delta_retention_norm_ratio": 0.8849789900127679, + "delta_cosine": 0.6997700140203119, + "relative_weight_rmse": 0.011615990118031959 + }, + { + "index": 54, + "total": 252, + "projection": "layers.15.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1272, + "parameters": 16777216, + "reference_delta_l2": 2.150801054633744, + "retained_delta_l2": 2.0668833790463297, + "delta_retention_norm_ratio": 0.9609830600526162, + "delta_cosine": 0.7886640611263671, + "relative_weight_rmse": 0.013840934538340514 + }, + { + "index": 55, + "total": 252, + "projection": "layers.15.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1271, + "parameters": 16777216, + "reference_delta_l2": 2.2391489123779404, + "retained_delta_l2": 2.164264373080386, + "delta_retention_norm_ratio": 0.966556695321336, + "delta_cosine": 0.8012388621399645, + "relative_weight_rmse": 0.013459137927281995 + }, + { + "index": 56, + "total": 252, + "projection": "layers.15.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.029, + "parameters": 4194304, + "reference_delta_l2": 0.8608909645259568, + "retained_delta_l2": 0.7658203983961419, + "delta_retention_norm_ratio": 0.8895672390031822, + "delta_cosine": 0.7060028338947187, + "relative_weight_rmse": 0.011251307486880139 + }, + { + "index": 57, + "total": 252, + "projection": "layers.16.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.4217, + "parameters": 50331648, + "reference_delta_l2": 2.491828145943435, + "retained_delta_l2": 2.129266484528405, + "delta_retention_norm_ratio": 0.8544997326540108, + "delta_cosine": 0.6692310954044279, + "relative_weight_rmse": 0.010666880270070934 + }, + { + "index": 58, + "total": 252, + "projection": "layers.16.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3774, + "parameters": 50331648, + "reference_delta_l2": 2.651126262430824, + "retained_delta_l2": 2.2698065505621257, + "delta_retention_norm_ratio": 0.8561668988488442, + "delta_cosine": 0.6695471818818178, + "relative_weight_rmse": 0.011105407194152962 + }, + { + "index": 59, + "total": 252, + "projection": "layers.16.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.377, + "parameters": 50331648, + "reference_delta_l2": 2.557631699985966, + "retained_delta_l2": 2.156089908294467, + "delta_retention_norm_ratio": 0.8430024965307936, + "delta_cosine": 0.657117244568733, + "relative_weight_rmse": 0.010891446288279935 + }, + { + "index": 60, + "total": 252, + "projection": "layers.16.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0302, + "parameters": 4194304, + "reference_delta_l2": 0.7413591744163778, + "retained_delta_l2": 0.6389776617706627, + "delta_retention_norm_ratio": 0.8619002553973747, + "delta_cosine": 0.675790074082963, + "relative_weight_rmse": 0.011228211433914847 + }, + { + "index": 61, + "total": 252, + "projection": "layers.16.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1275, + "parameters": 16777216, + "reference_delta_l2": 1.668741441122574, + "retained_delta_l2": 1.4928646978055056, + "delta_retention_norm_ratio": 0.8946051563274207, + "delta_cosine": 0.7097522509284444, + "relative_weight_rmse": 0.011948178323955503 + }, + { + "index": 62, + "total": 252, + "projection": "layers.16.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.127, + "parameters": 16777216, + "reference_delta_l2": 1.8142699819126562, + "retained_delta_l2": 1.6829732447041383, + "delta_retention_norm_ratio": 0.9276310921100612, + "delta_cosine": 0.7485220902672682, + "relative_weight_rmse": 0.01248137141490157 + }, + { + "index": 63, + "total": 252, + "projection": "layers.16.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0293, + "parameters": 4194304, + "reference_delta_l2": 0.9562417782073619, + "retained_delta_l2": 0.8727241623803927, + "delta_retention_norm_ratio": 0.912660565841897, + "delta_cosine": 0.7294498183615773, + "relative_weight_rmse": 0.012130416894435234 + }, + { + "index": 64, + "total": 252, + "projection": "layers.17.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.4839, + "parameters": 50331648, + "reference_delta_l2": 2.581820957597962, + "retained_delta_l2": 2.22135144083909, + "delta_retention_norm_ratio": 0.8603816753063154, + "delta_cosine": 0.6748947014142955, + "relative_weight_rmse": 0.010931993491070512 + }, + { + "index": 65, + "total": 252, + "projection": "layers.17.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.5646, + "parameters": 50331648, + "reference_delta_l2": 3.565795288989979, + "retained_delta_l2": 3.348620759535893, + "delta_retention_norm_ratio": 0.9390950652370172, + "delta_cosine": 0.75818235341218, + "relative_weight_rmse": 0.013358632156064528 + }, + { + "index": 66, + "total": 252, + "projection": "layers.17.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.4724, + "parameters": 50331648, + "reference_delta_l2": 3.1890801714299326, + "retained_delta_l2": 2.9054238256974156, + "delta_retention_norm_ratio": 0.9110538680483126, + "delta_cosine": 0.725669876957664, + "relative_weight_rmse": 0.01239152187476628 + }, + { + "index": 67, + "total": 252, + "projection": "layers.17.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0378, + "parameters": 4194304, + "reference_delta_l2": 0.7872495188121694, + "retained_delta_l2": 0.7046549880866457, + "delta_retention_norm_ratio": 0.8950846856659306, + "delta_cosine": 0.7147860219013704, + "relative_weight_rmse": 0.011556619859539647 + }, + { + "index": 68, + "total": 252, + "projection": "layers.17.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1434, + "parameters": 16777216, + "reference_delta_l2": 1.7303039270497635, + "retained_delta_l2": 1.570607219142342, + "delta_retention_norm_ratio": 0.9077059784637309, + "delta_cosine": 0.7248461301962547, + "relative_weight_rmse": 0.01220195104711847 + }, + { + "index": 69, + "total": 252, + "projection": "layers.17.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1413, + "parameters": 16777216, + "reference_delta_l2": 1.6625518827542718, + "retained_delta_l2": 1.4877936061477128, + "delta_retention_norm_ratio": 0.8948855200133393, + "delta_cosine": 0.7079726725027112, + "relative_weight_rmse": 0.011977287949676606 + }, + { + "index": 70, + "total": 252, + "projection": "layers.17.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0324, + "parameters": 4194304, + "reference_delta_l2": 1.0330150174042918, + "retained_delta_l2": 0.9687762178011139, + "delta_retention_norm_ratio": 0.9378142635674417, + "delta_cosine": 0.7589729310248537, + "relative_weight_rmse": 0.012756159432053974 + }, + { + "index": 71, + "total": 252, + "projection": "layers.18.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.4529, + "parameters": 50331648, + "reference_delta_l2": 2.9126262727027, + "retained_delta_l2": 2.6240751644108298, + "delta_retention_norm_ratio": 0.9009309532787686, + "delta_cosine": 0.7154604011431565, + "relative_weight_rmse": 0.011722422011536813 + }, + { + "index": 72, + "total": 252, + "projection": "layers.18.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3776, + "parameters": 50331648, + "reference_delta_l2": 4.76339438700651, + "retained_delta_l2": 4.729295804193338, + "delta_retention_norm_ratio": 0.9928415369287529, + "delta_cosine": 0.8285342481999447, + "relative_weight_rmse": 0.015494836538716457 + }, + { + "index": 73, + "total": 252, + "projection": "layers.18.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.377, + "parameters": 50331648, + "reference_delta_l2": 3.269296827597192, + "retained_delta_l2": 3.008229704736429, + "delta_retention_norm_ratio": 0.9201457877250512, + "delta_cosine": 0.7367260328556939, + "relative_weight_rmse": 0.012493611711513246 + }, + { + "index": 74, + "total": 252, + "projection": "layers.18.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0306, + "parameters": 4194304, + "reference_delta_l2": 1.1200741516742057, + "retained_delta_l2": 1.0870591777194687, + "delta_retention_norm_ratio": 0.9705242961768302, + "delta_cosine": 0.7994159923560071, + "relative_weight_rmse": 0.014016867668440455 + }, + { + "index": 75, + "total": 252, + "projection": "layers.18.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.131, + "parameters": 16777216, + "reference_delta_l2": 1.6272852709405714, + "retained_delta_l2": 1.4540871565024667, + "delta_retention_norm_ratio": 0.8935662249692726, + "delta_cosine": 0.7105534713486058, + "relative_weight_rmse": 0.011690703016201487 + }, + { + "index": 76, + "total": 252, + "projection": "layers.18.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1407, + "parameters": 16777216, + "reference_delta_l2": 2.0999408362102576, + "retained_delta_l2": 2.0096246836817837, + "delta_retention_norm_ratio": 0.9569910966199092, + "delta_cosine": 0.7836787707305709, + "relative_weight_rmse": 0.013546053503709429 + }, + { + "index": 77, + "total": 252, + "projection": "layers.18.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0323, + "parameters": 4194304, + "reference_delta_l2": 0.9277123799334632, + "retained_delta_l2": 0.8449661156489838, + "delta_retention_norm_ratio": 0.9108061225932825, + "delta_cosine": 0.7277616565032062, + "relative_weight_rmse": 0.011784664649381557 + }, + { + "index": 78, + "total": 252, + "projection": "layers.19.