diff --git "a/manifests/merge_step_00003010.json" "b/manifests/merge_step_00003010.json" new file mode 100644--- /dev/null +++ "b/manifests/merge_step_00003010.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_00003010/adapters.safetensors", + "adapter_sha256": "605e25f84b688ad82da52f79c11c042499db66cc68aa5ecc8f17bc4798546d45", + "adapter_global_step": "3010", + "output": "/workspace/Ideogram/artifacts/huggingface_release/s020_v1/qwen3vl_8b_ideogram4_nl_s020_v1_step_00003010_fp8_scaled.safetensors", + "output_bytes": 10588637800, + "output_sha256": "50356a534962cfc198d511be180558fe6d6f382140eb88692fd2dc29d5cff712", + "output_tensors": 1254, + "merged_projections": 252, + "lora_rank": 64, + "lora_alpha": 64.0, + "elapsed_seconds": 68.053, + "projections": [ + { + "index": 1, + "total": 252, + "projection": "layers.0.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.4371, + "parameters": 50331648, + "reference_delta_l2": 3.128466941417038, + "retained_delta_l2": 2.848897467132658, + "delta_retention_norm_ratio": 0.910636909540828, + "delta_cosine": 0.7259526523730795, + "relative_weight_rmse": 0.01231216238907918 + }, + { + "index": 2, + "total": 252, + "projection": "layers.0.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.4009, + "parameters": 50331648, + "reference_delta_l2": 3.3237639995881594, + "retained_delta_l2": 3.039333663337872, + "delta_retention_norm_ratio": 0.9144252310676899, + "delta_cosine": 0.7270897873544261, + "relative_weight_rmse": 0.013531932524432943 + }, + { + "index": 3, + "total": 252, + "projection": "layers.0.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3866, + "parameters": 50331648, + "reference_delta_l2": 2.946776224071399, + "retained_delta_l2": 2.6376769678691434, + "delta_retention_norm_ratio": 0.8951059623471544, + "delta_cosine": 0.7081998082627273, + "relative_weight_rmse": 0.013057582385661515 + }, + { + "index": 4, + "total": 252, + "projection": "layers.0.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0318, + "parameters": 4194304, + "reference_delta_l2": 0.603281485378585, + "retained_delta_l2": 0.4422823883738678, + "delta_retention_norm_ratio": 0.7331277340565436, + "delta_cosine": 0.5608919015198992, + "relative_weight_rmse": 0.009018581547077189 + }, + { + "index": 5, + "total": 252, + "projection": "layers.0.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1322, + "parameters": 16777216, + "reference_delta_l2": 1.9742149145809642, + "retained_delta_l2": 1.862422410622986, + "delta_retention_norm_ratio": 0.9433736909126196, + "delta_cosine": 0.7629146130818641, + "relative_weight_rmse": 0.013627355259654514 + }, + { + "index": 6, + "total": 252, + "projection": "layers.0.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1317, + "parameters": 16777216, + "reference_delta_l2": 1.2260947851948691, + "retained_delta_l2": 0.9542804876376539, + "delta_retention_norm_ratio": 0.7783089033251093, + "delta_cosine": 0.5989796145099707, + "relative_weight_rmse": 0.009810233253454835 + }, + { + "index": 7, + "total": 252, + "projection": "layers.0.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0307, + "parameters": 4194304, + "reference_delta_l2": 0.7955917157456153, + "retained_delta_l2": 0.6931053107236256, + "delta_retention_norm_ratio": 0.871182161661976, + "delta_cosine": 0.683996446494238, + "relative_weight_rmse": 0.01198669618547365 + }, + { + "index": 8, + "total": 252, + "projection": "layers.1.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.4334, + "parameters": 50331648, + "reference_delta_l2": 2.0371442740508208, + "retained_delta_l2": 1.9015960554875635, + "delta_retention_norm_ratio": 0.933461650070703, + "delta_cosine": 0.7565789134419134, + "relative_weight_rmse": 0.011987375923693484 + }, + { + "index": 9, + "total": 252, + "projection": "layers.1.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3866, + "parameters": 50331648, + "reference_delta_l2": 3.82866503351673, + "retained_delta_l2": 3.7957003779944833, + "delta_retention_norm_ratio": 0.9913900392868875, + "delta_cosine": 0.8456850538625545, + "relative_weight_rmse": 0.01435901310519096 + }, + { + "index": 10, + "total": 252, + "projection": "layers.1.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3856, + "parameters": 50331648, + "reference_delta_l2": 4.2347593166853645, + "retained_delta_l2": 4.23190542477896, + "delta_retention_norm_ratio": 0.9993260793133721, + "delta_cosine": 0.9296379303488191, + "relative_weight_rmse": 0.01576399248740128 + }, + { + "index": 11, + "total": 252, + "projection": "layers.1.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0311, + "parameters": 4194304, + "reference_delta_l2": 0.6767510634049011, + "retained_delta_l2": 0.5557797449990938, + "delta_retention_norm_ratio": 0.8212469474415439, + "delta_cosine": 0.6374782723546867, + "relative_weight_rmse": 0.01082281126423862 + }, + { + "index": 12, + "total": 252, + "projection": "layers.1.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1317, + "parameters": 16777216, + "reference_delta_l2": 1.713160047736516, + "retained_delta_l2": 1.532452915823821, + "delta_retention_norm_ratio": 0.8945182429677534, + "delta_cosine": 0.7051924813905717, + "relative_weight_rmse": 0.012659249725434132 + }, + { + "index": 13, + "total": 252, + "projection": "layers.1.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1319, + "parameters": 16777216, + "reference_delta_l2": 2.093133876528023, + "retained_delta_l2": 2.0238628178756524, + "delta_retention_norm_ratio": 0.966905576643156, + "delta_cosine": 0.7927284941978027, + "relative_weight_rmse": 0.01463924781793106 + }, + { + "index": 14, + "total": 252, + "projection": "layers.1.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0305, + "parameters": 4194304, + "reference_delta_l2": 0.9120458171238204, + "retained_delta_l2": 0.82884292893393, + "delta_retention_norm_ratio": 0.9087733460011092, + "delta_cosine": 0.7214643241759308, + "relative_weight_rmse": 0.012412040776157724 + }, + { + "index": 15, + "total": 252, + "projection": "layers.10.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.4305, + "parameters": 50331648, + "reference_delta_l2": 2.8135854474463997, + "retained_delta_l2": 2.4950584659798065, + "delta_retention_norm_ratio": 0.8867896541917052, + "delta_cosine": 0.7007580006466992, + "relative_weight_rmse": 0.011668626055166673 + }, + { + "index": 16, + "total": 252, + "projection": "layers.10.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3901, + "parameters": 50331648, + "reference_delta_l2": 2.392425091377587, + "retained_delta_l2": 1.8551706595412834, + "delta_retention_norm_ratio": 0.7754352126749574, + "delta_cosine": 0.596210563148029, + "relative_weight_rmse": 0.009659380812700635 + }, + { + "index": 17, + "total": 252, + "projection": "layers.10.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3883, + "parameters": 50331648, + "reference_delta_l2": 2.85620794767155, + "retained_delta_l2": 2.5276471077059974, + "delta_retention_norm_ratio": 0.8849660648016182, + "delta_cosine": 0.6974549874261605, + "relative_weight_rmse": 0.011908535945441735 + }, + { + "index": 18, + "total": 252, + "projection": "layers.10.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0337, + "parameters": 4194304, + "reference_delta_l2": 0.8985968351124203, + "retained_delta_l2": 0.8259778009118853, + "delta_retention_norm_ratio": 0.9191861896648569, + "delta_cosine": 0.7372781093798628, + "relative_weight_rmse": 0.011975392766938939 + }, + { + "index": 19, + "total": 252, + "projection": "layers.10.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1319, + "parameters": 16777216, + "reference_delta_l2": 1.775994068550649, + "retained_delta_l2": 1.5875193138047252, + "delta_retention_norm_ratio": 0.8938764728534628, + "delta_cosine": 0.7070655691104736, + "relative_weight_rmse": 0.012176717499931275 + }, + { + "index": 20, + "total": 252, + "projection": "layers.10.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1315, + "parameters": 16777216, + "reference_delta_l2": 2.120636571366303, + "retained_delta_l2": 2.0230820004805934, + "delta_retention_norm_ratio": 0.9539975061248444, + "delta_cosine": 0.7800795162676436, + "relative_weight_rmse": 0.013458991414516524 + }, + { + "index": 21, + "total": 252, + "projection": "layers.10.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0304, + "parameters": 4194304, + "reference_delta_l2": 0.9102292345903797, + "retained_delta_l2": 0.8130730209057979, + "delta_retention_norm_ratio": 0.8932618180207055, + "delta_cosine": 0.7113670197372001, + "relative_weight_rmse": 0.011207517681123344 + }, + { + "index": 22, + "total": 252, + "projection": "layers.11.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.4353, + "parameters": 50331648, + "reference_delta_l2": 2.61087131386716, + "retained_delta_l2": 2.235226936535596, + "delta_retention_norm_ratio": 0.856122982647058, + "delta_cosine": 0.6707405583461962, + "relative_weight_rmse": 0.011054031140988114 + }, + { + "index": 23, + "total": 252, + "projection": "layers.11.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3887, + "parameters": 50331648, + "reference_delta_l2": 2.7770980086111607, + "retained_delta_l2": 2.3451198043184065, + "delta_retention_norm_ratio": 0.8444497806871467, + "delta_cosine": 0.6598443204898431, + "relative_weight_rmse": 0.010821698560628682 + }, + { + "index": 24, + "total": 252, + "projection": "layers.11.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3858, + "parameters": 50331648, + "reference_delta_l2": 2.676426251493652, + "retained_delta_l2": 2.2748360712108977, + "delta_retention_norm_ratio": 0.849952831669232, + "delta_cosine": 0.6635457874047436, + "relative_weight_rmse": 0.011378674912726343 + }, + { + "index": 25, + "total": 252, + "projection": "layers.11.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0315, + "parameters": 4194304, + "reference_delta_l2": 1.1310468572541754, + "retained_delta_l2": 1.0884632199490039, + "delta_retention_norm_ratio": 0.9623502447913156, + "delta_cosine": 0.7874903278412039, + "relative_weight_rmse": 0.013676520795632042 + }, + { + "index": 26, + "total": 252, + "projection": "layers.11.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1298, + "parameters": 16777216, + "reference_delta_l2": 1.6709143595474434, + "retained_delta_l2": 1.4674897394989572, + "delta_retention_norm_ratio": 0.8782555078983327, + "delta_cosine": 0.6931695443430704, + "relative_weight_rmse": 0.011730766726139924 + }, + { + "index": 27, + "total": 252, + "projection": "layers.11.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1301, + "parameters": 16777216, + "reference_delta_l2": 1.4093751749869647, + "retained_delta_l2": 1.1728600783295207, + "delta_retention_norm_ratio": 0.8321844311897777, + "delta_cosine": 0.6469568877272976, + "relative_weight_rmse": 0.010767714798950374 + }, + { + "index": 28, + "total": 252, + "projection": "layers.11.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0303, + "parameters": 4194304, + "reference_delta_l2": 0.9980156028877636, + "retained_delta_l2": 0.9078748176489159, + "delta_retention_norm_ratio": 0.9096799839821894, + "delta_cosine": 0.7270654736594205, + "relative_weight_rmse": 0.011889830775574251 + }, + { + "index": 29, + "total": 252, + "projection": "layers.12.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.4286, + "parameters": 50331648, + "reference_delta_l2": 2.5691174520080495, + "retained_delta_l2": 2.1896290411009085, + "delta_retention_norm_ratio": 0.8522884149922656, + "delta_cosine": 0.6668695651142327, + "relative_weight_rmse": 0.010839231909001119 + }, + { + "index": 30, + "total": 252, + "projection": "layers.12.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3848, + "parameters": 50331648, + "reference_delta_l2": 3.2195405090351974, + "retained_delta_l2": 2.888105136566055, + "delta_retention_norm_ratio": 0.8970550699582703, + "delta_cosine": 0.7131680115083394, + "relative_weight_rmse": 0.011996902378243978 + }, + { + "index": 31, + "total": 252, + "projection": "layers.12.