diff --git "a/manifests/merge_step_00001000.json" "b/manifests/merge_step_00001000.json" new file mode 100644--- /dev/null +++ "b/manifests/merge_step_00001000.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_00001000/adapters.safetensors", + "adapter_sha256": "4a86181524ab17680c500add76f183acd257840b8a0d18e33807ae379d52925e", + "adapter_global_step": "1000", + "output": "/workspace/Ideogram/artifacts/huggingface_release/s020_v1/qwen3vl_8b_ideogram4_nl_s020_v1_step_00001000_fp8_scaled.safetensors", + "output_bytes": 10588637800, + "output_sha256": "32b551cc403a4056efb1b03fa2fa5f00c0af537e95f5d35fbe873bd634348431", + "output_tensors": 1254, + "merged_projections": 252, + "lora_rank": 64, + "lora_alpha": 64.0, + "elapsed_seconds": 63.154, + "projections": [ + { + "index": 1, + "total": 252, + "projection": "layers.0.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.4224, + "parameters": 50331648, + "reference_delta_l2": 3.0816885619998686, + "retained_delta_l2": 2.7982748486855002, + "delta_retention_norm_ratio": 0.9080329800976233, + "delta_cosine": 0.7232567857868553, + "relative_weight_rmse": 0.012175151807146295 + }, + { + "index": 2, + "total": 252, + "projection": "layers.0.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.405, + "parameters": 50331648, + "reference_delta_l2": 3.1480601580259853, + "retained_delta_l2": 2.832615243262361, + "delta_retention_norm_ratio": 0.8997970499517307, + "delta_cosine": 0.7116974133867178, + "relative_weight_rmse": 0.013097672001861951 + }, + { + "index": 3, + "total": 252, + "projection": "layers.0.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.4112, + "parameters": 50331648, + "reference_delta_l2": 2.804011535014854, + "retained_delta_l2": 2.4696998178344756, + "delta_retention_norm_ratio": 0.8807737725021137, + "delta_cosine": 0.693008894807113, + "relative_weight_rmse": 0.012674360524560205 + }, + { + "index": 4, + "total": 252, + "projection": "layers.0.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0346, + "parameters": 4194304, + "reference_delta_l2": 0.5889772111434513, + "retained_delta_l2": 0.4264678883491288, + "delta_retention_norm_ratio": 0.7240821550993053, + "delta_cosine": 0.5528368996524894, + "relative_weight_rmse": 0.008857766972819075 + }, + { + "index": 5, + "total": 252, + "projection": "layers.0.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1317, + "parameters": 16777216, + "reference_delta_l2": 1.953012788138372, + "retained_delta_l2": 1.8380204357780252, + "delta_retention_norm_ratio": 0.941120532820495, + "delta_cosine": 0.760871469990079, + "relative_weight_rmse": 0.013526435937534671 + }, + { + "index": 6, + "total": 252, + "projection": "layers.0.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1288, + "parameters": 16777216, + "reference_delta_l2": 1.1948266779405832, + "retained_delta_l2": 0.9188756904166978, + "delta_retention_norm_ratio": 0.7690451739832947, + "delta_cosine": 0.5909115937759196, + "relative_weight_rmse": 0.009624968518458206 + }, + { + "index": 7, + "total": 252, + "projection": "layers.0.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0311, + "parameters": 4194304, + "reference_delta_l2": 0.7817860145303532, + "retained_delta_l2": 0.6765826875245834, + "delta_retention_norm_ratio": 0.865432068302003, + "delta_cosine": 0.679117835122916, + "relative_weight_rmse": 0.011844264175045386 + }, + { + "index": 8, + "total": 252, + "projection": "layers.1.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.4346, + "parameters": 50331648, + "reference_delta_l2": 2.000792856247298, + "retained_delta_l2": 1.8609967514572074, + "delta_retention_norm_ratio": 0.930129646178219, + "delta_cosine": 0.7529486527281806, + "relative_weight_rmse": 0.011844969026974602 + }, + { + "index": 9, + "total": 252, + "projection": "layers.1.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3775, + "parameters": 50331648, + "reference_delta_l2": 3.8474750849362778, + "retained_delta_l2": 3.8115742317771417, + "delta_retention_norm_ratio": 0.9906689835888227, + "delta_cosine": 0.8487083486486127, + "relative_weight_rmse": 0.014282374751721142 + }, + { + "index": 10, + "total": 252, + "projection": "layers.1.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.4014, + "parameters": 50331648, + "reference_delta_l2": 3.952469385601115, + "retained_delta_l2": 3.9478783704965132, + "delta_retention_norm_ratio": 0.9988384438545364, + "delta_cosine": 0.9198098546606038, + "relative_weight_rmse": 0.015704980665568778 + }, + { + "index": 11, + "total": 252, + "projection": "layers.1.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0326, + "parameters": 4194304, + "reference_delta_l2": 0.6682698794480948, + "retained_delta_l2": 0.5467667468151203, + "delta_retention_norm_ratio": 0.8181825391661816, + "delta_cosine": 0.6351642691683482, + "relative_weight_rmse": 0.010709906067521315 + }, + { + "index": 12, + "total": 252, + "projection": "layers.1.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1344, + "parameters": 16777216, + "reference_delta_l2": 1.6857501342429857, + "retained_delta_l2": 1.5002057950336098, + "delta_retention_norm_ratio": 0.8899336648769139, + "delta_cosine": 0.7007274386238729, + "relative_weight_rmse": 0.012528648331599874 + }, + { + "index": 13, + "total": 252, + "projection": "layers.1.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1374, + "parameters": 16777216, + "reference_delta_l2": 2.0614481256874786, + "retained_delta_l2": 1.987957594333203, + "delta_retention_norm_ratio": 0.9643500457573887, + "delta_cosine": 0.7895502817137614, + "relative_weight_rmse": 0.01451153624066018 + }, + { + "index": 14, + "total": 252, + "projection": "layers.1.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0304, + "parameters": 4194304, + "reference_delta_l2": 0.8667964063303081, + "retained_delta_l2": 0.77575589081767, + "delta_retention_norm_ratio": 0.8949689744353354, + "delta_cosine": 0.706884798109923, + "relative_weight_rmse": 0.012035954867732342 + }, + { + "index": 15, + "total": 252, + "projection": "layers.10.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.414, + "parameters": 50331648, + "reference_delta_l2": 2.7788980063059525, + "retained_delta_l2": 2.4570278951823883, + "delta_retention_norm_ratio": 0.884173470781163, + "delta_cosine": 0.6983210734453389, + "relative_weight_rmse": 0.011560182532117278 + }, + { + "index": 16, + "total": 252, + "projection": "layers.10.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3779, + "parameters": 50331648, + "reference_delta_l2": 2.368764390066135, + "retained_delta_l2": 1.8310619378040711, + "delta_retention_norm_ratio": 0.7730029822649219, + "delta_cosine": 0.5941158911125419, + "relative_weight_rmse": 0.009580620878966087 + }, + { + "index": 17, + "total": 252, + "projection": "layers.10.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3804, + "parameters": 50331648, + "reference_delta_l2": 2.8477191151243963, + "retained_delta_l2": 2.5201258455709303, + "delta_retention_norm_ratio": 0.884962927764329, + "delta_cosine": 0.6975840407097988, + "relative_weight_rmse": 0.011870660911946867 + }, + { + "index": 18, + "total": 252, + "projection": "layers.10.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0325, + "parameters": 4194304, + "reference_delta_l2": 0.8841966844382879, + "retained_delta_l2": 0.8093074279674809, + "delta_retention_norm_ratio": 0.9153024911890699, + "delta_cosine": 0.732661207864237, + "relative_weight_rmse": 0.01186819485958493 + }, + { + "index": 19, + "total": 252, + "projection": "layers.10.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1304, + "parameters": 16777216, + "reference_delta_l2": 1.7629359486792742, + "retained_delta_l2": 1.573622608399111, + "delta_retention_norm_ratio": 0.892614737125311, + "delta_cosine": 0.7057376986654624, + "relative_weight_rmse": 0.012108665291259917 + }, + { + "index": 20, + "total": 252, + "projection": "layers.10.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1266, + "parameters": 16777216, + "reference_delta_l2": 2.1068966603030224, + "retained_delta_l2": 2.0082854319482006, + "delta_retention_norm_ratio": 0.9531959824073009, + "delta_cosine": 0.7789324690671694, + "relative_weight_rmse": 0.013402050492345592 + }, + { + "index": 21, + "total": 252, + "projection": "layers.10.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0306, + "parameters": 4194304, + "reference_delta_l2": 0.9027386910930574, + "retained_delta_l2": 0.8047880119788361, + "delta_retention_norm_ratio": 0.8914960884244141, + "delta_cosine": 0.7091814370016879, + "relative_weight_rmse": 0.011149653421646751 + }, + { + "index": 22, + "total": 252, + "projection": "layers.11.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.417, + "parameters": 50331648, + "reference_delta_l2": 2.5840047800340313, + "retained_delta_l2": 2.2079660414887887, + "delta_retention_norm_ratio": 0.8544744415912844, + "delta_cosine": 0.6692794347825356, + "relative_weight_rmse": 0.010957964912540056 + }, + { + "index": 23, + "total": 252, + "projection": "layers.11.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3827, + "parameters": 50331648, + "reference_delta_l2": 2.7383686211590557, + "retained_delta_l2": 2.302185080322085, + "delta_retention_norm_ratio": 0.8407140888678641, + "delta_cosine": 0.6564171180671049, + "relative_weight_rmse": 0.010709929420531077 + }, + { + "index": 24, + "total": 252, + "projection": "layers.11.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.4189, + "parameters": 50331648, + "reference_delta_l2": 2.6496623578996585, + "retained_delta_l2": 2.2453994501196055, + "delta_retention_norm_ratio": 0.847428519873564, + "delta_cosine": 0.6613952117246973, + "relative_weight_rmse": 0.011290561995860135 + }, + { + "index": 25, + "total": 252, + "projection": "layers.11.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0359, + "parameters": 4194304, + "reference_delta_l2": 1.0926010979238467, + "retained_delta_l2": 1.0447864068194332, + "delta_retention_norm_ratio": 0.9562377420311305, + "delta_cosine": 0.779335603836543, + "relative_weight_rmse": 0.013427349461668502 + }, + { + "index": 26, + "total": 252, + "projection": "layers.11.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1491, + "parameters": 16777216, + "reference_delta_l2": 1.674614888066037, + "retained_delta_l2": 1.473326613582539, + "delta_retention_norm_ratio": 0.8798002597982634, + "delta_cosine": 0.6945583937590933, + "relative_weight_rmse": 0.011736867215697059 + }, + { + "index": 27, + "total": 252, + "projection": "layers.11.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1375, + "parameters": 16777216, + "reference_delta_l2": 1.3849513751455016, + "retained_delta_l2": 1.1455167655182101, + "delta_retention_norm_ratio": 0.8271169559276861, + "delta_cosine": 0.6423088179518641, + "relative_weight_rmse": 0.010631178127543661 + }, + { + "index": 28, + "total": 252, + "projection": "layers.11.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0335, + "parameters": 4194304, + "reference_delta_l2": 0.9783124757473397, + "retained_delta_l2": 0.8855191820256861, + "delta_retention_norm_ratio": 0.9051496367244339, + "delta_cosine": 0.7225166647242568, + "relative_weight_rmse": 0.011731087388485384 + }, + { + "index": 29, + "total": 252, + "projection": "layers.12.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.4499, + "parameters": 50331648, + "reference_delta_l2": 2.5304979327583323, + "retained_delta_l2": 2.1471333458008868, + "delta_retention_norm_ratio": 0.8485023117408499, + "delta_cosine": 0.6634892219290347, + "relative_weight_rmse": 0.010715613067999084 + }, + { + "index": 30, + "total": 252, + "projection": "layers.12.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.4339, + "parameters": 50331648, + "reference_delta_l2": 3.1633003862648446, + "retained_delta_l2": 2.8246694645696486, + "delta_retention_norm_ratio": 0.8929501215990923, + "delta_cosine": 0.7087972874821045, + "relative_weight_rmse": 0.011858031862104884 + }, + { + "index": 31, + "total": 252, + "projection": "layers.12.