diff --git "a/manifests/merge_step_00000510.json" "b/manifests/merge_step_00000510.json" new file mode 100644--- /dev/null +++ "b/manifests/merge_step_00000510.json" @@ -0,0 +1,4052 @@ +{ + "schema_version": 1, + "format": "comfy_scaled_fp8_merged_text_encoder", + "scale_policy": "preserve_stock_per_tensor_scale", + "base": "F:\\AI\\Ideogram\\checkpoints\\text_encoders\\qwen3vl_8b_fp8_scaled.safetensors", + "base_sha256": "4ba424cf62e51392e4d1a39933e803706f4e823c1065f36aaf149c6453f66bcd", + "adapter": "F:\\AI\\Ideogram\\artifacts\\vast_runs\\48495290\\recovery_s020_v1\\step_00000510\\adapters.safetensors", + "adapter_sha256": "85359811e619276f85742072c2b3327915bbb56bcefb874f2f3101a5d1216eef", + "adapter_global_step": "510", + "output": "F:\\AI\\Ideogram\\artifacts\\merged_text_encoders\\qwen3vl_8b_ideogram_nl_s020_v1_step_00000510_fp8_scaled.safetensors", + "output_bytes": 10588637800, + "output_sha256": "f0d1009af7407b60bda169967bb5e4c3e8ea264042b70f3d7b667bb0c2404ece", + "output_tensors": 1254, + "merged_projections": 252, + "lora_rank": 64, + "lora_alpha": 64.0, + "elapsed_seconds": 52.564, + "projections": [ + { + "index": 1, + "total": 252, + "projection": "layers.0.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.433, + "parameters": 50331648, + "reference_delta_l2": 2.6327089901753045, + "retained_delta_l2": 2.2871515016030406, + "delta_retention_norm_ratio": 0.8687445175817726, + "delta_cosine": 0.6831931704979608, + "relative_weight_rmse": 0.01096298133808025 + }, + { + "index": 2, + "total": 252, + "projection": "layers.0.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3243, + "parameters": 50331648, + "reference_delta_l2": 2.6772801535866004, + "retained_delta_l2": 2.2880149418926226, + "delta_retention_norm_ratio": 0.8546042291567801, + "delta_cosine": 0.6670250460348756, + "relative_weight_rmse": 0.011768324045049742 + }, + { + "index": 3, + "total": 252, + "projection": "layers.0.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3036, + "parameters": 50331648, + "reference_delta_l2": 2.390616730980462, + "retained_delta_l2": 1.9866289619366921, + "delta_retention_norm_ratio": 0.8310110676427489, + "delta_cosine": 0.6447662799695758, + "relative_weight_rmse": 0.01141205272124229 + }, + { + "index": 4, + "total": 252, + "projection": "layers.0.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.02, + "parameters": 4194304, + "reference_delta_l2": 0.5230720972912708, + "retained_delta_l2": 0.3567391071841315, + "delta_retention_norm_ratio": 0.6820075263649222, + "delta_cosine": 0.5185684276018468, + "relative_weight_rmse": 0.008049897840969543 + }, + { + "index": 5, + "total": 252, + "projection": "layers.0.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.0814, + "parameters": 16777216, + "reference_delta_l2": 1.8613124987703769, + "retained_delta_l2": 1.7369299527282984, + "delta_retention_norm_ratio": 0.9331748182402213, + "delta_cosine": 0.7526839574109756, + "relative_weight_rmse": 0.013067631665388612 + }, + { + "index": 6, + "total": 252, + "projection": "layers.0.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.0816, + "parameters": 16777216, + "reference_delta_l2": 1.0563514233516575, + "retained_delta_l2": 0.7668640299186094, + "delta_retention_norm_ratio": 0.725955409313934, + "delta_cosine": 0.5543170897702486, + "relative_weight_rmse": 0.008753209330524843 + }, + { + "index": 7, + "total": 252, + "projection": "layers.0.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0204, + "parameters": 4194304, + "reference_delta_l2": 0.7022345644480543, + "retained_delta_l2": 0.5874381003072418, + "delta_retention_norm_ratio": 0.8365268957801288, + "delta_cosine": 0.6512962468872778, + "relative_weight_rmse": 0.010973664045303274 + }, + { + "index": 8, + "total": 252, + "projection": "layers.1.