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3835, + "parameters": 50331648, + "reference_delta_l2": 2.6897602984349396, + "retained_delta_l2": 2.2935535675213874, + "delta_retention_norm_ratio": 0.8526981266159336, + "delta_cosine": 0.6664514880370325, + "relative_weight_rmse": 0.011229094938473369 + }, + { + "index": 32, + "total": 252, + "projection": "layers.12.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0312, + "parameters": 4194304, + "reference_delta_l2": 0.9324732987397525, + "retained_delta_l2": 0.8657481171654666, + "delta_retention_norm_ratio": 0.9284427965235404, + "delta_cosine": 0.7477601539008357, + "relative_weight_rmse": 0.01229117627217462 + }, + { + "index": 33, + "total": 252, + "projection": "layers.12.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1302, + "parameters": 16777216, + "reference_delta_l2": 1.7739567241821477, + "retained_delta_l2": 1.606990686365855, + "delta_retention_norm_ratio": 0.9058793061069347, + "delta_cosine": 0.7219112197717785, + "relative_weight_rmse": 0.01203431373629876 + }, + { + "index": 34, + "total": 252, + "projection": "layers.12.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1319, + "parameters": 16777216, + "reference_delta_l2": 1.8071684201072193, + "retained_delta_l2": 1.6524540974370052, + "delta_retention_norm_ratio": 0.9143885423467975, + "delta_cosine": 0.7313398513460301, + "relative_weight_rmse": 0.012437720014468174 + }, + { + "index": 35, + "total": 252, + "projection": "layers.12.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0301, + "parameters": 4194304, + "reference_delta_l2": 1.0873764073683962, + "retained_delta_l2": 1.0243465383712873, + "delta_retention_norm_ratio": 0.9420349121334626, + "delta_cosine": 0.7660227170466977, + "relative_weight_rmse": 0.012194798441283202 + }, + { + "index": 36, + "total": 252, + "projection": "layers.13.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.4303, + "parameters": 50331648, + "reference_delta_l2": 2.3465366647937898, + "retained_delta_l2": 1.9221563954862428, + "delta_retention_norm_ratio": 0.8191461161997906, + "delta_cosine": 0.6359440133091215, + "relative_weight_rmse": 0.010197072553166034 + }, + { + "index": 37, + "total": 252, + "projection": "layers.13.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3835, + "parameters": 50331648, + "reference_delta_l2": 2.996415746873252, + "retained_delta_l2": 2.623109979553534, + "delta_retention_norm_ratio": 0.8754158972401408, + "delta_cosine": 0.6882491006998085, + "relative_weight_rmse": 0.011904435614526717 + }, + { + "index": 38, + "total": 252, + "projection": "layers.13.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.4482, + "parameters": 50331648, + "reference_delta_l2": 2.4028818556282885, + "retained_delta_l2": 1.9430271615235961, + "delta_retention_norm_ratio": 0.8086236770120132, + "delta_cosine": 0.6252200503360601, + "relative_weight_rmse": 0.010353410266772944 + }, + { + "index": 39, + "total": 252, + "projection": "layers.13.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.046, + "parameters": 4194304, + "reference_delta_l2": 1.3420515977679963, + "retained_delta_l2": 1.3277750701448803, + "delta_retention_norm_ratio": 0.9893621618968602, + "delta_cosine": 0.8345681503853014, + "relative_weight_rmse": 0.014781641479791092 + }, + { + "index": 40, + "total": 252, + "projection": "layers.13.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.194, + "parameters": 16777216, + "reference_delta_l2": 1.7312945922615102, + "retained_delta_l2": 1.5518425220460732, + "delta_retention_norm_ratio": 0.8963480443954792, + "delta_cosine": 0.7120905144858279, + "relative_weight_rmse": 0.012268268351629964 + }, + { + "index": 41, + "total": 252, + "projection": "layers.13.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1931, + "parameters": 16777216, + "reference_delta_l2": 1.4108104499042138, + "retained_delta_l2": 1.1748659737731313, + "delta_retention_norm_ratio": 0.8327596197298498, + "delta_cosine": 0.6472937916138816, + "relative_weight_rmse": 0.010756449097432487 + }, + { + "index": 42, + "total": 252, + "projection": "layers.13.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0453, + "parameters": 4194304, + "reference_delta_l2": 1.0303501773529953, + "retained_delta_l2": 0.9494857449227051, + "delta_retention_norm_ratio": 0.9215175246166951, + "delta_cosine": 0.7388953066744807, + "relative_weight_rmse": 0.012333213978898123 + }, + { + "index": 43, + "total": 252, + "projection": "layers.14.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.5484, + "parameters": 50331648, + "reference_delta_l2": 2.4702147919164936, + "retained_delta_l2": 2.0895445010819698, + "delta_retention_norm_ratio": 0.8458958742858209, + "delta_cosine": 0.6615558004736427, + "relative_weight_rmse": 0.010526226390915274 + }, + { + "index": 44, + "total": 252, + "projection": "layers.14.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.4345, + "parameters": 50331648, + "reference_delta_l2": 2.738402489233057, + "retained_delta_l2": 2.324150448981756, + "delta_retention_norm_ratio": 0.8487249256162777, + "delta_cosine": 0.6620074666458025, + "relative_weight_rmse": 0.011375445009352022 + }, + { + "index": 45, + "total": 252, + "projection": "layers.14.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.4185, + "parameters": 50331648, + "reference_delta_l2": 2.947150022299582, + "retained_delta_l2": 2.613153392731818, + "delta_retention_norm_ratio": 0.8866713173606426, + "delta_cosine": 0.7010917602200739, + "relative_weight_rmse": 0.01179029964880898 + }, + { + "index": 46, + "total": 252, + "projection": "layers.14.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0311, + "parameters": 4194304, + "reference_delta_l2": 0.729580415806346, + "retained_delta_l2": 0.6196952097535995, + "delta_retention_norm_ratio": 0.8493857514921102, + "delta_cosine": 0.6663638769163613, + "relative_weight_rmse": 0.01072642971743858 + }, + { + "index": 47, + "total": 252, + "projection": "layers.14.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1312, + "parameters": 16777216, + "reference_delta_l2": 1.5569801596646544, + "retained_delta_l2": 1.3556244185536042, + "delta_retention_norm_ratio": 0.8706754611732375, + "delta_cosine": 0.6850034695808919, + "relative_weight_rmse": 0.011449360756224521 + }, + { + "index": 48, + "total": 252, + "projection": "layers.14.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1308, + "parameters": 16777216, + "reference_delta_l2": 2.3203337195703373, + "retained_delta_l2": 2.261856605403634, + "delta_retention_norm_ratio": 0.9747979725185687, + "delta_cosine": 0.8056448974784804, + "relative_weight_rmse": 0.01392129885625326 + }, + { + "index": 49, + "total": 252, + "projection": "layers.14.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0309, + "parameters": 4194304, + "reference_delta_l2": 1.0172823632078256, + "retained_delta_l2": 0.944275672582837, + "delta_retention_norm_ratio": 0.9282336023257353, + "delta_cosine": 0.745753392239752, + "relative_weight_rmse": 0.012381717517466653 + }, + { + "index": 50, + "total": 252, + "projection": "layers.15.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.4256, + "parameters": 50331648, + "reference_delta_l2": 2.3730436734714417, + "retained_delta_l2": 1.9748352208552025, + "delta_retention_norm_ratio": 0.8321950594218462, + "delta_cosine": 0.6482920937785805, + "relative_weight_rmse": 0.010306385807353713 + }, + { + "index": 51, + "total": 252, + "projection": "layers.15.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3841, + "parameters": 50331648, + "reference_delta_l2": 2.882614652156324, + "retained_delta_l2": 2.5237526119199485, + "delta_retention_norm_ratio": 0.8755081467556093, + "delta_cosine": 0.6874013944661279, + "relative_weight_rmse": 0.01204510318209208 + }, + { + "index": 52, + "total": 252, + "projection": "layers.15.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3832, + "parameters": 50331648, + "reference_delta_l2": 2.7889548017648815, + "retained_delta_l2": 2.440223818349846, + "delta_retention_norm_ratio": 0.8749599731073611, + "delta_cosine": 0.6893239613664839, + "relative_weight_rmse": 0.011425990895298851 + }, + { + "index": 53, + "total": 252, + "projection": "layers.15.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0314, + "parameters": 4194304, + "reference_delta_l2": 0.8147907693002489, + "retained_delta_l2": 0.7206926790805357, + "delta_retention_norm_ratio": 0.8845125721042156, + "delta_cosine": 0.6992954284541983, + "relative_weight_rmse": 0.011596366956845006 + }, + { + "index": 54, + "total": 252, + "projection": "layers.15.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1312, + "parameters": 16777216, + "reference_delta_l2": 2.150069517275001, + "retained_delta_l2": 2.06624494894049, + "delta_retention_norm_ratio": 0.9610130892694343, + "delta_cosine": 0.7886049487920187, + "relative_weight_rmse": 0.013838330515567637 + }, + { + "index": 55, + "total": 252, + "projection": "layers.15.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1315, + "parameters": 16777216, + "reference_delta_l2": 2.236000406522123, + "retained_delta_l2": 2.160702291548502, + "delta_retention_norm_ratio": 0.9663246416440775, + "delta_cosine": 0.8009809560543107, + "relative_weight_rmse": 0.013447574784155885 + }, + { + "index": 56, + "total": 252, + "projection": "layers.15.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0306, + "parameters": 4194304, + "reference_delta_l2": 0.8583436021206987, + "retained_delta_l2": 0.7631955303973562, + "delta_retention_norm_ratio": 0.889149203782423, + "delta_cosine": 0.7054127844359597, + "relative_weight_rmse": 0.011227408884923578 + }, + { + "index": 57, + "total": 252, + "projection": "layers.16.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.4261, + "parameters": 50331648, + "reference_delta_l2": 2.4930011335012625, + "retained_delta_l2": 2.130319548975539, + "delta_retention_norm_ratio": 0.8545200883978901, + "delta_cosine": 0.6692683915172937, + "relative_weight_rmse": 0.010671389443382733 + }, + { + "index": 58, + "total": 252, + "projection": "layers.16.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3926, + "parameters": 50331648, + "reference_delta_l2": 2.642613120703143, + "retained_delta_l2": 2.2601582587880835, + "delta_retention_norm_ratio": 0.8552739866010746, + "delta_cosine": 0.6687272360032434, + "relative_weight_rmse": 0.01107984800109428 + }, + { + "index": 59, + "total": 252, + "projection": "layers.16.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3896, + "parameters": 50331648, + "reference_delta_l2": 2.557740659389756, + "retained_delta_l2": 2.156321054894793, + "delta_retention_norm_ratio": 0.8430569561377124, + "delta_cosine": 0.657184811279028, + "relative_weight_rmse": 0.01089106400982966 + }, + { + "index": 60, + "total": 252, + "projection": "layers.16.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0317, + "parameters": 4194304, + "reference_delta_l2": 0.7412295808504257, + "retained_delta_l2": 0.6387561783109527, + "delta_retention_norm_ratio": 0.8617521410547275, + "delta_cosine": 0.6757402605193612, + "relative_weight_rmse": 0.011226544678444318 + }, + { + "index": 61, + "total": 252, + "projection": "layers.16.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1326, + "parameters": 16777216, + "reference_delta_l2": 1.6670130916188086, + "retained_delta_l2": 1.4909452777293772, + "delta_retention_norm_ratio": 0.8943812650454623, + "delta_cosine": 0.7094261968986654, + "relative_weight_rmse": 0.0119414345371388 + }, + { + "index": 62, + "total": 252, + "projection": "layers.16.