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.4618, + "parameters": 50331648, + "reference_delta_l2": 2.69204546887291, + "retained_delta_l2": 2.3556572010134587, + "delta_retention_norm_ratio": 0.8750436158122217, + "delta_cosine": 0.6892420188448023, + "relative_weight_rmse": 0.011204105407353046 + }, + { + "index": 79, + "total": 252, + "projection": "layers.19.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.4332, + "parameters": 50331648, + "reference_delta_l2": 4.646558121825567, + "retained_delta_l2": 4.583357486482991, + "delta_retention_norm_ratio": 0.9863983977633436, + "delta_cosine": 0.8198661685329147, + "relative_weight_rmse": 0.015500685173111154 + }, + { + "index": 80, + "total": 252, + "projection": "layers.19.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.4461, + "parameters": 50331648, + "reference_delta_l2": 3.0966390602779152, + "retained_delta_l2": 2.777364179232524, + "delta_retention_norm_ratio": 0.8968963205493063, + "delta_cosine": 0.7111518928704179, + "relative_weight_rmse": 0.0122094136708107 + }, + { + "index": 81, + "total": 252, + "projection": "layers.19.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0376, + "parameters": 4194304, + "reference_delta_l2": 1.0452082409501329, + "retained_delta_l2": 1.0060081477725002, + "delta_retention_norm_ratio": 0.9624954227857997, + "delta_cosine": 0.7952802361204205, + "relative_weight_rmse": 0.013596075789164266 + }, + { + "index": 82, + "total": 252, + "projection": "layers.19.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1451, + "parameters": 16777216, + "reference_delta_l2": 1.6410590693578941, + "retained_delta_l2": 1.4617610198484243, + "delta_retention_norm_ratio": 0.8907424766985231, + "delta_cosine": 0.7054735492839265, + "relative_weight_rmse": 0.011799119986630215 + }, + { + "index": 83, + "total": 252, + "projection": "layers.19.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1431, + "parameters": 16777216, + "reference_delta_l2": 1.9427775069945303, + "retained_delta_l2": 1.8340507448747867, + "delta_retention_norm_ratio": 0.9440354020322462, + "delta_cosine": 0.7664055714765793, + "relative_weight_rmse": 0.012849164180343813 + }, + { + "index": 84, + "total": 252, + "projection": "layers.19.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0327, + "parameters": 4194304, + "reference_delta_l2": 1.2336913153947044, + "retained_delta_l2": 1.2001883069941015, + "delta_retention_norm_ratio": 0.9728432809872832, + "delta_cosine": 0.8055331507034024, + "relative_weight_rmse": 0.013855669213057621 + }, + { + "index": 85, + "total": 252, + "projection": "layers.2.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.4638, + "parameters": 50331648, + "reference_delta_l2": 1.5518642981607695, + "retained_delta_l2": 1.336564993481032, + "delta_retention_norm_ratio": 0.8612640905941938, + "delta_cosine": 0.6979839707431235, + "relative_weight_rmse": 0.010113011263998705 + }, + { + "index": 86, + "total": 252, + "projection": "layers.2.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.4306, + "parameters": 50331648, + "reference_delta_l2": 2.5114528976054897, + "retained_delta_l2": 2.368654370034365, + "delta_retention_norm_ratio": 0.9431410687784455, + "delta_cosine": 0.7975273949570989, + "relative_weight_rmse": 0.012669618502563918 + }, + { + "index": 87, + "total": 252, + "projection": "layers.2.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.5003, + "parameters": 50331648, + "reference_delta_l2": 3.3709012795534457, + "retained_delta_l2": 3.351069441650554, + "delta_retention_norm_ratio": 0.9941167550580067, + "delta_cosine": 0.8796380868082753, + "relative_weight_rmse": 0.014946718766317337 + }, + { + "index": 88, + "total": 252, + "projection": "layers.2.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0452, + "parameters": 4194304, + "reference_delta_l2": 0.5777347882140783, + "retained_delta_l2": 0.43485062771642585, + "delta_retention_norm_ratio": 0.7526820897537728, + "delta_cosine": 0.5763385778877209, + "relative_weight_rmse": 0.009298079222984458 + }, + { + "index": 89, + "total": 252, + "projection": "layers.2.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1907, + "parameters": 16777216, + "reference_delta_l2": 2.090544888405803, + "retained_delta_l2": 1.9729454899637549, + "delta_retention_norm_ratio": 0.9437470110810552, + "delta_cosine": 0.7611603233612839, + "relative_weight_rmse": 0.013963239284386296 + }, + { + "index": 90, + "total": 252, + "projection": "layers.2.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1905, + "parameters": 16777216, + "reference_delta_l2": 1.231582001030221, + "retained_delta_l2": 0.9736107898996961, + "delta_retention_norm_ratio": 0.7905367154483165, + "delta_cosine": 0.6086393106196143, + "relative_weight_rmse": 0.010197943915111418 + }, + { + "index": 91, + "total": 252, + "projection": "layers.2.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0437, + "parameters": 4194304, + "reference_delta_l2": 1.1642013080067461, + "retained_delta_l2": 1.1239344864992464, + "delta_retention_norm_ratio": 0.9654124924696732, + "delta_cosine": 0.7940087210427319, + "relative_weight_rmse": 0.013529149550340836 + }, + { + "index": 92, + "total": 252, + "projection": "layers.20.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.6079, + "parameters": 50331648, + "reference_delta_l2": 2.6481378159567726, + "retained_delta_l2": 2.2605324157763285, + "delta_retention_norm_ratio": 0.853630956121367, + "delta_cosine": 0.6660251825806656, + "relative_weight_rmse": 0.01125508356282899 + }, + { + "index": 93, + "total": 252, + "projection": "layers.20.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.5711, + "parameters": 50331648, + "reference_delta_l2": 4.588796501243996, + "retained_delta_l2": 4.525325172710448, + "delta_retention_norm_ratio": 0.9861681971479143, + "delta_cosine": 0.8187467420424862, + "relative_weight_rmse": 0.015497069578017638 + }, + { + "index": 94, + "total": 252, + "projection": "layers.20.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.4424, + "parameters": 50331648, + "reference_delta_l2": 3.131377931467238, + "retained_delta_l2": 2.8142274300363845, + "delta_retention_norm_ratio": 0.8987185487117969, + "delta_cosine": 0.7126364818511092, + "relative_weight_rmse": 0.012291275132750318 + }, + { + "index": 95, + "total": 252, + "projection": "layers.20.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0342, + "parameters": 4194304, + "reference_delta_l2": 0.8443767114111614, + "retained_delta_l2": 0.7765496685785003, + "delta_retention_norm_ratio": 0.9196720587907908, + "delta_cosine": 0.7401891092415827, + "relative_weight_rmse": 0.012201758395188113 + }, + { + "index": 96, + "total": 252, + "projection": "layers.20.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1427, + "parameters": 16777216, + "reference_delta_l2": 1.6103817779168306, + "retained_delta_l2": 1.4296096572873462, + "delta_retention_norm_ratio": 0.8877457984756082, + "delta_cosine": 0.7036362895995829, + "relative_weight_rmse": 0.011827662893954958 + }, + { + "index": 97, + "total": 252, + "projection": "layers.20.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1411, + "parameters": 16777216, + "reference_delta_l2": 1.8953051306455484, + "retained_delta_l2": 1.7717421064691508, + "delta_retention_norm_ratio": 0.9348057354045617, + "delta_cosine": 0.7572950262380886, + "relative_weight_rmse": 0.012349288130135085 + }, + { + "index": 98, + "total": 252, + "projection": "layers.20.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0322, + "parameters": 4194304, + "reference_delta_l2": 0.9458679444695625, + "retained_delta_l2": 0.8682304509007128, + "delta_retention_norm_ratio": 0.9179193099599243, + "delta_cosine": 0.737299013327211, + "relative_weight_rmse": 0.012036773253217456 + }, + { + "index": 99, + "total": 252, + "projection": "layers.21.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.4601, + "parameters": 50331648, + "reference_delta_l2": 3.1063557555245036, + "retained_delta_l2": 2.804620103416943, + "delta_retention_norm_ratio": 0.9028650689571089, + "delta_cosine": 0.7159237204892357, + "relative_weight_rmse": 0.012307140914190159 + }, + { + "index": 100, + "total": 252, + "projection": "layers.21.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3806, + "parameters": 50331648, + "reference_delta_l2": 4.658436495249149, + "retained_delta_l2": 4.599627434990953, + "delta_retention_norm_ratio": 0.9873757943640163, + "delta_cosine": 0.8183164944416749, + "relative_weight_rmse": 0.01597945182150837 + }, + { + "index": 101, + "total": 252, + "projection": "layers.21.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3825, + "parameters": 50331648, + "reference_delta_l2": 3.3486012468888253, + "retained_delta_l2": 3.076218336638799, + "delta_retention_norm_ratio": 0.9186577050632421, + "delta_cosine": 0.7336301532310867, + "relative_weight_rmse": 0.012777358540621694 + }, + { + "index": 102, + "total": 252, + "projection": "layers.21.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0301, + "parameters": 4194304, + "reference_delta_l2": 0.8894337662681111, + "retained_delta_l2": 0.8225220103515385, + "delta_retention_norm_ratio": 0.92477038937107, + "delta_cosine": 0.7420804268034151, + "relative_weight_rmse": 0.012826546977625941 + }, + { + "index": 103, + "total": 252, + "projection": "layers.21.