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.3191, + "parameters": 50331648, + "reference_delta_l2": 1.8898138296216929, + "retained_delta_l2": 1.7377354284360245, + "delta_retention_norm_ratio": 0.919527310679004, + "delta_cosine": 0.7448556189453461, + "relative_weight_rmse": 0.011322768425377668 + }, + { + "index": 9, + "total": 252, + "projection": "layers.1.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.2841, + "parameters": 50331648, + "reference_delta_l2": 3.657081033192418, + "retained_delta_l2": 3.6015521990413126, + "delta_retention_norm_ratio": 0.984816077727807, + "delta_cosine": 0.8418804183137666, + "relative_weight_rmse": 0.013840827955486369 + }, + { + "index": 10, + "total": 252, + "projection": "layers.1.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.2966, + "parameters": 50331648, + "reference_delta_l2": 3.3758522814584864, + "retained_delta_l2": 3.348022721023458, + "delta_retention_norm_ratio": 0.9917562860828125, + "delta_cosine": 0.9015911746554313, + "relative_weight_rmse": 0.014812390015197048 + }, + { + "index": 11, + "total": 252, + "projection": "layers.1.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0195, + "parameters": 4194304, + "reference_delta_l2": 0.6051762921808931, + "retained_delta_l2": 0.4756470663655935, + "delta_retention_norm_ratio": 0.7859644743376992, + "delta_cosine": 0.6060727303822177, + "relative_weight_rmse": 0.009964429077215512 + }, + { + "index": 12, + "total": 252, + "projection": "layers.1.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.0896, + "parameters": 16777216, + "reference_delta_l2": 1.4935938757174607, + "retained_delta_l2": 1.2785725581027818, + "delta_retention_norm_ratio": 0.8560376276908664, + "delta_cosine": 0.6677743891393897, + "relative_weight_rmse": 0.011547649685206057 + }, + { + "index": 13, + "total": 252, + "projection": "layers.1.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.09, + "parameters": 16777216, + "reference_delta_l2": 1.8196863010939506, + "retained_delta_l2": 1.7101196678521942, + "delta_retention_norm_ratio": 0.9397881749310925, + "delta_cosine": 0.7595529659149882, + "relative_weight_rmse": 0.013550407565098701 + }, + { + "index": 14, + "total": 252, + "projection": "layers.1.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0222, + "parameters": 4194304, + "reference_delta_l2": 0.7615395905357294, + "retained_delta_l2": 0.6517298088924471, + "delta_retention_norm_ratio": 0.8558055510074886, + "delta_cosine": 0.6681908747332516, + "relative_weight_rmse": 0.011085857994289851 + }, + { + "index": 15, + "total": 252, + "projection": "layers.10.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.3549, + "parameters": 50331648, + "reference_delta_l2": 2.531215055123725, + "retained_delta_l2": 2.1824833997654514, + "delta_retention_norm_ratio": 0.8622275674868615, + "delta_cosine": 0.6764698570780457, + "relative_weight_rmse": 0.01081504682034425 + }, + { + "index": 16, + "total": 252, + "projection": "layers.10.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.2811, + "parameters": 50331648, + "reference_delta_l2": 2.1752664087061495, + "retained_delta_l2": 1.6278962284509517, + "delta_retention_norm_ratio": 0.7483663710962309, + "delta_cosine": 0.5733297280350798, + "relative_weight_rmse": 0.008945254391297452 + }, + { + "index": 17, + "total": 252, + "projection": "layers.10.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.2791, + "parameters": 50331648, + "reference_delta_l2": 2.666534835939869, + "retained_delta_l2": 2.314993607953368, + "delta_retention_norm_ratio": 0.8681655220669211, + "delta_cosine": 0.6808506874174365, + "relative_weight_rmse": 0.011346629690113269 + }, + { + "index": 18, + "total": 252, + "projection": "layers.10.