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.399, + "parameters": 50331648, + "reference_delta_l2": 2.6580761108935107, + "retained_delta_l2": 2.2587677501269527, + "delta_retention_norm_ratio": 0.8497754224831694, + "delta_cosine": 0.6637711209384476, + "relative_weight_rmse": 0.01112945261763965 + }, + { + "index": 32, + "total": 252, + "projection": "layers.12.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0312, + "parameters": 4194304, + "reference_delta_l2": 0.9224668769685606, + "retained_delta_l2": 0.8550623115033129, + "delta_retention_norm_ratio": 0.9269300967350127, + "delta_cosine": 0.7462213863560913, + "relative_weight_rmse": 0.012188923906095057 + }, + { + "index": 33, + "total": 252, + "projection": "layers.12.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1265, + "parameters": 16777216, + "reference_delta_l2": 1.758802705798867, + "retained_delta_l2": 1.5903634728969844, + "delta_retention_norm_ratio": 0.9042307404085009, + "delta_cosine": 0.7203372747546253, + "relative_weight_rmse": 0.011957594930966502 + }, + { + "index": 34, + "total": 252, + "projection": "layers.12.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1353, + "parameters": 16777216, + "reference_delta_l2": 1.7918025744901438, + "retained_delta_l2": 1.6359286559612636, + "delta_retention_norm_ratio": 0.9130072024964948, + "delta_cosine": 0.730084159456328, + "relative_weight_rmse": 0.012354100880457046 + }, + { + "index": 35, + "total": 252, + "projection": "layers.12.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0308, + "parameters": 4194304, + "reference_delta_l2": 1.0804826074661804, + "retained_delta_l2": 1.0165835916893688, + "delta_retention_norm_ratio": 0.9408606715783607, + "delta_cosine": 0.7649719145411196, + "relative_weight_rmse": 0.012138865703742095 + }, + { + "index": 36, + "total": 252, + "projection": "layers.13.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.4224, + "parameters": 50331648, + "reference_delta_l2": 2.324236423547833, + "retained_delta_l2": 1.8986620410835808, + "delta_retention_norm_ratio": 0.8168971202100714, + "delta_cosine": 0.6341284918339433, + "relative_weight_rmse": 0.010117405580660927 + }, + { + "index": 37, + "total": 252, + "projection": "layers.13.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3865, + "parameters": 50331648, + "reference_delta_l2": 2.9355113214494977, + "retained_delta_l2": 2.5514820556484334, + "delta_retention_norm_ratio": 0.8691780668686236, + "delta_cosine": 0.6823762719120268, + "relative_weight_rmse": 0.011744441811130537 + }, + { + "index": 38, + "total": 252, + "projection": "layers.13.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3789, + "parameters": 50331648, + "reference_delta_l2": 2.3865673851492515, + "retained_delta_l2": 1.925716840298132, + "delta_retention_norm_ratio": 0.8068981635637752, + "delta_cosine": 0.6236413885350592, + "relative_weight_rmse": 0.010298461240813597 + }, + { + "index": 39, + "total": 252, + "projection": "layers.13.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0317, + "parameters": 4194304, + "reference_delta_l2": 1.3380128734896723, + "retained_delta_l2": 1.3237175722885999, + "delta_retention_norm_ratio": 0.9893160211801334, + "delta_cosine": 0.8335338176236261, + "relative_weight_rmse": 0.014782816548570533 + }, + { + "index": 40, + "total": 252, + "projection": "layers.13.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1278, + "parameters": 16777216, + "reference_delta_l2": 1.7131074458581192, + "retained_delta_l2": 1.531750375884282, + "delta_retention_norm_ratio": 0.8941356127939816, + "delta_cosine": 0.7096703108243391, + "relative_weight_rmse": 0.012179875360966609 + }, + { + "index": 41, + "total": 252, + "projection": "layers.13.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1268, + "parameters": 16777216, + "reference_delta_l2": 1.3735221604424783, + "retained_delta_l2": 1.1332531527367187, + "delta_retention_norm_ratio": 0.8250708910088809, + "delta_cosine": 0.6402408291163287, + "relative_weight_rmse": 0.010547182836815596 + }, + { + "index": 42, + "total": 252, + "projection": "layers.13.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0307, + "parameters": 4194304, + "reference_delta_l2": 0.9969963944070404, + "retained_delta_l2": 0.9115721762195691, + "delta_retention_norm_ratio": 0.914318428164149, + "delta_cosine": 0.731245724014459, + "relative_weight_rmse": 0.01207298339000756 + }, + { + "index": 43, + "total": 252, + "projection": "layers.14.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.4176, + "parameters": 50331648, + "reference_delta_l2": 2.45495155106801, + "retained_delta_l2": 2.076652965519915, + "delta_retention_norm_ratio": 0.84590384874051, + "delta_cosine": 0.6618875021201504, + "relative_weight_rmse": 0.01045629905580165 + }, + { + "index": 44, + "total": 252, + "projection": "layers.14.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3784, + "parameters": 50331648, + "reference_delta_l2": 2.678491958143725, + "retained_delta_l2": 2.2562722872591454, + "delta_retention_norm_ratio": 0.8423666460521352, + "delta_cosine": 0.6559101667040926, + "relative_weight_rmse": 0.011200407226753849 + }, + { + "index": 45, + "total": 252, + "projection": "layers.14.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3772, + "parameters": 50331648, + "reference_delta_l2": 2.8943822537465125, + "retained_delta_l2": 2.5535949867271546, + "delta_retention_norm_ratio": 0.8822590669984104, + "delta_cosine": 0.6966157399628781, + "relative_weight_rmse": 0.011646018845291059 + }, + { + "index": 46, + "total": 252, + "projection": "layers.14.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0326, + "parameters": 4194304, + "reference_delta_l2": 0.7193206751123594, + "retained_delta_l2": 0.6080653608867637, + "delta_retention_norm_ratio": 0.8453328006897376, + "delta_cosine": 0.6622707417399369, + "relative_weight_rmse": 0.010624440044558647 + }, + { + "index": 47, + "total": 252, + "projection": "layers.14.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1281, + "parameters": 16777216, + "reference_delta_l2": 1.5477825423303293, + "retained_delta_l2": 1.3465421238514126, + "delta_retention_norm_ratio": 0.8699814651120623, + "delta_cosine": 0.6845359781721311, + "relative_weight_rmse": 0.011387335697726245 + }, + { + "index": 48, + "total": 252, + "projection": "layers.14.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.127, + "parameters": 16777216, + "reference_delta_l2": 2.31726342833638, + "retained_delta_l2": 2.258437926215563, + "delta_retention_norm_ratio": 0.9746142361712197, + "delta_cosine": 0.8052658593739344, + "relative_weight_rmse": 0.013915283945923996 + }, + { + "index": 49, + "total": 252, + "projection": "layers.14.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0308, + "parameters": 4194304, + "reference_delta_l2": 1.0084847564065471, + "retained_delta_l2": 0.934072561923918, + "delta_retention_norm_ratio": 0.9262138629167028, + "delta_cosine": 0.7429330922555639, + "relative_weight_rmse": 0.012332294066236866 + }, + { + "index": 50, + "total": 252, + "projection": "layers.15.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.4231, + "parameters": 50331648, + "reference_delta_l2": 2.3508724083193986, + "retained_delta_l2": 1.9527586511552484, + "delta_retention_norm_ratio": 0.830652758628974, + "delta_cosine": 0.6470499357809475, + "relative_weight_rmse": 0.010222570843416734 + }, + { + "index": 51, + "total": 252, + "projection": "layers.15.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3805, + "parameters": 50331648, + "reference_delta_l2": 2.7941246177612578, + "retained_delta_l2": 2.4231315782214073, + "delta_retention_norm_ratio": 0.86722387499055, + "delta_cosine": 0.6795366829782085, + "relative_weight_rmse": 0.011784804982088613 + }, + { + "index": 52, + "total": 252, + "projection": "layers.15.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3762, + "parameters": 50331648, + "reference_delta_l2": 2.7679733593437232, + "retained_delta_l2": 2.417205448280628, + "delta_retention_norm_ratio": 0.873276269123392, + "delta_cosine": 0.6878619366042454, + "relative_weight_rmse": 0.011359617528472313 + }, + { + "index": 53, + "total": 252, + "projection": "layers.15.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0324, + "parameters": 4194304, + "reference_delta_l2": 0.7776479157002186, + "retained_delta_l2": 0.677901359807814, + "delta_retention_norm_ratio": 0.8717330119729186, + "delta_cosine": 0.6869978403240675, + "relative_weight_rmse": 0.011237756007861584 + }, + { + "index": 54, + "total": 252, + "projection": "layers.15.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1279, + "parameters": 16777216, + "reference_delta_l2": 2.1367385814187783, + "retained_delta_l2": 2.052169554173674, + "delta_retention_norm_ratio": 0.9604214441670488, + "delta_cosine": 0.7876284843094786, + "relative_weight_rmse": 0.013780714005269206 + }, + { + "index": 55, + "total": 252, + "projection": "layers.15.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1273, + "parameters": 16777216, + "reference_delta_l2": 2.2542125019847377, + "retained_delta_l2": 2.1816583557346045, + "delta_retention_norm_ratio": 0.9678139722025975, + "delta_cosine": 0.8030489099528074, + "relative_weight_rmse": 0.013495301422341281 + }, + { + "index": 56, + "total": 252, + "projection": "layers.15.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0311, + "parameters": 4194304, + "reference_delta_l2": 0.8159615265348684, + "retained_delta_l2": 0.7160163282309382, + "delta_retention_norm_ratio": 0.8775123641817207, + "delta_cosine": 0.6940623175429661, + "relative_weight_rmse": 0.010829061532462544 + }, + { + "index": 57, + "total": 252, + "projection": "layers.16.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.4321, + "parameters": 50331648, + "reference_delta_l2": 2.4573315682382946, + "retained_delta_l2": 2.091623194215147, + "delta_retention_norm_ratio": 0.8511766264064516, + "delta_cosine": 0.6661996603476956, + "relative_weight_rmse": 0.010554514481033535 + }, + { + "index": 58, + "total": 252, + "projection": "layers.16.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3753, + "parameters": 50331648, + "reference_delta_l2": 2.6011198700619484, + "retained_delta_l2": 2.213585965173769, + "delta_retention_norm_ratio": 0.8510126698317254, + "delta_cosine": 0.6647791380353715, + "relative_weight_rmse": 0.010953587049171 + }, + { + "index": 59, + "total": 252, + "projection": "layers.16.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3772, + "parameters": 50331648, + "reference_delta_l2": 2.534007189292717, + "retained_delta_l2": 2.130868616559504, + "delta_retention_norm_ratio": 0.8409086704897094, + "delta_cosine": 0.6550433329888625, + "relative_weight_rmse": 0.010815131821803752 + }, + { + "index": 60, + "total": 252, + "projection": "layers.16.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0335, + "parameters": 4194304, + "reference_delta_l2": 0.7291351979009432, + "retained_delta_l2": 0.6252688276208757, + "delta_retention_norm_ratio": 0.8575485443864441, + "delta_cosine": 0.6717687931868451, + "relative_weight_rmse": 0.011093600926568206 + }, + { + "index": 61, + "total": 252, + "projection": "layers.16.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1287, + "parameters": 16777216, + "reference_delta_l2": 1.654412246524243, + "retained_delta_l2": 1.4767944665425285, + "delta_retention_norm_ratio": 0.8926399509222253, + "delta_cosine": 0.7073607977623679, + "relative_weight_rmse": 0.011885141947948749 + }, + { + "index": 62, + "total": 252, + "projection": "layers.16.