diff --git "a/manifests/merge_step_00005000.json" "b/manifests/merge_step_00005000.json" new file mode 100644--- /dev/null +++ "b/manifests/merge_step_00005000.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_00005000/adapters.safetensors", + "adapter_sha256": "39925cea8660d30a05ffe4b899903351d4c53b80d950efc0e1667837c7a1cb9c", + "adapter_global_step": "5000", + "output": "/workspace/Ideogram/artifacts/huggingface_release/s020_v1/qwen3vl_8b_ideogram4_nl_s020_v1_step_00005000_fp8_scaled.safetensors", + "output_bytes": 10588637800, + "output_sha256": "372d0f8035190ec4034a7535d84850209f3dc4cc0befb1a8fa4e825429dd6704", + "output_tensors": 1254, + "merged_projections": 252, + "lora_rank": 64, + "lora_alpha": 64.0, + "elapsed_seconds": 62.58, + "projections": [ + { + "index": 1, + "total": 252, + "projection": "layers.0.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.424, + "parameters": 50331648, + "reference_delta_l2": 3.135186515348298, + "retained_delta_l2": 2.8560539722592195, + "delta_retention_norm_ratio": 0.9109678031202975, + "delta_cosine": 0.7262562008756012, + "relative_weight_rmse": 0.01233335680315245 + }, + { + "index": 2, + "total": 252, + "projection": "layers.0.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3849, + "parameters": 50331648, + "reference_delta_l2": 3.3614798551881813, + "retained_delta_l2": 3.0834443642268603, + "delta_retention_norm_ratio": 0.9172877711784604, + "delta_cosine": 0.7301667356431611, + "relative_weight_rmse": 0.013623613025763162 + }, + { + "index": 3, + "total": 252, + "projection": "layers.0.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3797, + "parameters": 50331648, + "reference_delta_l2": 2.9777154449787466, + "retained_delta_l2": 2.673958036520115, + "delta_retention_norm_ratio": 0.8979897797249732, + "delta_cosine": 0.7112452828018604, + "relative_weight_rmse": 0.013140300556275497 + }, + { + "index": 4, + "total": 252, + "projection": "layers.0.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.029, + "parameters": 4194304, + "reference_delta_l2": 0.6049897209190485, + "retained_delta_l2": 0.4440773425327372, + "delta_retention_norm_ratio": 0.7340246076547102, + "delta_cosine": 0.5617987633284574, + "relative_weight_rmse": 0.009037651463493121 + }, + { + "index": 5, + "total": 252, + "projection": "layers.0.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.128, + "parameters": 16777216, + "reference_delta_l2": 1.9775603155065633, + "retained_delta_l2": 1.865948352780388, + "delta_retention_norm_ratio": 0.94356077948622, + "delta_cosine": 0.7631682046102871, + "relative_weight_rmse": 0.013644203744777697 + }, + { + "index": 6, + "total": 252, + "projection": "layers.0.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1281, + "parameters": 16777216, + "reference_delta_l2": 1.2302239091902034, + "retained_delta_l2": 0.9587679575670257, + "delta_retention_norm_ratio": 0.77934427253827, + "delta_cosine": 0.5999597508273914, + "relative_weight_rmse": 0.00983482103229521 + }, + { + "index": 7, + "total": 252, + "projection": "layers.0.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0295, + "parameters": 4194304, + "reference_delta_l2": 0.7989491932896987, + "retained_delta_l2": 0.696997109080163, + "delta_retention_norm_ratio": 0.8723922809287225, + "delta_cosine": 0.6852699405629848, + "relative_weight_rmse": 0.012018497319098176 + }, + { + "index": 8, + "total": 252, + "projection": "layers.1.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.4177, + "parameters": 50331648, + "reference_delta_l2": 2.039540808786972, + "retained_delta_l2": 1.904364953570825, + "delta_retention_norm_ratio": 0.9337224072037353, + "delta_cosine": 0.7567914664358877, + "relative_weight_rmse": 0.011997490454341045 + }, + { + "index": 9, + "total": 252, + "projection": "layers.1.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3791, + "parameters": 50331648, + "reference_delta_l2": 3.8236958190838717, + "retained_delta_l2": 3.7907148274017604, + "delta_retention_norm_ratio": 0.9913745775703432, + "delta_cosine": 0.845240793486271, + "relative_weight_rmse": 0.01436090278758579 + }, + { + "index": 10, + "total": 252, + "projection": "layers.1.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3781, + "parameters": 50331648, + "reference_delta_l2": 4.233003496547231, + "retained_delta_l2": 4.230068095208527, + "delta_retention_norm_ratio": 0.9993065440789031, + "delta_cosine": 0.9295672671061014, + "relative_weight_rmse": 0.015765239811592596 + }, + { + "index": 11, + "total": 252, + "projection": "layers.1.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0272, + "parameters": 4194304, + "reference_delta_l2": 0.6780480021025355, + "retained_delta_l2": 0.5573929109150937, + "delta_retention_norm_ratio": 0.82205523677777, + "delta_cosine": 0.6380309659793468, + "relative_weight_rmse": 0.010838267313523747 + }, + { + "index": 12, + "total": 252, + "projection": "layers.1.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.126, + "parameters": 16777216, + "reference_delta_l2": 1.7197743724890977, + "retained_delta_l2": 1.5400788437254787, + "delta_retention_norm_ratio": 0.8955121487805761, + "delta_cosine": 0.7062720073601932, + "relative_weight_rmse": 0.012689735653416788 + }, + { + "index": 13, + "total": 252, + "projection": "layers.1.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1288, + "parameters": 16777216, + "reference_delta_l2": 2.093652038037236, + "retained_delta_l2": 2.0243876503438547, + "delta_retention_norm_ratio": 0.9669169535171108, + "delta_cosine": 0.7927280112007618, + "relative_weight_rmse": 0.014642959288955999 + }, + { + "index": 14, + "total": 252, + "projection": "layers.1.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0288, + "parameters": 4194304, + "reference_delta_l2": 0.9102291599732378, + "retained_delta_l2": 0.8269043748425299, + "delta_retention_norm_ratio": 0.9084573547026797, + "delta_cosine": 0.7210325993344192, + "relative_weight_rmse": 0.01239533413987135 + }, + { + "index": 15, + "total": 252, + "projection": "layers.10.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.4198, + "parameters": 50331648, + "reference_delta_l2": 2.815750646038315, + "retained_delta_l2": 2.497199691402181, + "delta_retention_norm_ratio": 0.8868681944246994, + "delta_cosine": 0.7007905993418094, + "relative_weight_rmse": 0.01167729700137916 + }, + { + "index": 16, + "total": 252, + "projection": "layers.10.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3784, + "parameters": 50331648, + "reference_delta_l2": 2.3989329841884732, + "retained_delta_l2": 1.861966849896403, + "delta_retention_norm_ratio": 0.7761645957468385, + "delta_cosine": 0.5968364007448302, + "relative_weight_rmse": 0.009680572880202264 + }, + { + "index": 17, + "total": 252, + "projection": "layers.10.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3783, + "parameters": 50331648, + "reference_delta_l2": 2.8571875232232324, + "retained_delta_l2": 2.5284986641654177, + "delta_retention_norm_ratio": 0.884960697753917, + "delta_cosine": 0.6974691920334167, + "relative_weight_rmse": 0.011912326044391911 + }, + { + "index": 18, + "total": 252, + "projection": "layers.10.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0273, + "parameters": 4194304, + "reference_delta_l2": 0.9003451220266855, + "retained_delta_l2": 0.8277931750356239, + "delta_retention_norm_ratio": 0.9194176264011443, + "delta_cosine": 0.737629177890454, + "relative_weight_rmse": 0.011991794454569911 + }, + { + "index": 19, + "total": 252, + "projection": "layers.10.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1279, + "parameters": 16777216, + "reference_delta_l2": 1.7763426698945382, + "retained_delta_l2": 1.5878093392627646, + "delta_retention_norm_ratio": 0.8938643236876324, + "delta_cosine": 0.7070843944769404, + "relative_weight_rmse": 0.012178674526479671 + }, + { + "index": 20, + "total": 252, + "projection": "layers.10.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1294, + "parameters": 16777216, + "reference_delta_l2": 2.123701389490391, + "retained_delta_l2": 2.0266935532719232, + "delta_retention_norm_ratio": 0.9543213388197922, + "delta_cosine": 0.7804707573447506, + "relative_weight_rmse": 0.013468296606334093 + }, + { + "index": 21, + "total": 252, + "projection": "layers.10.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0286, + "parameters": 4194304, + "reference_delta_l2": 0.9139935802992992, + "retained_delta_l2": 0.817229615603824, + "delta_retention_norm_ratio": 0.8941305860553326, + "delta_cosine": 0.7123253822564569, + "relative_weight_rmse": 0.011238932459847967 + }, + { + "index": 22, + "total": 252, + "projection": "layers.11.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.4206, + "parameters": 50331648, + "reference_delta_l2": 2.611804809629326, + "retained_delta_l2": 2.2360590955736663, + "delta_retention_norm_ratio": 0.8561356068147427, + "delta_cosine": 0.6706593698553941, + "relative_weight_rmse": 0.011059342359572706 + }, + { + "index": 23, + "total": 252, + "projection": "layers.11.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3802, + "parameters": 50331648, + "reference_delta_l2": 2.787149920660926, + "retained_delta_l2": 2.356436835146604, + "delta_retention_norm_ratio": 0.8454646869472361, + "delta_cosine": 0.6607703373976561, + "relative_weight_rmse": 0.0108500592682522 + }, + { + "index": 24, + "total": 252, + "projection": "layers.11.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3808, + "parameters": 50331648, + "reference_delta_l2": 2.680144008384116, + "retained_delta_l2": 2.2789481194653325, + "delta_retention_norm_ratio": 0.8503080850641798, + "delta_cosine": 0.6638271405022999, + "relative_weight_rmse": 0.01139116263399098 + }, + { + "index": 25, + "total": 252, + "projection": "layers.11.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0273, + "parameters": 4194304, + "reference_delta_l2": 1.1366927936122104, + "retained_delta_l2": 1.0948450976181021, + "delta_retention_norm_ratio": 0.9631846913877902, + "delta_cosine": 0.7886591368315472, + "relative_weight_rmse": 0.013711924887066174 + }, + { + "index": 26, + "total": 252, + "projection": "layers.11.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1275, + "parameters": 16777216, + "reference_delta_l2": 1.6714846962330223, + "retained_delta_l2": 1.4679641251206186, + "delta_retention_norm_ratio": 0.8782396443287382, + "delta_cosine": 0.6931109771143484, + "relative_weight_rmse": 0.01173579706583997 + }, + { + "index": 27, + "total": 252, + "projection": "layers.11.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1293, + "parameters": 16777216, + "reference_delta_l2": 1.4149194896492567, + "retained_delta_l2": 1.1792060860749967, + "delta_retention_norm_ratio": 0.8334086106675294, + "delta_cosine": 0.6480283418324282, + "relative_weight_rmse": 0.010798380947022401 + }, + { + "index": 28, + "total": 252, + "projection": "layers.11.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0288, + "parameters": 4194304, + "reference_delta_l2": 1.0015595060547162, + "retained_delta_l2": 0.912226688257879, + "delta_retention_norm_ratio": 0.9108062803490011, + "delta_cosine": 0.7280353129937391, + "relative_weight_rmse": 0.011916026681035733 + }, + { + "index": 29, + "total": 252, + "projection": "layers.12.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.4208, + "parameters": 50331648, + "reference_delta_l2": 2.570622595928691, + "retained_delta_l2": 2.190965410117967, + "delta_retention_norm_ratio": 0.8523092474126622, + "delta_cosine": 0.6669268187339376, + "relative_weight_rmse": 0.010844754236504215 + }, + { + "index": 30, + "total": 252, + "projection": "layers.12.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3817, + "parameters": 50331648, + "reference_delta_l2": 3.2308685741393934, + "retained_delta_l2": 2.9004585854706004, + "delta_retention_norm_ratio": 0.8977333862127758, + "delta_cosine": 0.7138159179539258, + "relative_weight_rmse": 0.01202862716931747 + }, + { + "index": 31, + "total": 252, + "projection": "layers.12.