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1419, + "parameters": 16777216, + "reference_delta_l2": 1.7390002767546273, + "retained_delta_l2": 1.590858009332058, + "delta_retention_norm_ratio": 0.9148118206749014, + "delta_cosine": 0.7325920089020045, + "relative_weight_rmse": 0.01239351230226023 + }, + { + "index": 104, + "total": 252, + "projection": "layers.21.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1428, + "parameters": 16777216, + "reference_delta_l2": 1.3490575147990649, + "retained_delta_l2": 1.1161839512677845, + "delta_retention_norm_ratio": 0.8273805519952456, + "delta_cosine": 0.644112057896957, + "relative_weight_rmse": 0.010346139449396108 + }, + { + "index": 105, + "total": 252, + "projection": "layers.21.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0328, + "parameters": 4194304, + "reference_delta_l2": 1.3408966424827091, + "retained_delta_l2": 1.3237425734462804, + "delta_retention_norm_ratio": 0.9872070161912947, + "delta_cosine": 0.8234264060066494, + "relative_weight_rmse": 0.014555834458189592 + }, + { + "index": 106, + "total": 252, + "projection": "layers.22.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.4751, + "parameters": 50331648, + "reference_delta_l2": 3.3631959694360236, + "retained_delta_l2": 3.12796512214615, + "delta_retention_norm_ratio": 0.9300573474077637, + "delta_cosine": 0.7476771306758753, + "relative_weight_rmse": 0.012611960494823589 + }, + { + "index": 107, + "total": 252, + "projection": "layers.22.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.4284, + "parameters": 50331648, + "reference_delta_l2": 4.231183303380896, + "retained_delta_l2": 4.11863707365216, + "delta_retention_norm_ratio": 0.9734007671946505, + "delta_cosine": 0.8028479015786785, + "relative_weight_rmse": 0.01469228330871762 + }, + { + "index": 108, + "total": 252, + "projection": "layers.22.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.4358, + "parameters": 50331648, + "reference_delta_l2": 3.430692303380956, + "retained_delta_l2": 3.1802809790513904, + "delta_retention_norm_ratio": 0.9270085154291497, + "delta_cosine": 0.7451873077927372, + "relative_weight_rmse": 0.012632709098973662 + }, + { + "index": 109, + "total": 252, + "projection": "layers.22.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0342, + "parameters": 4194304, + "reference_delta_l2": 1.1250802889071663, + "retained_delta_l2": 1.0999214302085918, + "delta_retention_norm_ratio": 0.9776381659632377, + "delta_cosine": 0.8174104175691086, + "relative_weight_rmse": 0.014423052832685745 + }, + { + "index": 110, + "total": 252, + "projection": "layers.22.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1452, + "parameters": 16777216, + "reference_delta_l2": 2.0977300669570895, + "retained_delta_l2": 1.9973119298106634, + "delta_retention_norm_ratio": 0.9521300958935627, + "delta_cosine": 0.7727093296784171, + "relative_weight_rmse": 0.01373144987681112 + }, + { + "index": 111, + "total": 252, + "projection": "layers.22.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1512, + "parameters": 16777216, + "reference_delta_l2": 1.8700917288546843, + "retained_delta_l2": 1.740740623429105, + "delta_retention_norm_ratio": 0.9308316787728915, + "delta_cosine": 0.7545796916735813, + "relative_weight_rmse": 0.012596689340508347 + }, + { + "index": 112, + "total": 252, + "projection": "layers.22.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0324, + "parameters": 4194304, + "reference_delta_l2": 1.1046241873952773, + "retained_delta_l2": 1.0522086529364827, + "delta_retention_norm_ratio": 0.9525489890073915, + "delta_cosine": 0.7775540582670679, + "relative_weight_rmse": 0.013233412221306171 + }, + { + "index": 113, + "total": 252, + "projection": "layers.23.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.4885, + "parameters": 50331648, + "reference_delta_l2": 3.163652780787419, + "retained_delta_l2": 2.904511398361662, + "delta_retention_norm_ratio": 0.91808791912959, + "delta_cosine": 0.7353718534542721, + "relative_weight_rmse": 0.011981638799415275 + }, + { + "index": 114, + "total": 252, + "projection": "layers.23.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.4442, + "parameters": 50331648, + "reference_delta_l2": 5.673471157116757, + "retained_delta_l2": 5.721161007202696, + "delta_retention_norm_ratio": 1.0084057623217344, + "delta_cosine": 0.8544888770040429, + "relative_weight_rmse": 0.016865416130790185 + }, + { + "index": 115, + "total": 252, + "projection": "layers.23.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.5614, + "parameters": 50331648, + "reference_delta_l2": 4.990220148592181, + "retained_delta_l2": 4.953764690573903, + "delta_retention_norm_ratio": 0.9926946192887776, + "delta_cosine": 0.8292728635880532, + "relative_weight_rmse": 0.01542512537608608 + }, + { + "index": 116, + "total": 252, + "projection": "layers.23.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0456, + "parameters": 4194304, + "reference_delta_l2": 0.7942768854802407, + "retained_delta_l2": 0.7067902854103598, + "delta_retention_norm_ratio": 0.8898537755923941, + "delta_cosine": 0.705752420298061, + "relative_weight_rmse": 0.011967553523098761 + }, + { + "index": 117, + "total": 252, + "projection": "layers.23.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1947, + "parameters": 16777216, + "reference_delta_l2": 1.960752770116258, + "retained_delta_l2": 1.8338049090397375, + "delta_retention_norm_ratio": 0.9352555492914115, + "delta_cosine": 0.7547622953587477, + "relative_weight_rmse": 0.012807511442686801 + }, + { + "index": 118, + "total": 252, + "projection": "layers.23.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1924, + "parameters": 16777216, + "reference_delta_l2": 1.7876419658591682, + "retained_delta_l2": 1.6201349102637372, + "delta_retention_norm_ratio": 0.9062972011205138, + "delta_cosine": 0.7240842190672281, + "relative_weight_rmse": 0.012033576639859586 + }, + { + "index": 119, + "total": 252, + "projection": "layers.23.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0438, + "parameters": 4194304, + "reference_delta_l2": 0.9295224837108503, + "retained_delta_l2": 0.8539101913832426, + "delta_retention_norm_ratio": 0.9186546924333154, + "delta_cosine": 0.7380886758391304, + "relative_weight_rmse": 0.011748968034076874 + }, + { + "index": 120, + "total": 252, + "projection": "layers.24.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.613, + "parameters": 50331648, + "reference_delta_l2": 3.252584613610017, + "retained_delta_l2": 2.9805501698215298, + "delta_retention_norm_ratio": 0.9163636073754408, + "delta_cosine": 0.7316670116924744, + "relative_weight_rmse": 0.012278691265320801 + }, + { + "index": 121, + "total": 252, + "projection": "layers.24.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.5406, + "parameters": 50331648, + "reference_delta_l2": 5.162360133417278, + "retained_delta_l2": 5.135506093193869, + "delta_retention_norm_ratio": 0.9947981079333121, + "delta_cosine": 0.8287570610137442, + "relative_weight_rmse": 0.01630549393936965 + }, + { + "index": 122, + "total": 252, + "projection": "layers.24.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.4292, + "parameters": 50331648, + "reference_delta_l2": 4.014104083373702, + "retained_delta_l2": 3.836623749147334, + "delta_retention_norm_ratio": 0.9557858165757368, + "delta_cosine": 0.7758899956499796, + "relative_weight_rmse": 0.013958052816711282 + }, + { + "index": 123, + "total": 252, + "projection": "layers.24.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0344, + "parameters": 4194304, + "reference_delta_l2": 0.9660670148452469, + "retained_delta_l2": 0.9164875452272386, + "delta_retention_norm_ratio": 0.9486790576055943, + "delta_cosine": 0.7724061527507058, + "relative_weight_rmse": 0.01314939133326335 + }, + { + "index": 124, + "total": 252, + "projection": "layers.24.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1442, + "parameters": 16777216, + "reference_delta_l2": 1.9196207631349165, + "retained_delta_l2": 1.7808698353420314, + "delta_retention_norm_ratio": 0.9277196150106795, + "delta_cosine": 0.746399241855516, + "relative_weight_rmse": 0.012584866132615998 + }, + { + "index": 125, + "total": 252, + "projection": "layers.24.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1451, + "parameters": 16777216, + "reference_delta_l2": 3.0356638472910538, + "retained_delta_l2": 3.033490399018169, + "delta_retention_norm_ratio": 0.9992840286730614, + "delta_cosine": 0.8494126002584376, + "relative_weight_rmse": 0.0155988092460959 + }, + { + "index": 126, + "total": 252, + "projection": "layers.24.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0332, + "parameters": 4194304, + "reference_delta_l2": 1.087439877127159, + "retained_delta_l2": 1.0359707514700809, + "delta_retention_norm_ratio": 0.9526694516729963, + "delta_cosine": 0.7788704392434199, + "relative_weight_rmse": 0.013049867425404758 + }, + { + "index": 127, + "total": 252, + "projection": "layers.25.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.5842, + "parameters": 50331648, + "reference_delta_l2": 3.6413084066401664, + "retained_delta_l2": 3.4087667317862116, + "delta_retention_norm_ratio": 0.9361378798813362, + "delta_cosine": 0.7517619463387702, + "relative_weight_rmse": 0.013271298801952719 + }, + { + "index": 128, + "total": 252, + "projection": "layers.25.