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1326, + "parameters": 16777216, + "reference_delta_l2": 1.8115666246471622, + "retained_delta_l2": 1.6798626235378304, + "delta_retention_norm_ratio": 0.927298284635276, + "delta_cosine": 0.7481804236182887, + "relative_weight_rmse": 0.012469574469276157 + }, + { + "index": 63, + "total": 252, + "projection": "layers.16.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0311, + "parameters": 4194304, + "reference_delta_l2": 0.9586279408077624, + "retained_delta_l2": 0.875329981037118, + "delta_retention_norm_ratio": 0.9131071021146583, + "delta_cosine": 0.7300955822635882, + "relative_weight_rmse": 0.0121483551055893 + }, + { + "index": 64, + "total": 252, + "projection": "layers.17.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.4276, + "parameters": 50331648, + "reference_delta_l2": 2.5826198380274024, + "retained_delta_l2": 2.2223135511785674, + "delta_retention_norm_ratio": 0.860488066596733, + "delta_cosine": 0.6749930952033166, + "relative_weight_rmse": 0.010934149299153798 + }, + { + "index": 65, + "total": 252, + "projection": "layers.17.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3871, + "parameters": 50331648, + "reference_delta_l2": 3.5468698103420495, + "retained_delta_l2": 3.327306695527173, + "delta_retention_norm_ratio": 0.9380966523849651, + "delta_cosine": 0.756978564183495, + "relative_weight_rmse": 0.01331527702457968 + }, + { + "index": 66, + "total": 252, + "projection": "layers.17.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3884, + "parameters": 50331648, + "reference_delta_l2": 3.186673756109155, + "retained_delta_l2": 2.902766191608826, + "delta_retention_norm_ratio": 0.9109078662489213, + "delta_cosine": 0.7255771160197569, + "relative_weight_rmse": 0.0123835749219045 + }, + { + "index": 67, + "total": 252, + "projection": "layers.17.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0315, + "parameters": 4194304, + "reference_delta_l2": 0.7883189976740391, + "retained_delta_l2": 0.7058264248553517, + "delta_retention_norm_ratio": 0.8953563556604821, + "delta_cosine": 0.7151037762037882, + "relative_weight_rmse": 0.011567089376913942 + }, + { + "index": 68, + "total": 252, + "projection": "layers.17.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1321, + "parameters": 16777216, + "reference_delta_l2": 1.7299423367298572, + "retained_delta_l2": 1.5699691628692967, + "delta_retention_norm_ratio": 0.9075268750500893, + "delta_cosine": 0.7247709848611007, + "relative_weight_rmse": 0.012200250860506863 + }, + { + "index": 69, + "total": 252, + "projection": "layers.17.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1336, + "parameters": 16777216, + "reference_delta_l2": 1.6616101263889567, + "retained_delta_l2": 1.4868413490993115, + "delta_retention_norm_ratio": 0.8948196243426513, + "delta_cosine": 0.7078313134028258, + "relative_weight_rmse": 0.011973064641087678 + }, + { + "index": 70, + "total": 252, + "projection": "layers.17.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0311, + "parameters": 4194304, + "reference_delta_l2": 1.0334048039951524, + "retained_delta_l2": 0.9689434022159047, + "delta_retention_norm_ratio": 0.9376223126406618, + "delta_cosine": 0.7589304834608148, + "relative_weight_rmse": 0.012761122103350817 + }, + { + "index": 71, + "total": 252, + "projection": "layers.18.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.4274, + "parameters": 50331648, + "reference_delta_l2": 2.9133508565108577, + "retained_delta_l2": 2.6252136555433325, + "delta_retention_norm_ratio": 0.9010976654859176, + "delta_cosine": 0.7156324671177533, + "relative_weight_rmse": 0.011722550865338218 + }, + { + "index": 72, + "total": 252, + "projection": "layers.18.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3863, + "parameters": 50331648, + "reference_delta_l2": 4.742728265476155, + "retained_delta_l2": 4.706476295467704, + "delta_retention_norm_ratio": 0.992356304645927, + "delta_cosine": 0.8278143289397117, + "relative_weight_rmse": 0.015456396205276569 + }, + { + "index": 73, + "total": 252, + "projection": "layers.18.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3847, + "parameters": 50331648, + "reference_delta_l2": 3.269382926329566, + "retained_delta_l2": 3.008309915491614, + "delta_retention_norm_ratio": 0.9201460897298284, + "delta_cosine": 0.7367722041590823, + "relative_weight_rmse": 0.01249286070997407 + }, + { + "index": 74, + "total": 252, + "projection": "layers.18.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0314, + "parameters": 4194304, + "reference_delta_l2": 1.1182723670823707, + "retained_delta_l2": 1.0849893815174072, + "delta_retention_norm_ratio": 0.9702371385141166, + "delta_cosine": 0.7991848885989181, + "relative_weight_rmse": 0.014000603857513817 + }, + { + "index": 75, + "total": 252, + "projection": "layers.18.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1339, + "parameters": 16777216, + "reference_delta_l2": 1.628716792550775, + "retained_delta_l2": 1.455750461977996, + "delta_retention_norm_ratio": 0.8938020831099236, + "delta_cosine": 0.7107906236695126, + "relative_weight_rmse": 0.011697246997117433 + }, + { + "index": 76, + "total": 252, + "projection": "layers.18.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1319, + "parameters": 16777216, + "reference_delta_l2": 2.0982255845203817, + "retained_delta_l2": 2.0079684495201096, + "delta_retention_norm_ratio": 0.9569840651709985, + "delta_cosine": 0.7835406465554546, + "relative_weight_rmse": 0.013539250555833047 + }, + { + "index": 77, + "total": 252, + "projection": "layers.18.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0309, + "parameters": 4194304, + "reference_delta_l2": 0.9282750221116398, + "retained_delta_l2": 0.8456048067115631, + "delta_retention_norm_ratio": 0.9109421093632161, + "delta_cosine": 0.7279578305545193, + "relative_weight_rmse": 0.011788206672603024 + }, + { + "index": 78, + "total": 252, + "projection": "layers.19.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.4285, + "parameters": 50331648, + "reference_delta_l2": 2.6912306249301237, + "retained_delta_l2": 2.3546250981874732, + "delta_retention_norm_ratio": 0.8749250533846796, + "delta_cosine": 0.6891156281526946, + "relative_weight_rmse": 0.011202486611129691 + }, + { + "index": 79, + "total": 252, + "projection": "layers.19.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3867, + "parameters": 50331648, + "reference_delta_l2": 4.628921362949428, + "retained_delta_l2": 4.563656410672661, + "delta_retention_norm_ratio": 0.9859006133050441, + "delta_cosine": 0.8191227844177474, + "relative_weight_rmse": 0.01547010211296256 + }, + { + "index": 80, + "total": 252, + "projection": "layers.19.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3848, + "parameters": 50331648, + "reference_delta_l2": 3.092950227547161, + "retained_delta_l2": 2.7732438356199482, + "delta_retention_norm_ratio": 0.8966338387602334, + "delta_cosine": 0.7108807915947286, + "relative_weight_rmse": 0.012199354921894282 + }, + { + "index": 81, + "total": 252, + "projection": "layers.19.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0312, + "parameters": 4194304, + "reference_delta_l2": 1.045892086255653, + "retained_delta_l2": 1.0068944854231603, + "delta_retention_norm_ratio": 0.9627135520528642, + "delta_cosine": 0.7955178900092555, + "relative_weight_rmse": 0.013598346534787564 + }, + { + "index": 82, + "total": 252, + "projection": "layers.19.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1314, + "parameters": 16777216, + "reference_delta_l2": 1.6400303062672876, + "retained_delta_l2": 1.4605086489661825, + "delta_retention_norm_ratio": 0.8905375976193411, + "delta_cosine": 0.705280469906104, + "relative_weight_rmse": 0.011794668218334548 + }, + { + "index": 83, + "total": 252, + "projection": "layers.19.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1317, + "parameters": 16777216, + "reference_delta_l2": 1.9352134197482955, + "retained_delta_l2": 1.8254082045639979, + "delta_retention_norm_ratio": 0.9432593769432522, + "delta_cosine": 0.765526472657706, + "relative_weight_rmse": 0.012819070038953157 + }, + { + "index": 84, + "total": 252, + "projection": "layers.19.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0307, + "parameters": 4194304, + "reference_delta_l2": 1.238934716812419, + "retained_delta_l2": 1.2059749967078048, + "delta_retention_norm_ratio": 0.9733967257052785, + "delta_cosine": 0.8061348457657889, + "relative_weight_rmse": 0.013896407827664898 + }, + { + "index": 85, + "total": 252, + "projection": "layers.2.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.4294, + "parameters": 50331648, + "reference_delta_l2": 1.5506319449815449, + "retained_delta_l2": 1.33552458767854, + "delta_retention_norm_ratio": 0.8612776178130621, + "delta_cosine": 0.6980396538475043, + "relative_weight_rmse": 0.010104122288460906 + }, + { + "index": 86, + "total": 252, + "projection": "layers.2.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3887, + "parameters": 50331648, + "reference_delta_l2": 2.5071655827311052, + "retained_delta_l2": 2.363836554944495, + "delta_retention_norm_ratio": 0.9428322449965673, + "delta_cosine": 0.7973100987789815, + "relative_weight_rmse": 0.012653250427764077 + }, + { + "index": 87, + "total": 252, + "projection": "layers.2.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.5182, + "parameters": 50331648, + "reference_delta_l2": 3.37076181332055, + "retained_delta_l2": 3.350879546206954, + "delta_retention_norm_ratio": 0.9941015508615811, + "delta_cosine": 0.8794590867496381, + "relative_weight_rmse": 0.014957108991746916 + }, + { + "index": 88, + "total": 252, + "projection": "layers.2.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0471, + "parameters": 4194304, + "reference_delta_l2": 0.5771590744001776, + "retained_delta_l2": 0.4342005060515799, + "delta_retention_norm_ratio": 0.7523064702791518, + "delta_cosine": 0.5761141115708854, + "relative_weight_rmse": 0.00929018122646379 + }, + { + "index": 89, + "total": 252, + "projection": "layers.2.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1958, + "parameters": 16777216, + "reference_delta_l2": 2.0834827393435082, + "retained_delta_l2": 1.9645847173226942, + "delta_retention_norm_ratio": 0.9429330419803345, + "delta_cosine": 0.7602359756343721, + "relative_weight_rmse": 0.01393824365053602 + }, + { + "index": 90, + "total": 252, + "projection": "layers.2.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1958, + "parameters": 16777216, + "reference_delta_l2": 1.2297332326760075, + "retained_delta_l2": 0.9716073063961379, + "delta_retention_norm_ratio": 0.7900959985295632, + "delta_cosine": 0.6082426868696985, + "relative_weight_rmse": 0.010186223431886576 + }, + { + "index": 91, + "total": 252, + "projection": "layers.2.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0464, + "parameters": 4194304, + "reference_delta_l2": 1.1603356637198619, + "retained_delta_l2": 1.1192718688127925, + "delta_retention_norm_ratio": 0.9646104173206012, + "delta_cosine": 0.7931693664764421, + "relative_weight_rmse": 0.013506959857138937 + }, + { + "index": 92, + "total": 252, + "projection": "layers.20.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.5038, + "parameters": 50331648, + "reference_delta_l2": 2.6484489109787934, + "retained_delta_l2": 2.260972169451217, + "delta_retention_norm_ratio": 0.8536967279522201, + "delta_cosine": 0.6660883770430748, + "relative_weight_rmse": 0.011255612688842639 + }, + { + "index": 93, + "total": 252, + "projection": "layers.20.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.434, + "parameters": 50331648, + "reference_delta_l2": 4.553321764650756, + "retained_delta_l2": 4.485639182490889, + "delta_retention_norm_ratio": 0.9851355591240412, + "delta_cosine": 0.8172332002329461, + "relative_weight_rmse": 0.015433938919068012 + }, + { + "index": 94, + "total": 252, + "projection": "layers.20.