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3801, + "parameters": 50331648, + "reference_delta_l2": 2.6925897553775506, + "retained_delta_l2": 2.296402796685541, + "delta_retention_norm_ratio": 0.8528602591981351, + "delta_cosine": 0.666576856045294, + "relative_weight_rmse": 0.01123944526908056 + }, + { + "index": 32, + "total": 252, + "projection": "layers.12.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0272, + "parameters": 4194304, + "reference_delta_l2": 0.9321993592192336, + "retained_delta_l2": 0.8654938512955905, + "delta_retention_norm_ratio": 0.9284428730142955, + "delta_cosine": 0.7476333638462151, + "relative_weight_rmse": 0.012290617430907062 + }, + { + "index": 33, + "total": 252, + "projection": "layers.12.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1293, + "parameters": 16777216, + "reference_delta_l2": 1.7746351817899557, + "retained_delta_l2": 1.607432307392857, + "delta_retention_norm_ratio": 0.905781832732262, + "delta_cosine": 0.7218634704923429, + "relative_weight_rmse": 0.01203951255127609 + }, + { + "index": 34, + "total": 252, + "projection": "layers.12.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1292, + "parameters": 16777216, + "reference_delta_l2": 1.807689998181352, + "retained_delta_l2": 1.652943713714136, + "delta_retention_norm_ratio": 0.9143955630540079, + "delta_cosine": 0.731369216769661, + "relative_weight_rmse": 0.01244067252691149 + }, + { + "index": 35, + "total": 252, + "projection": "layers.12.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.029, + "parameters": 4194304, + "reference_delta_l2": 1.088034545248337, + "retained_delta_l2": 1.025166821321755, + "delta_retention_norm_ratio": 0.9422190001216985, + "delta_cosine": 0.7662832512568094, + "relative_weight_rmse": 0.012196326546506273 + }, + { + "index": 36, + "total": 252, + "projection": "layers.13.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.424, + "parameters": 50331648, + "reference_delta_l2": 2.349516064651879, + "retained_delta_l2": 1.9253748764861496, + "delta_retention_norm_ratio": 0.819477212968717, + "delta_cosine": 0.6362778844407208, + "relative_weight_rmse": 0.010206561115358198 + }, + { + "index": 37, + "total": 252, + "projection": "layers.13.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3827, + "parameters": 50331648, + "reference_delta_l2": 3.006984984137078, + "retained_delta_l2": 2.6350765456604233, + "delta_retention_norm_ratio": 0.8763184916324473, + "delta_cosine": 0.6891116610958277, + "relative_weight_rmse": 0.01193392488554407 + }, + { + "index": 38, + "total": 252, + "projection": "layers.13.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3794, + "parameters": 50331648, + "reference_delta_l2": 2.4074461436722676, + "retained_delta_l2": 1.9481472708324796, + "delta_retention_norm_ratio": 0.8092173841367087, + "delta_cosine": 0.6257388809621606, + "relative_weight_rmse": 0.010368056418957074 + }, + { + "index": 39, + "total": 252, + "projection": "layers.13.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.027, + "parameters": 4194304, + "reference_delta_l2": 1.344359788194323, + "retained_delta_l2": 1.3301268872832719, + "delta_retention_norm_ratio": 0.9894128781327444, + "delta_cosine": 0.834898539661817, + "relative_weight_rmse": 0.01479261987327952 + }, + { + "index": 40, + "total": 252, + "projection": "layers.13.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1273, + "parameters": 16777216, + "reference_delta_l2": 1.7332866148034838, + "retained_delta_l2": 1.5541717587654111, + "delta_retention_norm_ratio": 0.8966617208554511, + "delta_cosine": 0.7124019684501789, + "relative_weight_rmse": 0.012277217442899068 + }, + { + "index": 41, + "total": 252, + "projection": "layers.13.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1293, + "parameters": 16777216, + "reference_delta_l2": 1.4129397612140535, + "retained_delta_l2": 1.1770620718772369, + "delta_retention_norm_ratio": 0.8330589202655453, + "delta_cosine": 0.6476767900380466, + "relative_weight_rmse": 0.01076806653608202 + }, + { + "index": 42, + "total": 252, + "projection": "layers.13.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0286, + "parameters": 4194304, + "reference_delta_l2": 1.035136168968896, + "retained_delta_l2": 0.9550400803251853, + "delta_retention_norm_ratio": 0.9226226548305284, + "delta_cosine": 0.7401943174497557, + "relative_weight_rmse": 0.012365144877185026 + }, + { + "index": 43, + "total": 252, + "projection": "layers.14.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.4201, + "parameters": 50331648, + "reference_delta_l2": 2.470701762151342, + "retained_delta_l2": 2.0898129981915057, + "delta_retention_norm_ratio": 0.8458378223569238, + "delta_cosine": 0.6614596710529592, + "relative_weight_rmse": 0.010529548316267456 + }, + { + "index": 44, + "total": 252, + "projection": "layers.14.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3778, + "parameters": 50331648, + "reference_delta_l2": 2.74722649797281, + "retained_delta_l2": 2.3340647564629458, + "delta_retention_norm_ratio": 0.8496076891312974, + "delta_cosine": 0.6627933337041668, + "relative_weight_rmse": 0.011402475625767456 + }, + { + "index": 45, + "total": 252, + "projection": "layers.14.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.38, + "parameters": 50331648, + "reference_delta_l2": 2.9501600583233736, + "retained_delta_l2": 2.6162905939812773, + "delta_retention_norm_ratio": 0.8868300506611021, + "delta_cosine": 0.7012459300014316, + "relative_weight_rmse": 0.011800006904719555 + }, + { + "index": 46, + "total": 252, + "projection": "layers.14.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0268, + "parameters": 4194304, + "reference_delta_l2": 0.729918443155983, + "retained_delta_l2": 0.6200392380407965, + "delta_retention_norm_ratio": 0.8494637227686747, + "delta_cosine": 0.6664010596012043, + "relative_weight_rmse": 0.010731083800672671 + }, + { + "index": 47, + "total": 252, + "projection": "layers.14.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1275, + "parameters": 16777216, + "reference_delta_l2": 1.5591070036567194, + "retained_delta_l2": 1.357860744696475, + "delta_retention_norm_ratio": 0.8709220993246503, + "delta_cosine": 0.685239627250001, + "relative_weight_rmse": 0.011461754476629643 + }, + { + "index": 48, + "total": 252, + "projection": "layers.14.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1286, + "parameters": 16777216, + "reference_delta_l2": 2.317079032780429, + "retained_delta_l2": 2.258352424426768, + "delta_retention_norm_ratio": 0.9746548963057203, + "delta_cosine": 0.8053928778272853, + "relative_weight_rmse": 0.013909891998684688 + }, + { + "index": 49, + "total": 252, + "projection": "layers.14.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0287, + "parameters": 4194304, + "reference_delta_l2": 1.019153393597738, + "retained_delta_l2": 0.9467406420311555, + "delta_retention_norm_ratio": 0.9289481328115325, + "delta_cosine": 0.7464038423557939, + "relative_weight_rmse": 0.012392165813170427 + }, + { + "index": 50, + "total": 252, + "projection": "layers.15.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.4189, + "parameters": 50331648, + "reference_delta_l2": 2.3712801332115747, + "retained_delta_l2": 1.97260584174988, + "delta_retention_norm_ratio": 0.8318738111630257, + "delta_cosine": 0.6479866019547492, + "relative_weight_rmse": 0.010302002598645308 + }, + { + "index": 51, + "total": 252, + "projection": "layers.15.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3789, + "parameters": 50331648, + "reference_delta_l2": 2.8999020234196347, + "retained_delta_l2": 2.5435002854085984, + "delta_retention_norm_ratio": 0.877098696737775, + "delta_cosine": 0.6889267286180097, + "relative_weight_rmse": 0.012094963529580971 + }, + { + "index": 52, + "total": 252, + "projection": "layers.15.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3795, + "parameters": 50331648, + "reference_delta_l2": 2.789907870604423, + "retained_delta_l2": 2.4409558059046863, + "delta_retention_norm_ratio": 0.8749234451874075, + "delta_cosine": 0.6892638530383427, + "relative_weight_rmse": 0.011430830430350996 + }, + { + "index": 53, + "total": 252, + "projection": "layers.15.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0272, + "parameters": 4194304, + "reference_delta_l2": 0.8181995214514396, + "retained_delta_l2": 0.7244323583273169, + "delta_retention_norm_ratio": 0.8853981691925399, + "delta_cosine": 0.7001481056499137, + "relative_weight_rmse": 0.011632250841184306 + }, + { + "index": 54, + "total": 252, + "projection": "layers.15.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1266, + "parameters": 16777216, + "reference_delta_l2": 2.1489820829006403, + "retained_delta_l2": 2.064650844097545, + "delta_retention_norm_ratio": 0.9607575886862365, + "delta_cosine": 0.7884630546129239, + "relative_weight_rmse": 0.01383446779953585 + }, + { + "index": 55, + "total": 252, + "projection": "layers.15.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1287, + "parameters": 16777216, + "reference_delta_l2": 2.239629875906978, + "retained_delta_l2": 2.164612144458897, + "delta_retention_norm_ratio": 0.9665044067079603, + "delta_cosine": 0.8012332235464946, + "relative_weight_rmse": 0.013461916248059555 + }, + { + "index": 56, + "total": 252, + "projection": "layers.15.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0286, + "parameters": 4194304, + "reference_delta_l2": 0.8622840472553119, + "retained_delta_l2": 0.7672109892395579, + "delta_retention_norm_ratio": 0.8897427613112225, + "delta_cosine": 0.7062963769690268, + "relative_weight_rmse": 0.011264688962582484 + }, + { + "index": 57, + "total": 252, + "projection": "layers.16.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.423, + "parameters": 50331648, + "reference_delta_l2": 2.492997818948942, + "retained_delta_l2": 2.130565572114366, + "delta_retention_norm_ratio": 0.85461991018212, + "delta_cosine": 0.6693559838294143, + "relative_weight_rmse": 0.010670349074915835 + }, + { + "index": 58, + "total": 252, + "projection": "layers.16.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3799, + "parameters": 50331648, + "reference_delta_l2": 2.6530217062891728, + "retained_delta_l2": 2.271999747972484, + "delta_retention_norm_ratio": 0.856381891858084, + "delta_cosine": 0.6697329841804793, + "relative_weight_rmse": 0.011111090959197583 + }, + { + "index": 59, + "total": 252, + "projection": "layers.16.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3799, + "parameters": 50331648, + "reference_delta_l2": 2.5590487926250303, + "retained_delta_l2": 2.1577905132080555, + "delta_retention_norm_ratio": 0.8432002232339733, + "delta_cosine": 0.6572806546391785, + "relative_weight_rmse": 0.010895653258080075 + }, + { + "index": 60, + "total": 252, + "projection": "layers.16.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0274, + "parameters": 4194304, + "reference_delta_l2": 0.7421347571068885, + "retained_delta_l2": 0.6397912949145651, + "delta_retention_norm_ratio": 0.86209585090544, + "delta_cosine": 0.67602087587244, + "relative_weight_rmse": 0.011236792558610478 + }, + { + "index": 61, + "total": 252, + "projection": "layers.16.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1272, + "parameters": 16777216, + "reference_delta_l2": 1.6688992374074818, + "retained_delta_l2": 1.4931286300640905, + "delta_retention_norm_ratio": 0.8946787179216171, + "delta_cosine": 0.7097863074539114, + "relative_weight_rmse": 0.01194892788211655 + }, + { + "index": 62, + "total": 252, + "projection": "layers.16.