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.5577, + "parameters": 50331648, + "reference_delta_l2": 5.618125664041613, + "retained_delta_l2": 5.634205348706462, + "delta_retention_norm_ratio": 1.002862108401698, + "delta_cosine": 0.8412324904758905, + "relative_weight_rmse": 0.016978569929669337 + }, + { + "index": 129, + "total": 252, + "projection": "layers.25.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.4408, + "parameters": 50331648, + "reference_delta_l2": 3.750783696134975, + "retained_delta_l2": 3.5016894236315097, + "delta_retention_norm_ratio": 0.9335887396652207, + "delta_cosine": 0.7486386618798719, + "relative_weight_rmse": 0.013633263078187501 + }, + { + "index": 130, + "total": 252, + "projection": "layers.25.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0363, + "parameters": 4194304, + "reference_delta_l2": 1.028321782511848, + "retained_delta_l2": 0.9817620151176812, + "delta_retention_norm_ratio": 0.9547225701273809, + "delta_cosine": 0.7798391993358456, + "relative_weight_rmse": 0.01354292405773642 + }, + { + "index": 131, + "total": 252, + "projection": "layers.25.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1473, + "parameters": 16777216, + "reference_delta_l2": 1.9561465982758552, + "retained_delta_l2": 1.8092118698350346, + "delta_retention_norm_ratio": 0.9248856253563364, + "delta_cosine": 0.7402634152259523, + "relative_weight_rmse": 0.013199110733930028 + }, + { + "index": 132, + "total": 252, + "projection": "layers.25.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1453, + "parameters": 16777216, + "reference_delta_l2": 2.4513099043687303, + "retained_delta_l2": 2.4041984374006553, + "delta_retention_norm_ratio": 0.9807811052841124, + "delta_cosine": 0.8152699970141684, + "relative_weight_rmse": 0.01460027497498254 + }, + { + "index": 133, + "total": 252, + "projection": "layers.25.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0333, + "parameters": 4194304, + "reference_delta_l2": 1.189276842761756, + "retained_delta_l2": 1.1459512122709345, + "delta_retention_norm_ratio": 0.9635697686753827, + "delta_cosine": 0.7874865637698173, + "relative_weight_rmse": 0.013464022144077914 + }, + { + "index": 134, + "total": 252, + "projection": "layers.26.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.4499, + "parameters": 50331648, + "reference_delta_l2": 3.3018911806101747, + "retained_delta_l2": 3.024148539135432, + "delta_retention_norm_ratio": 0.9158837689425559, + "delta_cosine": 0.7288061856627296, + "relative_weight_rmse": 0.012351683008350083 + }, + { + "index": 135, + "total": 252, + "projection": "layers.26.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3942, + "parameters": 50331648, + "reference_delta_l2": 5.265746606259319, + "retained_delta_l2": 5.22881619858765, + "delta_retention_norm_ratio": 0.9929866720841124, + "delta_cosine": 0.8223831512490348, + "relative_weight_rmse": 0.016664350522507733 + }, + { + "index": 136, + "total": 252, + "projection": "layers.26.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.4094, + "parameters": 50331648, + "reference_delta_l2": 3.606153259776773, + "retained_delta_l2": 3.334838409915596, + "delta_retention_norm_ratio": 0.9247633613115014, + "delta_cosine": 0.7385433579936497, + "relative_weight_rmse": 0.013186302879476907 + }, + { + "index": 137, + "total": 252, + "projection": "layers.26.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0348, + "parameters": 4194304, + "reference_delta_l2": 1.103863773268816, + "retained_delta_l2": 1.0657053185834187, + "delta_retention_norm_ratio": 0.9654319168637986, + "delta_cosine": 0.791585055187683, + "relative_weight_rmse": 0.014453820078573135 + }, + { + "index": 138, + "total": 252, + "projection": "layers.26.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1488, + "parameters": 16777216, + "reference_delta_l2": 2.148398831961565, + "retained_delta_l2": 2.0244152936192146, + "delta_retention_norm_ratio": 0.9422902598447472, + "delta_cosine": 0.7607306015808954, + "relative_weight_rmse": 0.013821713478567725 + }, + { + "index": 139, + "total": 252, + "projection": "layers.26.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1472, + "parameters": 16777216, + "reference_delta_l2": 2.7177157728331176, + "retained_delta_l2": 2.693831052432037, + "delta_retention_norm_ratio": 0.991211472281304, + "delta_cosine": 0.832184560960097, + "relative_weight_rmse": 0.015379613038605862 + }, + { + "index": 140, + "total": 252, + "projection": "layers.26.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0344, + "parameters": 4194304, + "reference_delta_l2": 1.3032695648582875, + "retained_delta_l2": 1.2682607502239467, + "delta_retention_norm_ratio": 0.9731377026070984, + "delta_cosine": 0.8073968854063354, + "relative_weight_rmse": 0.014266547957942074 + }, + { + "index": 141, + "total": 252, + "projection": "layers.27.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.4815, + "parameters": 50331648, + "reference_delta_l2": 3.1291582486293152, + "retained_delta_l2": 2.803564762887972, + "delta_retention_norm_ratio": 0.8959485395524612, + "delta_cosine": 0.7084046905945871, + "relative_weight_rmse": 0.011926185212224393 + }, + { + "index": 142, + "total": 252, + "projection": "layers.27.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.4392, + "parameters": 50331648, + "reference_delta_l2": 5.242473178766284, + "retained_delta_l2": 5.210976294240727, + "delta_retention_norm_ratio": 0.9939919798439544, + "delta_cosine": 0.825381251160355, + "relative_weight_rmse": 0.016479813329711532 + }, + { + "index": 143, + "total": 252, + "projection": "layers.27.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.4438, + "parameters": 50331648, + "reference_delta_l2": 3.1495516466110596, + "retained_delta_l2": 2.766994496462717, + "delta_retention_norm_ratio": 0.8785359971601111, + "delta_cosine": 0.6903104525042502, + "relative_weight_rmse": 0.012199682235341219 + }, + { + "index": 144, + "total": 252, + "projection": "layers.27.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0361, + "parameters": 4194304, + "reference_delta_l2": 0.9598316171323823, + "retained_delta_l2": 0.9039917464042844, + "delta_retention_norm_ratio": 0.9418232638606691, + "delta_cosine": 0.7669163480271755, + "relative_weight_rmse": 0.013519748691148765 + }, + { + "index": 145, + "total": 252, + "projection": "layers.27.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1476, + "parameters": 16777216, + "reference_delta_l2": 1.836071449976087, + "retained_delta_l2": 1.671271717811431, + "delta_retention_norm_ratio": 0.9102432902778361, + "delta_cosine": 0.7241318906576018, + "relative_weight_rmse": 0.01275325748682899 + }, + { + "index": 146, + "total": 252, + "projection": "layers.27.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.149, + "parameters": 16777216, + "reference_delta_l2": 2.505142600316082, + "retained_delta_l2": 2.4507606604416377, + "delta_retention_norm_ratio": 0.9782918785271614, + "delta_cosine": 0.8133658565639071, + "relative_weight_rmse": 0.014896257276405912 + }, + { + "index": 147, + "total": 252, + "projection": "layers.27.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.034, + "parameters": 4194304, + "reference_delta_l2": 1.302699657346531, + "retained_delta_l2": 1.2790988423697518, + "delta_retention_norm_ratio": 0.9818831494706526, + "delta_cosine": 0.8137518980492485, + "relative_weight_rmse": 0.014299891075395718 + }, + { + "index": 148, + "total": 252, + "projection": "layers.28.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.481, + "parameters": 50331648, + "reference_delta_l2": 3.7694347637465806, + "retained_delta_l2": 3.555372192174322, + "delta_retention_norm_ratio": 0.9432109626538558, + "delta_cosine": 0.7614285691445255, + "relative_weight_rmse": 0.01311733328029739 + }, + { + "index": 149, + "total": 252, + "projection": "layers.28.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.4911, + "parameters": 50331648, + "reference_delta_l2": 6.463534007011093, + "retained_delta_l2": 6.580855431367806, + "delta_retention_norm_ratio": 1.018151281362402, + "delta_cosine": 0.8679283430685545, + "relative_weight_rmse": 0.017989242546877723 + }, + { + "index": 150, + "total": 252, + "projection": "layers.28.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.5909, + "parameters": 50331648, + "reference_delta_l2": 3.2507767334340634, + "retained_delta_l2": 2.8840407424406, + "delta_retention_norm_ratio": 0.8871851187989616, + "delta_cosine": 0.6984463383508485, + "relative_weight_rmse": 0.01234291510575387 + }, + { + "index": 151, + "total": 252, + "projection": "layers.28.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0454, + "parameters": 4194304, + "reference_delta_l2": 1.0322954248089176, + "retained_delta_l2": 0.9823414185691245, + "delta_retention_norm_ratio": 0.9516088078671474, + "delta_cosine": 0.7712204011943862, + "relative_weight_rmse": 0.014133461510404306 + }, + { + "index": 152, + "total": 252, + "projection": "layers.28.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1872, + "parameters": 16777216, + "reference_delta_l2": 1.8685004021396905, + "retained_delta_l2": 1.716826483856002, + "delta_retention_norm_ratio": 0.9188258572971133, + "delta_cosine": 0.7338742439373884, + "relative_weight_rmse": 0.012838882513384468 + }, + { + "index": 153, + "total": 252, + "projection": "layers.28.