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0192, + "parameters": 4194304, + "reference_delta_l2": 0.8598177590242261, + "retained_delta_l2": 0.7820980451680212, + "delta_retention_norm_ratio": 0.9096090851340335, + "delta_cosine": 0.7270726588216819, + "relative_weight_rmse": 0.011634985351763842 + }, + { + "index": 19, + "total": 252, + "projection": "layers.10.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.0875, + "parameters": 16777216, + "reference_delta_l2": 1.6733573890753237, + "retained_delta_l2": 1.4703711717179522, + "delta_retention_norm_ratio": 0.8786952394732966, + "delta_cosine": 0.6914507221186205, + "relative_weight_rmse": 0.011706738980892818 + }, + { + "index": 20, + "total": 252, + "projection": "layers.10.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.0864, + "parameters": 16777216, + "reference_delta_l2": 2.027281987832174, + "retained_delta_l2": 1.9186377269978645, + "delta_retention_norm_ratio": 0.9464089053785331, + "delta_cosine": 0.7704910918797055, + "relative_weight_rmse": 0.013102069197817139 + }, + { + "index": 21, + "total": 252, + "projection": "layers.10.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0199, + "parameters": 4194304, + "reference_delta_l2": 0.8179661631941778, + "retained_delta_l2": 0.7086405890578836, + "delta_retention_norm_ratio": 0.8663446251744997, + "delta_cosine": 0.6851426874031002, + "relative_weight_rmse": 0.010413497936023764 + }, + { + "index": 22, + "total": 252, + "projection": "layers.11.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.3252, + "parameters": 50331648, + "reference_delta_l2": 2.34500506277976, + "retained_delta_l2": 1.9435492747038492, + "delta_retention_norm_ratio": 0.8288038714935538, + "delta_cosine": 0.6447956411947068, + "relative_weight_rmse": 0.010210191435923827 + }, + { + "index": 23, + "total": 252, + "projection": "layers.11.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.2686, + "parameters": 50331648, + "reference_delta_l2": 2.5517632567783286, + "retained_delta_l2": 2.1012649758237125, + "delta_retention_norm_ratio": 0.8234560828642926, + "delta_cosine": 0.6404681709972034, + "relative_weight_rmse": 0.010146022133471601 + }, + { + "index": 24, + "total": 252, + "projection": "layers.11.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.2805, + "parameters": 50331648, + "reference_delta_l2": 2.4056315041822263, + "retained_delta_l2": 1.9713382785726346, + "delta_retention_norm_ratio": 0.8194681002245912, + "delta_cosine": 0.6359068240802416, + "relative_weight_rmse": 0.010523245330967618 + }, + { + "index": 25, + "total": 252, + "projection": "layers.11.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0193, + "parameters": 4194304, + "reference_delta_l2": 0.8617108467783874, + "retained_delta_l2": 0.7790011437447393, + "delta_retention_norm_ratio": 0.9040168713868828, + "delta_cosine": 0.7186376496198527, + "relative_weight_rmse": 0.011705958411568637 + }, + { + "index": 26, + "total": 252, + "projection": "layers.11.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.086, + "parameters": 16777216, + "reference_delta_l2": 1.5926304671746783, + "retained_delta_l2": 1.3857014199563344, + "delta_retention_norm_ratio": 0.8700708974973741, + "delta_cosine": 0.6847607651679083, + "relative_weight_rmse": 0.011298482092380215 + }, + { + "index": 27, + "total": 252, + "projection": "layers.11.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.0814, + "parameters": 16777216, + "reference_delta_l2": 1.2443009836930687, + "retained_delta_l2": 0.9880804106015009, + "delta_retention_norm_ratio": 0.7940847299412167, + "delta_cosine": 0.6130805571685117, + "relative_weight_rmse": 0.00981917000253394 + }, + { + "index": 28, + "total": 252, + "projection": "layers.11.