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1281, + "parameters": 16777216, + "reference_delta_l2": 1.8139515232656178, + "retained_delta_l2": 1.6833357740492332, + "delta_retention_norm_ratio": 0.9279938038359261, + "delta_cosine": 0.7488159261124129, + "relative_weight_rmse": 0.01247368051553322 + }, + { + "index": 63, + "total": 252, + "projection": "layers.16.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.032, + "parameters": 4194304, + "reference_delta_l2": 0.9342047823832472, + "retained_delta_l2": 0.8463807340885396, + "delta_retention_norm_ratio": 0.9059905815610791, + "delta_cosine": 0.7229628350748435, + "relative_weight_rmse": 0.011960945516098463 + }, + { + "index": 64, + "total": 252, + "projection": "layers.17.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.4183, + "parameters": 50331648, + "reference_delta_l2": 2.5324195438776362, + "retained_delta_l2": 2.165307921272776, + "delta_retention_norm_ratio": 0.8550352276768725, + "delta_cosine": 0.669880256413046, + "relative_weight_rmse": 0.010784003147264142 + }, + { + "index": 65, + "total": 252, + "projection": "layers.17.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3771, + "parameters": 50331648, + "reference_delta_l2": 3.4293703351448817, + "retained_delta_l2": 3.1932583538525163, + "delta_retention_norm_ratio": 0.9311500484877232, + "delta_cosine": 0.7490466617957803, + "relative_weight_rmse": 0.013045466838710296 + }, + { + "index": 66, + "total": 252, + "projection": "layers.17.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.376, + "parameters": 50331648, + "reference_delta_l2": 3.1344403318131016, + "retained_delta_l2": 2.8428468910673423, + "delta_retention_norm_ratio": 0.906971130448322, + "delta_cosine": 0.7214019043408542, + "relative_weight_rmse": 0.012253945090414364 + }, + { + "index": 67, + "total": 252, + "projection": "layers.17.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0333, + "parameters": 4194304, + "reference_delta_l2": 0.7828475923209274, + "retained_delta_l2": 0.6983824803773275, + "delta_retention_norm_ratio": 0.8921052925599681, + "delta_cosine": 0.7116379990114124, + "relative_weight_rmse": 0.011542060532710544 + }, + { + "index": 68, + "total": 252, + "projection": "layers.17.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1288, + "parameters": 16777216, + "reference_delta_l2": 1.7174285248377603, + "retained_delta_l2": 1.5551782228154287, + "delta_retention_norm_ratio": 0.9055271880745903, + "delta_cosine": 0.7228138335412242, + "relative_weight_rmse": 0.012145564745262539 + }, + { + "index": 69, + "total": 252, + "projection": "layers.17.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1291, + "parameters": 16777216, + "reference_delta_l2": 1.6536853726050285, + "retained_delta_l2": 1.477901179425521, + "delta_retention_norm_ratio": 0.8937015492236005, + "delta_cosine": 0.7065054241140166, + "relative_weight_rmse": 0.01193774066627311 + }, + { + "index": 70, + "total": 252, + "projection": "layers.17.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0321, + "parameters": 4194304, + "reference_delta_l2": 1.0194408732968017, + "retained_delta_l2": 0.9529750905075698, + "delta_retention_norm_ratio": 0.9348017285452898, + "delta_cosine": 0.7554989899505795, + "relative_weight_rmse": 0.012663269085123486 + }, + { + "index": 71, + "total": 252, + "projection": "layers.18.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.4179, + "parameters": 50331648, + "reference_delta_l2": 2.899431101930452, + "retained_delta_l2": 2.611353903809252, + "delta_retention_norm_ratio": 0.9006435442009997, + "delta_cosine": 0.7149613293676941, + "relative_weight_rmse": 0.011678172729971216 + }, + { + "index": 72, + "total": 252, + "projection": "layers.18.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3754, + "parameters": 50331648, + "reference_delta_l2": 4.5890100404179845, + "retained_delta_l2": 4.534477602671866, + "delta_retention_norm_ratio": 0.9881167316554508, + "delta_cosine": 0.8220542425353964, + "relative_weight_rmse": 0.015173122032415403 + }, + { + "index": 73, + "total": 252, + "projection": "layers.18.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3767, + "parameters": 50331648, + "reference_delta_l2": 3.266588304278785, + "retained_delta_l2": 3.0069138257067127, + "delta_retention_norm_ratio": 0.9205059057390446, + "delta_cosine": 0.7369795766243502, + "relative_weight_rmse": 0.012479011234955232 + }, + { + "index": 74, + "total": 252, + "projection": "layers.18.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0346, + "parameters": 4194304, + "reference_delta_l2": 1.1146092689035678, + "retained_delta_l2": 1.0808960531489715, + "delta_retention_norm_ratio": 0.9697533326743643, + "delta_cosine": 0.7983662545363905, + "relative_weight_rmse": 0.01398004908416685 + }, + { + "index": 75, + "total": 252, + "projection": "layers.18.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.129, + "parameters": 16777216, + "reference_delta_l2": 1.6260721310282324, + "retained_delta_l2": 1.452712507352149, + "delta_retention_norm_ratio": 0.8933874947069778, + "delta_cosine": 0.7099197632942137, + "relative_weight_rmse": 0.011693765517108823 + }, + { + "index": 76, + "total": 252, + "projection": "layers.18.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.129, + "parameters": 16777216, + "reference_delta_l2": 2.101630652715759, + "retained_delta_l2": 2.0120008520790047, + "delta_retention_norm_ratio": 0.9573522585802918, + "delta_cosine": 0.7847540097428316, + "relative_weight_rmse": 0.013525379648372384 + }, + { + "index": 77, + "total": 252, + "projection": "layers.18.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0323, + "parameters": 4194304, + "reference_delta_l2": 0.9173633033004055, + "retained_delta_l2": 0.8325380615782747, + "delta_retention_norm_ratio": 0.9075336440677818, + "delta_cosine": 0.7242506539370204, + "relative_weight_rmse": 0.011713017960162778 + }, + { + "index": 78, + "total": 252, + "projection": "layers.19.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.419, + "parameters": 50331648, + "reference_delta_l2": 2.6779400857457345, + "retained_delta_l2": 2.3403128602042385, + "delta_retention_norm_ratio": 0.8739227858985218, + "delta_cosine": 0.6881088113005049, + "relative_weight_rmse": 0.011160997185037408 + }, + { + "index": 79, + "total": 252, + "projection": "layers.19.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3781, + "parameters": 50331648, + "reference_delta_l2": 4.448558989953332, + "retained_delta_l2": 4.36185324358659, + "delta_retention_norm_ratio": 0.9805092510715135, + "delta_cosine": 0.8113188310815217, + "relative_weight_rmse": 0.01514704434869321 + }, + { + "index": 80, + "total": 252, + "projection": "layers.19.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3765, + "parameters": 50331648, + "reference_delta_l2": 3.063246627652322, + "retained_delta_l2": 2.7395466596887146, + "delta_retention_norm_ratio": 0.8943278138163848, + "delta_cosine": 0.7085337178302815, + "relative_weight_rmse": 0.012120377731919669 + }, + { + "index": 81, + "total": 252, + "projection": "layers.19.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0327, + "parameters": 4194304, + "reference_delta_l2": 1.038542610002628, + "retained_delta_l2": 0.9987469325634057, + "delta_retention_norm_ratio": 0.9616812280440554, + "delta_cosine": 0.7934637827745186, + "relative_weight_rmse": 0.01356428686058323 + }, + { + "index": 82, + "total": 252, + "projection": "layers.19.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1293, + "parameters": 16777216, + "reference_delta_l2": 1.6305255320970735, + "retained_delta_l2": 1.4495263143497212, + "delta_retention_norm_ratio": 0.8889933250449853, + "delta_cosine": 0.7037037396863515, + "relative_weight_rmse": 0.01175068926467507 + }, + { + "index": 83, + "total": 252, + "projection": "layers.19.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1292, + "parameters": 16777216, + "reference_delta_l2": 1.9187319565958694, + "retained_delta_l2": 1.8063254844946643, + "delta_retention_norm_ratio": 0.9414162714521982, + "delta_cosine": 0.7636972271279521, + "relative_weight_rmse": 0.012749695224232683 + }, + { + "index": 84, + "total": 252, + "projection": "layers.19.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.032, + "parameters": 4194304, + "reference_delta_l2": 1.238333822391516, + "retained_delta_l2": 1.2056878115530523, + "delta_retention_norm_ratio": 0.9736371483616458, + "delta_cosine": 0.8049945357439888, + "relative_weight_rmse": 0.013931864650010494 + }, + { + "index": 85, + "total": 252, + "projection": "layers.2.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.4325, + "parameters": 50331648, + "reference_delta_l2": 1.5163221557782192, + "retained_delta_l2": 1.2998390823120969, + "delta_retention_norm_ratio": 0.8572314777297328, + "delta_cosine": 0.6941018406592098, + "relative_weight_rmse": 0.009930361765633434 + }, + { + "index": 86, + "total": 252, + "projection": "layers.2.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3768, + "parameters": 50331648, + "reference_delta_l2": 2.507124496785796, + "retained_delta_l2": 2.3613625027298766, + "delta_retention_norm_ratio": 0.941860887146693, + "delta_cosine": 0.7994646319421327, + "relative_weight_rmse": 0.01258178701267469 + }, + { + "index": 87, + "total": 252, + "projection": "layers.2.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3771, + "parameters": 50331648, + "reference_delta_l2": 3.3657371303030876, + "retained_delta_l2": 3.3428196209156122, + "delta_retention_norm_ratio": 0.9931909390126936, + "delta_cosine": 0.8803031766681577, + "relative_weight_rmse": 0.014876112756496833 + }, + { + "index": 88, + "total": 252, + "projection": "layers.2.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0335, + "parameters": 4194304, + "reference_delta_l2": 0.5654454172264339, + "retained_delta_l2": 0.42165701508364567, + "delta_retention_norm_ratio": 0.7457077239248224, + "delta_cosine": 0.5708347089338801, + "relative_weight_rmse": 0.009137987951469532 + }, + { + "index": 89, + "total": 252, + "projection": "layers.2.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1294, + "parameters": 16777216, + "reference_delta_l2": 2.01865574307719, + "retained_delta_l2": 1.8910245853196903, + "delta_retention_norm_ratio": 0.936774183416266, + "delta_cosine": 0.7529442522388894, + "relative_weight_rmse": 0.013673422112398305 + }, + { + "index": 90, + "total": 252, + "projection": "layers.2.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1299, + "parameters": 16777216, + "reference_delta_l2": 1.2106949183435032, + "retained_delta_l2": 0.9504159178632374, + "delta_retention_norm_ratio": 0.7850168555787905, + "delta_cosine": 0.603823434820368, + "relative_weight_rmse": 0.010067183217040766 + }, + { + "index": 91, + "total": 252, + "projection": "layers.2.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0349, + "parameters": 4194304, + "reference_delta_l2": 1.1196199495350254, + "retained_delta_l2": 1.0737372487164127, + "delta_retention_norm_ratio": 0.959019396860812, + "delta_cosine": 0.7853926611761529, + "relative_weight_rmse": 0.013243675264669859 + }, + { + "index": 92, + "total": 252, + "projection": "layers.20.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.4265, + "parameters": 50331648, + "reference_delta_l2": 2.6131003917687843, + "retained_delta_l2": 2.2232143033125533, + "delta_retention_norm_ratio": 0.8507955952689897, + "delta_cosine": 0.6634199678359055, + "relative_weight_rmse": 0.011137848124095295 + }, + { + "index": 93, + "total": 252, + "projection": "layers.20.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3838, + "parameters": 50331648, + "reference_delta_l2": 4.31757704185912, + "retained_delta_l2": 4.2196783628573655, + "delta_retention_norm_ratio": 0.9773255513329302, + "delta_cosine": 0.8062174730963484, + "relative_weight_rmse": 0.015015690410046452 + }, + { + "index": 94, + "total": 252, + "projection": "layers.20.