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.5203, + "parameters": 50331648, + "reference_delta_l2": 3.1250292041680536, + "retained_delta_l2": 2.807337479103815, + "delta_retention_norm_ratio": 0.8983395980298319, + "delta_cosine": 0.7121608786864989, + "relative_weight_rmse": 0.012274666659700928 + }, + { + "index": 95, + "total": 252, + "projection": "layers.20.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0464, + "parameters": 4194304, + "reference_delta_l2": 0.8440082402273854, + "retained_delta_l2": 0.7759200650893181, + "delta_retention_norm_ratio": 0.9193275943375581, + "delta_cosine": 0.7400741333245867, + "relative_weight_rmse": 0.01219753914450794 + }, + { + "index": 96, + "total": 252, + "projection": "layers.20.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.197, + "parameters": 16777216, + "reference_delta_l2": 1.6099755273188814, + "retained_delta_l2": 1.4290227337410972, + "delta_retention_norm_ratio": 0.8876052520629752, + "delta_cosine": 0.7035251064207005, + "relative_weight_rmse": 0.011826276846087162 + }, + { + "index": 97, + "total": 252, + "projection": "layers.20.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.196, + "parameters": 16777216, + "reference_delta_l2": 1.8927138419704896, + "retained_delta_l2": 1.7688167288870376, + "delta_retention_norm_ratio": 0.9345399656640839, + "delta_cosine": 0.7569561481973495, + "relative_weight_rmse": 0.012339664664295173 + }, + { + "index": 98, + "total": 252, + "projection": "layers.20.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0461, + "parameters": 4194304, + "reference_delta_l2": 0.9451652005666981, + "retained_delta_l2": 0.8676563899192017, + "delta_retention_norm_ratio": 0.9179944303905562, + "delta_cosine": 0.7372785677127263, + "relative_weight_rmse": 0.012028624407446963 + }, + { + "index": 99, + "total": 252, + "projection": "layers.21.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.5125, + "parameters": 50331648, + "reference_delta_l2": 3.105831811743238, + "retained_delta_l2": 2.8037985139223083, + "delta_retention_norm_ratio": 0.9027528481487846, + "delta_cosine": 0.7158225649506773, + "relative_weight_rmse": 0.012306721185022003 + }, + { + "index": 100, + "total": 252, + "projection": "layers.21.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.4326, + "parameters": 50331648, + "reference_delta_l2": 4.631992561760132, + "retained_delta_l2": 4.570159205590307, + "delta_retention_norm_ratio": 0.9866508084058044, + "delta_cosine": 0.8172349827625934, + "relative_weight_rmse": 0.015930576808165023 + }, + { + "index": 101, + "total": 252, + "projection": "layers.21.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.4137, + "parameters": 50331648, + "reference_delta_l2": 3.3410382535706633, + "retained_delta_l2": 3.0674403397188623, + "delta_retention_norm_ratio": 0.9181099128214415, + "delta_cosine": 0.7330786753108589, + "relative_weight_rmse": 0.012758917553166282 + }, + { + "index": 102, + "total": 252, + "projection": "layers.21.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0322, + "parameters": 4194304, + "reference_delta_l2": 0.8904404008877999, + "retained_delta_l2": 0.8236720774140686, + "delta_retention_norm_ratio": 0.9250165160889364, + "delta_cosine": 0.7424900007287104, + "relative_weight_rmse": 0.012832174675203442 + }, + { + "index": 103, + "total": 252, + "projection": "layers.21.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1332, + "parameters": 16777216, + "reference_delta_l2": 1.7388620331740472, + "retained_delta_l2": 1.5907312020741902, + "delta_retention_norm_ratio": 0.9148116249168629, + "delta_cosine": 0.7325832201537938, + "relative_weight_rmse": 0.01239272918621332 + }, + { + "index": 104, + "total": 252, + "projection": "layers.21.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.133, + "parameters": 16777216, + "reference_delta_l2": 1.3477118162774981, + "retained_delta_l2": 1.1148559619377523, + "delta_retention_norm_ratio": 0.8272213306084125, + "delta_cosine": 0.643909005017572, + "relative_weight_rmse": 0.010338139240074241 + }, + { + "index": 105, + "total": 252, + "projection": "layers.21.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0315, + "parameters": 4194304, + "reference_delta_l2": 1.3424533813651953, + "retained_delta_l2": 1.3257706458726028, + "delta_retention_norm_ratio": 0.9875729498512439, + "delta_cosine": 0.8235578999378094, + "relative_weight_rmse": 0.014569797845618397 + }, + { + "index": 106, + "total": 252, + "projection": "layers.22.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.4264, + "parameters": 50331648, + "reference_delta_l2": 3.3618399885983146, + "retained_delta_l2": 3.126512027707271, + "delta_retention_norm_ratio": 0.9300002493607196, + "delta_cosine": 0.7475680532810279, + "relative_weight_rmse": 0.01260929497904248 + }, + { + "index": 107, + "total": 252, + "projection": "layers.22.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.431, + "parameters": 50331648, + "reference_delta_l2": 4.216612798620095, + "retained_delta_l2": 4.102572933926655, + "delta_retention_norm_ratio": 0.9729546272945051, + "delta_cosine": 0.8022093491998286, + "relative_weight_rmse": 0.014662467372207407 + }, + { + "index": 108, + "total": 252, + "projection": "layers.22.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.4293, + "parameters": 50331648, + "reference_delta_l2": 3.4261120492346593, + "retained_delta_l2": 3.1752163604901824, + "delta_retention_norm_ratio": 0.9267695611996918, + "delta_cosine": 0.7449269716537344, + "relative_weight_rmse": 0.01262107358914896 + }, + { + "index": 109, + "total": 252, + "projection": "layers.22.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0353, + "parameters": 4194304, + "reference_delta_l2": 1.125807263291138, + "retained_delta_l2": 1.100649015285995, + "delta_retention_norm_ratio": 0.9776531482559486, + "delta_cosine": 0.8176819899707022, + "relative_weight_rmse": 0.01442174223575734 + }, + { + "index": 110, + "total": 252, + "projection": "layers.22.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1463, + "parameters": 16777216, + "reference_delta_l2": 2.099426243477152, + "retained_delta_l2": 1.9990457362955465, + "delta_retention_norm_ratio": 0.9521866950584787, + "delta_cosine": 0.7729369553466884, + "relative_weight_rmse": 0.013736021150640399 + }, + { + "index": 111, + "total": 252, + "projection": "layers.22.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.146, + "parameters": 16777216, + "reference_delta_l2": 1.8704669727023573, + "retained_delta_l2": 1.7411331316454428, + "delta_retention_norm_ratio": 0.9308547849577588, + "delta_cosine": 0.754684048204232, + "relative_weight_rmse": 0.012596676041165988 + }, + { + "index": 112, + "total": 252, + "projection": "layers.22.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0342, + "parameters": 4194304, + "reference_delta_l2": 1.1041240599891748, + "retained_delta_l2": 1.0518398406647753, + "delta_retention_norm_ratio": 0.9526464269559419, + "delta_cosine": 0.7776064187613364, + "relative_weight_rmse": 0.013226405553858848 + }, + { + "index": 113, + "total": 252, + "projection": "layers.23.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.4698, + "parameters": 50331648, + "reference_delta_l2": 3.161205806332495, + "retained_delta_l2": 2.901733948997141, + "delta_retention_norm_ratio": 0.9179199731901092, + "delta_cosine": 0.7352040404153808, + "relative_weight_rmse": 0.01197537150032579 + }, + { + "index": 114, + "total": 252, + "projection": "layers.23.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.4271, + "parameters": 50331648, + "reference_delta_l2": 5.644697395185492, + "retained_delta_l2": 5.689266581260737, + "delta_retention_norm_ratio": 1.0078957618017326, + "delta_cosine": 0.8536415134657893, + "relative_weight_rmse": 0.016824252505465966 + }, + { + "index": 115, + "total": 252, + "projection": "layers.23.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.4337, + "parameters": 50331648, + "reference_delta_l2": 4.984003733044327, + "retained_delta_l2": 4.946891272306707, + "delta_retention_norm_ratio": 0.9925536852046154, + "delta_cosine": 0.8290333634990239, + "relative_weight_rmse": 0.015415676002855193 + }, + { + "index": 116, + "total": 252, + "projection": "layers.23.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0356, + "parameters": 4194304, + "reference_delta_l2": 0.7948172661022419, + "retained_delta_l2": 0.7074683419386653, + "delta_retention_norm_ratio": 0.890101878898614, + "delta_cosine": 0.7057162449783906, + "relative_weight_rmse": 0.011977436630420743 + }, + { + "index": 117, + "total": 252, + "projection": "layers.23.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1466, + "parameters": 16777216, + "reference_delta_l2": 1.9606870799219778, + "retained_delta_l2": 1.8338586324299864, + "delta_retention_norm_ratio": 0.9353142840636058, + "delta_cosine": 0.7547703114515996, + "relative_weight_rmse": 0.012807168537045768 + }, + { + "index": 118, + "total": 252, + "projection": "layers.23.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1468, + "parameters": 16777216, + "reference_delta_l2": 1.7877365182079712, + "retained_delta_l2": 1.6202185102969995, + "delta_retention_norm_ratio": 0.9062960306483576, + "delta_cosine": 0.7240803966408368, + "relative_weight_rmse": 0.012034289859878202 + }, + { + "index": 119, + "total": 252, + "projection": "layers.23.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0339, + "parameters": 4194304, + "reference_delta_l2": 0.9274534607521585, + "retained_delta_l2": 0.8517049795559939, + "delta_retention_norm_ratio": 0.9183263803504133, + "delta_cosine": 0.7377274053139314, + "relative_weight_rmse": 0.011729376308224649 + }, + { + "index": 120, + "total": 252, + "projection": "layers.24.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.5673, + "parameters": 50331648, + "reference_delta_l2": 3.25021600221689, + "retained_delta_l2": 2.977850368680503, + "delta_retention_norm_ratio": 0.9162007591647406, + "delta_cosine": 0.7315237105470652, + "relative_weight_rmse": 0.012272240937957282 + }, + { + "index": 121, + "total": 252, + "projection": "layers.24.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.5666, + "parameters": 50331648, + "reference_delta_l2": 5.133618041769099, + "retained_delta_l2": 5.103397989223682, + "delta_retention_norm_ratio": 0.9941133032688575, + "delta_cosine": 0.8276718023182057, + "relative_weight_rmse": 0.01626066271117759 + }, + { + "index": 122, + "total": 252, + "projection": "layers.24.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3864, + "parameters": 50331648, + "reference_delta_l2": 4.009391553973875, + "retained_delta_l2": 3.8313755613607, + "delta_retention_norm_ratio": 0.9556002475146794, + "delta_cosine": 0.7756364299462929, + "relative_weight_rmse": 0.013948441084571748 + }, + { + "index": 123, + "total": 252, + "projection": "layers.24.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0318, + "parameters": 4194304, + "reference_delta_l2": 0.9665741357404588, + "retained_delta_l2": 0.9172442227548325, + "delta_retention_norm_ratio": 0.9489641702983949, + "delta_cosine": 0.7724891920777631, + "relative_weight_rmse": 0.013155425649971826 + }, + { + "index": 124, + "total": 252, + "projection": "layers.24.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1316, + "parameters": 16777216, + "reference_delta_l2": 1.9179278110579951, + "retained_delta_l2": 1.7787224033031426, + "delta_retention_norm_ratio": 0.9274188491598847, + "delta_cosine": 0.746114262372815, + "relative_weight_rmse": 0.012579315372017813 + }, + { + "index": 125, + "total": 252, + "projection": "layers.24.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1318, + "parameters": 16777216, + "reference_delta_l2": 3.025439662489526, + "retained_delta_l2": 3.022386024150926, + "delta_retention_norm_ratio": 0.9989906794782722, + "delta_cosine": 0.8489524003836729, + "relative_weight_rmse": 0.015567782458987304 + }, + { + "index": 126, + "total": 252, + "projection": "layers.24.