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1322, + "parameters": 16777216, + "reference_delta_l2": 2.110743747110102, + "retained_delta_l2": 2.0115605874320037, + "delta_retention_norm_ratio": 0.9530103264245631, + "delta_cosine": 0.7774594180099803, + "relative_weight_rmse": 0.013262301572717601 + }, + { + "index": 154, + "total": 252, + "projection": "layers.28.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.03, + "parameters": 4194304, + "reference_delta_l2": 1.0575864044302496, + "retained_delta_l2": 0.9881498139970877, + "delta_retention_norm_ratio": 0.9343442860627833, + "delta_cosine": 0.7519250586828223, + "relative_weight_rmse": 0.012589780660289391 + }, + { + "index": 155, + "total": 252, + "projection": "layers.29.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.4296, + "parameters": 50331648, + "reference_delta_l2": 3.7503386724276506, + "retained_delta_l2": 3.5210119859431863, + "delta_retention_norm_ratio": 0.9388517393987734, + "delta_cosine": 0.7565333709142951, + "relative_weight_rmse": 0.013018956595986977 + }, + { + "index": 156, + "total": 252, + "projection": "layers.29.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3857, + "parameters": 50331648, + "reference_delta_l2": 6.161868581061987, + "retained_delta_l2": 6.249541397470819, + "delta_retention_norm_ratio": 1.0142282840432992, + "delta_cosine": 0.8578239645710001, + "relative_weight_rmse": 0.017756600006409558 + }, + { + "index": 157, + "total": 252, + "projection": "layers.29.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3889, + "parameters": 50331648, + "reference_delta_l2": 3.58893045533757, + "retained_delta_l2": 3.290133769661884, + "delta_retention_norm_ratio": 0.9167449218105337, + "delta_cosine": 0.7299541980400778, + "relative_weight_rmse": 0.01291761347732998 + }, + { + "index": 158, + "total": 252, + "projection": "layers.29.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0312, + "parameters": 4194304, + "reference_delta_l2": 0.9627503667306867, + "retained_delta_l2": 0.9124917994270688, + "delta_retention_norm_ratio": 0.9477968858383444, + "delta_cosine": 0.7819753291236863, + "relative_weight_rmse": 0.013135656632060767 + }, + { + "index": 159, + "total": 252, + "projection": "layers.29.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1326, + "parameters": 16777216, + "reference_delta_l2": 1.9473056969704086, + "retained_delta_l2": 1.7957115161873454, + "delta_retention_norm_ratio": 0.9221518321345686, + "delta_cosine": 0.7377141704623403, + "relative_weight_rmse": 0.01293879820563313 + }, + { + "index": 160, + "total": 252, + "projection": "layers.29.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1313, + "parameters": 16777216, + "reference_delta_l2": 2.949713538213498, + "retained_delta_l2": 2.937162075911486, + "delta_retention_norm_ratio": 0.9957448538173597, + "delta_cosine": 0.8482304594765492, + "relative_weight_rmse": 0.015287151010537427 + }, + { + "index": 161, + "total": 252, + "projection": "layers.29.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0303, + "parameters": 4194304, + "reference_delta_l2": 1.070649555367137, + "retained_delta_l2": 1.011136701125239, + "delta_retention_norm_ratio": 0.9444142539979009, + "delta_cosine": 0.7677832739636724, + "relative_weight_rmse": 0.012962639447336012 + }, + { + "index": 162, + "total": 252, + "projection": "layers.3.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.429, + "parameters": 50331648, + "reference_delta_l2": 2.097452268427944, + "retained_delta_l2": 1.8650699519260572, + "delta_retention_norm_ratio": 0.8892073397808193, + "delta_cosine": 0.7209604336152523, + "relative_weight_rmse": 0.0114248068866949 + }, + { + "index": 163, + "total": 252, + "projection": "layers.3.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.388, + "parameters": 50331648, + "reference_delta_l2": 2.935610397323825, + "retained_delta_l2": 2.715579339212817, + "delta_retention_norm_ratio": 0.9250475954467277, + "delta_cosine": 0.7468871380662229, + "relative_weight_rmse": 0.012279867291566805 + }, + { + "index": 164, + "total": 252, + "projection": "layers.3.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3886, + "parameters": 50331648, + "reference_delta_l2": 3.194564290558053, + "retained_delta_l2": 3.1293385190445773, + "delta_retention_norm_ratio": 0.9795822636262921, + "delta_cosine": 0.8437490848781236, + "relative_weight_rmse": 0.014685508152813161 + }, + { + "index": 165, + "total": 252, + "projection": "layers.3.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0311, + "parameters": 4194304, + "reference_delta_l2": 0.5960714704777751, + "retained_delta_l2": 0.46289853683824933, + "delta_retention_norm_ratio": 0.776582272033952, + "delta_cosine": 0.5977316787014598, + "relative_weight_rmse": 0.009655597625645933 + }, + { + "index": 166, + "total": 252, + "projection": "layers.3.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1313, + "parameters": 16777216, + "reference_delta_l2": 1.8539803433661013, + "retained_delta_l2": 1.7008303835316216, + "delta_retention_norm_ratio": 0.9173939678581384, + "delta_cosine": 0.730887404192904, + "relative_weight_rmse": 0.012830916819152128 + }, + { + "index": 167, + "total": 252, + "projection": "layers.3.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1298, + "parameters": 16777216, + "reference_delta_l2": 1.512110233150118, + "retained_delta_l2": 1.3099561432701632, + "delta_retention_norm_ratio": 0.8663099518486721, + "delta_cosine": 0.6796332434963573, + "relative_weight_rmse": 0.011745710378391625 + }, + { + "index": 168, + "total": 252, + "projection": "layers.3.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0305, + "parameters": 4194304, + "reference_delta_l2": 0.9491608293721302, + "retained_delta_l2": 0.8561494882558944, + "delta_retention_norm_ratio": 0.902006764040439, + "delta_cosine": 0.7163494148267664, + "relative_weight_rmse": 0.012404859459595699 + }, + { + "index": 169, + "total": 252, + "projection": "layers.30.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.4299, + "parameters": 50331648, + "reference_delta_l2": 3.8950640372795693, + "retained_delta_l2": 3.6959250354891973, + "delta_retention_norm_ratio": 0.9488740108289833, + "delta_cosine": 0.7671682691074769, + "relative_weight_rmse": 0.013150206063540658 + }, + { + "index": 170, + "total": 252, + "projection": "layers.30.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3996, + "parameters": 50331648, + "reference_delta_l2": 5.423733080902422, + "retained_delta_l2": 5.4237625699772, + "delta_retention_norm_ratio": 1.0000054370438842, + "delta_cosine": 0.8316221280480262, + "relative_weight_rmse": 0.01696628617910767 + }, + { + "index": 171, + "total": 252, + "projection": "layers.30.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.404, + "parameters": 50331648, + "reference_delta_l2": 3.8312705294252947, + "retained_delta_l2": 3.5546148145573113, + "delta_retention_norm_ratio": 0.9277900861494417, + "delta_cosine": 0.7412507800975485, + "relative_weight_rmse": 0.013449023238115166 + }, + { + "index": 172, + "total": 252, + "projection": "layers.30.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0316, + "parameters": 4194304, + "reference_delta_l2": 1.497861100481724, + "retained_delta_l2": 1.510062629983289, + "delta_retention_norm_ratio": 1.0081459686065959, + "delta_cosine": 0.8531201706199472, + "relative_weight_rmse": 0.016617887592308447 + }, + { + "index": 173, + "total": 252, + "projection": "layers.30.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1306, + "parameters": 16777216, + "reference_delta_l2": 1.8733553805023273, + "retained_delta_l2": 1.6979135826582805, + "delta_retention_norm_ratio": 0.9063488969204534, + "delta_cosine": 0.7182008820523232, + "relative_weight_rmse": 0.01288385736718186 + }, + { + "index": 174, + "total": 252, + "projection": "layers.30.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1308, + "parameters": 16777216, + "reference_delta_l2": 3.0296228868115977, + "retained_delta_l2": 3.0415871953000733, + "delta_retention_norm_ratio": 1.0039491081680687, + "delta_cosine": 0.8556895409101121, + "relative_weight_rmse": 0.016319885240090704 + }, + { + "index": 175, + "total": 252, + "projection": "layers.30.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0305, + "parameters": 4194304, + "reference_delta_l2": 1.3112404855000326, + "retained_delta_l2": 1.273721000881319, + "delta_retention_norm_ratio": 0.9713862674058558, + "delta_cosine": 0.8017335299100133, + "relative_weight_rmse": 0.013932060608436915 + }, + { + "index": 176, + "total": 252, + "projection": "layers.31.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.4255, + "parameters": 50331648, + "reference_delta_l2": 4.115829321037284, + "retained_delta_l2": 3.9506936392168406, + "delta_retention_norm_ratio": 0.959877908207616, + "delta_cosine": 0.780756883338157, + "relative_weight_rmse": 0.013389823088724431 + }, + { + "index": 177, + "total": 252, + "projection": "layers.31.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3859, + "parameters": 50331648, + "reference_delta_l2": 4.232274668008745, + "retained_delta_l2": 4.086995580761939, + "delta_retention_norm_ratio": 0.9656735210630462, + "delta_cosine": 0.7867727770298929, + "relative_weight_rmse": 0.014790841779742112 + }, + { + "index": 178, + "total": 252, + "projection": "layers.31.