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3792, + "parameters": 50331648, + "reference_delta_l2": 3.0842895980982585, + "retained_delta_l2": 2.76239977588676, + "delta_retention_norm_ratio": 0.8956356684502088, + "delta_cosine": 0.7095376044716241, + "relative_weight_rmse": 0.012157092563078684 + }, + { + "index": 95, + "total": 252, + "projection": "layers.20.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0336, + "parameters": 4194304, + "reference_delta_l2": 0.8379474530400505, + "retained_delta_l2": 0.7689408142812618, + "delta_retention_norm_ratio": 0.9176480118073819, + "delta_cosine": 0.7373170826253173, + "relative_weight_rmse": 0.012165603932648003 + }, + { + "index": 96, + "total": 252, + "projection": "layers.20.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1291, + "parameters": 16777216, + "reference_delta_l2": 1.5955174518101594, + "retained_delta_l2": 1.4123102945700292, + "delta_retention_norm_ratio": 0.8851738305762332, + "delta_cosine": 0.7011995151432837, + "relative_weight_rmse": 0.011755138657372448 + }, + { + "index": 97, + "total": 252, + "projection": "layers.20.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1295, + "parameters": 16777216, + "reference_delta_l2": 1.869906772204829, + "retained_delta_l2": 1.7437207907199404, + "delta_retention_norm_ratio": 0.9325175012141909, + "delta_cosine": 0.7548889661402391, + "relative_weight_rmse": 0.012232723171194271 + }, + { + "index": 98, + "total": 252, + "projection": "layers.20.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0325, + "parameters": 4194304, + "reference_delta_l2": 0.9286100794562515, + "retained_delta_l2": 0.8487618239822285, + "delta_retention_norm_ratio": 0.9140131501471768, + "delta_cosine": 0.7327040594611609, + "relative_weight_rmse": 0.011901526821837701 + }, + { + "index": 99, + "total": 252, + "projection": "layers.21.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.423, + "parameters": 50331648, + "reference_delta_l2": 3.064181667470853, + "retained_delta_l2": 2.7575754533586934, + "delta_retention_norm_ratio": 0.8999386304777324, + "delta_cosine": 0.7129984995145269, + "relative_weight_rmse": 0.0121885584297972 + }, + { + "index": 100, + "total": 252, + "projection": "layers.21.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3786, + "parameters": 50331648, + "reference_delta_l2": 4.352146801117139, + "retained_delta_l2": 4.2567664708244, + "delta_retention_norm_ratio": 0.9780843030689459, + "delta_cosine": 0.8050088360589763, + "relative_weight_rmse": 0.015399961397248134 + }, + { + "index": 101, + "total": 252, + "projection": "layers.21.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3798, + "parameters": 50331648, + "reference_delta_l2": 3.2990502232160317, + "retained_delta_l2": 3.0202128027104185, + "delta_retention_norm_ratio": 0.915479486021952, + "delta_cosine": 0.7300971583014276, + "relative_weight_rmse": 0.012655458776477037 + }, + { + "index": 102, + "total": 252, + "projection": "layers.21.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0334, + "parameters": 4194304, + "reference_delta_l2": 0.8730228901885023, + "retained_delta_l2": 0.804018654738106, + "delta_retention_norm_ratio": 0.920959420164233, + "delta_cosine": 0.7372712252656706, + "relative_weight_rmse": 0.0126871581732406 + }, + { + "index": 103, + "total": 252, + "projection": "layers.21.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1297, + "parameters": 16777216, + "reference_delta_l2": 1.7177655640157186, + "retained_delta_l2": 1.566250194997175, + "delta_retention_norm_ratio": 0.9117950829888932, + "delta_cosine": 0.7298742053065208, + "relative_weight_rmse": 0.012289786567341105 + }, + { + "index": 104, + "total": 252, + "projection": "layers.21.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1294, + "parameters": 16777216, + "reference_delta_l2": 1.3281743780431459, + "retained_delta_l2": 1.0929817623753093, + "delta_retention_norm_ratio": 0.8229203788629355, + "delta_cosine": 0.6402893061159396, + "relative_weight_rmse": 0.010224438757938017 + }, + { + "index": 105, + "total": 252, + "projection": "layers.21.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0321, + "parameters": 4194304, + "reference_delta_l2": 1.3232957768297005, + "retained_delta_l2": 1.3034358920124598, + "delta_retention_norm_ratio": 0.9849921044372858, + "delta_cosine": 0.8203647238304811, + "relative_weight_rmse": 0.014473819454316998 + }, + { + "index": 106, + "total": 252, + "projection": "layers.22.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.418, + "parameters": 50331648, + "reference_delta_l2": 3.2908776463167944, + "retained_delta_l2": 3.0443212802975403, + "delta_retention_norm_ratio": 0.9250788414162999, + "delta_cosine": 0.7423013623433464, + "relative_weight_rmse": 0.01244650725153509 + }, + { + "index": 107, + "total": 252, + "projection": "layers.22.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3773, + "parameters": 50331648, + "reference_delta_l2": 4.156579561474297, + "retained_delta_l2": 4.035627263664332, + "delta_retention_norm_ratio": 0.9709010026101691, + "delta_cosine": 0.7994464871552224, + "relative_weight_rmse": 0.014541043360316317 + }, + { + "index": 108, + "total": 252, + "projection": "layers.22.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3777, + "parameters": 50331648, + "reference_delta_l2": 3.3739982936572726, + "retained_delta_l2": 3.1173599715665126, + "delta_retention_norm_ratio": 0.9239364398692168, + "delta_cosine": 0.7418640638148511, + "relative_weight_rmse": 0.012489290827240337 + }, + { + "index": 109, + "total": 252, + "projection": "layers.22.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0331, + "parameters": 4194304, + "reference_delta_l2": 1.111751268571817, + "retained_delta_l2": 1.0850456015892047, + "delta_retention_norm_ratio": 0.9759787393659383, + "delta_cosine": 0.8156693178467094, + "relative_weight_rmse": 0.01430937717218605 + }, + { + "index": 110, + "total": 252, + "projection": "layers.22.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1288, + "parameters": 16777216, + "reference_delta_l2": 2.0783969533374167, + "retained_delta_l2": 1.9751661319280827, + "delta_retention_norm_ratio": 0.9503315181233452, + "delta_cosine": 0.770842062340233, + "relative_weight_rmse": 0.013650318249712263 + }, + { + "index": 111, + "total": 252, + "projection": "layers.22.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1287, + "parameters": 16777216, + "reference_delta_l2": 1.8598561924125512, + "retained_delta_l2": 1.7303239024680177, + "delta_retention_norm_ratio": 0.9303535991261196, + "delta_cosine": 0.7542930687655398, + "relative_weight_rmse": 0.012532727549685888 + }, + { + "index": 112, + "total": 252, + "projection": "layers.22.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0324, + "parameters": 4194304, + "reference_delta_l2": 1.0968905396436142, + "retained_delta_l2": 1.0445944209465525, + "delta_retention_norm_ratio": 0.9523233022740328, + "delta_cosine": 0.7770764680561785, + "relative_weight_rmse": 0.013153597555521881 + }, + { + "index": 113, + "total": 252, + "projection": "layers.23.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.4234, + "parameters": 50331648, + "reference_delta_l2": 3.150983045577694, + "retained_delta_l2": 2.889498471314519, + "delta_retention_norm_ratio": 0.917014921857431, + "delta_cosine": 0.7340642309489184, + "relative_weight_rmse": 0.011957939553488443 + }, + { + "index": 114, + "total": 252, + "projection": "layers.23.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3778, + "parameters": 50331648, + "reference_delta_l2": 5.498566390098913, + "retained_delta_l2": 5.529099999695977, + "delta_retention_norm_ratio": 1.0055530128093106, + "delta_cosine": 0.8489577297620059, + "relative_weight_rmse": 0.016629036114430826 + }, + { + "index": 115, + "total": 252, + "projection": "layers.23.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3791, + "parameters": 50331648, + "reference_delta_l2": 4.892409398048379, + "retained_delta_l2": 4.845441489980659, + "delta_retention_norm_ratio": 0.9903998410095329, + "delta_cosine": 0.8255309676122626, + "relative_weight_rmse": 0.015270962752763314 + }, + { + "index": 116, + "total": 252, + "projection": "layers.23.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0333, + "parameters": 4194304, + "reference_delta_l2": 0.788090022318649, + "retained_delta_l2": 0.6997193193971544, + "delta_retention_norm_ratio": 0.8878672481330266, + "delta_cosine": 0.7037733963313199, + "relative_weight_rmse": 0.011905207296088429 + }, + { + "index": 117, + "total": 252, + "projection": "layers.23.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1293, + "parameters": 16777216, + "reference_delta_l2": 1.957345037571751, + "retained_delta_l2": 1.8304433542378304, + "delta_retention_norm_ratio": 0.9351664214034779, + "delta_cosine": 0.754286297850627, + "relative_weight_rmse": 0.01279709757413271 + }, + { + "index": 118, + "total": 252, + "projection": "layers.23.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1285, + "parameters": 16777216, + "reference_delta_l2": 1.7772523774440465, + "retained_delta_l2": 1.610545615747031, + "delta_retention_norm_ratio": 0.9061997250290561, + "delta_cosine": 0.7240522406626962, + "relative_weight_rmse": 0.011963931737173772 + }, + { + "index": 119, + "total": 252, + "projection": "layers.23.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0316, + "parameters": 4194304, + "reference_delta_l2": 0.9271681995718138, + "retained_delta_l2": 0.8519849881978571, + "delta_retention_norm_ratio": 0.9189109253222036, + "delta_cosine": 0.7381867516067402, + "relative_weight_rmse": 0.011718181948495339 + }, + { + "index": 120, + "total": 252, + "projection": "layers.24.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.4212, + "parameters": 50331648, + "reference_delta_l2": 3.229563782023289, + "retained_delta_l2": 2.9549708221397792, + "delta_retention_norm_ratio": 0.9149752169590284, + "delta_cosine": 0.7301365155712533, + "relative_weight_rmse": 0.012219770988900034 + }, + { + "index": 121, + "total": 252, + "projection": "layers.24.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3772, + "parameters": 50331648, + "reference_delta_l2": 4.959174317167857, + "retained_delta_l2": 4.906939572842355, + "delta_retention_norm_ratio": 0.9894670481445522, + "delta_cosine": 0.8209189413142773, + "relative_weight_rmse": 0.015977585311792136 + }, + { + "index": 122, + "total": 252, + "projection": "layers.24.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3812, + "parameters": 50331648, + "reference_delta_l2": 3.985967023074872, + "retained_delta_l2": 3.80268421919501, + "delta_retention_norm_ratio": 0.9540179828837436, + "delta_cosine": 0.7736464236524154, + "relative_weight_rmse": 0.013918867092406802 + }, + { + "index": 123, + "total": 252, + "projection": "layers.24.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0329, + "parameters": 4194304, + "reference_delta_l2": 0.9425593444012208, + "retained_delta_l2": 0.889527553659845, + "delta_retention_norm_ratio": 0.9437363906512802, + "delta_cosine": 0.7662777796154222, + "relative_weight_rmse": 0.012974176122436301 + }, + { + "index": 124, + "total": 252, + "projection": "layers.24.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1292, + "parameters": 16777216, + "reference_delta_l2": 1.9015276475048024, + "retained_delta_l2": 1.7601973080592979, + "delta_retention_norm_ratio": 0.925675369679237, + "delta_cosine": 0.7439690983948056, + "relative_weight_rmse": 0.01251547548185933 + }, + { + "index": 125, + "total": 252, + "projection": "layers.24.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1287, + "parameters": 16777216, + "reference_delta_l2": 2.9529138341435077, + "retained_delta_l2": 2.943117944407057, + "delta_retention_norm_ratio": 0.9966826361056715, + "delta_cosine": 0.8450159080661319, + "relative_weight_rmse": 0.015374080366834101 + }, + { + "index": 126, + "total": 252, + "projection": "layers.24.