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0309, + "parameters": 4194304, + "reference_delta_l2": 1.0884633110273467, + "retained_delta_l2": 1.037194330745601, + "delta_retention_norm_ratio": 0.9528978333377581, + "delta_cosine": 0.7791504430838209, + "relative_weight_rmse": 0.013055144546528667 + }, + { + "index": 127, + "total": 252, + "projection": "layers.25.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.427, + "parameters": 50331648, + "reference_delta_l2": 3.6372634698137727, + "retained_delta_l2": 3.4041002556738786, + "delta_retention_norm_ratio": 0.9358959789207044, + "delta_cosine": 0.7514438032951944, + "relative_weight_rmse": 0.013263716593532362 + }, + { + "index": 128, + "total": 252, + "projection": "layers.25.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3846, + "parameters": 50331648, + "reference_delta_l2": 5.607015174453435, + "retained_delta_l2": 5.621550020553077, + "delta_retention_norm_ratio": 1.0025922608816655, + "delta_cosine": 0.8408692704226695, + "relative_weight_rmse": 0.016962074591348374 + }, + { + "index": 129, + "total": 252, + "projection": "layers.25.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3879, + "parameters": 50331648, + "reference_delta_l2": 3.745353683273597, + "retained_delta_l2": 3.495572689887468, + "delta_retention_norm_ratio": 0.9333091038900739, + "delta_cosine": 0.748287645310684, + "relative_weight_rmse": 0.013621461318509673 + }, + { + "index": 130, + "total": 252, + "projection": "layers.25.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0317, + "parameters": 4194304, + "reference_delta_l2": 1.029701528133221, + "retained_delta_l2": 0.9834244826109858, + "delta_retention_norm_ratio": 0.9550578063080742, + "delta_cosine": 0.7802292195819164, + "relative_weight_rmse": 0.013551009504157704 + }, + { + "index": 131, + "total": 252, + "projection": "layers.25.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1325, + "parameters": 16777216, + "reference_delta_l2": 1.955257532356347, + "retained_delta_l2": 1.8081089901020486, + "delta_retention_norm_ratio": 0.9247421171793342, + "delta_cosine": 0.7401296636506898, + "relative_weight_rmse": 0.01319575220635778 + }, + { + "index": 132, + "total": 252, + "projection": "layers.25.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1321, + "parameters": 16777216, + "reference_delta_l2": 2.452481148873038, + "retained_delta_l2": 2.4055864951580226, + "delta_retention_norm_ratio": 0.9808786894298598, + "delta_cosine": 0.8154168271342893, + "relative_weight_rmse": 0.014602094190322861 + }, + { + "index": 133, + "total": 252, + "projection": "layers.25.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0308, + "parameters": 4194304, + "reference_delta_l2": 1.1881853587496678, + "retained_delta_l2": 1.1448003307596157, + "delta_retention_norm_ratio": 0.963486313250227, + "delta_cosine": 0.7872584929524116, + "relative_weight_rmse": 0.01345838899926889 + }, + { + "index": 134, + "total": 252, + "projection": "layers.26.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.4275, + "parameters": 50331648, + "reference_delta_l2": 3.296143428993007, + "retained_delta_l2": 3.017379274200927, + "delta_retention_norm_ratio": 0.9154271770032639, + "delta_cosine": 0.7283414827056016, + "relative_weight_rmse": 0.012338506520400626 + }, + { + "index": 135, + "total": 252, + "projection": "layers.26.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3848, + "parameters": 50331648, + "reference_delta_l2": 5.262198078002477, + "retained_delta_l2": 5.224837041527631, + "delta_retention_norm_ratio": 0.9929001082967542, + "delta_cosine": 0.8222129114626838, + "relative_weight_rmse": 0.01666040547640635 + }, + { + "index": 136, + "total": 252, + "projection": "layers.26.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3879, + "parameters": 50331648, + "reference_delta_l2": 3.598935173132037, + "retained_delta_l2": 3.3267227355060474, + "delta_retention_norm_ratio": 0.9243630617027502, + "delta_cosine": 0.7381324813507778, + "relative_weight_rmse": 0.013168130882825398 + }, + { + "index": 137, + "total": 252, + "projection": "layers.26.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0319, + "parameters": 4194304, + "reference_delta_l2": 1.102731112913039, + "retained_delta_l2": 1.0645370680697435, + "delta_retention_norm_ratio": 0.9653641360109989, + "delta_cosine": 0.7914369705331658, + "relative_weight_rmse": 0.014443687847254976 + }, + { + "index": 138, + "total": 252, + "projection": "layers.26.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1314, + "parameters": 16777216, + "reference_delta_l2": 2.145420810454555, + "retained_delta_l2": 2.020976648079422, + "delta_retention_norm_ratio": 0.9419954529345843, + "delta_cosine": 0.7603852879952234, + "relative_weight_rmse": 0.013810822996967245 + }, + { + "index": 139, + "total": 252, + "projection": "layers.26.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1307, + "parameters": 16777216, + "reference_delta_l2": 2.712891710355561, + "retained_delta_l2": 2.6885197657892053, + "delta_retention_norm_ratio": 0.9910162486496148, + "delta_cosine": 0.8319306416481287, + "relative_weight_rmse": 0.015362506902790398 + }, + { + "index": 140, + "total": 252, + "projection": "layers.26.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0303, + "parameters": 4194304, + "reference_delta_l2": 1.3011344185588933, + "retained_delta_l2": 1.2658965447651351, + "delta_retention_norm_ratio": 0.9729175761618951, + "delta_cosine": 0.806983788179549, + "relative_weight_rmse": 0.014257043592580667 + }, + { + "index": 141, + "total": 252, + "projection": "layers.27.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.4272, + "parameters": 50331648, + "reference_delta_l2": 3.123011383482389, + "retained_delta_l2": 2.7963383794443883, + "delta_retention_norm_ratio": 0.895398074510463, + "delta_cosine": 0.7078548518864215, + "relative_weight_rmse": 0.011911445809095231 + }, + { + "index": 142, + "total": 252, + "projection": "layers.27.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.4007, + "parameters": 50331648, + "reference_delta_l2": 5.231988258843969, + "retained_delta_l2": 5.199668161745496, + "delta_retention_norm_ratio": 0.9938225975481042, + "delta_cosine": 0.825069634808694, + "relative_weight_rmse": 0.01646019428619956 + }, + { + "index": 143, + "total": 252, + "projection": "layers.27.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3882, + "parameters": 50331648, + "reference_delta_l2": 3.1442957530656317, + "retained_delta_l2": 2.7610954614368444, + "delta_retention_norm_ratio": 0.8781284199314986, + "delta_cosine": 0.6898880597820134, + "relative_weight_rmse": 0.012185740763517383 + }, + { + "index": 144, + "total": 252, + "projection": "layers.27.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0319, + "parameters": 4194304, + "reference_delta_l2": 0.9581309963591961, + "retained_delta_l2": 0.9020189071919141, + "delta_retention_norm_ratio": 0.9414358898934464, + "delta_cosine": 0.7666384008707281, + "relative_weight_rmse": 0.013501718945248388 + }, + { + "index": 145, + "total": 252, + "projection": "layers.27.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1316, + "parameters": 16777216, + "reference_delta_l2": 1.834858768399369, + "retained_delta_l2": 1.6698680560604757, + "delta_retention_norm_ratio": 0.9100798845227625, + "delta_cosine": 0.7240175455536444, + "relative_weight_rmse": 0.012746677317509142 + }, + { + "index": 146, + "total": 252, + "projection": "layers.27.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.132, + "parameters": 16777216, + "reference_delta_l2": 2.505486662757741, + "retained_delta_l2": 2.4510638799011137, + "delta_retention_norm_ratio": 0.9782785581477712, + "delta_cosine": 0.8134164664490362, + "relative_weight_rmse": 0.014896191307883081 + }, + { + "index": 147, + "total": 252, + "projection": "layers.27.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0304, + "parameters": 4194304, + "reference_delta_l2": 1.3047011093368113, + "retained_delta_l2": 1.2814081896946499, + "delta_retention_norm_ratio": 0.982146930453672, + "delta_cosine": 0.8141221414471734, + "relative_weight_rmse": 0.014309363488148276 + }, + { + "index": 148, + "total": 252, + "projection": "layers.28.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.4382, + "parameters": 50331648, + "reference_delta_l2": 3.762789462488106, + "retained_delta_l2": 3.5481016609195812, + "delta_retention_norm_ratio": 0.9429445086660351, + "delta_cosine": 0.7610996582309881, + "relative_weight_rmse": 0.01310177561141105 + }, + { + "index": 149, + "total": 252, + "projection": "layers.28.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3899, + "parameters": 50331648, + "reference_delta_l2": 6.438360569200188, + "retained_delta_l2": 6.553508545352642, + "delta_retention_norm_ratio": 1.0178846734218798, + "delta_cosine": 0.8672698433210566, + "relative_weight_rmse": 0.017961164684581265 + }, + { + "index": 150, + "total": 252, + "projection": "layers.28.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3883, + "parameters": 50331648, + "reference_delta_l2": 3.2448978161914703, + "retained_delta_l2": 2.8773300518030918, + "delta_retention_norm_ratio": 0.8867243946622048, + "delta_cosine": 0.697975544585092, + "relative_weight_rmse": 0.012328033999554563 + }, + { + "index": 151, + "total": 252, + "projection": "layers.28.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0318, + "parameters": 4194304, + "reference_delta_l2": 1.0320638572183087, + "retained_delta_l2": 0.9821626085605092, + "delta_retention_norm_ratio": 0.9516490686997828, + "delta_cosine": 0.7710948278561396, + "relative_weight_rmse": 0.014134383568436673 + }, + { + "index": 152, + "total": 252, + "projection": "layers.28.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1321, + "parameters": 16777216, + "reference_delta_l2": 1.8683109186858797, + "retained_delta_l2": 1.7167334247067823, + "delta_retention_norm_ratio": 0.9188692350598083, + "delta_cosine": 0.7339151321122213, + "relative_weight_rmse": 0.012836818361648275 + }, + { + "index": 153, + "total": 252, + "projection": "layers.28.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1329, + "parameters": 16777216, + "reference_delta_l2": 2.1068458268519263, + "retained_delta_l2": 2.0070207707992296, + "delta_retention_norm_ratio": 0.95261871809488, + "delta_cosine": 0.7769981399785503, + "relative_weight_rmse": 0.013249324767482331 + }, + { + "index": 154, + "total": 252, + "projection": "layers.28.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0308, + "parameters": 4194304, + "reference_delta_l2": 1.056848088619748, + "retained_delta_l2": 0.9875918527299122, + "delta_retention_norm_ratio": 0.9344690721063942, + "delta_cosine": 0.7518514915046068, + "relative_weight_rmse": 0.012583457380717504 + }, + { + "index": 155, + "total": 252, + "projection": "layers.29.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.4301, + "parameters": 50331648, + "reference_delta_l2": 3.741068557922537, + "retained_delta_l2": 3.5107556859373794, + "delta_retention_norm_ratio": 0.9384366075041797, + "delta_cosine": 0.7559837396886837, + "relative_weight_rmse": 0.012999186322999736 + }, + { + "index": 156, + "total": 252, + "projection": "layers.29.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3919, + "parameters": 50331648, + "reference_delta_l2": 6.13810226974935, + "retained_delta_l2": 6.223313811643556, + "delta_retention_norm_ratio": 1.013882392008057, + "delta_cosine": 0.8571541670520612, + "relative_weight_rmse": 0.017726461236553998 + }, + { + "index": 157, + "total": 252, + "projection": "layers.29.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.4407, + "parameters": 50331648, + "reference_delta_l2": 3.5820628193425765, + "retained_delta_l2": 3.2821423568744894, + "delta_retention_norm_ratio": 0.9162715793680212, + "delta_cosine": 0.7294835399933917, + "relative_weight_rmse": 0.012901708808166195 + }, + { + "index": 158, + "total": 252, + "projection": "layers.29.