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1293, + "parameters": 16777216, + "reference_delta_l2": 1.8168048948215056, + "retained_delta_l2": 1.6859507799228413, + "delta_retention_norm_ratio": 0.9279756922322029, + "delta_cosine": 0.7488695954478054, + "relative_weight_rmse": 0.012491895612114922 + }, + { + "index": 63, + "total": 252, + "projection": "layers.16.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0285, + "parameters": 4194304, + "reference_delta_l2": 0.9559863902201, + "retained_delta_l2": 0.8724007989316481, + "delta_retention_norm_ratio": 0.9125661284056484, + "delta_cosine": 0.7294168827875759, + "relative_weight_rmse": 0.012127487179049738 + }, + { + "index": 64, + "total": 252, + "projection": "layers.17.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.4181, + "parameters": 50331648, + "reference_delta_l2": 2.5826459749642634, + "retained_delta_l2": 2.222327270085148, + "delta_retention_norm_ratio": 0.8604846702289108, + "delta_cosine": 0.674990062797647, + "relative_weight_rmse": 0.010934296981534891 + }, + { + "index": 65, + "total": 252, + "projection": "layers.17.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3788, + "parameters": 50331648, + "reference_delta_l2": 3.5677109849255655, + "retained_delta_l2": 3.3508333514846402, + "delta_retention_norm_ratio": 0.9392109858793818, + "delta_cosine": 0.7582646180832158, + "relative_weight_rmse": 0.013364162354051858 + }, + { + "index": 66, + "total": 252, + "projection": "layers.17.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3795, + "parameters": 50331648, + "reference_delta_l2": 3.189207585068944, + "retained_delta_l2": 2.905526488680171, + "delta_retention_norm_ratio": 0.9110496608258127, + "delta_cosine": 0.7256624417792032, + "relative_weight_rmse": 0.012392163139724533 + }, + { + "index": 67, + "total": 252, + "projection": "layers.17.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0273, + "parameters": 4194304, + "reference_delta_l2": 0.7881034605462751, + "retained_delta_l2": 0.7054711603223036, + "delta_retention_norm_ratio": 0.8951504410769942, + "delta_cosine": 0.7150171541628376, + "relative_weight_rmse": 0.01156482498208805 + }, + { + "index": 68, + "total": 252, + "projection": "layers.17.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1274, + "parameters": 16777216, + "reference_delta_l2": 1.7306393697451647, + "retained_delta_l2": 1.570861065501436, + "delta_retention_norm_ratio": 0.9076767193460669, + "delta_cosine": 0.7248865736065294, + "relative_weight_rmse": 0.012203306164663034 + }, + { + "index": 69, + "total": 252, + "projection": "layers.17.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1302, + "parameters": 16777216, + "reference_delta_l2": 1.6634187594720447, + "retained_delta_l2": 1.4887031861118358, + "delta_retention_norm_ratio": 0.8949659715178022, + "delta_cosine": 0.7080650198934098, + "relative_weight_rmse": 0.011982013430126896 + }, + { + "index": 70, + "total": 252, + "projection": "layers.17.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0287, + "parameters": 4194304, + "reference_delta_l2": 1.033310004312815, + "retained_delta_l2": 0.969074009228285, + "delta_retention_norm_ratio": 0.9378347303167273, + "delta_cosine": 0.7591014945813579, + "relative_weight_rmse": 0.012756530218560696 + }, + { + "index": 71, + "total": 252, + "projection": "layers.18.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.4214, + "parameters": 50331648, + "reference_delta_l2": 2.91300001332646, + "retained_delta_l2": 2.624356604242212, + "delta_retention_norm_ratio": 0.9009119781106227, + "delta_cosine": 0.7154699872689531, + "relative_weight_rmse": 0.011723653165008202 + }, + { + "index": 72, + "total": 252, + "projection": "layers.18.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3798, + "parameters": 50331648, + "reference_delta_l2": 4.773282049472027, + "retained_delta_l2": 4.740512890063904, + "delta_retention_norm_ratio": 0.9931348788802983, + "delta_cosine": 0.8289113415462789, + "relative_weight_rmse": 0.015512092849871293 + }, + { + "index": 73, + "total": 252, + "projection": "layers.18.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3821, + "parameters": 50331648, + "reference_delta_l2": 3.272086783487216, + "retained_delta_l2": 3.0113484819898617, + "delta_retention_norm_ratio": 0.9203143685512298, + "delta_cosine": 0.736889755775823, + "relative_weight_rmse": 0.012501219592350362 + }, + { + "index": 74, + "total": 252, + "projection": "layers.18.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0274, + "parameters": 4194304, + "reference_delta_l2": 1.1198283608502044, + "retained_delta_l2": 1.0866557126337533, + "delta_retention_norm_ratio": 0.9703770243939298, + "delta_cosine": 0.7993391464160701, + "relative_weight_rmse": 0.014015566126521639 + }, + { + "index": 75, + "total": 252, + "projection": "layers.18.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1273, + "parameters": 16777216, + "reference_delta_l2": 1.626564664291748, + "retained_delta_l2": 1.453394490798149, + "delta_retention_norm_ratio": 0.893536250174842, + "delta_cosine": 0.7104582955821651, + "relative_weight_rmse": 0.011687285777003188 + }, + { + "index": 76, + "total": 252, + "projection": "layers.18.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1298, + "parameters": 16777216, + "reference_delta_l2": 2.1012849876408812, + "retained_delta_l2": 2.011378937645608, + "delta_retention_norm_ratio": 0.9572137760826954, + "delta_cosine": 0.7838343385285703, + "relative_weight_rmse": 0.013551126227561699 + }, + { + "index": 77, + "total": 252, + "projection": "layers.18.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0287, + "parameters": 4194304, + "reference_delta_l2": 0.9281509567125034, + "retained_delta_l2": 0.8454780872916859, + "delta_retention_norm_ratio": 0.9109273455757203, + "delta_cosine": 0.7278002651864842, + "relative_weight_rmse": 0.01178993087135816 + }, + { + "index": 78, + "total": 252, + "projection": "layers.19.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.4211, + "parameters": 50331648, + "reference_delta_l2": 2.6922750035667775, + "retained_delta_l2": 2.3558934173555723, + "delta_retention_norm_ratio": 0.8750567509761966, + "delta_cosine": 0.6892908042194223, + "relative_weight_rmse": 0.011204255373374137 + }, + { + "index": 79, + "total": 252, + "projection": "layers.19.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3844, + "parameters": 50331648, + "reference_delta_l2": 4.656655274733582, + "retained_delta_l2": 4.5944078048499355, + "delta_retention_norm_ratio": 0.9866325793490892, + "delta_cosine": 0.8202697964519513, + "relative_weight_rmse": 0.015518642190223074 + }, + { + "index": 80, + "total": 252, + "projection": "layers.19.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3843, + "parameters": 50331648, + "reference_delta_l2": 3.0977015640031476, + "retained_delta_l2": 2.7784333744300307, + "delta_retention_norm_ratio": 0.8969338449890801, + "delta_cosine": 0.711227319018263, + "relative_weight_rmse": 0.01221220014459275 + }, + { + "index": 81, + "total": 252, + "projection": "layers.19.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0273, + "parameters": 4194304, + "reference_delta_l2": 1.047128698342103, + "retained_delta_l2": 1.0081357593174012, + "delta_retention_norm_ratio": 0.9627620376688761, + "delta_cosine": 0.7956820746471963, + "relative_weight_rmse": 0.013609249885393589 + }, + { + "index": 82, + "total": 252, + "projection": "layers.19.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1273, + "parameters": 16777216, + "reference_delta_l2": 1.640882895333352, + "retained_delta_l2": 1.4613969515577732, + "delta_retention_norm_ratio": 0.890616238193454, + "delta_cosine": 0.7054189075118479, + "relative_weight_rmse": 0.011798410714126939 + }, + { + "index": 83, + "total": 252, + "projection": "layers.19.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.129, + "parameters": 16777216, + "reference_delta_l2": 1.943267319773534, + "retained_delta_l2": 1.8345260607109972, + "delta_retention_norm_ratio": 0.9440420481752303, + "delta_cosine": 0.766395739841044, + "relative_weight_rmse": 0.012852704864556646 + }, + { + "index": 84, + "total": 252, + "projection": "layers.19.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0289, + "parameters": 4194304, + "reference_delta_l2": 1.234666897955104, + "retained_delta_l2": 1.2011835954797547, + "delta_retention_norm_ratio": 0.9728806996196258, + "delta_cosine": 0.8056964980148436, + "relative_weight_rmse": 0.013861031420803058 + }, + { + "index": 85, + "total": 252, + "projection": "layers.2.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.4205, + "parameters": 50331648, + "reference_delta_l2": 1.5526634376141815, + "retained_delta_l2": 1.33742076785188, + "delta_retention_norm_ratio": 0.8613719724777941, + "delta_cosine": 0.6980385160317994, + "relative_weight_rmse": 0.010117667896620617 + }, + { + "index": 86, + "total": 252, + "projection": "layers.2.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3801, + "parameters": 50331648, + "reference_delta_l2": 2.515285664613831, + "retained_delta_l2": 2.3728132874325953, + "delta_retention_norm_ratio": 0.9433573771816056, + "delta_cosine": 0.7979933008894563, + "relative_weight_rmse": 0.012675496269100575 + }, + { + "index": 87, + "total": 252, + "projection": "layers.2.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3861, + "parameters": 50331648, + "reference_delta_l2": 3.3820200106405816, + "retained_delta_l2": 3.362728819973336, + "delta_retention_norm_ratio": 0.9942959560834794, + "delta_cosine": 0.8802202170527957, + "relative_weight_rmse": 0.014960959971704624 + }, + { + "index": 88, + "total": 252, + "projection": "layers.2.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0276, + "parameters": 4194304, + "reference_delta_l2": 0.5780582396299822, + "retained_delta_l2": 0.4352584308054825, + "delta_retention_norm_ratio": 0.7529663984793185, + "delta_cosine": 0.5765613771409978, + "relative_weight_rmse": 0.009301718024074838 + }, + { + "index": 89, + "total": 252, + "projection": "layers.2.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1296, + "parameters": 16777216, + "reference_delta_l2": 2.0915892526659636, + "retained_delta_l2": 1.9742645164339088, + "delta_retention_norm_ratio": 0.9439064165765285, + "delta_cosine": 0.7612765502356699, + "relative_weight_rmse": 0.013967732208073852 + }, + { + "index": 90, + "total": 252, + "projection": "layers.2.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1312, + "parameters": 16777216, + "reference_delta_l2": 1.2320945095750178, + "retained_delta_l2": 0.9741780923035718, + "delta_retention_norm_ratio": 0.7906683170267448, + "delta_cosine": 0.6087487381000445, + "relative_weight_rmse": 0.01020122359884597 + }, + { + "index": 91, + "total": 252, + "projection": "layers.2.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0293, + "parameters": 4194304, + "reference_delta_l2": 1.1641128777895646, + "retained_delta_l2": 1.1237200857579628, + "delta_retention_norm_ratio": 0.9653016534716976, + "delta_cosine": 0.7939388350240377, + "relative_weight_rmse": 0.013529768739609014 + }, + { + "index": 92, + "total": 252, + "projection": "layers.20.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.4224, + "parameters": 50331648, + "reference_delta_l2": 2.6482357286299263, + "retained_delta_l2": 2.2605104788025727, + "delta_retention_norm_ratio": 0.853591111381937, + "delta_cosine": 0.6660232040520053, + "relative_weight_rmse": 0.011255390535106997 + }, + { + "index": 93, + "total": 252, + "projection": "layers.20.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3855, + "parameters": 50331648, + "reference_delta_l2": 4.595450009542096, + "retained_delta_l2": 4.532970757221847, + "delta_retention_norm_ratio": 0.9864041057588451, + "delta_cosine": 0.8190441377530484, + "relative_weight_rmse": 0.015508505324454483 + }, + { + "index": 94, + "total": 252, + "projection": "layers.20.