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.386, + "parameters": 50331648, + "reference_delta_l2": 3.7522602192357497, + "retained_delta_l2": 3.4655694510341983, + "delta_retention_norm_ratio": 0.923595179584868, + "delta_cosine": 0.7371230249679259, + "relative_weight_rmse": 0.013140419710767343 + }, + { + "index": 179, + "total": 252, + "projection": "layers.31.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.031, + "parameters": 4194304, + "reference_delta_l2": 1.1315771011121951, + "retained_delta_l2": 1.1078994284262837, + "delta_retention_norm_ratio": 0.9790755109283854, + "delta_cosine": 0.8130743081793506, + "relative_weight_rmse": 0.014434169550537385 + }, + { + "index": 180, + "total": 252, + "projection": "layers.31.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1307, + "parameters": 16777216, + "reference_delta_l2": 1.9324255331889064, + "retained_delta_l2": 1.7842959594246393, + "delta_retention_norm_ratio": 0.9233452615792017, + "delta_cosine": 0.7393483538920406, + "relative_weight_rmse": 0.0125643137153554 + }, + { + "index": 181, + "total": 252, + "projection": "layers.31.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1301, + "parameters": 16777216, + "reference_delta_l2": 3.099272549999048, + "retained_delta_l2": 3.1232015156320574, + "delta_retention_norm_ratio": 1.0077208329525638, + "delta_cosine": 0.8595104156720689, + "relative_weight_rmse": 0.016371860325143907 + }, + { + "index": 182, + "total": 252, + "projection": "layers.31.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0296, + "parameters": 4194304, + "reference_delta_l2": 0.9155097993063135, + "retained_delta_l2": 0.8083075638047963, + "delta_retention_norm_ratio": 0.8829043276404633, + "delta_cosine": 0.700113519431438, + "relative_weight_rmse": 0.01126498892688826 + }, + { + "index": 183, + "total": 252, + "projection": "layers.32.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.4279, + "parameters": 50331648, + "reference_delta_l2": 3.056876645101006, + "retained_delta_l2": 2.7049628845810765, + "delta_retention_norm_ratio": 0.8848779975849168, + "delta_cosine": 0.6985512680251854, + "relative_weight_rmse": 0.011299068537040777 + }, + { + "index": 184, + "total": 252, + "projection": "layers.32.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3914, + "parameters": 50331648, + "reference_delta_l2": 5.774789303352165, + "retained_delta_l2": 5.830146247138633, + "delta_retention_norm_ratio": 1.009585967708698, + "delta_cosine": 0.8496400852164547, + "relative_weight_rmse": 0.0175333218981735 + }, + { + "index": 185, + "total": 252, + "projection": "layers.32.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.4067, + "parameters": 50331648, + "reference_delta_l2": 4.190777341990573, + "retained_delta_l2": 3.975929014215143, + "delta_retention_norm_ratio": 0.948733060660918, + "delta_cosine": 0.7678586543003513, + "relative_weight_rmse": 0.013860843971770948 + }, + { + "index": 186, + "total": 252, + "projection": "layers.32.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0308, + "parameters": 4194304, + "reference_delta_l2": 1.0783365439280528, + "retained_delta_l2": 1.044233761644472, + "delta_retention_norm_ratio": 0.9683746391832787, + "delta_cosine": 0.7992451047041467, + "relative_weight_rmse": 0.01410586055434709 + }, + { + "index": 187, + "total": 252, + "projection": "layers.32.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1315, + "parameters": 16777216, + "reference_delta_l2": 1.9553335281852247, + "retained_delta_l2": 1.7974152937075847, + "delta_retention_norm_ratio": 0.9192371878243164, + "delta_cosine": 0.73290126070321, + "relative_weight_rmse": 0.01292046760864544 + }, + { + "index": 188, + "total": 252, + "projection": "layers.32.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1307, + "parameters": 16777216, + "reference_delta_l2": 3.5218932383788344, + "retained_delta_l2": 3.5825260751259425, + "delta_retention_norm_ratio": 1.017215978067245, + "delta_cosine": 0.8845172274918165, + "relative_weight_rmse": 0.017048707578256873 + }, + { + "index": 189, + "total": 252, + "projection": "layers.32.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0302, + "parameters": 4194304, + "reference_delta_l2": 1.1079466741589432, + "retained_delta_l2": 1.038371528096713, + "delta_retention_norm_ratio": 0.9372035246055089, + "delta_cosine": 0.7588609209107184, + "relative_weight_rmse": 0.012618995536003064 + }, + { + "index": 190, + "total": 252, + "projection": "layers.33.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.5602, + "parameters": 50331648, + "reference_delta_l2": 4.687324672463539, + "retained_delta_l2": 4.601612864748442, + "delta_retention_norm_ratio": 0.9817141304039326, + "delta_cosine": 0.8101883225516318, + "relative_weight_rmse": 0.014416737330240039 + }, + { + "index": 191, + "total": 252, + "projection": "layers.33.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.5672, + "parameters": 50331648, + "reference_delta_l2": 4.02782065921811, + "retained_delta_l2": 3.846869787981053, + "delta_retention_norm_ratio": 0.9550747447449205, + "delta_cosine": 0.7723561530845422, + "relative_weight_rmse": 0.014779589810549918 + }, + { + "index": 192, + "total": 252, + "projection": "layers.33.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.4299, + "parameters": 50331648, + "reference_delta_l2": 4.562896573504985, + "retained_delta_l2": 4.427774845848602, + "delta_retention_norm_ratio": 0.970386852851107, + "delta_cosine": 0.7960995135956899, + "relative_weight_rmse": 0.01424624046643528 + }, + { + "index": 193, + "total": 252, + "projection": "layers.33.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0344, + "parameters": 4194304, + "reference_delta_l2": 1.1713897832764697, + "retained_delta_l2": 1.1620524076735506, + "delta_retention_norm_ratio": 0.9920288056663755, + "delta_cosine": 0.8276977541120957, + "relative_weight_rmse": 0.015548934151994313 + }, + { + "index": 194, + "total": 252, + "projection": "layers.33.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1457, + "parameters": 16777216, + "reference_delta_l2": 1.9603680458460517, + "retained_delta_l2": 1.8125899541779231, + "delta_retention_norm_ratio": 0.924617169729294, + "delta_cosine": 0.741645917604003, + "relative_weight_rmse": 0.012314534198132183 + }, + { + "index": 195, + "total": 252, + "projection": "layers.33.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.145, + "parameters": 16777216, + "reference_delta_l2": 1.6961464180653134, + "retained_delta_l2": 1.5640088128830405, + "delta_retention_norm_ratio": 0.9220954018032277, + "delta_cosine": 0.7428344203124573, + "relative_weight_rmse": 0.01192750420709463 + }, + { + "index": 196, + "total": 252, + "projection": "layers.33.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0334, + "parameters": 4194304, + "reference_delta_l2": 1.2357353784065346, + "retained_delta_l2": 1.1863228587248493, + "delta_retention_norm_ratio": 0.9600136723888231, + "delta_cosine": 0.7866107364371089, + "relative_weight_rmse": 0.012737525311644961 + }, + { + "index": 197, + "total": 252, + "projection": "layers.34.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.5467, + "parameters": 50331648, + "reference_delta_l2": 3.2565833651505516, + "retained_delta_l2": 3.0255489951779886, + "delta_retention_norm_ratio": 0.9290562088952137, + "delta_cosine": 0.7455103372977274, + "relative_weight_rmse": 0.011843588377196124 + }, + { + "index": 198, + "total": 252, + "projection": "layers.34.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.5743, + "parameters": 50331648, + "reference_delta_l2": 3.518525086669968, + "retained_delta_l2": 3.3155130236000065, + "delta_retention_norm_ratio": 0.9423019424135759, + "delta_cosine": 0.7574008311791389, + "relative_weight_rmse": 0.013121291271205936 + }, + { + "index": 199, + "total": 252, + "projection": "layers.34.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.4473, + "parameters": 50331648, + "reference_delta_l2": 3.122839585398148, + "retained_delta_l2": 2.7992683076510696, + "delta_retention_norm_ratio": 0.8963855590725693, + "delta_cosine": 0.7099281078472282, + "relative_weight_rmse": 0.011421056356851617 + }, + { + "index": 200, + "total": 252, + "projection": "layers.34.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0342, + "parameters": 4194304, + "reference_delta_l2": 0.789778602037118, + "retained_delta_l2": 0.7284479516672997, + "delta_retention_norm_ratio": 0.9223445023559452, + "delta_cosine": 0.7527854359891812, + "relative_weight_rmse": 0.011641206148687635 + }, + { + "index": 201, + "total": 252, + "projection": "layers.34.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1457, + "parameters": 16777216, + "reference_delta_l2": 2.3701434474977137, + "retained_delta_l2": 2.300165870829211, + "delta_retention_norm_ratio": 0.9704753833602852, + "delta_cosine": 0.7933415277491485, + "relative_weight_rmse": 0.014057024596917099 + }, + { + "index": 202, + "total": 252, + "projection": "layers.34.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1454, + "parameters": 16777216, + "reference_delta_l2": 2.79645424905211, + "retained_delta_l2": 2.810282234280324, + "delta_retention_norm_ratio": 1.0049448279845457, + "delta_cosine": 0.8600068332513529, + "relative_weight_rmse": 0.015296338846627629 + }, + { + "index": 203, + "total": 252, + "projection": "layers.34.