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0467, + "parameters": 4194304, + "reference_delta_l2": 0.9628755695901171, + "retained_delta_l2": 0.9126935236004589, + "delta_retention_norm_ratio": 0.9478831454712056, + "delta_cosine": 0.7820864106255127, + "relative_weight_rmse": 0.01313449066827085 + }, + { + "index": 159, + "total": 252, + "projection": "layers.29.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1963, + "parameters": 16777216, + "reference_delta_l2": 1.9460512594459791, + "retained_delta_l2": 1.7943990197662905, + "delta_retention_norm_ratio": 0.922071816482952, + "delta_cosine": 0.7376092764159529, + "relative_weight_rmse": 0.012932633062480688 + }, + { + "index": 160, + "total": 252, + "projection": "layers.29.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1968, + "parameters": 16777216, + "reference_delta_l2": 2.947773580817007, + "retained_delta_l2": 2.93511516979154, + "delta_retention_norm_ratio": 0.9957057722791727, + "delta_cosine": 0.8480690788527846, + "relative_weight_rmse": 0.015284934669139631 + }, + { + "index": 161, + "total": 252, + "projection": "layers.29.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0454, + "parameters": 4194304, + "reference_delta_l2": 1.0700284614203275, + "retained_delta_l2": 1.010683508397803, + "delta_retention_norm_ratio": 0.9445389023169051, + "delta_cosine": 0.7676898065872572, + "relative_weight_rmse": 0.0129583597724709 + }, + { + "index": 162, + "total": 252, + "projection": "layers.3.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.5733, + "parameters": 50331648, + "reference_delta_l2": 2.0743569009756007, + "retained_delta_l2": 1.8382691756052478, + "delta_retention_norm_ratio": 0.8861875093628694, + "delta_cosine": 0.7173073529092404, + "relative_weight_rmse": 0.011359621859503791 + }, + { + "index": 163, + "total": 252, + "projection": "layers.3.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.4326, + "parameters": 50331648, + "reference_delta_l2": 2.926958575992499, + "retained_delta_l2": 2.7059936619403366, + "delta_retention_norm_ratio": 0.9245069896565804, + "delta_cosine": 0.7461013369494565, + "relative_weight_rmse": 0.012259968483905708 + }, + { + "index": 164, + "total": 252, + "projection": "layers.3.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.4285, + "parameters": 50331648, + "reference_delta_l2": 3.1802686879413384, + "retained_delta_l2": 3.1142541585175083, + "delta_retention_norm_ratio": 0.9792424678851387, + "delta_cosine": 0.8425646980142314, + "relative_weight_rmse": 0.014672863151850154 + }, + { + "index": 165, + "total": 252, + "projection": "layers.3.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0318, + "parameters": 4194304, + "reference_delta_l2": 0.5947120352689738, + "retained_delta_l2": 0.4615478574992086, + "delta_retention_norm_ratio": 0.776086290721293, + "delta_cosine": 0.5971063879847892, + "relative_weight_rmse": 0.009639244029858614 + }, + { + "index": 166, + "total": 252, + "projection": "layers.3.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1341, + "parameters": 16777216, + "reference_delta_l2": 1.853556905749305, + "retained_delta_l2": 1.7003902293228157, + "delta_retention_norm_ratio": 0.9173660781865386, + "delta_cosine": 0.7308417605121308, + "relative_weight_rmse": 0.012828925539646631 + }, + { + "index": 167, + "total": 252, + "projection": "layers.3.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1306, + "parameters": 16777216, + "reference_delta_l2": 1.509794211282041, + "retained_delta_l2": 1.3072280448613307, + "delta_retention_norm_ratio": 0.8658319359638414, + "delta_cosine": 0.6792206253764826, + "relative_weight_rmse": 0.011733211443473323 + }, + { + "index": 168, + "total": 252, + "projection": "layers.3.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0303, + "parameters": 4194304, + "reference_delta_l2": 0.9474064313172714, + "retained_delta_l2": 0.8539360623494765, + "delta_retention_norm_ratio": 0.901340791155667, + "delta_cosine": 0.7156835757924905, + "relative_weight_rmse": 0.012393251473286376 + }, + { + "index": 169, + "total": 252, + "projection": "layers.30.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.4301, + "parameters": 50331648, + "reference_delta_l2": 3.887550185815536, + "retained_delta_l2": 3.6872367488804443, + "delta_retention_norm_ratio": 0.9484730929864331, + "delta_cosine": 0.7666490475410297, + "relative_weight_rmse": 0.013137235869439547 + }, + { + "index": 170, + "total": 252, + "projection": "layers.30.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.4159, + "parameters": 50331648, + "reference_delta_l2": 5.406724511969585, + "retained_delta_l2": 5.4045390351384865, + "delta_retention_norm_ratio": 0.9995957854286343, + "delta_cosine": 0.8309819772108276, + "relative_weight_rmse": 0.016941780124070223 + }, + { + "index": 171, + "total": 252, + "projection": "layers.30.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.4313, + "parameters": 50331648, + "reference_delta_l2": 3.8128369928338883, + "retained_delta_l2": 3.5330640598507848, + "delta_retention_norm_ratio": 0.926623421481451, + "delta_cosine": 0.7399615116788848, + "relative_weight_rmse": 0.013411276714095241 + }, + { + "index": 172, + "total": 252, + "projection": "layers.30.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0348, + "parameters": 4194304, + "reference_delta_l2": 1.4959604277170342, + "retained_delta_l2": 1.5078898358472888, + "delta_retention_norm_ratio": 1.007974414235315, + "delta_cosine": 0.8528712844814826, + "relative_weight_rmse": 0.016609391188736788 + }, + { + "index": 173, + "total": 252, + "projection": "layers.30.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1454, + "parameters": 16777216, + "reference_delta_l2": 1.8688273581688304, + "retained_delta_l2": 1.6928385267418848, + "delta_retention_norm_ratio": 0.9058292727481322, + "delta_cosine": 0.7175848846442401, + "relative_weight_rmse": 0.012864115480060978 + }, + { + "index": 174, + "total": 252, + "projection": "layers.30.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1453, + "parameters": 16777216, + "reference_delta_l2": 3.0218163002013974, + "retained_delta_l2": 3.033255695310336, + "delta_retention_norm_ratio": 1.0037856024233427, + "delta_cosine": 0.855246497937382, + "relative_weight_rmse": 0.016301478296790135 + }, + { + "index": 175, + "total": 252, + "projection": "layers.30.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0304, + "parameters": 4194304, + "reference_delta_l2": 1.3097866963734044, + "retained_delta_l2": 1.2717793784547025, + "delta_retention_norm_ratio": 0.9709820553041666, + "delta_cosine": 0.8015100539645267, + "relative_weight_rmse": 0.013921982367461434 + }, + { + "index": 176, + "total": 252, + "projection": "layers.31.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.4256, + "parameters": 50331648, + "reference_delta_l2": 4.111855477331991, + "retained_delta_l2": 3.946481597377216, + "delta_retention_norm_ratio": 0.9597812031900306, + "delta_cosine": 0.7805752015994533, + "relative_weight_rmse": 0.01338187372349077 + }, + { + "index": 177, + "total": 252, + "projection": "layers.31.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3876, + "parameters": 50331648, + "reference_delta_l2": 4.217842761635248, + "retained_delta_l2": 4.0700160479109995, + "delta_retention_norm_ratio": 0.9649520567554452, + "delta_cosine": 0.7859478836623668, + "relative_weight_rmse": 0.014764225147167017 + }, + { + "index": 178, + "total": 252, + "projection": "layers.31.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.4142, + "parameters": 50331648, + "reference_delta_l2": 3.746610597758799, + "retained_delta_l2": 3.459116745774283, + "delta_retention_norm_ratio": 0.9232656171536767, + "delta_cosine": 0.7367602923704923, + "relative_weight_rmse": 0.013127943116183337 + }, + { + "index": 179, + "total": 252, + "projection": "layers.31.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0348, + "parameters": 4194304, + "reference_delta_l2": 1.1313048726490362, + "retained_delta_l2": 1.1078314198905324, + "delta_retention_norm_ratio": 0.9792509929674934, + "delta_cosine": 0.8130234486479346, + "relative_weight_rmse": 0.014433804273277513 + }, + { + "index": 180, + "total": 252, + "projection": "layers.31.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1479, + "parameters": 16777216, + "reference_delta_l2": 1.9278459560552061, + "retained_delta_l2": 1.7792393123094077, + "delta_retention_norm_ratio": 0.9229157063721626, + "delta_cosine": 0.7388878065229529, + "relative_weight_rmse": 0.012543449992171508 + }, + { + "index": 181, + "total": 252, + "projection": "layers.31.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1476, + "parameters": 16777216, + "reference_delta_l2": 3.096086590658917, + "retained_delta_l2": 3.1196112552948527, + "delta_retention_norm_ratio": 1.0075981933796396, + "delta_cosine": 0.8593676477956623, + "relative_weight_rmse": 0.01636230118241256 + }, + { + "index": 182, + "total": 252, + "projection": "layers.31.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.034, + "parameters": 4194304, + "reference_delta_l2": 0.9139796154536786, + "retained_delta_l2": 0.8064729269280674, + "delta_retention_norm_ratio": 0.8823751791529318, + "delta_cosine": 0.6996030420922704, + "relative_weight_rmse": 0.011253490135860623 + }, + { + "index": 183, + "total": 252, + "projection": "layers.32.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.4659, + "parameters": 50331648, + "reference_delta_l2": 3.053831443049286, + "retained_delta_l2": 2.701723451449301, + "delta_retention_norm_ratio": 0.8846995984662464, + "delta_cosine": 0.6983672545938556, + "relative_weight_rmse": 0.01129049061966828 + }, + { + "index": 184, + "total": 252, + "projection": "layers.32.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3921, + "parameters": 50331648, + "reference_delta_l2": 5.744040422204762, + "retained_delta_l2": 5.796206138948864, + "delta_retention_norm_ratio": 1.0090817112885286, + "delta_cosine": 0.8486326161192361, + "relative_weight_rmse": 0.01749373845032408 + }, + { + "index": 185, + "total": 252, + "projection": "layers.32.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3895, + "parameters": 50331648, + "reference_delta_l2": 4.183211149438155, + "retained_delta_l2": 3.9676960266992682, + "delta_retention_norm_ratio": 0.9484809360465027, + "delta_cosine": 0.7674993693179356, + "relative_weight_rmse": 0.013845045613414502 + }, + { + "index": 186, + "total": 252, + "projection": "layers.32.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0316, + "parameters": 4194304, + "reference_delta_l2": 1.0756438480769024, + "retained_delta_l2": 1.0412425021487295, + "delta_retention_norm_ratio": 0.968017902961396, + "delta_cosine": 0.7987848584940211, + "relative_weight_rmse": 0.014084555923825329 + }, + { + "index": 187, + "total": 252, + "projection": "layers.32.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1372, + "parameters": 16777216, + "reference_delta_l2": 1.9529995006127592, + "retained_delta_l2": 1.7949668005355779, + "delta_retention_norm_ratio": 0.9190820581225965, + "delta_cosine": 0.7327240550048206, + "relative_weight_rmse": 0.012908527612575873 + }, + { + "index": 188, + "total": 252, + "projection": "layers.32.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1469, + "parameters": 16777216, + "reference_delta_l2": 3.505475846335067, + "retained_delta_l2": 3.565156670039804, + "delta_retention_norm_ratio": 1.017025027791058, + "delta_cosine": 0.8838348012627846, + "relative_weight_rmse": 0.017017504730019232 + }, + { + "index": 189, + "total": 252, + "projection": "layers.32.