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0324, + "parameters": 4194304, + "reference_delta_l2": 1.0812206083030709, + "retained_delta_l2": 1.029671547967984, + "delta_retention_norm_ratio": 0.9523232724762889, + "delta_cosine": 0.7777162975928628, + "relative_weight_rmse": 0.01300701810000996 + }, + { + "index": 127, + "total": 252, + "projection": "layers.25.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.4192, + "parameters": 50331648, + "reference_delta_l2": 3.6146112629687503, + "retained_delta_l2": 3.378615393372034, + "delta_retention_norm_ratio": 0.934710580909636, + "delta_cosine": 0.7501352166654595, + "relative_weight_rmse": 0.013209345333199856 + }, + { + "index": 128, + "total": 252, + "projection": "layers.25.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3821, + "parameters": 50331648, + "reference_delta_l2": 5.414258447636641, + "retained_delta_l2": 5.408162145190015, + "delta_retention_norm_ratio": 0.9988740281784503, + "delta_cosine": 0.8340880034972216, + "relative_weight_rmse": 0.01669364608022774 + }, + { + "index": 129, + "total": 252, + "projection": "layers.25.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3769, + "parameters": 50331648, + "reference_delta_l2": 3.7171822984048513, + "retained_delta_l2": 3.4643499195094227, + "delta_retention_norm_ratio": 0.9319827873376224, + "delta_cosine": 0.7464982147792217, + "relative_weight_rmse": 0.013559557974612967 + }, + { + "index": 130, + "total": 252, + "projection": "layers.25.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0335, + "parameters": 4194304, + "reference_delta_l2": 1.0405920648254727, + "retained_delta_l2": 0.995596902235951, + "delta_retention_norm_ratio": 0.9567600367997536, + "delta_cosine": 0.7832947018990778, + "relative_weight_rmse": 0.013608494459009265 + }, + { + "index": 131, + "total": 252, + "projection": "layers.25.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1301, + "parameters": 16777216, + "reference_delta_l2": 1.939740845014259, + "retained_delta_l2": 1.7901875617894292, + "delta_retention_norm_ratio": 0.9229003793938615, + "delta_cosine": 0.7379099758757841, + "relative_weight_rmse": 0.013137021406277078 + }, + { + "index": 132, + "total": 252, + "projection": "layers.25.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.129, + "parameters": 16777216, + "reference_delta_l2": 2.4699216072651056, + "retained_delta_l2": 2.425107654734864, + "delta_retention_norm_ratio": 0.9818561235310366, + "delta_cosine": 0.8173100852418096, + "relative_weight_rmse": 0.014636873507821186 + }, + { + "index": 133, + "total": 252, + "projection": "layers.25.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0317, + "parameters": 4194304, + "reference_delta_l2": 1.1867217840520754, + "retained_delta_l2": 1.1432698784549291, + "delta_retention_norm_ratio": 0.9633849262893118, + "delta_cosine": 0.7870851672658014, + "relative_weight_rmse": 0.013446708396739022 + }, + { + "index": 134, + "total": 252, + "projection": "layers.26.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.4187, + "parameters": 50331648, + "reference_delta_l2": 3.2602575434231946, + "retained_delta_l2": 2.975224274599929, + "delta_retention_norm_ratio": 0.9125733887501453, + "delta_cosine": 0.7253935078520014, + "relative_weight_rmse": 0.012256360561517467 + }, + { + "index": 135, + "total": 252, + "projection": "layers.26.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3774, + "parameters": 50331648, + "reference_delta_l2": 5.176009025454817, + "retained_delta_l2": 5.1281907444925725, + "delta_retention_norm_ratio": 0.9907615537903662, + "delta_cosine": 0.8190993674331786, + "relative_weight_rmse": 0.01651358465871949 + }, + { + "index": 136, + "total": 252, + "projection": "layers.26.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3767, + "parameters": 50331648, + "reference_delta_l2": 3.5471542618653533, + "retained_delta_l2": 3.2678319175628037, + "delta_retention_norm_ratio": 0.9212545258306128, + "delta_cosine": 0.7346146619754262, + "relative_weight_rmse": 0.013049206896867354 + }, + { + "index": 137, + "total": 252, + "projection": "layers.26.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0329, + "parameters": 4194304, + "reference_delta_l2": 1.0986270978420647, + "retained_delta_l2": 1.0591068147615998, + "delta_retention_norm_ratio": 0.9640275729971608, + "delta_cosine": 0.7902424195946403, + "relative_weight_rmse": 0.014422722081515689 + }, + { + "index": 138, + "total": 252, + "projection": "layers.26.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1295, + "parameters": 16777216, + "reference_delta_l2": 2.126285478410601, + "retained_delta_l2": 1.999536302129749, + "delta_retention_norm_ratio": 0.9403893891164619, + "delta_cosine": 0.758816612319269, + "relative_weight_rmse": 0.013723288225917417 + }, + { + "index": 139, + "total": 252, + "projection": "layers.26.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1294, + "parameters": 16777216, + "reference_delta_l2": 2.703623687644987, + "retained_delta_l2": 2.679250180912468, + "delta_retention_norm_ratio": 0.9909848745430434, + "delta_cosine": 0.8315686412507582, + "relative_weight_rmse": 0.015326307545327001 + }, + { + "index": 140, + "total": 252, + "projection": "layers.26.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0327, + "parameters": 4194304, + "reference_delta_l2": 1.3063660284838896, + "retained_delta_l2": 1.27086842098662, + "delta_retention_norm_ratio": 0.9728272117283495, + "delta_cosine": 0.8070248194093773, + "relative_weight_rmse": 0.014312182274555583 + }, + { + "index": 141, + "total": 252, + "projection": "layers.27.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.4213, + "parameters": 50331648, + "reference_delta_l2": 3.084811817177797, + "retained_delta_l2": 2.752909414750815, + "delta_retention_norm_ratio": 0.8924075690520955, + "delta_cosine": 0.7048422903409844, + "relative_weight_rmse": 0.011812657790110415 + }, + { + "index": 142, + "total": 252, + "projection": "layers.27.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3803, + "parameters": 50331648, + "reference_delta_l2": 5.113703499652078, + "retained_delta_l2": 5.0674462084363086, + "delta_retention_norm_ratio": 0.9909542484778564, + "delta_cosine": 0.8207084554376873, + "relative_weight_rmse": 0.016265134100379552 + }, + { + "index": 143, + "total": 252, + "projection": "layers.27.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3771, + "parameters": 50331648, + "reference_delta_l2": 3.1102131217099194, + "retained_delta_l2": 2.723123315057206, + "delta_retention_norm_ratio": 0.8755423530462437, + "delta_cosine": 0.6873033589978299, + "relative_weight_rmse": 0.012091975347910403 + }, + { + "index": 144, + "total": 252, + "projection": "layers.27.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0333, + "parameters": 4194304, + "reference_delta_l2": 0.9514556845610295, + "retained_delta_l2": 0.8948159007635973, + "delta_retention_norm_ratio": 0.9404703921406871, + "delta_cosine": 0.7659651477371985, + "relative_weight_rmse": 0.013421774348384962 + }, + { + "index": 145, + "total": 252, + "projection": "layers.27.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1298, + "parameters": 16777216, + "reference_delta_l2": 1.8210524056350437, + "retained_delta_l2": 1.6538660611392224, + "delta_retention_norm_ratio": 0.9081924583946723, + "delta_cosine": 0.7220787136833681, + "relative_weight_rmse": 0.012685708987067926 + }, + { + "index": 146, + "total": 252, + "projection": "layers.27.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1296, + "parameters": 16777216, + "reference_delta_l2": 2.480814010093733, + "retained_delta_l2": 2.4240899851797493, + "delta_retention_norm_ratio": 0.9771349143131287, + "delta_cosine": 0.8117471467323384, + "relative_weight_rmse": 0.01480771964964978 + }, + { + "index": 147, + "total": 252, + "projection": "layers.27.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.033, + "parameters": 4194304, + "reference_delta_l2": 1.294105392147078, + "retained_delta_l2": 1.2691599471567527, + "delta_retention_norm_ratio": 0.9807237917856614, + "delta_cosine": 0.811726414007371, + "relative_weight_rmse": 0.01427501257156657 + }, + { + "index": 148, + "total": 252, + "projection": "layers.28.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.4221, + "parameters": 50331648, + "reference_delta_l2": 3.743912161119115, + "retained_delta_l2": 3.527636920014464, + "delta_retention_norm_ratio": 0.9422328217657748, + "delta_cosine": 0.7603507593120301, + "relative_weight_rmse": 0.013052627725093087 + }, + { + "index": 149, + "total": 252, + "projection": "layers.28.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3756, + "parameters": 50331648, + "reference_delta_l2": 6.349154983510624, + "retained_delta_l2": 6.4566844818673355, + "delta_retention_norm_ratio": 1.0169360330053332, + "delta_cosine": 0.8653585744984237, + "relative_weight_rmse": 0.017830159358243242 + }, + { + "index": 150, + "total": 252, + "projection": "layers.28.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.381, + "parameters": 50331648, + "reference_delta_l2": 3.2190661318529354, + "retained_delta_l2": 2.848179102652828, + "delta_retention_norm_ratio": 0.8847842777971697, + "delta_cosine": 0.6959143144906602, + "relative_weight_rmse": 0.012262428248752077 + }, + { + "index": 151, + "total": 252, + "projection": "layers.28.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0333, + "parameters": 4194304, + "reference_delta_l2": 1.0327169221950787, + "retained_delta_l2": 0.9824886382873864, + "delta_retention_norm_ratio": 0.9513629700180275, + "delta_cosine": 0.7715584392145897, + "relative_weight_rmse": 0.014127425543861507 + }, + { + "index": 152, + "total": 252, + "projection": "layers.28.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1305, + "parameters": 16777216, + "reference_delta_l2": 1.8458373458271067, + "retained_delta_l2": 1.691014300772356, + "delta_retention_norm_ratio": 0.9161231376075687, + "delta_cosine": 0.7307393520488142, + "relative_weight_rmse": 0.012743851521728798 + }, + { + "index": 153, + "total": 252, + "projection": "layers.28.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1296, + "parameters": 16777216, + "reference_delta_l2": 2.0909157187798595, + "retained_delta_l2": 1.9887292649809938, + "delta_retention_norm_ratio": 0.9511283726641571, + "delta_cosine": 0.7752259818981031, + "relative_weight_rmse": 0.013192975890512321 + }, + { + "index": 154, + "total": 252, + "projection": "layers.28.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0324, + "parameters": 4194304, + "reference_delta_l2": 1.052380748021918, + "retained_delta_l2": 0.9828669542038823, + "delta_retention_norm_ratio": 0.9339461559433735, + "delta_cosine": 0.7510604571083496, + "relative_weight_rmse": 0.012547507369907656 + }, + { + "index": 155, + "total": 252, + "projection": "layers.29.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.4209, + "parameters": 50331648, + "reference_delta_l2": 3.7108834987108525, + "retained_delta_l2": 3.478094632706491, + "delta_retention_norm_ratio": 0.9372686137721027, + "delta_cosine": 0.75461328207414, + "relative_weight_rmse": 0.012924251128278644 + }, + { + "index": 156, + "total": 252, + "projection": "layers.29.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3763, + "parameters": 50331648, + "reference_delta_l2": 5.932195322790751, + "retained_delta_l2": 5.9949666451675, + "delta_retention_norm_ratio": 1.0105814658758097, + "delta_cosine": 0.85070938987256, + "relative_weight_rmse": 0.017483580029996466 + }, + { + "index": 157, + "total": 252, + "projection": "layers.29.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3803, + "parameters": 50331648, + "reference_delta_l2": 3.54319853671956, + "retained_delta_l2": 3.23744389783926, + "delta_retention_norm_ratio": 0.913706602745614, + "delta_cosine": 0.7267214176385923, + "relative_weight_rmse": 0.012813658284754881 + }, + { + "index": 158, + "total": 252, + "projection": "layers.29.