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3851, + "parameters": 50331648, + "reference_delta_l2": 3.1309412335852236, + "retained_delta_l2": 2.8137809312180577, + "delta_retention_norm_ratio": 0.8987012918144148, + "delta_cosine": 0.7126249167235419, + "relative_weight_rmse": 0.012289729218413594 + }, + { + "index": 95, + "total": 252, + "projection": "layers.20.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0272, + "parameters": 4194304, + "reference_delta_l2": 0.8441918524059813, + "retained_delta_l2": 0.7764582601431815, + "delta_retention_norm_ratio": 0.9197651670414064, + "delta_cosine": 0.7401758309254866, + "relative_weight_rmse": 0.012199816379483584 + }, + { + "index": 96, + "total": 252, + "projection": "layers.20.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1287, + "parameters": 16777216, + "reference_delta_l2": 1.6106790746244113, + "retained_delta_l2": 1.4299995478416576, + "delta_retention_norm_ratio": 0.8878240056450191, + "delta_cosine": 0.7037405306978761, + "relative_weight_rmse": 0.011828130266026518 + }, + { + "index": 97, + "total": 252, + "projection": "layers.20.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1337, + "parameters": 16777216, + "reference_delta_l2": 1.89756966268521, + "retained_delta_l2": 1.7741818818002881, + "delta_retention_norm_ratio": 0.9349758887321593, + "delta_cosine": 0.7575586370695429, + "relative_weight_rmse": 0.012358198652403663 + }, + { + "index": 98, + "total": 252, + "projection": "layers.20.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0295, + "parameters": 4194304, + "reference_delta_l2": 0.9470939301798534, + "retained_delta_l2": 0.8695552463826534, + "delta_retention_norm_ratio": 0.9181298904719246, + "delta_cosine": 0.7375911903164596, + "relative_weight_rmse": 0.012046697274210143 + }, + { + "index": 99, + "total": 252, + "projection": "layers.21.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.4299, + "parameters": 50331648, + "reference_delta_l2": 3.1066514376986794, + "retained_delta_l2": 2.8050756971559947, + "delta_retention_norm_ratio": 0.9029257879132769, + "delta_cosine": 0.7159739154341643, + "relative_weight_rmse": 0.012307512521779558 + }, + { + "index": 100, + "total": 252, + "projection": "layers.21.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3882, + "parameters": 50331648, + "reference_delta_l2": 4.6665958222580395, + "retained_delta_l2": 4.608779402644631, + "delta_retention_norm_ratio": 0.9876105791425852, + "delta_cosine": 0.8186339309377361, + "relative_weight_rmse": 0.01599520405836275 + }, + { + "index": 101, + "total": 252, + "projection": "layers.21.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3863, + "parameters": 50331648, + "reference_delta_l2": 3.34791742619495, + "retained_delta_l2": 3.0753668378522954, + "delta_retention_norm_ratio": 0.9185910063939601, + "delta_cosine": 0.7335714757219073, + "relative_weight_rmse": 0.012775820497452597 + }, + { + "index": 102, + "total": 252, + "projection": "layers.21.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0282, + "parameters": 4194304, + "reference_delta_l2": 0.8906854892964806, + "retained_delta_l2": 0.8238880013714661, + "delta_retention_norm_ratio": 0.9250044053397846, + "delta_cosine": 0.7423613421872413, + "relative_weight_rmse": 0.012838813818728211 + }, + { + "index": 103, + "total": 252, + "projection": "layers.21.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.13, + "parameters": 16777216, + "reference_delta_l2": 1.7393285246147288, + "retained_delta_l2": 1.5911743044917448, + "delta_retention_norm_ratio": 0.9148210254553256, + "delta_cosine": 0.7326442129169288, + "relative_weight_rmse": 0.012394702418153382 + }, + { + "index": 104, + "total": 252, + "projection": "layers.21.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1316, + "parameters": 16777216, + "reference_delta_l2": 1.35023526346365, + "retained_delta_l2": 1.1176851682557851, + "delta_retention_norm_ratio": 0.8277706844869961, + "delta_cosine": 0.6443744276873155, + "relative_weight_rmse": 0.01035273223411896 + }, + { + "index": 105, + "total": 252, + "projection": "layers.21.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0299, + "parameters": 4194304, + "reference_delta_l2": 1.340558070538281, + "retained_delta_l2": 1.3237576668026891, + "delta_retention_norm_ratio": 0.9874676046455445, + "delta_cosine": 0.823507805084283, + "relative_weight_rmse": 0.0145505935204622 + }, + { + "index": 106, + "total": 252, + "projection": "layers.22.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.4239, + "parameters": 50331648, + "reference_delta_l2": 3.3651014455456614, + "retained_delta_l2": 3.1301175587978727, + "delta_retention_norm_ratio": 0.9301703409093851, + "delta_cosine": 0.7478149278286402, + "relative_weight_rmse": 0.012616241203419263 + }, + { + "index": 107, + "total": 252, + "projection": "layers.22.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3884, + "parameters": 50331648, + "reference_delta_l2": 4.236275956939207, + "retained_delta_l2": 4.124291183097534, + "delta_retention_norm_ratio": 0.973565278801482, + "delta_cosine": 0.8030949736407592, + "relative_weight_rmse": 0.014701826077672196 + }, + { + "index": 108, + "total": 252, + "projection": "layers.22.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3897, + "parameters": 50331648, + "reference_delta_l2": 3.432581665075316, + "retained_delta_l2": 3.182276464768107, + "delta_retention_norm_ratio": 0.9270796080821817, + "delta_cosine": 0.745293924915482, + "relative_weight_rmse": 0.012637390341163894 + }, + { + "index": 109, + "total": 252, + "projection": "layers.22.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.028, + "parameters": 4194304, + "reference_delta_l2": 1.1250070410829698, + "retained_delta_l2": 1.0998477214110505, + "delta_retention_norm_ratio": 0.9776363002602187, + "delta_cosine": 0.8172957846306897, + "relative_weight_rmse": 0.014426618318959437 + }, + { + "index": 110, + "total": 252, + "projection": "layers.22.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1301, + "parameters": 16777216, + "reference_delta_l2": 2.0978269166758747, + "retained_delta_l2": 1.9973749588672103, + "delta_retention_norm_ratio": 0.9521161841283664, + "delta_cosine": 0.7727043603398874, + "relative_weight_rmse": 0.01373215505568896 + }, + { + "index": 111, + "total": 252, + "projection": "layers.22.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1313, + "parameters": 16777216, + "reference_delta_l2": 1.871279276226468, + "retained_delta_l2": 1.7421599999103519, + "delta_retention_norm_ratio": 0.9309994622627938, + "delta_cosine": 0.7547089819512466, + "relative_weight_rmse": 0.012602207687986134 + }, + { + "index": 112, + "total": 252, + "projection": "layers.22.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0291, + "parameters": 4194304, + "reference_delta_l2": 1.1049204569108604, + "retained_delta_l2": 1.0524277699980864, + "delta_retention_norm_ratio": 0.9524918861041517, + "delta_cosine": 0.7775562635452063, + "relative_weight_rmse": 0.01323658225526853 + }, + { + "index": 113, + "total": 252, + "projection": "layers.23.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.4295, + "parameters": 50331648, + "reference_delta_l2": 3.164064877389265, + "retained_delta_l2": 2.904799180491484, + "delta_retention_norm_ratio": 0.9180592981039927, + "delta_cosine": 0.7354081507082781, + "relative_weight_rmse": 0.011982262330114096 + }, + { + "index": 114, + "total": 252, + "projection": "layers.23.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3863, + "parameters": 50331648, + "reference_delta_l2": 5.681356830205064, + "retained_delta_l2": 5.729787391982792, + "delta_retention_norm_ratio": 1.0085244710418901, + "delta_cosine": 0.8547443822259567, + "relative_weight_rmse": 0.016875053895657025 + }, + { + "index": 115, + "total": 252, + "projection": "layers.23.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3863, + "parameters": 50331648, + "reference_delta_l2": 4.994572694078467, + "retained_delta_l2": 4.958950032711758, + "delta_retention_norm_ratio": 0.9928677259199885, + "delta_cosine": 0.8294620342905394, + "relative_weight_rmse": 0.01543130444925714 + }, + { + "index": 116, + "total": 252, + "projection": "layers.23.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0283, + "parameters": 4194304, + "reference_delta_l2": 0.7939351098558397, + "retained_delta_l2": 0.706441633084495, + "delta_retention_norm_ratio": 0.8897976979664857, + "delta_cosine": 0.7057072256232827, + "relative_weight_rmse": 0.011963073195813807 + }, + { + "index": 117, + "total": 252, + "projection": "layers.23.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1283, + "parameters": 16777216, + "reference_delta_l2": 1.960288229105135, + "retained_delta_l2": 1.8332819433258007, + "delta_retention_norm_ratio": 0.9352104022798157, + "delta_cosine": 0.7547145276610445, + "relative_weight_rmse": 0.012805489955986052 + }, + { + "index": 118, + "total": 252, + "projection": "layers.23.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1292, + "parameters": 16777216, + "reference_delta_l2": 1.7881764262115596, + "retained_delta_l2": 1.6206762884554973, + "delta_retention_norm_ratio": 0.9063290762025484, + "delta_cosine": 0.7241064277641575, + "relative_weight_rmse": 0.012036833781657983 + }, + { + "index": 119, + "total": 252, + "projection": "layers.23.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0299, + "parameters": 4194304, + "reference_delta_l2": 0.9296093087596113, + "retained_delta_l2": 0.854002764164332, + "delta_retention_norm_ratio": 0.9186684729995206, + "delta_cosine": 0.7381432465761304, + "relative_weight_rmse": 0.011748920526745516 + }, + { + "index": 120, + "total": 252, + "projection": "layers.24.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.4267, + "parameters": 50331648, + "reference_delta_l2": 3.2534230405335545, + "retained_delta_l2": 2.981800644478014, + "delta_retention_norm_ratio": 0.9165118115069981, + "delta_cosine": 0.7317666128734056, + "relative_weight_rmse": 0.012280282010517366 + }, + { + "index": 121, + "total": 252, + "projection": "layers.24.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3848, + "parameters": 50331648, + "reference_delta_l2": 5.165655340422529, + "retained_delta_l2": 5.138772758574253, + "delta_retention_norm_ratio": 0.9947959009889968, + "delta_cosine": 0.8288703561669849, + "relative_weight_rmse": 0.01631047714886263 + }, + { + "index": 122, + "total": 252, + "projection": "layers.24.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3869, + "parameters": 50331648, + "reference_delta_l2": 4.0141745870852334, + "retained_delta_l2": 3.8368578938450653, + "delta_retention_norm_ratio": 0.9558273589268769, + "delta_cosine": 0.7759200295888791, + "relative_weight_rmse": 0.013957609399077614 + }, + { + "index": 123, + "total": 252, + "projection": "layers.24.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0282, + "parameters": 4194304, + "reference_delta_l2": 0.9669618716241911, + "retained_delta_l2": 0.9176098077360659, + "delta_retention_norm_ratio": 0.9489617270997156, + "delta_cosine": 0.7726282893544615, + "relative_weight_rmse": 0.013156688473918708 + }, + { + "index": 124, + "total": 252, + "projection": "layers.24.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1297, + "parameters": 16777216, + "reference_delta_l2": 1.9193056791500216, + "retained_delta_l2": 1.7804441285586727, + "delta_retention_norm_ratio": 0.927650112173458, + "delta_cosine": 0.7463482254981281, + "relative_weight_rmse": 0.01258371936003536 + }, + { + "index": 125, + "total": 252, + "projection": "layers.24.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1309, + "parameters": 16777216, + "reference_delta_l2": 3.039059120168847, + "retained_delta_l2": 3.0371101854864153, + "delta_retention_norm_ratio": 0.9993587045840941, + "delta_cosine": 0.8496008974508179, + "relative_weight_rmse": 0.015607056499383726 + }, + { + "index": 126, + "total": 252, + "projection": "layers.24.