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0342, + "parameters": 4194304, + "reference_delta_l2": 1.1045105107116564, + "retained_delta_l2": 1.0348550150008564, + "delta_retention_norm_ratio": 0.9369354161546912, + "delta_cosine": 0.7580528592700045, + "relative_weight_rmse": 0.012599439574745123 + }, + { + "index": 190, + "total": 252, + "projection": "layers.33.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.4688, + "parameters": 50331648, + "reference_delta_l2": 4.678264151276083, + "retained_delta_l2": 4.5917712503631485, + "delta_retention_norm_ratio": 0.9815117534803285, + "delta_cosine": 0.8098173119099811, + "relative_weight_rmse": 0.014401589246758412 + }, + { + "index": 191, + "total": 252, + "projection": "layers.33.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.4303, + "parameters": 50331648, + "reference_delta_l2": 4.01159353165815, + "retained_delta_l2": 3.8282471221634697, + "delta_retention_norm_ratio": 0.9542958657082847, + "delta_cosine": 0.7714066860695666, + "relative_weight_rmse": 0.014745803469471288 + }, + { + "index": 192, + "total": 252, + "projection": "layers.33.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.4308, + "parameters": 50331648, + "reference_delta_l2": 4.553582099114154, + "retained_delta_l2": 4.417418331222436, + "delta_retention_norm_ratio": 0.9700974386915727, + "delta_cosine": 0.7956922427163521, + "relative_weight_rmse": 0.01422952810296179 + }, + { + "index": 193, + "total": 252, + "projection": "layers.33.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0353, + "parameters": 4194304, + "reference_delta_l2": 1.171864142331482, + "retained_delta_l2": 1.1625983617559128, + "delta_retention_norm_ratio": 0.9920931273166749, + "delta_cosine": 0.8277975258194085, + "relative_weight_rmse": 0.01555121427324402 + }, + { + "index": 194, + "total": 252, + "projection": "layers.33.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1462, + "parameters": 16777216, + "reference_delta_l2": 1.9573452017526336, + "retained_delta_l2": 1.8092278968319213, + "delta_retention_norm_ratio": 0.9243274488383111, + "delta_cosine": 0.7413609007930597, + "relative_weight_rmse": 0.0123009063116843 + }, + { + "index": 195, + "total": 252, + "projection": "layers.33.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1464, + "parameters": 16777216, + "reference_delta_l2": 1.6955939867236154, + "retained_delta_l2": 1.563413023761108, + "delta_retention_norm_ratio": 0.9220444493213144, + "delta_cosine": 0.7428051155316843, + "relative_weight_rmse": 0.011924065759423506 + }, + { + "index": 196, + "total": 252, + "projection": "layers.33.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0341, + "parameters": 4194304, + "reference_delta_l2": 1.2335489621729434, + "retained_delta_l2": 1.1838319813016203, + "delta_retention_norm_ratio": 0.9596959809493537, + "delta_cosine": 0.7862561679189393, + "relative_weight_rmse": 0.012723818963755568 + }, + { + "index": 197, + "total": 252, + "projection": "layers.34.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.4326, + "parameters": 50331648, + "reference_delta_l2": 3.252748128233362, + "retained_delta_l2": 3.0211996385106383, + "delta_retention_norm_ratio": 0.9288145037382642, + "delta_cosine": 0.7452421823975507, + "relative_weight_rmse": 0.011834710350399152 + }, + { + "index": 198, + "total": 252, + "projection": "layers.34.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3865, + "parameters": 50331648, + "reference_delta_l2": 3.5141024024549754, + "retained_delta_l2": 3.3104077992582472, + "delta_retention_norm_ratio": 0.9420350974819556, + "delta_cosine": 0.7571188297576488, + "relative_weight_rmse": 0.013110959157848892 + }, + { + "index": 199, + "total": 252, + "projection": "layers.34.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.399, + "parameters": 50331648, + "reference_delta_l2": 3.1198018847302547, + "retained_delta_l2": 2.7957385412191966, + "delta_retention_norm_ratio": 0.8961269479651343, + "delta_cosine": 0.7096561401166236, + "relative_weight_rmse": 0.01141415214434449 + }, + { + "index": 200, + "total": 252, + "projection": "layers.34.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0346, + "parameters": 4194304, + "reference_delta_l2": 0.7891379146299576, + "retained_delta_l2": 0.7278324850298318, + "delta_retention_norm_ratio": 0.9223134151032736, + "delta_cosine": 0.7527277500802979, + "relative_weight_rmse": 0.011632973760519227 + }, + { + "index": 201, + "total": 252, + "projection": "layers.34.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1461, + "parameters": 16777216, + "reference_delta_l2": 2.3668129283754418, + "retained_delta_l2": 2.29638818093873, + "delta_retention_norm_ratio": 0.9702449033498179, + "delta_cosine": 0.7930587439203136, + "relative_weight_rmse": 0.014045439094688958 + }, + { + "index": 202, + "total": 252, + "projection": "layers.34.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1462, + "parameters": 16777216, + "reference_delta_l2": 2.7951850043391318, + "retained_delta_l2": 2.808792609847545, + "delta_retention_norm_ratio": 1.0048682307207892, + "delta_cosine": 0.8599799296958386, + "relative_weight_rmse": 0.015290268071276977 + }, + { + "index": 203, + "total": 252, + "projection": "layers.34.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0339, + "parameters": 4194304, + "reference_delta_l2": 0.9817039826831607, + "retained_delta_l2": 0.8771706618028705, + "delta_retention_norm_ratio": 0.8935184915980648, + "delta_cosine": 0.7154696473562105, + "relative_weight_rmse": 0.011407674885279172 + }, + { + "index": 204, + "total": 252, + "projection": "layers.35.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.4678, + "parameters": 50331648, + "reference_delta_l2": 2.3247118593602805, + "retained_delta_l2": 2.077241279031564, + "delta_retention_norm_ratio": 0.8935478479484267, + "delta_cosine": 0.713891705186287, + "relative_weight_rmse": 0.009790703119852036 + }, + { + "index": 205, + "total": 252, + "projection": "layers.35.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3895, + "parameters": 50331648, + "reference_delta_l2": 2.620668661924457, + "retained_delta_l2": 2.286002393746483, + "delta_retention_norm_ratio": 0.8722973746966487, + "delta_cosine": 0.6870305896206982, + "relative_weight_rmse": 0.010515925872027135 + }, + { + "index": 206, + "total": 252, + "projection": "layers.35.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3892, + "parameters": 50331648, + "reference_delta_l2": 2.6604885496698265, + "retained_delta_l2": 2.2876053523445137, + "delta_retention_norm_ratio": 0.8598440886462043, + "delta_cosine": 0.6745012238913933, + "relative_weight_rmse": 0.010341996642101553 + }, + { + "index": 207, + "total": 252, + "projection": "layers.35.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0308, + "parameters": 4194304, + "reference_delta_l2": 0.8327469708556873, + "retained_delta_l2": 0.7626626606082086, + "delta_retention_norm_ratio": 0.91583960950892, + "delta_cosine": 0.7341530315974766, + "relative_weight_rmse": 0.012229577559635408 + }, + { + "index": 208, + "total": 252, + "projection": "layers.35.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1316, + "parameters": 16777216, + "reference_delta_l2": 1.9782298089460089, + "retained_delta_l2": 1.876763858217896, + "delta_retention_norm_ratio": 0.9487087140891011, + "delta_cosine": 0.767549205706205, + "relative_weight_rmse": 0.012984556767950172 + }, + { + "index": 209, + "total": 252, + "projection": "layers.35.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1362, + "parameters": 16777216, + "reference_delta_l2": 1.9213623452775879, + "retained_delta_l2": 1.8334149703375806, + "delta_retention_norm_ratio": 0.9542265543216415, + "delta_cosine": 0.7745283451972568, + "relative_weight_rmse": 0.013082516912719225 + }, + { + "index": 210, + "total": 252, + "projection": "layers.35.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0341, + "parameters": 4194304, + "reference_delta_l2": 0.9021829580956073, + "retained_delta_l2": 0.7955235445478319, + "delta_retention_norm_ratio": 0.881776293166832, + "delta_cosine": 0.6987772350844439, + "relative_weight_rmse": 0.011400123966470789 + }, + { + "index": 211, + "total": 252, + "projection": "layers.4.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.4705, + "parameters": 50331648, + "reference_delta_l2": 2.2260016272118324, + "retained_delta_l2": 1.909627666903947, + "delta_retention_norm_ratio": 0.8578734370899099, + "delta_cosine": 0.6731615833498001, + "relative_weight_rmse": 0.01083998019039234 + }, + { + "index": 212, + "total": 252, + "projection": "layers.4.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.4336, + "parameters": 50331648, + "reference_delta_l2": 3.3169942648288515, + "retained_delta_l2": 2.9262862948112773, + "delta_retention_norm_ratio": 0.8822102364901937, + "delta_cosine": 0.709347452002898, + "relative_weight_rmse": 0.010835394375379706 + }, + { + "index": 213, + "total": 252, + "projection": "layers.4.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3871, + "parameters": 50331648, + "reference_delta_l2": 3.302637021239384, + "retained_delta_l2": 3.180670097434266, + "delta_retention_norm_ratio": 0.9630698369149427, + "delta_cosine": 0.7953906244948861, + "relative_weight_rmse": 0.014022962820982353 + }, + { + "index": 214, + "total": 252, + "projection": "layers.4.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0314, + "parameters": 4194304, + "reference_delta_l2": 0.5936335633616754, + "retained_delta_l2": 0.4484424713150318, + "delta_retention_norm_ratio": 0.7554196713129832, + "delta_cosine": 0.5783864908583726, + "relative_weight_rmse": 0.00946943014537174 + }, + { + "index": 215, + "total": 252, + "projection": "layers.4.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1324, + "parameters": 16777216, + "reference_delta_l2": 1.6292055103524448, + "retained_delta_l2": 1.4156739956673485, + "delta_retention_norm_ratio": 0.8689351875326624, + "delta_cosine": 0.6811352287595004, + "relative_weight_rmse": 0.011807318380884849 + }, + { + "index": 216, + "total": 252, + "projection": "layers.4.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1325, + "parameters": 16777216, + "reference_delta_l2": 1.2624481434789505, + "retained_delta_l2": 0.998231827120063, + "delta_retention_norm_ratio": 0.7907111529897917, + "delta_cosine": 0.6085039151512668, + "relative_weight_rmse": 0.010246638558303461 + }, + { + "index": 217, + "total": 252, + "projection": "layers.4.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0311, + "parameters": 4194304, + "reference_delta_l2": 1.134759460682007, + "retained_delta_l2": 1.0769977887516942, + "delta_retention_norm_ratio": 0.9490978714593863, + "delta_cosine": 0.7690369144559485, + "relative_weight_rmse": 0.013529102135591874 + }, + { + "index": 218, + "total": 252, + "projection": "layers.5.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.4438, + "parameters": 50331648, + "reference_delta_l2": 1.8075376864954238, + "retained_delta_l2": 1.3462648062957494, + "delta_retention_norm_ratio": 0.7448059403430634, + "delta_cosine": 0.5699311139981033, + "relative_weight_rmse": 0.009057626133657137 + }, + { + "index": 219, + "total": 252, + "projection": "layers.5.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.4282, + "parameters": 50331648, + "reference_delta_l2": 2.9547763398015277, + "retained_delta_l2": 2.491105461832083, + "delta_retention_norm_ratio": 0.8430775041333283, + "delta_cosine": 0.6605351907380126, + "relative_weight_rmse": 0.009998123812329477 + }, + { + "index": 220, + "total": 252, + "projection": "layers.5.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.4276, + "parameters": 50331648, + "reference_delta_l2": 3.179110191236641, + "retained_delta_l2": 2.9652745298795473, + "delta_retention_norm_ratio": 0.9327372602728459, + "delta_cosine": 0.7495243085720829, + "relative_weight_rmse": 0.013132384995557108 + }, + { + "index": 221, + "total": 252, + "projection": "layers.5.