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0335, + "parameters": 4194304, + "reference_delta_l2": 0.961132595911601, + "retained_delta_l2": 0.9109052865551184, + "delta_retention_norm_ratio": 0.9477415399601095, + "delta_cosine": 0.7820198227407719, + "relative_weight_rmse": 0.013111966927430765 + }, + { + "index": 159, + "total": 252, + "projection": "layers.29.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1309, + "parameters": 16777216, + "reference_delta_l2": 1.9316333035249935, + "retained_delta_l2": 1.7774873511356866, + "delta_retention_norm_ratio": 0.9201991640400848, + "delta_cosine": 0.7356934157891413, + "relative_weight_rmse": 0.012873993469682769 + }, + { + "index": 160, + "total": 252, + "projection": "layers.29.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1347, + "parameters": 16777216, + "reference_delta_l2": 2.9198813467073346, + "retained_delta_l2": 2.904689464038373, + "delta_retention_norm_ratio": 0.9947970890371648, + "delta_cosine": 0.8466813608297601, + "relative_weight_rmse": 0.015202681225306708 + }, + { + "index": 161, + "total": 252, + "projection": "layers.29.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0322, + "parameters": 4194304, + "reference_delta_l2": 1.060919515766402, + "retained_delta_l2": 0.9995206890330894, + "delta_retention_norm_ratio": 0.9421267817012883, + "delta_cosine": 0.7653361477996109, + "relative_weight_rmse": 0.01290014368582686 + }, + { + "index": 162, + "total": 252, + "projection": "layers.3.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.4223, + "parameters": 50331648, + "reference_delta_l2": 1.9428858586294158, + "retained_delta_l2": 1.6887811879936636, + "delta_retention_norm_ratio": 0.8692127643489016, + "delta_cosine": 0.6981307961538187, + "relative_weight_rmse": 0.01092470252519349 + }, + { + "index": 163, + "total": 252, + "projection": "layers.3.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3761, + "parameters": 50331648, + "reference_delta_l2": 2.880089840962753, + "retained_delta_l2": 2.6525041698857796, + "delta_retention_norm_ratio": 0.9209796625646595, + "delta_cosine": 0.7416474527699114, + "relative_weight_rmse": 0.012151772578522492 + }, + { + "index": 164, + "total": 252, + "projection": "layers.3.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3776, + "parameters": 50331648, + "reference_delta_l2": 3.074882962002297, + "retained_delta_l2": 2.9993109652543275, + "delta_retention_norm_ratio": 0.9754228054589893, + "delta_cosine": 0.8377552136536656, + "relative_weight_rmse": 0.014377580749460666 + }, + { + "index": 165, + "total": 252, + "projection": "layers.3.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0333, + "parameters": 4194304, + "reference_delta_l2": 0.5752064770349931, + "retained_delta_l2": 0.43929183411803424, + "delta_retention_norm_ratio": 0.7637115569046515, + "delta_cosine": 0.5861665494321837, + "relative_weight_rmse": 0.009408533173434519 + }, + { + "index": 166, + "total": 252, + "projection": "layers.3.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1314, + "parameters": 16777216, + "reference_delta_l2": 1.8108847146075846, + "retained_delta_l2": 1.6524473584934873, + "delta_retention_norm_ratio": 0.9125083143967945, + "delta_cosine": 0.7258433663652634, + "relative_weight_rmse": 0.012625176027044745 + }, + { + "index": 167, + "total": 252, + "projection": "layers.3.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1307, + "parameters": 16777216, + "reference_delta_l2": 1.4765530165504193, + "retained_delta_l2": 1.2697094997913443, + "delta_retention_norm_ratio": 0.8599146021574553, + "delta_cosine": 0.6736389435961885, + "relative_weight_rmse": 0.01154899639293874 + }, + { + "index": 168, + "total": 252, + "projection": "layers.3.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0325, + "parameters": 4194304, + "reference_delta_l2": 0.9302378043324021, + "retained_delta_l2": 0.8350247186136616, + "delta_retention_norm_ratio": 0.8976465100909639, + "delta_cosine": 0.7123176945693385, + "relative_weight_rmse": 0.01222328533761647 + }, + { + "index": 169, + "total": 252, + "projection": "layers.30.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.4222, + "parameters": 50331648, + "reference_delta_l2": 3.8247761023981672, + "retained_delta_l2": 3.6150402985313304, + "delta_retention_norm_ratio": 0.9451639002514864, + "delta_cosine": 0.762606632376754, + "relative_weight_rmse": 0.013018458427999504 + }, + { + "index": 170, + "total": 252, + "projection": "layers.30.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3777, + "parameters": 50331648, + "reference_delta_l2": 5.2954186929117455, + "retained_delta_l2": 5.278639138480981, + "delta_retention_norm_ratio": 0.996831307323588, + "delta_cosine": 0.826463649170075, + "relative_weight_rmse": 0.016790597145143017 + }, + { + "index": 171, + "total": 252, + "projection": "layers.30.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3983, + "parameters": 50331648, + "reference_delta_l2": 3.70801012603116, + "retained_delta_l2": 3.4102694195168053, + "delta_retention_norm_ratio": 0.9197033728618645, + "delta_cosine": 0.7324017772816978, + "relative_weight_rmse": 0.013193976075075341 + }, + { + "index": 172, + "total": 252, + "projection": "layers.30.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0334, + "parameters": 4194304, + "reference_delta_l2": 1.4699289869173329, + "retained_delta_l2": 1.479142737238952, + "delta_retention_norm_ratio": 1.006268160165303, + "delta_cosine": 0.8501008995493529, + "relative_weight_rmse": 0.016458976723748953 + }, + { + "index": 173, + "total": 252, + "projection": "layers.30.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1298, + "parameters": 16777216, + "reference_delta_l2": 1.8588530436150195, + "retained_delta_l2": 1.6816886532995579, + "delta_retention_norm_ratio": 0.9046915564820983, + "delta_cosine": 0.7163547524914659, + "relative_weight_rmse": 0.012817577233499798 + }, + { + "index": 174, + "total": 252, + "projection": "layers.30.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1352, + "parameters": 16777216, + "reference_delta_l2": 2.980989192119819, + "retained_delta_l2": 2.9880321836896293, + "delta_retention_norm_ratio": 1.0023626357279083, + "delta_cosine": 0.8533756815276609, + "relative_weight_rmse": 0.016173299194076656 + }, + { + "index": 175, + "total": 252, + "projection": "layers.30.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.032, + "parameters": 4194304, + "reference_delta_l2": 1.3007350350801326, + "retained_delta_l2": 1.260939906255664, + "delta_retention_norm_ratio": 0.9694056608369769, + "delta_cosine": 0.7999852397428837, + "relative_weight_rmse": 0.013869153200357816 + }, + { + "index": 176, + "total": 252, + "projection": "layers.31.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.4237, + "parameters": 50331648, + "reference_delta_l2": 4.0748849255669635, + "retained_delta_l2": 3.9061340868348093, + "delta_retention_norm_ratio": 0.9585875817809322, + "delta_cosine": 0.7790557141602221, + "relative_weight_rmse": 0.013300556270801038 + }, + { + "index": 177, + "total": 252, + "projection": "layers.31.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3749, + "parameters": 50331648, + "reference_delta_l2": 4.16961603439327, + "retained_delta_l2": 4.015885039591248, + "delta_retention_norm_ratio": 0.9631306591460784, + "delta_cosine": 0.7836748385045768, + "relative_weight_rmse": 0.01466103950662985 + }, + { + "index": 178, + "total": 252, + "projection": "layers.31.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3776, + "parameters": 50331648, + "reference_delta_l2": 3.7239459071583116, + "retained_delta_l2": 3.433468728002582, + "delta_retention_norm_ratio": 0.921997476226128, + "delta_cosine": 0.7354459784846422, + "relative_weight_rmse": 0.01307441817986089 + }, + { + "index": 179, + "total": 252, + "projection": "layers.31.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0344, + "parameters": 4194304, + "reference_delta_l2": 1.1203967261495993, + "retained_delta_l2": 1.0954359473607782, + "delta_retention_norm_ratio": 0.977721481858839, + "delta_cosine": 0.8110008067027378, + "relative_weight_rmse": 0.014361827775112319 + }, + { + "index": 180, + "total": 252, + "projection": "layers.31.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1296, + "parameters": 16777216, + "reference_delta_l2": 1.9204288782097914, + "retained_delta_l2": 1.771155945613586, + "delta_retention_norm_ratio": 0.9222710435726438, + "delta_cosine": 0.7382532877675653, + "relative_weight_rmse": 0.012507161329326328 + }, + { + "index": 181, + "total": 252, + "projection": "layers.31.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1307, + "parameters": 16777216, + "reference_delta_l2": 3.1010568050237532, + "retained_delta_l2": 3.1261135640599997, + "delta_retention_norm_ratio": 1.008080070960214, + "delta_cosine": 0.8599307548208713, + "relative_weight_rmse": 0.016359843360553798 + }, + { + "index": 182, + "total": 252, + "projection": "layers.31.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0316, + "parameters": 4194304, + "reference_delta_l2": 0.9038364683743074, + "retained_delta_l2": 0.7949829672732996, + "delta_retention_norm_ratio": 0.879565048645583, + "delta_cosine": 0.696821647199585, + "relative_weight_rmse": 0.011168201839925737 + }, + { + "index": 183, + "total": 252, + "projection": "layers.32.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.5677, + "parameters": 50331648, + "reference_delta_l2": 3.0296443680611262, + "retained_delta_l2": 2.6745707092561255, + "delta_retention_norm_ratio": 0.882800218220914, + "delta_cosine": 0.6964783971412194, + "relative_weight_rmse": 0.011227996195838055 + }, + { + "index": 184, + "total": 252, + "projection": "layers.32.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.5443, + "parameters": 50331648, + "reference_delta_l2": 5.558472404415834, + "retained_delta_l2": 5.591466773219058, + "delta_retention_norm_ratio": 1.0059358698582388, + "delta_cosine": 0.84238944726041, + "relative_weight_rmse": 0.017246425424279965 + }, + { + "index": 185, + "total": 252, + "projection": "layers.32.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3777, + "parameters": 50331648, + "reference_delta_l2": 4.139198395433476, + "retained_delta_l2": 3.9187356273952725, + "delta_retention_norm_ratio": 0.9467378108086276, + "delta_cosine": 0.7654119071650936, + "relative_weight_rmse": 0.013750680280926506 + }, + { + "index": 186, + "total": 252, + "projection": "layers.32.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0334, + "parameters": 4194304, + "reference_delta_l2": 1.0704219240640445, + "retained_delta_l2": 1.035505217311008, + "delta_retention_norm_ratio": 0.9673804263832068, + "delta_cosine": 0.7977664836220791, + "relative_weight_rmse": 0.01404771022028685 + }, + { + "index": 187, + "total": 252, + "projection": "layers.32.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1331, + "parameters": 16777216, + "reference_delta_l2": 1.9415121030535512, + "retained_delta_l2": 1.7825813857154955, + "delta_retention_norm_ratio": 0.9181407537516277, + "delta_cosine": 0.7318194331930072, + "relative_weight_rmse": 0.012849526829844983 + }, + { + "index": 188, + "total": 252, + "projection": "layers.32.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1283, + "parameters": 16777216, + "reference_delta_l2": 3.4373532809404046, + "retained_delta_l2": 3.4923341324118504, + "delta_retention_norm_ratio": 1.0159951122208783, + "delta_cosine": 0.88074410162474, + "relative_weight_rmse": 0.0168977432320535 + }, + { + "index": 189, + "total": 252, + "projection": "layers.32.