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0183, + "parameters": 4194304, + "reference_delta_l2": 0.897717225216816, + "retained_delta_l2": 0.7899933152657633, + "delta_retention_norm_ratio": 0.8800024028445759, + "delta_cosine": 0.6969767151975313, + "relative_weight_rmse": 0.011141484998529136 + }, + { + "index": 29, + "total": 252, + "projection": "layers.12.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.3295, + "parameters": 50331648, + "reference_delta_l2": 2.320277198795607, + "retained_delta_l2": 1.9160150920903574, + "delta_retention_norm_ratio": 0.8257699093388104, + "delta_cosine": 0.6425239669217114, + "relative_weight_rmse": 0.010044208576518672 + }, + { + "index": 30, + "total": 252, + "projection": "layers.12.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.2946, + "parameters": 50331648, + "reference_delta_l2": 2.901090413343313, + "retained_delta_l2": 2.5319461880405907, + "delta_retention_norm_ratio": 0.8727567318809247, + "delta_cosine": 0.6884221568080938, + "relative_weight_rmse": 0.011163459806453873 + }, + { + "index": 31, + "total": 252, + "projection": "layers.12.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.2935, + "parameters": 50331648, + "reference_delta_l2": 2.4673848569621684, + "retained_delta_l2": 2.0462148511281737, + "delta_retention_norm_ratio": 0.8293051022642096, + "delta_cosine": 0.6449839710058615, + "relative_weight_rmse": 0.010537524426169257 + }, + { + "index": 32, + "total": 252, + "projection": "layers.12.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0195, + "parameters": 4194304, + "reference_delta_l2": 0.8282236259084472, + "retained_delta_l2": 0.7464401813193698, + "delta_retention_norm_ratio": 0.9012543931002062, + "delta_cosine": 0.7199601632722067, + "relative_weight_rmse": 0.01138058043089329 + }, + { + "index": 33, + "total": 252, + "projection": "layers.12.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.0907, + "parameters": 16777216, + "reference_delta_l2": 1.6453911434628496, + "retained_delta_l2": 1.463020090119954, + "delta_retention_norm_ratio": 0.8891624924155833, + "delta_cosine": 0.703984320068952, + "relative_weight_rmse": 0.01144197761146913 + }, + { + "index": 34, + "total": 252, + "projection": "layers.12.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.0883, + "parameters": 16777216, + "reference_delta_l2": 1.6210399311215336, + "retained_delta_l2": 1.4405291201316213, + "delta_retention_norm_ratio": 0.8886450558531127, + "delta_cosine": 0.7040551826089303, + "relative_weight_rmse": 0.011592997050092375 + }, + { + "index": 35, + "total": 252, + "projection": "layers.12.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.02, + "parameters": 4194304, + "reference_delta_l2": 1.0237886771272462, + "retained_delta_l2": 0.9535994191334317, + "delta_retention_norm_ratio": 0.9314416543551098, + "delta_cosine": 0.7544783126687498, + "relative_weight_rmse": 0.011710835840793916 + }, + { + "index": 36, + "total": 252, + "projection": "layers.13.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.3261, + "parameters": 50331648, + "reference_delta_l2": 2.0846221826380544, + "retained_delta_l2": 1.6348804269860413, + "delta_retention_norm_ratio": 0.7842574259269981, + "delta_cosine": 0.6048871053089986, + "relative_weight_rmse": 0.009322682612643638 + }, + { + "index": 37, + "total": 252, + "projection": "layers.13.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.2838, + "parameters": 50331648, + "reference_delta_l2": 2.5562534815687137, + "retained_delta_l2": 2.1167548474306748, + "delta_retention_norm_ratio": 0.8280692281469956, + "delta_cosine": 0.6432194070144006, + "relative_weight_rmse": 0.010677268993343826 + }, + { + "index": 38, + "total": 252, + "projection": "layers.13.