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.035, + "parameters": 4194304, + "reference_delta_l2": 0.6206844406136691, + "retained_delta_l2": 0.4888766787001676, + "delta_retention_norm_ratio": 0.7876412661751541, + "delta_cosine": 0.6083199970294374, + "relative_weight_rmse": 0.00986655306999503 + }, + { + "index": 222, + "total": 252, + "projection": "layers.5.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1457, + "parameters": 16777216, + "reference_delta_l2": 1.4688147290322306, + "retained_delta_l2": 1.2341536590443796, + "delta_retention_norm_ratio": 0.8402378017120894, + "delta_cosine": 0.6522332288079532, + "relative_weight_rmse": 0.011355749239293295 + }, + { + "index": 223, + "total": 252, + "projection": "layers.5.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1463, + "parameters": 16777216, + "reference_delta_l2": 1.4698201348999005, + "retained_delta_l2": 1.2554210359850437, + "delta_retention_norm_ratio": 0.8541324248973784, + "delta_cosine": 0.6707189309437211, + "relative_weight_rmse": 0.011193760056149171 + }, + { + "index": 224, + "total": 252, + "projection": "layers.5.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0358, + "parameters": 4194304, + "reference_delta_l2": 1.045526494743739, + "retained_delta_l2": 0.9808957900976146, + "delta_retention_norm_ratio": 0.9381835802621477, + "delta_cosine": 0.7607268259212184, + "relative_weight_rmse": 0.012869417216814834 + }, + { + "index": 225, + "total": 252, + "projection": "layers.6.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.4692, + "parameters": 50331648, + "reference_delta_l2": 4.767971618211114, + "retained_delta_l2": 4.780978603586194, + "delta_retention_norm_ratio": 1.002727991359135, + "delta_cosine": 0.8442329974814251, + "relative_weight_rmse": 0.015436940953544165 + }, + { + "index": 226, + "total": 252, + "projection": "layers.6.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.4296, + "parameters": 50331648, + "reference_delta_l2": 3.108221903546128, + "retained_delta_l2": 2.718122861103922, + "delta_retention_norm_ratio": 0.8744944683656121, + "delta_cosine": 0.6922136647043337, + "relative_weight_rmse": 0.010687526629909457 + }, + { + "index": 227, + "total": 252, + "projection": "layers.6.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.4336, + "parameters": 50331648, + "reference_delta_l2": 3.3861711107000456, + "retained_delta_l2": 3.1980358257326156, + "delta_retention_norm_ratio": 0.9444401127949686, + "delta_cosine": 0.7630505913583782, + "relative_weight_rmse": 0.013219754197903279 + }, + { + "index": 228, + "total": 252, + "projection": "layers.6.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0361, + "parameters": 4194304, + "reference_delta_l2": 0.5927062606906872, + "retained_delta_l2": 0.44756971552174746, + "delta_retention_norm_ratio": 0.7551290499280191, + "delta_cosine": 0.578259429925641, + "relative_weight_rmse": 0.009580118204043437 + }, + { + "index": 229, + "total": 252, + "projection": "layers.6.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1477, + "parameters": 16777216, + "reference_delta_l2": 1.364260871460181, + "retained_delta_l2": 1.105711838917299, + "delta_retention_norm_ratio": 0.8104841691558929, + "delta_cosine": 0.6264129271257128, + "relative_weight_rmse": 0.010391485250505374 + }, + { + "index": 230, + "total": 252, + "projection": "layers.6.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1634, + "parameters": 16777216, + "reference_delta_l2": 1.4026430871778701, + "retained_delta_l2": 1.163348604980614, + "delta_retention_norm_ratio": 0.8293974537180955, + "delta_cosine": 0.6444989695957347, + "relative_weight_rmse": 0.01072133496416448 + }, + { + "index": 231, + "total": 252, + "projection": "layers.6.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0464, + "parameters": 4194304, + "reference_delta_l2": 0.9858122694984375, + "retained_delta_l2": 0.9100168957351484, + "delta_retention_norm_ratio": 0.9231137853438848, + "delta_cosine": 0.7395032745856821, + "relative_weight_rmse": 0.012395624830624017 + }, + { + "index": 232, + "total": 252, + "projection": "layers.7.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.6151, + "parameters": 50331648, + "reference_delta_l2": 1.8470529972822454, + "retained_delta_l2": 1.367158888138189, + "delta_retention_norm_ratio": 0.7401838984316245, + "delta_cosine": 0.5657485318848586, + "relative_weight_rmse": 0.008967581176751666 + }, + { + "index": 233, + "total": 252, + "projection": "layers.7.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.576, + "parameters": 50331648, + "reference_delta_l2": 2.8987939693909643, + "retained_delta_l2": 2.483962990247949, + "delta_retention_norm_ratio": 0.8568953214601275, + "delta_cosine": 0.6739190887778956, + "relative_weight_rmse": 0.010357019868688137 + }, + { + "index": 234, + "total": 252, + "projection": "layers.7.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.5756, + "parameters": 50331648, + "reference_delta_l2": 2.9765437095874767, + "retained_delta_l2": 2.6935023761457253, + "delta_retention_norm_ratio": 0.9049093979268397, + "delta_cosine": 0.7166398843761006, + "relative_weight_rmse": 0.012386486627953898 + }, + { + "index": 235, + "total": 252, + "projection": "layers.7.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0468, + "parameters": 4194304, + "reference_delta_l2": 0.7224742538590713, + "retained_delta_l2": 0.6132539257508068, + "delta_retention_norm_ratio": 0.8488246085934996, + "delta_cosine": 0.6625368131529569, + "relative_weight_rmse": 0.01077088169072488 + }, + { + "index": 236, + "total": 252, + "projection": "layers.7.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1654, + "parameters": 16777216, + "reference_delta_l2": 1.366782587331481, + "retained_delta_l2": 1.107535765437666, + "delta_retention_norm_ratio": 0.810323291870457, + "delta_cosine": 0.6254448182797966, + "relative_weight_rmse": 0.010554290745063385 + }, + { + "index": 237, + "total": 252, + "projection": "layers.7.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1467, + "parameters": 16777216, + "reference_delta_l2": 1.6869748671129567, + "retained_delta_l2": 1.525353345117514, + "delta_retention_norm_ratio": 0.9041944695524496, + "delta_cosine": 0.7181050472258073, + "relative_weight_rmse": 0.012122600315732675 + }, + { + "index": 238, + "total": 252, + "projection": "layers.7.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0346, + "parameters": 4194304, + "reference_delta_l2": 0.989475627612034, + "retained_delta_l2": 0.8931585963392329, + "delta_retention_norm_ratio": 0.9026585106444216, + "delta_cosine": 0.7177665120113694, + "relative_weight_rmse": 0.011984108502319991 + }, + { + "index": 239, + "total": 252, + "projection": "layers.8.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.4723, + "parameters": 50331648, + "reference_delta_l2": 2.6443507154044354, + "retained_delta_l2": 2.2829722080496935, + "delta_retention_norm_ratio": 0.8633394181605478, + "delta_cosine": 0.6776105213940143, + "relative_weight_rmse": 0.01106639119753695 + }, + { + "index": 240, + "total": 252, + "projection": "layers.8.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.414, + "parameters": 50331648, + "reference_delta_l2": 2.2063846516091514, + "retained_delta_l2": 1.6638270824407644, + "delta_retention_norm_ratio": 0.7540965630028784, + "delta_cosine": 0.5774548213994682, + "relative_weight_rmse": 0.009479961256890872 + }, + { + "index": 241, + "total": 252, + "projection": "layers.8.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3886, + "parameters": 50331648, + "reference_delta_l2": 2.6807214209526298, + "retained_delta_l2": 2.2855658790940634, + "delta_retention_norm_ratio": 0.8525935821715699, + "delta_cosine": 0.6658783264574605, + "relative_weight_rmse": 0.011430911218074097 + }, + { + "index": 242, + "total": 252, + "projection": "layers.8.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0319, + "parameters": 4194304, + "reference_delta_l2": 0.8937173848383411, + "retained_delta_l2": 0.8104368929187749, + "delta_retention_norm_ratio": 0.9068156294904901, + "delta_cosine": 0.7199882969838584, + "relative_weight_rmse": 0.012130448195302889 + }, + { + "index": 243, + "total": 252, + "projection": "layers.8.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1339, + "parameters": 16777216, + "reference_delta_l2": 1.5739130851067147, + "retained_delta_l2": 1.3486808133142838, + "delta_retention_norm_ratio": 0.8568966266792556, + "delta_cosine": 0.6708153509637819, + "relative_weight_rmse": 0.01131327048918274 + }, + { + "index": 244, + "total": 252, + "projection": "layers.8.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.135, + "parameters": 16777216, + "reference_delta_l2": 1.5947410494592953, + "retained_delta_l2": 1.3839922985120297, + "delta_retention_norm_ratio": 0.867847666542025, + "delta_cosine": 0.6792454533204817, + "relative_weight_rmse": 0.01183956423545222 + }, + { + "index": 245, + "total": 252, + "projection": "layers.8.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0321, + "parameters": 4194304, + "reference_delta_l2": 0.7962851912237163, + "retained_delta_l2": 0.6603594434695064, + "delta_retention_norm_ratio": 0.8293001687682754, + "delta_cosine": 0.6433962239937907, + "relative_weight_rmse": 0.010626495192800725 + }, + { + "index": 246, + "total": 252, + "projection": "layers.9.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.4674, + "parameters": 50331648, + "reference_delta_l2": 2.4696335096990847, + "retained_delta_l2": 2.098786160386188, + "delta_retention_norm_ratio": 0.8498370920800784, + "delta_cosine": 0.664259211083334, + "relative_weight_rmse": 0.010884464944285042 + }, + { + "index": 247, + "total": 252, + "projection": "layers.9.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.4325, + "parameters": 50331648, + "reference_delta_l2": 2.702771022309509, + "retained_delta_l2": 2.2144799807195312, + "delta_retention_norm_ratio": 0.8193368814600007, + "delta_cosine": 0.6354458596298957, + "relative_weight_rmse": 0.01038149804780606 + }, + { + "index": 248, + "total": 252, + "projection": "layers.9.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.4304, + "parameters": 50331648, + "reference_delta_l2": 3.0266516531031895, + "retained_delta_l2": 2.7453503666443257, + "delta_retention_norm_ratio": 0.9070585852949251, + "delta_cosine": 0.7220161726284356, + "relative_weight_rmse": 0.012388641790906324 + }, + { + "index": 249, + "total": 252, + "projection": "layers.9.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0354, + "parameters": 4194304, + "reference_delta_l2": 0.8388777296193708, + "retained_delta_l2": 0.7569098012929493, + "delta_retention_norm_ratio": 0.9022885869629496, + "delta_cosine": 0.7183301203636718, + "relative_weight_rmse": 0.011789731040025026 + }, + { + "index": 250, + "total": 252, + "projection": "layers.9.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1478, + "parameters": 16777216, + "reference_delta_l2": 1.5261649101131054, + "retained_delta_l2": 1.300762183663171, + "delta_retention_norm_ratio": 0.8523077519629058, + "delta_cosine": 0.6668761770880343, + "relative_weight_rmse": 0.011161277092899484 + }, + { + "index": 251, + "total": 252, + "projection": "layers.9.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.1474, + "parameters": 16777216, + "reference_delta_l2": 1.8048646015695378, + "retained_delta_l2": 1.6581732449153583, + "delta_retention_norm_ratio": 0.9187244536090883, + "delta_cosine": 0.7349623458503459, + "relative_weight_rmse": 0.0127022390017426 + }, + { + "index": 252, + "total": 252, + "projection": "layers.9.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0365, + "parameters": 4194304, + "reference_delta_l2": 0.8013131569957732, + "retained_delta_l2": 0.6711979224430136, + "delta_retention_norm_ratio": 0.8376224907618158, + "delta_cosine": 0.6536087656796943, + "relative_weight_rmse": 0.010429187562476724 + } + ] +}