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.034, + "parameters": 4194304, + "reference_delta_l2": 0.9822567585383027, + "retained_delta_l2": 0.8778790775636134, + "delta_retention_norm_ratio": 0.8937368665907539, + "delta_cosine": 0.7155763314870319, + "relative_weight_rmse": 0.011412855542293352 + }, + { + "index": 204, + "total": 252, + "projection": "layers.35.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.5198, + "parameters": 50331648, + "reference_delta_l2": 2.3328259869038397, + "retained_delta_l2": 2.086362517056803, + "delta_retention_norm_ratio": 0.8943498266777511, + "delta_cosine": 0.7148426779507352, + "relative_weight_rmse": 0.009811587448131986 + }, + { + "index": 205, + "total": 252, + "projection": "layers.35.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.5749, + "parameters": 50331648, + "reference_delta_l2": 2.6254547609667505, + "retained_delta_l2": 2.291767053716837, + "delta_retention_norm_ratio": 0.8729028920204884, + "delta_cosine": 0.6876135248167063, + "relative_weight_rmse": 0.010527697572972752 + }, + { + "index": 206, + "total": 252, + "projection": "layers.35.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.4672, + "parameters": 50331648, + "reference_delta_l2": 2.6644939513948387, + "retained_delta_l2": 2.292344013435714, + "delta_retention_norm_ratio": 0.8603299745663496, + "delta_cosine": 0.67494142597945, + "relative_weight_rmse": 0.010352404487708527 + }, + { + "index": 207, + "total": 252, + "projection": "layers.35.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0341, + "parameters": 4194304, + "reference_delta_l2": 0.8330018642105433, + "retained_delta_l2": 0.7629345502165182, + "delta_retention_norm_ratio": 0.9158857656814133, + "delta_cosine": 0.7342284696895491, + "relative_weight_rmse": 0.012231814531927804 + }, + { + "index": 208, + "total": 252, + "projection": "layers.35.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.145, + "parameters": 16777216, + "reference_delta_l2": 1.97763190430294, + "retained_delta_l2": 1.8761931323228476, + "delta_retention_norm_ratio": 0.948706950085412, + "delta_cosine": 0.7675604966314904, + "relative_weight_rmse": 0.012980312464296371 + }, + { + "index": 209, + "total": 252, + "projection": "layers.35.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1453, + "parameters": 16777216, + "reference_delta_l2": 1.9258966242155413, + "retained_delta_l2": 1.838641371460967, + "delta_retention_norm_ratio": 0.954693698686909, + "delta_cosine": 0.7750370803706031, + "relative_weight_rmse": 0.01310119066491629 + }, + { + "index": 210, + "total": 252, + "projection": "layers.35.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0332, + "parameters": 4194304, + "reference_delta_l2": 0.9017170614304261, + "retained_delta_l2": 0.7952812095703247, + "delta_retention_norm_ratio": 0.8819631385356528, + "delta_cosine": 0.6987705269640657, + "relative_weight_rmse": 0.01139507748203064 + }, + { + "index": 211, + "total": 252, + "projection": "layers.4.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.4704, + "parameters": 50331648, + "reference_delta_l2": 2.22932606793877, + "retained_delta_l2": 1.9133400379966317, + "delta_retention_norm_ratio": 0.8582593930576076, + "delta_cosine": 0.6734965603575204, + "relative_weight_rmse": 0.010852124685479955 + }, + { + "index": 212, + "total": 252, + "projection": "layers.4.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3953, + "parameters": 50331648, + "reference_delta_l2": 3.327042619010718, + "retained_delta_l2": 2.9373369945513406, + "delta_retention_norm_ratio": 0.8828672580770082, + "delta_cosine": 0.7104127938935952, + "relative_weight_rmse": 0.010851148580997205 + }, + { + "index": 213, + "total": 252, + "projection": "layers.4.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3872, + "parameters": 50331648, + "reference_delta_l2": 3.3108983130289054, + "retained_delta_l2": 3.1892568937112924, + "delta_retention_norm_ratio": 0.9632602974126584, + "delta_cosine": 0.7961670120639696, + "relative_weight_rmse": 0.014032548417868468 + }, + { + "index": 214, + "total": 252, + "projection": "layers.4.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0308, + "parameters": 4194304, + "reference_delta_l2": 0.5944270448677177, + "retained_delta_l2": 0.4492600234490653, + "delta_retention_norm_ratio": 0.7557866475423278, + "delta_cosine": 0.5787230047675087, + "relative_weight_rmse": 0.009479510985493263 + }, + { + "index": 215, + "total": 252, + "projection": "layers.4.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1412, + "parameters": 16777216, + "reference_delta_l2": 1.6304020009660627, + "retained_delta_l2": 1.416930836773217, + "delta_retention_norm_ratio": 0.8690683867743307, + "delta_cosine": 0.6812520125552476, + "relative_weight_rmse": 0.011814405088465107 + }, + { + "index": 216, + "total": 252, + "projection": "layers.4.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1451, + "parameters": 16777216, + "reference_delta_l2": 1.2640944606248181, + "retained_delta_l2": 0.9998931013773171, + "delta_retention_norm_ratio": 0.790995556521218, + "delta_cosine": 0.6087871075666293, + "relative_weight_rmse": 0.010257334794911078 + }, + { + "index": 217, + "total": 252, + "projection": "layers.4.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0334, + "parameters": 4194304, + "reference_delta_l2": 1.1347308271421606, + "retained_delta_l2": 1.0768493892373605, + "delta_retention_norm_ratio": 0.9489910412934005, + "delta_cosine": 0.7689547489767203, + "relative_weight_rmse": 0.013530562159350277 + }, + { + "index": 218, + "total": 252, + "projection": "layers.5.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.4713, + "parameters": 50331648, + "reference_delta_l2": 1.810935893075941, + "retained_delta_l2": 1.3499338899161304, + "delta_retention_norm_ratio": 0.7454343884162671, + "delta_cosine": 0.5704132363093911, + "relative_weight_rmse": 0.009071446413057554 + }, + { + "index": 219, + "total": 252, + "projection": "layers.5.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.4156, + "parameters": 50331648, + "reference_delta_l2": 2.959942527378491, + "retained_delta_l2": 2.496841727204632, + "delta_retention_norm_ratio": 0.8435439891517049, + "delta_cosine": 0.660969789060384, + "relative_weight_rmse": 0.010010880134384415 + }, + { + "index": 220, + "total": 252, + "projection": "layers.5.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3841, + "parameters": 50331648, + "reference_delta_l2": 3.1870400685654765, + "retained_delta_l2": 2.974385603406863, + "delta_retention_norm_ratio": 0.9332752458131687, + "delta_cosine": 0.7501308407978079, + "relative_weight_rmse": 0.01315208292183413 + }, + { + "index": 221, + "total": 252, + "projection": "layers.5.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0313, + "parameters": 4194304, + "reference_delta_l2": 0.6209587806459246, + "retained_delta_l2": 0.4889663014395137, + "delta_retention_norm_ratio": 0.7874376153130299, + "delta_cosine": 0.608188679409699, + "relative_weight_rmse": 0.009871909845280627 + }, + { + "index": 222, + "total": 252, + "projection": "layers.5.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1323, + "parameters": 16777216, + "reference_delta_l2": 1.4712027123441913, + "retained_delta_l2": 1.2367591682757846, + "delta_retention_norm_ratio": 0.8406449756370773, + "delta_cosine": 0.6526329998062314, + "relative_weight_rmse": 0.011369367401890903 + }, + { + "index": 223, + "total": 252, + "projection": "layers.5.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1323, + "parameters": 16777216, + "reference_delta_l2": 1.4727506650418816, + "retained_delta_l2": 1.2586922536897038, + "delta_retention_norm_ratio": 0.8546540046233214, + "delta_cosine": 0.6711865950328052, + "relative_weight_rmse": 0.011210233891934967 + }, + { + "index": 224, + "total": 252, + "projection": "layers.5.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0301, + "parameters": 4194304, + "reference_delta_l2": 1.0486084194273977, + "retained_delta_l2": 0.9844592573104753, + "delta_retention_norm_ratio": 0.9388244830687593, + "delta_cosine": 0.7614519153311239, + "relative_weight_rmse": 0.012891174674649803 + }, + { + "index": 225, + "total": 252, + "projection": "layers.6.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.4564, + "parameters": 50331648, + "reference_delta_l2": 4.767720688111135, + "retained_delta_l2": 4.780773825148703, + "delta_retention_norm_ratio": 1.0027378149626336, + "delta_cosine": 0.8442279588688661, + "relative_weight_rmse": 0.015436456019849167 + }, + { + "index": 226, + "total": 252, + "projection": "layers.6.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.4283, + "parameters": 50331648, + "reference_delta_l2": 3.1144168145379054, + "retained_delta_l2": 2.7248036794985953, + "delta_retention_norm_ratio": 0.8749001311511613, + "delta_cosine": 0.6926207132666661, + "relative_weight_rmse": 0.010703368543273093 + }, + { + "index": 227, + "total": 252, + "projection": "layers.6.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.4189, + "parameters": 50331648, + "reference_delta_l2": 3.390041280945176, + "retained_delta_l2": 3.202277510368192, + "delta_retention_norm_ratio": 0.9446131315177868, + "delta_cosine": 0.7632713560900984, + "relative_weight_rmse": 0.013229654669870288 + }, + { + "index": 228, + "total": 252, + "projection": "layers.6.