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0301, + "parameters": 4194304, + "reference_delta_l2": 1.0872845874154113, + "retained_delta_l2": 1.0359463798430317, + "delta_retention_norm_ratio": 0.9527831000580852, + "delta_cosine": 0.778837353510625, + "relative_weight_rmse": 0.01304958405394525 + }, + { + "index": 127, + "total": 252, + "projection": "layers.25.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.4309, + "parameters": 50331648, + "reference_delta_l2": 3.642132638444929, + "retained_delta_l2": 3.409572774394396, + "delta_retention_norm_ratio": 0.9361473380744781, + "delta_cosine": 0.7517974948501079, + "relative_weight_rmse": 0.01327340848604111 + }, + { + "index": 128, + "total": 252, + "projection": "layers.25.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.391, + "parameters": 50331648, + "reference_delta_l2": 5.620100516508087, + "retained_delta_l2": 5.636423640519522, + "delta_retention_norm_ratio": 1.0029044185176916, + "delta_cosine": 0.841261441562537, + "relative_weight_rmse": 0.01698334918493056 + }, + { + "index": 129, + "total": 252, + "projection": "layers.25.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3893, + "parameters": 50331648, + "reference_delta_l2": 3.751771880231379, + "retained_delta_l2": 3.502705847373873, + "delta_retention_norm_ratio": 0.9336137588295625, + "delta_cosine": 0.7486844572560383, + "relative_weight_rmse": 0.013635755223338066 + }, + { + "index": 130, + "total": 252, + "projection": "layers.25.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0282, + "parameters": 4194304, + "reference_delta_l2": 1.028613932268488, + "retained_delta_l2": 0.9820507213318386, + "delta_retention_norm_ratio": 0.9547320821972928, + "delta_cosine": 0.7798742584487988, + "relative_weight_rmse": 0.013545750227291677 + }, + { + "index": 131, + "total": 252, + "projection": "layers.25.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1318, + "parameters": 16777216, + "reference_delta_l2": 1.9564144509544819, + "retained_delta_l2": 1.8095128691918798, + "delta_retention_norm_ratio": 0.9249128518290526, + "delta_cosine": 0.7403126992938154, + "relative_weight_rmse": 0.013199817134663757 + }, + { + "index": 132, + "total": 252, + "projection": "layers.25.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1321, + "parameters": 16777216, + "reference_delta_l2": 2.4514761557007434, + "retained_delta_l2": 2.40423426143534, + "delta_retention_norm_ratio": 0.9807292050728106, + "delta_cosine": 0.8152564312400812, + "relative_weight_rmse": 0.014601455683392647 + }, + { + "index": 133, + "total": 252, + "projection": "layers.25.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.029, + "parameters": 4194304, + "reference_delta_l2": 1.1897973765773928, + "retained_delta_l2": 1.1464713861908846, + "delta_retention_norm_ratio": 0.9635854043390639, + "delta_cosine": 0.7875521812962474, + "relative_weight_rmse": 0.013467930845111446 + }, + { + "index": 134, + "total": 252, + "projection": "layers.26.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.4284, + "parameters": 50331648, + "reference_delta_l2": 3.305241479049125, + "retained_delta_l2": 3.0279613748243444, + "delta_retention_norm_ratio": 0.9161089723754312, + "delta_cosine": 0.7290675691972286, + "relative_weight_rmse": 0.012359369142667078 + }, + { + "index": 135, + "total": 252, + "projection": "layers.26.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3909, + "parameters": 50331648, + "reference_delta_l2": 5.269251371509404, + "retained_delta_l2": 5.232637302627289, + "delta_retention_norm_ratio": 0.9930513717603062, + "delta_cosine": 0.8225288220968671, + "relative_weight_rmse": 0.016669117081924718 + }, + { + "index": 136, + "total": 252, + "projection": "layers.26.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.397, + "parameters": 50331648, + "reference_delta_l2": 3.6084143824322066, + "retained_delta_l2": 3.337343838080122, + "delta_retention_norm_ratio": 0.9248782108640824, + "delta_cosine": 0.7386694926000766, + "relative_weight_rmse": 0.013191997575593736 + }, + { + "index": 137, + "total": 252, + "projection": "layers.26.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0324, + "parameters": 4194304, + "reference_delta_l2": 1.1033783937516002, + "retained_delta_l2": 1.0650332196521097, + "delta_retention_norm_ratio": 0.9652474850725389, + "delta_cosine": 0.7914633519220408, + "relative_weight_rmse": 0.014450524331764551 + }, + { + "index": 138, + "total": 252, + "projection": "layers.26.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1385, + "parameters": 16777216, + "reference_delta_l2": 2.147096551460236, + "retained_delta_l2": 2.0228782430093153, + "delta_retention_norm_ratio": 0.9421459140407821, + "delta_cosine": 0.7605701256673978, + "relative_weight_rmse": 0.013817141329754716 + }, + { + "index": 139, + "total": 252, + "projection": "layers.26.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1317, + "parameters": 16777216, + "reference_delta_l2": 2.717881073870093, + "retained_delta_l2": 2.6939693177412587, + "delta_retention_norm_ratio": 0.9912020594430258, + "delta_cosine": 0.8321627533618473, + "relative_weight_rmse": 0.015381478917734464 + }, + { + "index": 140, + "total": 252, + "projection": "layers.26.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0297, + "parameters": 4194304, + "reference_delta_l2": 1.3035700311997667, + "retained_delta_l2": 1.2685645617688444, + "delta_retention_norm_ratio": 0.9731464604178539, + "delta_cosine": 0.8073929079759188, + "relative_weight_rmse": 0.014270040450847576 + }, + { + "index": 141, + "total": 252, + "projection": "layers.27.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.4336, + "parameters": 50331648, + "reference_delta_l2": 3.132261250787498, + "retained_delta_l2": 2.807137353424932, + "delta_retention_norm_ratio": 0.8962015389741755, + "delta_cosine": 0.7086808991494553, + "relative_weight_rmse": 0.011933529319931057 + }, + { + "index": 142, + "total": 252, + "projection": "layers.27.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.395, + "parameters": 50331648, + "reference_delta_l2": 5.247325963699111, + "retained_delta_l2": 5.216431893889305, + "delta_retention_norm_ratio": 0.9941124164910795, + "delta_cosine": 0.8255585582563386, + "relative_weight_rmse": 0.01648764474977256 + }, + { + "index": 143, + "total": 252, + "projection": "layers.27.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3869, + "parameters": 50331648, + "reference_delta_l2": 3.1510008911701286, + "retained_delta_l2": 2.768647223024788, + "delta_retention_norm_ratio": 0.8786564392232358, + "delta_cosine": 0.6904335192986641, + "relative_weight_rmse": 0.01220342793629529 + }, + { + "index": 144, + "total": 252, + "projection": "layers.27.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0282, + "parameters": 4194304, + "reference_delta_l2": 0.960118298135331, + "retained_delta_l2": 0.9043235411909332, + "delta_retention_norm_ratio": 0.9418876225432239, + "delta_cosine": 0.7669786023398018, + "relative_weight_rmse": 0.013522336301050803 + }, + { + "index": 145, + "total": 252, + "projection": "layers.27.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.131, + "parameters": 16777216, + "reference_delta_l2": 1.8368636487582302, + "retained_delta_l2": 1.6722199073501667, + "delta_retention_norm_ratio": 0.9103669227057941, + "delta_cosine": 0.7242585869216763, + "relative_weight_rmse": 0.012756450906067583 + }, + { + "index": 146, + "total": 252, + "projection": "layers.27.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.133, + "parameters": 16777216, + "reference_delta_l2": 2.505280677634362, + "retained_delta_l2": 2.4508958805780723, + "delta_retention_norm_ratio": 0.9782919344958811, + "delta_cosine": 0.8133856650561131, + "relative_weight_rmse": 0.014896287903625344 + }, + { + "index": 147, + "total": 252, + "projection": "layers.27.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0293, + "parameters": 4194304, + "reference_delta_l2": 1.3026741570714229, + "retained_delta_l2": 1.2790188773078266, + "delta_retention_norm_ratio": 0.9818409848424596, + "delta_cosine": 0.8137638637475447, + "relative_weight_rmse": 0.014298877186667228 + }, + { + "index": 148, + "total": 252, + "projection": "layers.28.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.4297, + "parameters": 50331648, + "reference_delta_l2": 3.770779313567698, + "retained_delta_l2": 3.5568828920765014, + "delta_retention_norm_ratio": 0.9432752744978808, + "delta_cosine": 0.7614841636602006, + "relative_weight_rmse": 0.013120830359412057 + }, + { + "index": 149, + "total": 252, + "projection": "layers.28.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3902, + "parameters": 50331648, + "reference_delta_l2": 6.474007977311115, + "retained_delta_l2": 6.592055973272809, + "delta_retention_norm_ratio": 1.0182341443469651, + "delta_cosine": 0.8681988272925049, + "relative_weight_rmse": 0.01800075192492093 + }, + { + "index": 150, + "total": 252, + "projection": "layers.28.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3879, + "parameters": 50331648, + "reference_delta_l2": 3.251831424096289, + "retained_delta_l2": 2.8852405574526903, + "delta_retention_norm_ratio": 0.8872663373854082, + "delta_cosine": 0.6985082000056504, + "relative_weight_rmse": 0.012346026593839841 + }, + { + "index": 151, + "total": 252, + "projection": "layers.28.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0281, + "parameters": 4194304, + "reference_delta_l2": 1.0323824251540816, + "retained_delta_l2": 0.9825007436597037, + "delta_retention_norm_ratio": 0.951682942019346, + "delta_cosine": 0.771248140491772, + "relative_weight_rmse": 0.014134231309731925 + }, + { + "index": 152, + "total": 252, + "projection": "layers.28.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.131, + "parameters": 16777216, + "reference_delta_l2": 1.8689306710458202, + "retained_delta_l2": 1.71735194200812, + "delta_retention_norm_ratio": 0.9188954778333862, + "delta_cosine": 0.7339179675507641, + "relative_weight_rmse": 0.012841131456311331 + }, + { + "index": 153, + "total": 252, + "projection": "layers.28.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1328, + "parameters": 16777216, + "reference_delta_l2": 2.112653591950212, + "retained_delta_l2": 2.0137616228372126, + "delta_retention_norm_ratio": 0.9531906368891688, + "delta_cosine": 0.777666777366398, + "relative_weight_rmse": 0.013269128681704517 + }, + { + "index": 154, + "total": 252, + "projection": "layers.28.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0298, + "parameters": 4194304, + "reference_delta_l2": 1.0579417102970974, + "retained_delta_l2": 0.9885788527580183, + "delta_retention_norm_ratio": 0.9344360309609117, + "delta_cosine": 0.7520058519219315, + "relative_weight_rmse": 0.012592426270211971 + }, + { + "index": 155, + "total": 252, + "projection": "layers.29.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.4302, + "parameters": 50331648, + "reference_delta_l2": 3.7534517264286573, + "retained_delta_l2": 3.5245574147557397, + "delta_retention_norm_ratio": 0.9390176487255089, + "delta_cosine": 0.756712105571667, + "relative_weight_rmse": 0.013025868946782511 + }, + { + "index": 156, + "total": 252, + "projection": "layers.29.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3902, + "parameters": 50331648, + "reference_delta_l2": 6.17579873844958, + "retained_delta_l2": 6.26468819573951, + "delta_retention_norm_ratio": 1.0143931920476161, + "delta_cosine": 0.8582469283492269, + "relative_weight_rmse": 0.01777181309605028 + }, + { + "index": 157, + "total": 252, + "projection": "layers.29.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3888, + "parameters": 50331648, + "reference_delta_l2": 3.59159804521513, + "retained_delta_l2": 3.293281299130373, + "delta_retention_norm_ratio": 0.9169403863324331, + "delta_cosine": 0.7301450210404935, + "relative_weight_rmse": 0.01292364590979042 + }, + { + "index": 158, + "total": 252, + "projection": "layers.29.