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0312, + "parameters": 4194304, + "reference_delta_l2": 1.0930150621581207, + "retained_delta_l2": 1.0219018793504238, + "delta_retention_norm_ratio": 0.9349385152412389, + "delta_cosine": 0.7558854539747729, + "relative_weight_rmse": 0.012513844391968194 + }, + { + "index": 190, + "total": 252, + "projection": "layers.33.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.4235, + "parameters": 50331648, + "reference_delta_l2": 4.5917956304527365, + "retained_delta_l2": 4.495425746165861, + "delta_retention_norm_ratio": 0.9790125928846327, + "delta_cosine": 0.806173058419255, + "relative_weight_rmse": 0.014253919019543627 + }, + { + "index": 191, + "total": 252, + "projection": "layers.33.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3866, + "parameters": 50331648, + "reference_delta_l2": 3.8734768375561504, + "retained_delta_l2": 3.6683491295191026, + "delta_retention_norm_ratio": 0.9470430012519537, + "delta_cosine": 0.7627524637308845, + "relative_weight_rmse": 0.01446062186255901 + }, + { + "index": 192, + "total": 252, + "projection": "layers.33.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3848, + "parameters": 50331648, + "reference_delta_l2": 4.476323264714016, + "retained_delta_l2": 4.329331684120589, + "delta_retention_norm_ratio": 0.967162429542984, + "delta_cosine": 0.7915884545938644, + "relative_weight_rmse": 0.014109572426896169 + }, + { + "index": 193, + "total": 252, + "projection": "layers.33.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0335, + "parameters": 4194304, + "reference_delta_l2": 1.170745586581659, + "retained_delta_l2": 1.1613882041024812, + "delta_retention_norm_ratio": 0.9920073305537719, + "delta_cosine": 0.8275751091485186, + "relative_weight_rmse": 0.01554576623822128 + }, + { + "index": 194, + "total": 252, + "projection": "layers.33.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1295, + "parameters": 16777216, + "reference_delta_l2": 1.9404892006292953, + "retained_delta_l2": 1.7915944157566739, + "delta_retention_norm_ratio": 0.9232694596680386, + "delta_cosine": 0.7399393979635713, + "relative_weight_rmse": 0.01222320696729086 + }, + { + "index": 195, + "total": 252, + "projection": "layers.33.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1295, + "parameters": 16777216, + "reference_delta_l2": 1.6874781519959392, + "retained_delta_l2": 1.5538392960744047, + "delta_retention_norm_ratio": 0.9208055785709184, + "delta_cosine": 0.7416672272586584, + "relative_weight_rmse": 0.011887412938477365 + }, + { + "index": 196, + "total": 252, + "projection": "layers.33.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0322, + "parameters": 4194304, + "reference_delta_l2": 1.2331104951670842, + "retained_delta_l2": 1.1828705554870533, + "delta_retention_norm_ratio": 0.9592575524440546, + "delta_cosine": 0.7857794388694083, + "relative_weight_rmse": 0.012731066532737175 + }, + { + "index": 197, + "total": 252, + "projection": "layers.34.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.4278, + "parameters": 50331648, + "reference_delta_l2": 3.228387208943387, + "retained_delta_l2": 2.9937475369420956, + "delta_retention_norm_ratio": 0.927319848328204, + "delta_cosine": 0.7434345212481395, + "relative_weight_rmse": 0.01178051960673429 + }, + { + "index": 198, + "total": 252, + "projection": "layers.34.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3803, + "parameters": 50331648, + "reference_delta_l2": 3.461909129416847, + "retained_delta_l2": 3.2513629627236718, + "delta_retention_norm_ratio": 0.9391820643401343, + "delta_cosine": 0.7539698159993119, + "relative_weight_rmse": 0.012984334547971913 + }, + { + "index": 199, + "total": 252, + "projection": "layers.34.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.377, + "parameters": 50331648, + "reference_delta_l2": 3.1118365223596998, + "retained_delta_l2": 2.7876437635536395, + "delta_retention_norm_ratio": 0.8958194762235694, + "delta_cosine": 0.7093271369414286, + "relative_weight_rmse": 0.011390108300238953 + }, + { + "index": 200, + "total": 252, + "projection": "layers.34.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0331, + "parameters": 4194304, + "reference_delta_l2": 0.7743710850557672, + "retained_delta_l2": 0.7110120937639662, + "delta_retention_norm_ratio": 0.9181800657145687, + "delta_cosine": 0.7481236546081707, + "relative_weight_rmse": 0.011502373268333867 + }, + { + "index": 201, + "total": 252, + "projection": "layers.34.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.129, + "parameters": 16777216, + "reference_delta_l2": 2.3251026730813797, + "retained_delta_l2": 2.248571538677158, + "delta_retention_norm_ratio": 0.9670848366008725, + "delta_cosine": 0.7888430858698025, + "relative_weight_rmse": 0.013918284105438097 + }, + { + "index": 202, + "total": 252, + "projection": "layers.34.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1289, + "parameters": 16777216, + "reference_delta_l2": 2.744006251442009, + "retained_delta_l2": 2.753636373536942, + "delta_retention_norm_ratio": 1.003509511718449, + "delta_cosine": 0.8564804806539661, + "relative_weight_rmse": 0.015186372012452388 + }, + { + "index": 203, + "total": 252, + "projection": "layers.34.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0315, + "parameters": 4194304, + "reference_delta_l2": 0.9721044615564538, + "retained_delta_l2": 0.8660658809365851, + "delta_retention_norm_ratio": 0.8909185331275115, + "delta_cosine": 0.7128542922271096, + "relative_weight_rmse": 0.0113367528326248 + }, + { + "index": 204, + "total": 252, + "projection": "layers.35.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.4211, + "parameters": 50331648, + "reference_delta_l2": 2.2394233029017703, + "retained_delta_l2": 1.9768490156933187, + "delta_retention_norm_ratio": 0.8827491493599193, + "delta_cosine": 0.7021811984534005, + "relative_weight_rmse": 0.009582939309797481 + }, + { + "index": 205, + "total": 252, + "projection": "layers.35.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3792, + "parameters": 50331648, + "reference_delta_l2": 2.5963392202827382, + "retained_delta_l2": 2.2578957313422663, + "delta_retention_norm_ratio": 0.8696458897602696, + "delta_cosine": 0.6842175282089173, + "relative_weight_rmse": 0.010454534944894235 + }, + { + "index": 206, + "total": 252, + "projection": "layers.35.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3788, + "parameters": 50331648, + "reference_delta_l2": 2.61023725701357, + "retained_delta_l2": 2.2306341307491717, + "delta_retention_norm_ratio": 0.8545714090761578, + "delta_cosine": 0.6694915099507697, + "relative_weight_rmse": 0.010204622654379214 + }, + { + "index": 207, + "total": 252, + "projection": "layers.35.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0329, + "parameters": 4194304, + "reference_delta_l2": 0.8266326407178699, + "retained_delta_l2": 0.7567485422564463, + "delta_retention_norm_ratio": 0.9154593043885438, + "delta_cosine": 0.7331230345727645, + "relative_weight_rmse": 0.012161272052373374 + }, + { + "index": 208, + "total": 252, + "projection": "layers.35.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1311, + "parameters": 16777216, + "reference_delta_l2": 1.9684078434627321, + "retained_delta_l2": 1.864749601288414, + "delta_retention_norm_ratio": 0.9473390423033635, + "delta_cosine": 0.7658705212612438, + "relative_weight_rmse": 0.012959151358305117 + }, + { + "index": 209, + "total": 252, + "projection": "layers.35.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1289, + "parameters": 16777216, + "reference_delta_l2": 1.8861064378325199, + "retained_delta_l2": 1.7932843157964347, + "delta_retention_norm_ratio": 0.9507863818423977, + "delta_cosine": 0.7697950214118542, + "relative_weight_rmse": 0.01295752483478403 + }, + { + "index": 210, + "total": 252, + "projection": "layers.35.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0315, + "parameters": 4194304, + "reference_delta_l2": 0.877203610291361, + "retained_delta_l2": 0.7662730612608324, + "delta_retention_norm_ratio": 0.8735407062521284, + "delta_cosine": 0.690023816174484, + "relative_weight_rmse": 0.011209386574030592 + }, + { + "index": 211, + "total": 252, + "projection": "layers.4.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.4276, + "parameters": 50331648, + "reference_delta_l2": 2.2004278134254696, + "retained_delta_l2": 1.8814457605690653, + "delta_retention_norm_ratio": 0.855036347518333, + "delta_cosine": 0.6702071935839145, + "relative_weight_rmse": 0.010752406576308568 + }, + { + "index": 212, + "total": 252, + "projection": "layers.4.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.379, + "parameters": 50331648, + "reference_delta_l2": 3.2816080412276745, + "retained_delta_l2": 2.889403259527361, + "delta_retention_norm_ratio": 0.8804839649425084, + "delta_cosine": 0.7073277076358653, + "relative_weight_rmse": 0.010749917186029764 + }, + { + "index": 213, + "total": 252, + "projection": "layers.4.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3765, + "parameters": 50331648, + "reference_delta_l2": 3.272960008909287, + "retained_delta_l2": 3.14625422505864, + "delta_retention_norm_ratio": 0.9612870968463585, + "delta_cosine": 0.7936507066963475, + "relative_weight_rmse": 0.013945259967970398 + }, + { + "index": 214, + "total": 252, + "projection": "layers.4.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0335, + "parameters": 4194304, + "reference_delta_l2": 0.57745607591256, + "retained_delta_l2": 0.4314742483627321, + "delta_retention_norm_ratio": 0.7471983867878934, + "delta_cosine": 0.5715363636567732, + "relative_weight_rmse": 0.009260143170868899 + }, + { + "index": 215, + "total": 252, + "projection": "layers.4.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1303, + "parameters": 16777216, + "reference_delta_l2": 1.5996132639589897, + "retained_delta_l2": 1.3816557914721903, + "delta_retention_norm_ratio": 0.8637436451687316, + "delta_cosine": 0.6759927280790385, + "relative_weight_rmse": 0.011663318191791757 + }, + { + "index": 216, + "total": 252, + "projection": "layers.4.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1292, + "parameters": 16777216, + "reference_delta_l2": 1.2310672526950397, + "retained_delta_l2": 0.9626843955624181, + "delta_retention_norm_ratio": 0.7819917177188488, + "delta_cosine": 0.6007705015331178, + "relative_weight_rmse": 0.010059531667273722 + }, + { + "index": 217, + "total": 252, + "projection": "layers.4.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0318, + "parameters": 4194304, + "reference_delta_l2": 1.1325985586505807, + "retained_delta_l2": 1.073863702455374, + "delta_retention_norm_ratio": 0.9481415054375617, + "delta_cosine": 0.7687626649008501, + "relative_weight_rmse": 0.013506144545122181 + }, + { + "index": 218, + "total": 252, + "projection": "layers.5.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.4279, + "parameters": 50331648, + "reference_delta_l2": 1.7832499315727321, + "retained_delta_l2": 1.3211806174506775, + "delta_retention_norm_ratio": 0.7408835935215576, + "delta_cosine": 0.5666654623095203, + "relative_weight_rmse": 0.008957948546127192 + }, + { + "index": 219, + "total": 252, + "projection": "layers.5.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.379, + "parameters": 50331648, + "reference_delta_l2": 2.9165240394310072, + "retained_delta_l2": 2.447778361549577, + "delta_retention_norm_ratio": 0.839279336791313, + "delta_cosine": 0.6566805077163231, + "relative_weight_rmse": 0.009910760846970084 + }, + { + "index": 220, + "total": 252, + "projection": "layers.5.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3767, + "parameters": 50331648, + "reference_delta_l2": 3.1558577050375884, + "retained_delta_l2": 2.940076178767105, + "delta_retention_norm_ratio": 0.9316250774152337, + "delta_cosine": 0.7481897280346822, + "relative_weight_rmse": 0.013065073904239007 + }, + { + "index": 221, + "total": 252, + "projection": "layers.5.