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0311, + "parameters": 4194304, + "reference_delta_l2": 0.5939044653381573, + "retained_delta_l2": 0.44887094795012406, + "delta_retention_norm_ratio": 0.755796553397769, + "delta_cosine": 0.5788145863006533, + "relative_weight_rmse": 0.009595333248779364 + }, + { + "index": 229, + "total": 252, + "projection": "layers.6.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1315, + "parameters": 16777216, + "reference_delta_l2": 1.3669632941100178, + "retained_delta_l2": 1.108979268810084, + "delta_retention_norm_ratio": 0.8112721633334725, + "delta_cosine": 0.6269790068485096, + "relative_weight_rmse": 0.010406969547164474 + }, + { + "index": 230, + "total": 252, + "projection": "layers.6.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1314, + "parameters": 16777216, + "reference_delta_l2": 1.4041762978890737, + "retained_delta_l2": 1.1651334773217623, + "delta_retention_norm_ratio": 0.8297629571680784, + "delta_cosine": 0.6448494218474633, + "relative_weight_rmse": 0.010729181201098165 + }, + { + "index": 231, + "total": 252, + "projection": "layers.6.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0303, + "parameters": 4194304, + "reference_delta_l2": 0.9871587580367326, + "retained_delta_l2": 0.9114799363593897, + "delta_retention_norm_ratio": 0.9233367266802622, + "delta_cosine": 0.7398639008846931, + "relative_weight_rmse": 0.012405099189583267 + }, + { + "index": 232, + "total": 252, + "projection": "layers.7.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.4273, + "parameters": 50331648, + "reference_delta_l2": 1.8509328174395354, + "retained_delta_l2": 1.371207388640461, + "delta_retention_norm_ratio": 0.7408196427881718, + "delta_cosine": 0.5662870907172969, + "relative_weight_rmse": 0.008982772613857194 + }, + { + "index": 233, + "total": 252, + "projection": "layers.7.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.4221, + "parameters": 50331648, + "reference_delta_l2": 2.903415555498348, + "retained_delta_l2": 2.489190538369936, + "delta_retention_norm_ratio": 0.8573318186079244, + "delta_cosine": 0.6743226108497193, + "relative_weight_rmse": 0.010368765252329331 + }, + { + "index": 234, + "total": 252, + "projection": "layers.7.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.4326, + "parameters": 50331648, + "reference_delta_l2": 2.9793293298215597, + "retained_delta_l2": 2.6967054131669843, + "delta_retention_norm_ratio": 0.9051384102369433, + "delta_cosine": 0.7168754766014962, + "relative_weight_rmse": 0.012394034156126544 + }, + { + "index": 235, + "total": 252, + "projection": "layers.7.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.035, + "parameters": 4194304, + "reference_delta_l2": 0.7233149380725827, + "retained_delta_l2": 0.6142220006851181, + "delta_retention_norm_ratio": 0.8491764352632277, + "delta_cosine": 0.6628629509564983, + "relative_weight_rmse": 0.010779587582608923 + }, + { + "index": 236, + "total": 252, + "projection": "layers.7.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1457, + "parameters": 16777216, + "reference_delta_l2": 1.3687782675178408, + "retained_delta_l2": 1.1095628123085044, + "delta_retention_norm_ratio": 0.8106227565408377, + "delta_cosine": 0.6257796839251223, + "relative_weight_rmse": 0.010566147208911206 + }, + { + "index": 237, + "total": 252, + "projection": "layers.7.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1463, + "parameters": 16777216, + "reference_delta_l2": 1.6879029091575957, + "retained_delta_l2": 1.5262817016236274, + "delta_retention_norm_ratio": 0.9042473316106608, + "delta_cosine": 0.7181181553334522, + "relative_weight_rmse": 0.012129219555857563 + }, + { + "index": 238, + "total": 252, + "projection": "layers.7.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0333, + "parameters": 4194304, + "reference_delta_l2": 0.9915028517782294, + "retained_delta_l2": 0.8953762433861933, + "delta_retention_norm_ratio": 0.903049589600639, + "delta_cosine": 0.7182410000696634, + "relative_weight_rmse": 0.012000415390473564 + }, + { + "index": 239, + "total": 252, + "projection": "layers.8.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.4713, + "parameters": 50331648, + "reference_delta_l2": 2.6473504917942394, + "retained_delta_l2": 2.2863272165665895, + "delta_retention_norm_ratio": 0.8636284555646553, + "delta_cosine": 0.6778262138844748, + "relative_weight_rmse": 0.011076389763901337 + }, + { + "index": 240, + "total": 252, + "projection": "layers.8.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.4285, + "parameters": 50331648, + "reference_delta_l2": 2.2109602018958143, + "retained_delta_l2": 1.6686687050895774, + "delta_retention_norm_ratio": 0.7547257990708098, + "delta_cosine": 0.5779876357757265, + "relative_weight_rmse": 0.009495656840626325 + }, + { + "index": 241, + "total": 252, + "projection": "layers.8.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.4312, + "parameters": 50331648, + "reference_delta_l2": 2.6841490883977883, + "retained_delta_l2": 2.2893303403846996, + "delta_retention_norm_ratio": 0.8529072957535446, + "delta_cosine": 0.6661517839052589, + "relative_weight_rmse": 0.011442145437871401 + }, + { + "index": 242, + "total": 252, + "projection": "layers.8.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0341, + "parameters": 4194304, + "reference_delta_l2": 0.8951759855086093, + "retained_delta_l2": 0.8121689642323364, + "delta_retention_norm_ratio": 0.9072729579211053, + "delta_cosine": 0.7203602186120661, + "relative_weight_rmse": 0.01214431881927836 + }, + { + "index": 243, + "total": 252, + "projection": "layers.8.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1463, + "parameters": 16777216, + "reference_delta_l2": 1.574753691969974, + "retained_delta_l2": 1.349399729505425, + "delta_retention_norm_ratio": 0.8568957395599832, + "delta_cosine": 0.6708681427720062, + "relative_weight_rmse": 0.011318433107251004 + }, + { + "index": 244, + "total": 252, + "projection": "layers.8.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1475, + "parameters": 16777216, + "reference_delta_l2": 1.5969319851881418, + "retained_delta_l2": 1.386485111750515, + "delta_retention_norm_ratio": 0.8682180109174573, + "delta_cosine": 0.6796174143683574, + "relative_weight_rmse": 0.011850598231748189 + }, + { + "index": 245, + "total": 252, + "projection": "layers.8.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0348, + "parameters": 4194304, + "reference_delta_l2": 0.796218156170733, + "retained_delta_l2": 0.6602160870038808, + "delta_retention_norm_ratio": 0.8291899423382537, + "delta_cosine": 0.643318340596573, + "relative_weight_rmse": 0.01062635570519243 + }, + { + "index": 246, + "total": 252, + "projection": "layers.9.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.56, + "parameters": 50331648, + "reference_delta_l2": 2.472573355563781, + "retained_delta_l2": 2.1019195610207904, + "delta_retention_norm_ratio": 0.8500939138129324, + "delta_cosine": 0.6644626423954203, + "relative_weight_rmse": 0.010895117125517034 + }, + { + "index": 247, + "total": 252, + "projection": "layers.9.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.5764, + "parameters": 50331648, + "reference_delta_l2": 2.715301678180105, + "retained_delta_l2": 2.228991511458173, + "delta_retention_norm_ratio": 0.8209001339962065, + "delta_cosine": 0.6368404026661941, + "relative_weight_rmse": 0.010415436832995234 + }, + { + "index": 248, + "total": 252, + "projection": "layers.9.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.5713, + "parameters": 50331648, + "reference_delta_l2": 3.033076984806482, + "retained_delta_l2": 2.752774106779336, + "delta_retention_norm_ratio": 0.90758464772531, + "delta_cosine": 0.7225313110499669, + "relative_weight_rmse": 0.012405972152059165 + }, + { + "index": 249, + "total": 252, + "projection": "layers.9.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0454, + "parameters": 4194304, + "reference_delta_l2": 0.8393531099281533, + "retained_delta_l2": 0.7571706014921384, + "delta_retention_norm_ratio": 0.902088277908389, + "delta_cosine": 0.7181965567767787, + "relative_weight_rmse": 0.0117983148448792 + }, + { + "index": 250, + "total": 252, + "projection": "layers.9.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1946, + "parameters": 16777216, + "reference_delta_l2": 1.5267188967900203, + "retained_delta_l2": 1.3014151099198887, + "delta_retention_norm_ratio": 0.8524261490809864, + "delta_cosine": 0.6668987375072588, + "relative_weight_rmse": 0.011165379692474192 + }, + { + "index": 251, + "total": 252, + "projection": "layers.9.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1936, + "parameters": 16777216, + "reference_delta_l2": 1.8066357912508113, + "retained_delta_l2": 1.6600406497686675, + "delta_retention_norm_ratio": 0.9188573910734661, + "delta_cosine": 0.7351219258583319, + "relative_weight_rmse": 0.012711553632753296 + }, + { + "index": 252, + "total": 252, + "projection": "layers.9.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0351, + "parameters": 4194304, + "reference_delta_l2": 0.8017069173020834, + "retained_delta_l2": 0.6713766264992046, + "delta_retention_norm_ratio": 0.8374339949049358, + "delta_cosine": 0.6535165562030031, + "relative_weight_rmse": 0.010435042306830964 + } + ] +}