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0282, + "parameters": 4194304, + "reference_delta_l2": 0.9631567917349719, + "retained_delta_l2": 0.9129046112541209, + "delta_retention_norm_ratio": 0.9478255452154059, + "delta_cosine": 0.7820175303265533, + "relative_weight_rmse": 0.013140085712954792 + }, + { + "index": 159, + "total": 252, + "projection": "layers.29.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1299, + "parameters": 16777216, + "reference_delta_l2": 1.9484483835817932, + "retained_delta_l2": 1.796999556748854, + "delta_retention_norm_ratio": 0.9222720868004037, + "delta_cosine": 0.7378988525244624, + "relative_weight_rmse": 0.01294247061943119 + }, + { + "index": 160, + "total": 252, + "projection": "layers.29.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1325, + "parameters": 16777216, + "reference_delta_l2": 2.9486054506181407, + "retained_delta_l2": 2.9360501658337737, + "delta_retention_norm_ratio": 0.995741958361457, + "delta_cosine": 0.8480940212651022, + "relative_weight_rmse": 0.015288258290702357 + }, + { + "index": 161, + "total": 252, + "projection": "layers.29.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0296, + "parameters": 4194304, + "reference_delta_l2": 1.0699779495571091, + "retained_delta_l2": 1.0105511699123044, + "delta_retention_norm_ratio": 0.9444598090368098, + "delta_cosine": 0.7676695688460679, + "relative_weight_rmse": 0.012957896364334942 + }, + { + "index": 162, + "total": 252, + "projection": "layers.3.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.4333, + "parameters": 50331648, + "reference_delta_l2": 2.1032100566399197, + "retained_delta_l2": 1.8715532853851151, + "delta_retention_norm_ratio": 0.8898556183090438, + "delta_cosine": 0.7218372867640742, + "relative_weight_rmse": 0.011441034776611706 + }, + { + "index": 163, + "total": 252, + "projection": "layers.3.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3902, + "parameters": 50331648, + "reference_delta_l2": 2.938720610342676, + "retained_delta_l2": 2.7192042076684735, + "delta_retention_norm_ratio": 0.925302050864711, + "delta_cosine": 0.7471542319968987, + "relative_weight_rmse": 0.012287635001598943 + }, + { + "index": 164, + "total": 252, + "projection": "layers.3.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3871, + "parameters": 50331648, + "reference_delta_l2": 3.196659265953979, + "retained_delta_l2": 3.1313538641926626, + "delta_retention_norm_ratio": 0.9795707342171712, + "delta_cosine": 0.8439606316876027, + "relative_weight_rmse": 0.014685120384091997 + }, + { + "index": 165, + "total": 252, + "projection": "layers.3.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.028, + "parameters": 4194304, + "reference_delta_l2": 0.5966929387519279, + "retained_delta_l2": 0.463749829103853, + "delta_retention_norm_ratio": 0.7772001292219994, + "delta_cosine": 0.5981156326647796, + "relative_weight_rmse": 0.009662974031804931 + }, + { + "index": 166, + "total": 252, + "projection": "layers.3.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1314, + "parameters": 16777216, + "reference_delta_l2": 1.855006646219326, + "retained_delta_l2": 1.701967156048477, + "delta_retention_norm_ratio": 0.9174992227209765, + "delta_cosine": 0.7310224175084274, + "relative_weight_rmse": 0.012835343971560898 + }, + { + "index": 167, + "total": 252, + "projection": "layers.3.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1308, + "parameters": 16777216, + "reference_delta_l2": 1.5126081346122637, + "retained_delta_l2": 1.310500384839273, + "delta_retention_norm_ratio": 0.8663845941666853, + "delta_cosine": 0.6796911912848108, + "relative_weight_rmse": 0.011748833378422311 + }, + { + "index": 168, + "total": 252, + "projection": "layers.3.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0299, + "parameters": 4194304, + "reference_delta_l2": 0.9493398299256175, + "retained_delta_l2": 0.8565163941492758, + "delta_retention_norm_ratio": 0.9022231735672415, + "delta_cosine": 0.7164521198474684, + "relative_weight_rmse": 0.012405948503876288 + }, + { + "index": 169, + "total": 252, + "projection": "layers.30.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.4262, + "parameters": 50331648, + "reference_delta_l2": 3.8978984993184693, + "retained_delta_l2": 3.6994383763582808, + "delta_retention_norm_ratio": 0.949085353814398, + "delta_cosine": 0.7673530694212525, + "relative_weight_rmse": 0.013155715168103967 + }, + { + "index": 170, + "total": 252, + "projection": "layers.30.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3909, + "parameters": 50331648, + "reference_delta_l2": 5.433652967437842, + "retained_delta_l2": 5.434993583742712, + "delta_retention_norm_ratio": 1.0002467246828062, + "delta_cosine": 0.8319933390667529, + "relative_weight_rmse": 0.016980591988578438 + }, + { + "index": 171, + "total": 252, + "projection": "layers.30.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3907, + "parameters": 50331648, + "reference_delta_l2": 3.8348783996802394, + "retained_delta_l2": 3.5590722971279973, + "delta_retention_norm_ratio": 0.9280795702478494, + "delta_cosine": 0.7415143180677388, + "relative_weight_rmse": 0.013456396568094107 + }, + { + "index": 172, + "total": 252, + "projection": "layers.30.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0282, + "parameters": 4194304, + "reference_delta_l2": 1.4984205802451858, + "retained_delta_l2": 1.5108422212234183, + "delta_retention_norm_ratio": 1.0082898227253392, + "delta_cosine": 0.8531969004133971, + "relative_weight_rmse": 0.016620995319562116 + }, + { + "index": 173, + "total": 252, + "projection": "layers.30.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1316, + "parameters": 16777216, + "reference_delta_l2": 1.8728191309298317, + "retained_delta_l2": 1.6973609956354359, + "delta_retention_norm_ratio": 0.9063133580831786, + "delta_cosine": 0.7181223959925678, + "relative_weight_rmse": 0.012881768328302717 + }, + { + "index": 174, + "total": 252, + "projection": "layers.30.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.133, + "parameters": 16777216, + "reference_delta_l2": 3.0332081303291325, + "retained_delta_l2": 3.0455803742269714, + "delta_retention_norm_ratio": 1.0040789300853208, + "delta_cosine": 0.8559045647858767, + "relative_weight_rmse": 0.01632808793865334 + }, + { + "index": 175, + "total": 252, + "projection": "layers.30.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0299, + "parameters": 4194304, + "reference_delta_l2": 1.3123151133807753, + "retained_delta_l2": 1.274843676062311, + "delta_retention_norm_ratio": 0.9714463112278493, + "delta_cosine": 0.801919063498299, + "relative_weight_rmse": 0.013937328553195962 + }, + { + "index": 176, + "total": 252, + "projection": "layers.31.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.4328, + "parameters": 50331648, + "reference_delta_l2": 4.117060720903275, + "retained_delta_l2": 3.952201024686551, + "delta_retention_norm_ratio": 0.9599569432192989, + "delta_cosine": 0.7808196240294195, + "relative_weight_rmse": 0.01339236697860167 + }, + { + "index": 177, + "total": 252, + "projection": "layers.31.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3893, + "parameters": 50331648, + "reference_delta_l2": 4.23682193488538, + "retained_delta_l2": 4.0921481609531325, + "delta_retention_norm_ratio": 0.9658532324096455, + "delta_cosine": 0.7870133171237784, + "relative_weight_rmse": 0.014799547618956832 + }, + { + "index": 178, + "total": 252, + "projection": "layers.31.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3891, + "parameters": 50331648, + "reference_delta_l2": 3.7533803077341537, + "retained_delta_l2": 3.466890820615773, + "delta_retention_norm_ratio": 0.923671607023662, + "delta_cosine": 0.7371873467107354, + "relative_weight_rmse": 0.013143132916215957 + }, + { + "index": 179, + "total": 252, + "projection": "layers.31.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0282, + "parameters": 4194304, + "reference_delta_l2": 1.1300260386779475, + "retained_delta_l2": 1.1061474945023615, + "delta_retention_norm_ratio": 0.9788690318998999, + "delta_cosine": 0.812792858989726, + "relative_weight_rmse": 0.014423882978006959 + }, + { + "index": 180, + "total": 252, + "projection": "layers.31.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1321, + "parameters": 16777216, + "reference_delta_l2": 1.9332114332159547, + "retained_delta_l2": 1.785208152819234, + "delta_retention_norm_ratio": 0.9234417519709612, + "delta_cosine": 0.7394358609471232, + "relative_weight_rmse": 0.012567795219459207 + }, + { + "index": 181, + "total": 252, + "projection": "layers.31.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1332, + "parameters": 16777216, + "reference_delta_l2": 3.097948358974668, + "retained_delta_l2": 3.1217002046644615, + "delta_retention_norm_ratio": 1.0076669598513432, + "delta_cosine": 0.8594629372319006, + "relative_weight_rmse": 0.016367174602252094 + }, + { + "index": 182, + "total": 252, + "projection": "layers.31.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0301, + "parameters": 4194304, + "reference_delta_l2": 0.9159050398617947, + "retained_delta_l2": 0.808662779418089, + "delta_retention_norm_ratio": 0.8829111580607876, + "delta_cosine": 0.7002209721035524, + "relative_weight_rmse": 0.011267909505400815 + }, + { + "index": 183, + "total": 252, + "projection": "layers.32.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.4312, + "parameters": 50331648, + "reference_delta_l2": 3.057429050751837, + "retained_delta_l2": 2.7055068754815284, + "delta_retention_norm_ratio": 0.8848960451972648, + "delta_cosine": 0.6985648933470412, + "relative_weight_rmse": 0.011300930284001908 + }, + { + "index": 184, + "total": 252, + "projection": "layers.32.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3854, + "parameters": 50331648, + "reference_delta_l2": 5.781916303688982, + "retained_delta_l2": 5.838110125536839, + "delta_retention_norm_ratio": 1.009718892300811, + "delta_cosine": 0.8498731301381173, + "relative_weight_rmse": 0.017542559791707914 + }, + { + "index": 185, + "total": 252, + "projection": "layers.32.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3868, + "parameters": 50331648, + "reference_delta_l2": 4.1927812903409505, + "retained_delta_l2": 3.9781009068246798, + "delta_retention_norm_ratio": 0.9487976193724111, + "delta_cosine": 0.7679477233425701, + "relative_weight_rmse": 0.013865189763080822 + }, + { + "index": 186, + "total": 252, + "projection": "layers.32.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0282, + "parameters": 4194304, + "reference_delta_l2": 1.0783175272646806, + "retained_delta_l2": 1.044111151228919, + "delta_retention_norm_ratio": 0.9682780116516038, + "delta_cosine": 0.7992171502076999, + "relative_weight_rmse": 0.014106002121241187 + }, + { + "index": 187, + "total": 252, + "projection": "layers.32.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1287, + "parameters": 16777216, + "reference_delta_l2": 1.955353845683428, + "retained_delta_l2": 1.7974128413490449, + "delta_retention_norm_ratio": 0.9192263821287139, + "delta_cosine": 0.7328830838255879, + "relative_weight_rmse": 0.01292098369173976 + }, + { + "index": 188, + "total": 252, + "projection": "layers.32.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1322, + "parameters": 16777216, + "reference_delta_l2": 3.521193846583141, + "retained_delta_l2": 3.5817414138144987, + "delta_retention_norm_ratio": 1.0171951814837208, + "delta_cosine": 0.8844759311883627, + "relative_weight_rmse": 0.017048171080786798 + }, + { + "index": 189, + "total": 252, + "projection": "layers.32.