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0328, + "parameters": 4194304, + "reference_delta_l2": 0.6035103873018789, + "retained_delta_l2": 0.46929695527635834, + "delta_retention_norm_ratio": 0.7776120596274239, + "delta_cosine": 0.5990772188812813, + "relative_weight_rmse": 0.009672104248142337 + }, + { + "index": 222, + "total": 252, + "projection": "layers.5.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1304, + "parameters": 16777216, + "reference_delta_l2": 1.4473759005984814, + "retained_delta_l2": 1.2112161722034411, + "delta_retention_norm_ratio": 0.8368359399259102, + "delta_cosine": 0.6492662493933895, + "relative_weight_rmse": 0.011223911774261313 + }, + { + "index": 223, + "total": 252, + "projection": "layers.5.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1292, + "parameters": 16777216, + "reference_delta_l2": 1.4280117823422716, + "retained_delta_l2": 1.2059811471537756, + "delta_retention_norm_ratio": 0.8445176447883965, + "delta_cosine": 0.6609766269323848, + "relative_weight_rmse": 0.010996026387990127 + }, + { + "index": 224, + "total": 252, + "projection": "layers.5.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0323, + "parameters": 4194304, + "reference_delta_l2": 1.0241498569620409, + "retained_delta_l2": 0.9577223591464533, + "delta_retention_norm_ratio": 0.9351388887436523, + "delta_cosine": 0.7571151109324145, + "relative_weight_rmse": 0.012685297092594399 + }, + { + "index": 225, + "total": 252, + "projection": "layers.6.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.4228, + "parameters": 50331648, + "reference_delta_l2": 4.727214998902797, + "retained_delta_l2": 4.735376760010324, + "delta_retention_norm_ratio": 1.0017265474723323, + "delta_cosine": 0.8426650516382252, + "relative_weight_rmse": 0.015374115095133636 + }, + { + "index": 226, + "total": 252, + "projection": "layers.6.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3816, + "parameters": 50331648, + "reference_delta_l2": 3.0924891073338148, + "retained_delta_l2": 2.702282053230638, + "delta_retention_norm_ratio": 0.8738210417046243, + "delta_cosine": 0.6914526087818395, + "relative_weight_rmse": 0.010643847579584732 + }, + { + "index": 227, + "total": 252, + "projection": "layers.6.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3783, + "parameters": 50331648, + "reference_delta_l2": 3.3566667646793156, + "retained_delta_l2": 3.16312210050322, + "delta_retention_norm_ratio": 0.9423402208963134, + "delta_cosine": 0.760478893939443, + "relative_weight_rmse": 0.013163935041648131 + }, + { + "index": 228, + "total": 252, + "projection": "layers.6.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0333, + "parameters": 4194304, + "reference_delta_l2": 0.5835276094698191, + "retained_delta_l2": 0.4381900053820814, + "delta_retention_norm_ratio": 0.7509327721103233, + "delta_cosine": 0.5747555670740172, + "relative_weight_rmse": 0.009457423038368874 + }, + { + "index": 229, + "total": 252, + "projection": "layers.6.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1295, + "parameters": 16777216, + "reference_delta_l2": 1.3475179764823484, + "retained_delta_l2": 1.0873416149342006, + "delta_retention_norm_ratio": 0.8069217879917789, + "delta_cosine": 0.6233097959273775, + "relative_weight_rmse": 0.010293604875755653 + }, + { + "index": 230, + "total": 252, + "projection": "layers.6.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1304, + "parameters": 16777216, + "reference_delta_l2": 1.3773303126655463, + "retained_delta_l2": 1.134798425924826, + "delta_retention_norm_ratio": 0.8239116031133095, + "delta_cosine": 0.6399410620320066, + "relative_weight_rmse": 0.01057466814510744 + }, + { + "index": 231, + "total": 252, + "projection": "layers.6.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0321, + "parameters": 4194304, + "reference_delta_l2": 0.983580890066979, + "retained_delta_l2": 0.9067566853669329, + "delta_retention_norm_ratio": 0.9218933536876519, + "delta_cosine": 0.7387668206297414, + "relative_weight_rmse": 0.012379133560948358 + }, + { + "index": 232, + "total": 252, + "projection": "layers.7.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.4219, + "parameters": 50331648, + "reference_delta_l2": 1.817480715727761, + "retained_delta_l2": 1.3370486467922988, + "delta_retention_norm_ratio": 0.7356604310692308, + "delta_cosine": 0.5620666423083053, + "relative_weight_rmse": 0.00884796073744926 + }, + { + "index": 233, + "total": 252, + "projection": "layers.7.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3868, + "parameters": 50331648, + "reference_delta_l2": 2.861486594101161, + "retained_delta_l2": 2.4411476613978973, + "delta_retention_norm_ratio": 0.8531046996446618, + "delta_cosine": 0.6704452644816503, + "relative_weight_rmse": 0.010263860795565866 + }, + { + "index": 234, + "total": 252, + "projection": "layers.7.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3784, + "parameters": 50331648, + "reference_delta_l2": 2.9436459704922333, + "retained_delta_l2": 2.654398734568261, + "delta_retention_norm_ratio": 0.9017384431336338, + "delta_cosine": 0.7134595815933792, + "relative_weight_rmse": 0.012302921171883023 + }, + { + "index": 235, + "total": 252, + "projection": "layers.7.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0342, + "parameters": 4194304, + "reference_delta_l2": 0.7079816241979522, + "retained_delta_l2": 0.5964634958602616, + "delta_retention_norm_ratio": 0.8424844310556426, + "delta_cosine": 0.6566465949901449, + "relative_weight_rmse": 0.010622029377689806 + }, + { + "index": 236, + "total": 252, + "projection": "layers.7.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1298, + "parameters": 16777216, + "reference_delta_l2": 1.3546260487578903, + "retained_delta_l2": 1.094032141014056, + "delta_retention_norm_ratio": 0.8076266819298338, + "delta_cosine": 0.6231247669535989, + "relative_weight_rmse": 0.010482850284672652 + }, + { + "index": 237, + "total": 252, + "projection": "layers.7.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1299, + "parameters": 16777216, + "reference_delta_l2": 1.649083656182199, + "retained_delta_l2": 1.4823165022298672, + "delta_retention_norm_ratio": 0.8988728356338118, + "delta_cosine": 0.7125481353152594, + "relative_weight_rmse": 0.011941834365392923 + }, + { + "index": 238, + "total": 252, + "projection": "layers.7.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0326, + "parameters": 4194304, + "reference_delta_l2": 0.9440331634936212, + "retained_delta_l2": 0.8405868050822033, + "delta_retention_norm_ratio": 0.8904208428138374, + "delta_cosine": 0.7056974033043314, + "relative_weight_rmse": 0.011620875459841685 + }, + { + "index": 239, + "total": 252, + "projection": "layers.8.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.4219, + "parameters": 50331648, + "reference_delta_l2": 2.5820322199583727, + "retained_delta_l2": 2.213317487468125, + "delta_retention_norm_ratio": 0.8571997941620605, + "delta_cosine": 0.6718104953340863, + "relative_weight_rmse": 0.010877725233878834 + }, + { + "index": 240, + "total": 252, + "projection": "layers.8.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3814, + "parameters": 50331648, + "reference_delta_l2": 2.1761535947069937, + "retained_delta_l2": 1.6323376655853048, + "delta_retention_norm_ratio": 0.7501022306309632, + "delta_cosine": 0.5740928494032511, + "relative_weight_rmse": 0.009374506055517337 + }, + { + "index": 241, + "total": 252, + "projection": "layers.8.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3793, + "parameters": 50331648, + "reference_delta_l2": 2.624971162138117, + "retained_delta_l2": 2.221722683781127, + "delta_retention_norm_ratio": 0.846379844406164, + "delta_cosine": 0.6600957236213177, + "relative_weight_rmse": 0.011264043126852539 + }, + { + "index": 242, + "total": 252, + "projection": "layers.8.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0331, + "parameters": 4194304, + "reference_delta_l2": 0.8861856016536126, + "retained_delta_l2": 0.8018405825993075, + "delta_retention_norm_ratio": 0.9048223996226996, + "delta_cosine": 0.7177011898528882, + "relative_weight_rmse": 0.012067736526154414 + }, + { + "index": 243, + "total": 252, + "projection": "layers.8.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1298, + "parameters": 16777216, + "reference_delta_l2": 1.553493041750593, + "retained_delta_l2": 1.325872147900529, + "delta_retention_norm_ratio": 0.8534780087630367, + "delta_cosine": 0.6674120532117493, + "relative_weight_rmse": 0.011209884942483626 + }, + { + "index": 244, + "total": 252, + "projection": "layers.8.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1295, + "parameters": 16777216, + "reference_delta_l2": 1.5787108889697208, + "retained_delta_l2": 1.364887900268177, + "delta_retention_norm_ratio": 0.864558488703979, + "delta_cosine": 0.6764236768245829, + "relative_weight_rmse": 0.011757768726608352 + }, + { + "index": 245, + "total": 252, + "projection": "layers.8.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.033, + "parameters": 4194304, + "reference_delta_l2": 0.7850002799510442, + "retained_delta_l2": 0.6464697716423436, + "delta_retention_norm_ratio": 0.8235280778277687, + "delta_cosine": 0.6388843372816533, + "relative_weight_rmse": 0.0105207034812092 + }, + { + "index": 246, + "total": 252, + "projection": "layers.9.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.5288, + "parameters": 50331648, + "reference_delta_l2": 2.436072625306775, + "retained_delta_l2": 2.0628749806480653, + "delta_retention_norm_ratio": 0.8468035637436249, + "delta_cosine": 0.6615567325568914, + "relative_weight_rmse": 0.0107678500634607 + }, + { + "index": 247, + "total": 252, + "projection": "layers.9.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.5662, + "parameters": 50331648, + "reference_delta_l2": 2.6471349576095133, + "retained_delta_l2": 2.1518202020815624, + "delta_retention_norm_ratio": 0.8128864740711056, + "delta_cosine": 0.6297265823608571, + "relative_weight_rmse": 0.01022391549444434 + }, + { + "index": 248, + "total": 252, + "projection": "layers.9.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3982, + "parameters": 50331648, + "reference_delta_l2": 3.013112514856541, + "retained_delta_l2": 2.7305695653409066, + "delta_retention_norm_ratio": 0.9062288752502536, + "delta_cosine": 0.7212432739499626, + "relative_weight_rmse": 0.012346394404510654 + }, + { + "index": 249, + "total": 252, + "projection": "layers.9.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0331, + "parameters": 4194304, + "reference_delta_l2": 0.8361129146060564, + "retained_delta_l2": 0.7539174469170958, + "delta_retention_norm_ratio": 0.9016933403932795, + "delta_cosine": 0.7175202823222869, + "relative_weight_rmse": 0.01176500119841116 + }, + { + "index": 250, + "total": 252, + "projection": "layers.9.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1283, + "parameters": 16777216, + "reference_delta_l2": 1.5131908112794805, + "retained_delta_l2": 1.2867376307715035, + "delta_retention_norm_ratio": 0.85034724053307, + "delta_cosine": 0.6649543031763168, + "relative_weight_rmse": 0.01109029327227482 + }, + { + "index": 251, + "total": 252, + "projection": "layers.9.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1277, + "parameters": 16777216, + "reference_delta_l2": 1.7630325000070284, + "retained_delta_l2": 1.609591996624781, + "delta_retention_norm_ratio": 0.9129678531838548, + "delta_cosine": 0.7288171812420224, + "relative_weight_rmse": 0.012522261943971564 + }, + { + "index": 252, + "total": 252, + "projection": "layers.9.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0311, + "parameters": 4194304, + "reference_delta_l2": 0.7589389731359304, + "retained_delta_l2": 0.6223879631922467, + "delta_retention_norm_ratio": 0.8200764293610382, + "delta_cosine": 0.6379893630410488, + "relative_weight_rmse": 0.010035020892086629 + } + ] +}