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0291, + "parameters": 4194304, + "reference_delta_l2": 1.1078237330254415, + "retained_delta_l2": 1.0383064489336704, + "delta_retention_norm_ratio": 0.9372487860483717, + "delta_cosine": 0.7589941781019551, + "relative_weight_rmse": 0.012614362339079541 + }, + { + "index": 190, + "total": 252, + "projection": "layers.33.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.4241, + "parameters": 50331648, + "reference_delta_l2": 4.689707248703135, + "retained_delta_l2": 4.604233147135592, + "delta_retention_norm_ratio": 0.9817741072022823, + "delta_cosine": 0.8102813971155682, + "relative_weight_rmse": 0.014420925141106806 + }, + { + "index": 191, + "total": 252, + "projection": "layers.33.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3857, + "parameters": 50331648, + "reference_delta_l2": 4.034563257265208, + "retained_delta_l2": 3.8543050664855696, + "delta_retention_norm_ratio": 0.9553215108339075, + "delta_cosine": 0.7727336337657214, + "relative_weight_rmse": 0.014793621153844716 + }, + { + "index": 192, + "total": 252, + "projection": "layers.33.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3864, + "parameters": 50331648, + "reference_delta_l2": 4.566171345371535, + "retained_delta_l2": 4.43130048233319, + "delta_retention_norm_ratio": 0.9704630306580467, + "delta_cosine": 0.796240684876951, + "relative_weight_rmse": 0.014252010655975832 + }, + { + "index": 193, + "total": 252, + "projection": "layers.33.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0282, + "parameters": 4194304, + "reference_delta_l2": 1.171552776418305, + "retained_delta_l2": 1.1623227521475235, + "delta_retention_norm_ratio": 0.9921215463301621, + "delta_cosine": 0.8277549320058807, + "relative_weight_rmse": 0.01554920977100678 + }, + { + "index": 194, + "total": 252, + "projection": "layers.33.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1287, + "parameters": 16777216, + "reference_delta_l2": 1.959212980651289, + "retained_delta_l2": 1.8112129064554086, + "delta_retention_norm_ratio": 0.924459425464463, + "delta_cosine": 0.7415341212291183, + "relative_weight_rmse": 0.012309177676342303 + }, + { + "index": 195, + "total": 252, + "projection": "layers.33.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1305, + "parameters": 16777216, + "reference_delta_l2": 1.6985174788058186, + "retained_delta_l2": 1.5665997536990643, + "delta_retention_norm_ratio": 0.9223336075413825, + "delta_cosine": 0.7431240448205082, + "relative_weight_rmse": 0.011938592566423342 + }, + { + "index": 196, + "total": 252, + "projection": "layers.33.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0287, + "parameters": 4194304, + "reference_delta_l2": 1.2343662918195417, + "retained_delta_l2": 1.184980602290347, + "delta_retention_norm_ratio": 0.9599910578760243, + "delta_cosine": 0.7864342520216658, + "relative_weight_rmse": 0.012728537950190457 + }, + { + "index": 197, + "total": 252, + "projection": "layers.34.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.4285, + "parameters": 50331648, + "reference_delta_l2": 3.261489432925536, + "retained_delta_l2": 3.0313032324464326, + "delta_retention_norm_ratio": 0.9294229813669432, + "delta_cosine": 0.7458831088239839, + "relative_weight_rmse": 0.011854496813221605 + }, + { + "index": 198, + "total": 252, + "projection": "layers.34.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3903, + "parameters": 50331648, + "reference_delta_l2": 3.5196454509206627, + "retained_delta_l2": 3.316919358673675, + "delta_retention_norm_ratio": 0.9424015586018873, + "delta_cosine": 0.7574977644879192, + "relative_weight_rmse": 0.0131233892592756 + }, + { + "index": 199, + "total": 252, + "projection": "layers.34.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3897, + "parameters": 50331648, + "reference_delta_l2": 3.1239829376209345, + "retained_delta_l2": 2.8006109007510576, + "delta_retention_norm_ratio": 0.8964872589489427, + "delta_cosine": 0.7100234638649024, + "relative_weight_rmse": 0.011423804865420405 + }, + { + "index": 200, + "total": 252, + "projection": "layers.34.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0281, + "parameters": 4194304, + "reference_delta_l2": 0.7901579176804271, + "retained_delta_l2": 0.7289411020474216, + "delta_retention_norm_ratio": 0.9225258467159166, + "delta_cosine": 0.7529023074456173, + "relative_weight_rmse": 0.011644851883442688 + }, + { + "index": 201, + "total": 252, + "projection": "layers.34.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1293, + "parameters": 16777216, + "reference_delta_l2": 2.3687711363412887, + "retained_delta_l2": 2.298505710434156, + "delta_retention_norm_ratio": 0.9703367603441582, + "delta_cosine": 0.79321241066586, + "relative_weight_rmse": 0.014052412268970403 + }, + { + "index": 202, + "total": 252, + "projection": "layers.34.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1301, + "parameters": 16777216, + "reference_delta_l2": 2.798669977547175, + "retained_delta_l2": 2.8128553327431063, + "delta_retention_norm_ratio": 1.0050686059127143, + "delta_cosine": 0.8601519965070576, + "relative_weight_rmse": 0.015301485033859027 + }, + { + "index": 203, + "total": 252, + "projection": "layers.34.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0296, + "parameters": 4194304, + "reference_delta_l2": 0.9811930089310823, + "retained_delta_l2": 0.8765817347135941, + "delta_retention_norm_ratio": 0.8933835919484869, + "delta_cosine": 0.7153690562087868, + "relative_weight_rmse": 0.01140318267189611 + }, + { + "index": 204, + "total": 252, + "projection": "layers.35.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.4241, + "parameters": 50331648, + "reference_delta_l2": 2.333473272753201, + "retained_delta_l2": 2.086970023664544, + "delta_retention_norm_ratio": 0.8943620859227521, + "delta_cosine": 0.7148638840512689, + "relative_weight_rmse": 0.0098139939294425 + }, + { + "index": 205, + "total": 252, + "projection": "layers.35.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3871, + "parameters": 50331648, + "reference_delta_l2": 2.627449696619369, + "retained_delta_l2": 2.294180125270862, + "delta_retention_norm_ratio": 0.8731585340045479, + "delta_cosine": 0.6878599929217832, + "relative_weight_rmse": 0.010532547008695962 + }, + { + "index": 206, + "total": 252, + "projection": "layers.35.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3845, + "parameters": 50331648, + "reference_delta_l2": 2.664437084904742, + "retained_delta_l2": 2.2924939981946424, + "delta_retention_norm_ratio": 0.8604046277477041, + "delta_cosine": 0.6749879160129268, + "relative_weight_rmse": 0.010351714937813263 + }, + { + "index": 207, + "total": 252, + "projection": "layers.35.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0281, + "parameters": 4194304, + "reference_delta_l2": 0.8330723387143064, + "retained_delta_l2": 0.7630192621151713, + "delta_retention_norm_ratio": 0.9159099716271351, + "delta_cosine": 0.7342410909336075, + "relative_weight_rmse": 0.012232671019988367 + }, + { + "index": 208, + "total": 252, + "projection": "layers.35.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1306, + "parameters": 16777216, + "reference_delta_l2": 1.9768898560640722, + "retained_delta_l2": 1.875411782222538, + "delta_retention_norm_ratio": 0.9486678159988267, + "delta_cosine": 0.7674952628310036, + "relative_weight_rmse": 0.01297704657162937 + }, + { + "index": 209, + "total": 252, + "projection": "layers.35.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1313, + "parameters": 16777216, + "reference_delta_l2": 1.9270585177821875, + "retained_delta_l2": 1.8399410652305475, + "delta_retention_norm_ratio": 0.954792523554551, + "delta_cosine": 0.7751783116171622, + "relative_weight_rmse": 0.013105533957306424 + }, + { + "index": 210, + "total": 252, + "projection": "layers.35.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0294, + "parameters": 4194304, + "reference_delta_l2": 0.900220271344646, + "retained_delta_l2": 0.7938819338328972, + "delta_retention_norm_ratio": 0.8818752022181051, + "delta_cosine": 0.6984764421348353, + "relative_weight_rmse": 0.011381243323434816 + }, + { + "index": 211, + "total": 252, + "projection": "layers.4.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.4266, + "parameters": 50331648, + "reference_delta_l2": 2.2298700594055836, + "retained_delta_l2": 1.9139909072946737, + "delta_retention_norm_ratio": 0.8583419016823278, + "delta_cosine": 0.6735711698792769, + "relative_weight_rmse": 0.010853858960267986 + }, + { + "index": 212, + "total": 252, + "projection": "layers.4.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3858, + "parameters": 50331648, + "reference_delta_l2": 3.3294753338867764, + "retained_delta_l2": 2.939987162403425, + "delta_retention_norm_ratio": 0.8830181537856088, + "delta_cosine": 0.7106284120654582, + "relative_weight_rmse": 0.010855684224578223 + }, + { + "index": 213, + "total": 252, + "projection": "layers.4.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.388, + "parameters": 50331648, + "reference_delta_l2": 3.312872016246803, + "retained_delta_l2": 3.1913194087885497, + "delta_retention_norm_ratio": 0.9633089938693249, + "delta_cosine": 0.796326371531978, + "relative_weight_rmse": 0.01403572649186527 + }, + { + "index": 214, + "total": 252, + "projection": "layers.4.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0273, + "parameters": 4194304, + "reference_delta_l2": 0.5946416784174827, + "retained_delta_l2": 0.4495808225440149, + "delta_retention_norm_ratio": 0.7560533323874679, + "delta_cosine": 0.5788912869023318, + "relative_weight_rmse": 0.009481844112531713 + }, + { + "index": 215, + "total": 252, + "projection": "layers.4.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1296, + "parameters": 16777216, + "reference_delta_l2": 1.6306689002006027, + "retained_delta_l2": 1.4171534861141628, + "delta_retention_norm_ratio": 0.8690626809279471, + "delta_cosine": 0.6812818573272754, + "relative_weight_rmse": 0.011815779929245714 + }, + { + "index": 216, + "total": 252, + "projection": "layers.4.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1314, + "parameters": 16777216, + "reference_delta_l2": 1.2641816961468235, + "retained_delta_l2": 0.999941832271729, + "delta_retention_norm_ratio": 0.7909795208390634, + "delta_cosine": 0.6087861609751001, + "relative_weight_rmse": 0.01025800939966456 + }, + { + "index": 217, + "total": 252, + "projection": "layers.4.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0297, + "parameters": 4194304, + "reference_delta_l2": 1.1351469229770412, + "retained_delta_l2": 1.077428948694092, + "delta_retention_norm_ratio": 0.9491537411460554, + "delta_cosine": 0.7691064312538114, + "relative_weight_rmse": 0.013532001808611497 + }, + { + "index": 218, + "total": 252, + "projection": "layers.5.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.4294, + "parameters": 50331648, + "reference_delta_l2": 1.8122803189862504, + "retained_delta_l2": 1.3513121930987764, + "delta_retention_norm_ratio": 0.7456419290889119, + "delta_cosine": 0.5706043859449802, + "relative_weight_rmse": 0.009076813205507014 + }, + { + "index": 219, + "total": 252, + "projection": "layers.5.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3871, + "parameters": 50331648, + "reference_delta_l2": 2.96157753494215, + "retained_delta_l2": 2.498763914915212, + "delta_retention_norm_ratio": 0.8437273329614252, + "delta_cosine": 0.6611374243301872, + "relative_weight_rmse": 0.010014595520091093 + }, + { + "index": 220, + "total": 252, + "projection": "layers.5.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3883, + "parameters": 50331648, + "reference_delta_l2": 3.189171842325098, + "retained_delta_l2": 2.976767244156863, + "delta_retention_norm_ratio": 0.9333981959362279, + "delta_cosine": 0.7502917705142815, + "relative_weight_rmse": 0.013157307556074232 + }, + { + "index": 221, + "total": 252, + "projection": "layers.5.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0275, + "parameters": 4194304, + "reference_delta_l2": 0.6215712114964702, + "retained_delta_l2": 0.48966192521097135, + "delta_retention_norm_ratio": 0.7877808948585001, + "delta_cosine": 0.6084530317620194, + "relative_weight_rmse": 0.009879456921352362 + }, + { + "index": 222, + "total": 252, + "projection": "layers.5.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1319, + "parameters": 16777216, + "reference_delta_l2": 1.4717689451304197, + "retained_delta_l2": 1.2374487199650834, + "delta_retention_norm_ratio": 0.8407900737811993, + "delta_cosine": 0.6527281679967601, + "relative_weight_rmse": 0.011372757364125982 + }, + { + "index": 223, + "total": 252, + "projection": "layers.5.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1316, + "parameters": 16777216, + "reference_delta_l2": 1.4738879837391277, + "retained_delta_l2": 1.259869599627624, + "delta_retention_norm_ratio": 0.8547933177604465, + "delta_cosine": 0.6712932410960587, + "relative_weight_rmse": 0.011217628045088336 + }, + { + "index": 224, + "total": 252, + "projection": "layers.5.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0299, + "parameters": 4194304, + "reference_delta_l2": 1.0490845583401458, + "retained_delta_l2": 0.9849090395496732, + "delta_retention_norm_ratio": 0.9388271247724678, + "delta_cosine": 0.7615308167888781, + "relative_weight_rmse": 0.012894924200935487 + }, + { + "index": 225, + "total": 252, + "projection": "layers.6.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.4276, + "parameters": 50331648, + "reference_delta_l2": 4.771066717444592, + "retained_delta_l2": 4.784661491987646, + "delta_retention_norm_ratio": 1.0028494203389249, + "delta_cosine": 0.8443594307082539, + "relative_weight_rmse": 0.015441635418925712 + }, + { + "index": 226, + "total": 252, + "projection": "layers.6.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.388, + "parameters": 50331648, + "reference_delta_l2": 3.1162700027819596, + "retained_delta_l2": 2.7267308062243303, + "delta_retention_norm_ratio": 0.8749982523305492, + "delta_cosine": 0.6927384968601482, + "relative_weight_rmse": 0.010708087927262619 + }, + { + "index": 227, + "total": 252, + "projection": "layers.6.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3897, + "parameters": 50331648, + "reference_delta_l2": 3.3907502317088767, + "retained_delta_l2": 3.2029933284858823, + "delta_retention_norm_ratio": 0.944626737331706, + "delta_cosine": 0.7633216672956021, + "relative_weight_rmse": 0.013231096334501414 + }, + { + "index": 228, + "total": 252, + "projection": "layers.6.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0284, + "parameters": 4194304, + "reference_delta_l2": 0.5943120478492087, + "retained_delta_l2": 0.4493094061782143, + "delta_retention_norm_ratio": 0.7560159815104656, + "delta_cosine": 0.5790233630960934, + "relative_weight_rmse": 0.009600276669995444 + }, + { + "index": 229, + "total": 252, + "projection": "layers.6.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1301, + "parameters": 16777216, + "reference_delta_l2": 1.3678280162024512, + "retained_delta_l2": 1.1097997193645424, + "delta_retention_norm_ratio": 0.8113591081762738, + "delta_cosine": 0.62711186565333, + "relative_weight_rmse": 0.01041206033866853 + }, + { + "index": 230, + "total": 252, + "projection": "layers.6.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1335, + "parameters": 16777216, + "reference_delta_l2": 1.4042828456838277, + "retained_delta_l2": 1.1652349058479508, + "delta_retention_norm_ratio": 0.829772228172829, + "delta_cosine": 0.6448680307395254, + "relative_weight_rmse": 0.010729757482986754 + }, + { + "index": 231, + "total": 252, + "projection": "layers.6.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0297, + "parameters": 4194304, + "reference_delta_l2": 0.9871926467769184, + "retained_delta_l2": 0.9114173856916332, + "delta_retention_norm_ratio": 0.923241667841952, + "delta_cosine": 0.7398151642235392, + "relative_weight_rmse": 0.01240622946152155 + }, + { + "index": 232, + "total": 252, + "projection": "layers.7.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.4281, + "parameters": 50331648, + "reference_delta_l2": 1.8521771475877542, + "retained_delta_l2": 1.3724603366167916, + "delta_retention_norm_ratio": 0.7409984182151593, + "delta_cosine": 0.5664671489968615, + "relative_weight_rmse": 0.008987516352000395 + }, + { + "index": 233, + "total": 252, + "projection": "layers.7.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3954, + "parameters": 50331648, + "reference_delta_l2": 2.905173671022292, + "retained_delta_l2": 2.4911625372470114, + "delta_retention_norm_ratio": 0.8574917782352076, + "delta_cosine": 0.6744849299735817, + "relative_weight_rmse": 0.010373072438816326 + }, + { + "index": 234, + "total": 252, + "projection": "layers.7.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3895, + "parameters": 50331648, + "reference_delta_l2": 2.9784221401381137, + "retained_delta_l2": 2.6957056205607213, + "delta_retention_norm_ratio": 0.9050784253288278, + "delta_cosine": 0.716828131113773, + "relative_weight_rmse": 0.012391011395935268 + }, + { + "index": 235, + "total": 252, + "projection": "layers.7.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.028, + "parameters": 4194304, + "reference_delta_l2": 0.723251167971211, + "retained_delta_l2": 0.6140592733678869, + "delta_retention_norm_ratio": 0.8490263141784923, + "delta_cosine": 0.6627401093051608, + "relative_weight_rmse": 0.010780021563793991 + }, + { + "index": 236, + "total": 252, + "projection": "layers.7.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1333, + "parameters": 16777216, + "reference_delta_l2": 1.3692825527464774, + "retained_delta_l2": 1.1101798495001234, + "delta_retention_norm_ratio": 0.8107748450262137, + "delta_cosine": 0.6258894885968138, + "relative_weight_rmse": 0.010569037607867555 + }, + { + "index": 237, + "total": 252, + "projection": "layers.7.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1314, + "parameters": 16777216, + "reference_delta_l2": 1.689175224407534, + "retained_delta_l2": 1.5277574225939996, + "delta_retention_norm_ratio": 0.904439871316399, + "delta_cosine": 0.7182779183237219, + "relative_weight_rmse": 0.012135819144340742 + }, + { + "index": 238, + "total": 252, + "projection": "layers.7.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0289, + "parameters": 4194304, + "reference_delta_l2": 0.9909627130040701, + "retained_delta_l2": 0.8947364555088995, + "delta_retention_norm_ratio": 0.9028961874827117, + "delta_cosine": 0.7180679728890625, + "relative_weight_rmse": 0.011996835799134424 + }, + { + "index": 239, + "total": 252, + "projection": "layers.8.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.4279, + "parameters": 50331648, + "reference_delta_l2": 2.6493503506226146, + "retained_delta_l2": 2.28859717462612, + "delta_retention_norm_ratio": 0.8638333446870416, + "delta_cosine": 0.67802243763419, + "relative_weight_rmse": 0.011082221751495574 + }, + { + "index": 240, + "total": 252, + "projection": "layers.8.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3876, + "parameters": 50331648, + "reference_delta_l2": 2.2123296606827147, + "retained_delta_l2": 1.6700506358803586, + "delta_retention_norm_ratio": 0.7548832642622477, + "delta_cosine": 0.5781258031286636, + "relative_weight_rmse": 0.00950049550406276 + }, + { + "index": 241, + "total": 252, + "projection": "layers.8.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.388, + "parameters": 50331648, + "reference_delta_l2": 2.6842542293325784, + "retained_delta_l2": 2.289448996749444, + "delta_retention_norm_ratio": 0.8529180923815476, + "delta_cosine": 0.6661503520663868, + "relative_weight_rmse": 0.011442656000368902 + }, + { + "index": 242, + "total": 252, + "projection": "layers.8.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0286, + "parameters": 4194304, + "reference_delta_l2": 0.8958509707361921, + "retained_delta_l2": 0.8129741795036367, + "delta_retention_norm_ratio": 0.9074881939744409, + "delta_cosine": 0.7206105831298275, + "relative_weight_rmse": 0.0121490709753579 + }, + { + "index": 243, + "total": 252, + "projection": "layers.8.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1285, + "parameters": 16777216, + "reference_delta_l2": 1.5746017238305277, + "retained_delta_l2": 1.3492451483752443, + "delta_retention_norm_ratio": 0.8568802688040636, + "delta_cosine": 0.670854960229517, + "relative_weight_rmse": 0.011317503801783183 + }, + { + "index": 244, + "total": 252, + "projection": "layers.8.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1313, + "parameters": 16777216, + "reference_delta_l2": 1.597507777771925, + "retained_delta_l2": 1.387246064626345, + "delta_retention_norm_ratio": 0.8683814150571233, + "delta_cosine": 0.679701985246401, + "relative_weight_rmse": 0.01185399009814743 + }, + { + "index": 245, + "total": 252, + "projection": "layers.8.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0297, + "parameters": 4194304, + "reference_delta_l2": 0.7959943619191734, + "retained_delta_l2": 0.659812128400203, + "delta_retention_norm_ratio": 0.8289155802678932, + "delta_cosine": 0.6431923544299393, + "relative_weight_rmse": 0.010624284532260794 + }, + { + "index": 246, + "total": 252, + "projection": "layers.9.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.4326, + "parameters": 50331648, + "reference_delta_l2": 2.4732992821388473, + "retained_delta_l2": 2.102661109084502, + "delta_retention_norm_ratio": 0.8501442281041595, + "delta_cosine": 0.66449962717924, + "relative_weight_rmse": 0.010897908991001034 + }, + { + "index": 247, + "total": 252, + "projection": "layers.9.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3909, + "parameters": 50331648, + "reference_delta_l2": 2.719512980178627, + "retained_delta_l2": 2.2337422557531568, + "delta_retention_norm_ratio": 0.8213758389954207, + "delta_cosine": 0.6372758864098026, + "relative_weight_rmse": 0.010427102510948057 + }, + { + "index": 248, + "total": 252, + "projection": "layers.9.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3898, + "parameters": 50331648, + "reference_delta_l2": 3.0350806499741614, + "retained_delta_l2": 2.7550311170101582, + "delta_retention_norm_ratio": 0.9077291297131141, + "delta_cosine": 0.7226854140229847, + "relative_weight_rmse": 0.012411422986188504 + }, + { + "index": 249, + "total": 252, + "projection": "layers.9.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.028, + "parameters": 4194304, + "reference_delta_l2": 0.8391053738870392, + "retained_delta_l2": 0.756872036454735, + "delta_retention_norm_ratio": 0.9019987953939924, + "delta_cosine": 0.7180783501603136, + "relative_weight_rmse": 0.011796885085392174 + }, + { + "index": 250, + "total": 252, + "projection": "layers.9.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1322, + "parameters": 16777216, + "reference_delta_l2": 1.5264005224754331, + "retained_delta_l2": 1.300963109455754, + "delta_retention_norm_ratio": 0.8523078250431433, + "delta_cosine": 0.6668298069238295, + "relative_weight_rmse": 0.011163748919664512 + }, + { + "index": 251, + "total": 252, + "projection": "layers.9.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1327, + "parameters": 16777216, + "reference_delta_l2": 1.807348681333385, + "retained_delta_l2": 1.660795021636992, + "delta_retention_norm_ratio": 0.9189123486740411, + "delta_cosine": 0.7352203254320391, + "relative_weight_rmse": 0.012714498697596345 + }, + { + "index": 252, + "total": 252, + "projection": "layers.9.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0289, + "parameters": 4194304, + "reference_delta_l2": 0.8014616733642863, + "retained_delta_l2": 0.6708871042318548, + "delta_retention_norm_ratio": 0.8370794593529094, + "delta_cosine": 0.6533762250475577, + "relative_weight_rmse": 0.01043275206420006 + } + ] +}