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.2769, + "parameters": 50331648, + "reference_delta_l2": 2.203330089333929, + "retained_delta_l2": 1.726298107952274, + "delta_retention_norm_ratio": 0.7834949998228078, + "delta_cosine": 0.6030605391711824, + "relative_weight_rmse": 0.009684847344538314 + }, + { + "index": 39, + "total": 252, + "projection": "layers.13.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0205, + "parameters": 4194304, + "reference_delta_l2": 1.2882447085921234, + "retained_delta_l2": 1.2714993722220769, + "delta_retention_norm_ratio": 0.9870014320583961, + "delta_cosine": 0.8276057019002796, + "relative_weight_rmse": 0.014468699445840155 + }, + { + "index": 40, + "total": 252, + "projection": "layers.13.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.0838, + "parameters": 16777216, + "reference_delta_l2": 1.6166214467944622, + "retained_delta_l2": 1.42378514680623, + "delta_retention_norm_ratio": 0.8807164779543163, + "delta_cosine": 0.6955817661003701, + "relative_weight_rmse": 0.01170919935545192 + }, + { + "index": 41, + "total": 252, + "projection": "layers.13.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.0806, + "parameters": 16777216, + "reference_delta_l2": 1.2623715176154386, + "retained_delta_l2": 1.0099243645887293, + "delta_retention_norm_ratio": 0.8000215075324495, + "delta_cosine": 0.6184279126413518, + "relative_weight_rmse": 0.00989565906685087 + }, + { + "index": 42, + "total": 252, + "projection": "layers.13.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0183, + "parameters": 4194304, + "reference_delta_l2": 0.8825092886380478, + "retained_delta_l2": 0.7817498199472648, + "delta_retention_norm_ratio": 0.8858261663780533, + "delta_cosine": 0.7006507079130242, + "relative_weight_rmse": 0.011154325923963038 + }, + { + "index": 43, + "total": 252, + "projection": "layers.14.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.3534, + "parameters": 50331648, + "reference_delta_l2": 2.13420138255605, + "retained_delta_l2": 1.7189831026670974, + "delta_retention_norm_ratio": 0.8054455951145238, + "delta_cosine": 0.6249848772544618, + "relative_weight_rmse": 0.009436160658905499 + }, + { + "index": 44, + "total": 252, + "projection": "layers.14.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.2954, + "parameters": 50331648, + "reference_delta_l2": 2.4254956227466296, + "retained_delta_l2": 1.9749517522158966, + "delta_retention_norm_ratio": 0.8142466775427376, + "delta_cosine": 0.6301317568289051, + "relative_weight_rmse": 0.010409978211722815 + }, + { + "index": 45, + "total": 252, + "projection": "layers.14.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3067, + "parameters": 50331648, + "reference_delta_l2": 2.651475706897975, + "retained_delta_l2": 2.2754130914762443, + "delta_retention_norm_ratio": 0.858168560834489, + "delta_cosine": 0.6730766377166094, + "relative_weight_rmse": 0.0109756270465385 + }, + { + "index": 46, + "total": 252, + "projection": "layers.14.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0223, + "parameters": 4194304, + "reference_delta_l2": 0.6661733648606436, + "retained_delta_l2": 0.5479984811762695, + "delta_retention_norm_ratio": 0.8226064116071422, + "delta_cosine": 0.6406453360097045, + "relative_weight_rmse": 0.010066646015560256 + }, + { + "index": 47, + "total": 252, + "projection": "layers.14.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.0832, + "parameters": 16777216, + "reference_delta_l2": 1.4582571531833697, + "retained_delta_l2": 1.2460505381967344, + "delta_retention_norm_ratio": 0.854479290896404, + "delta_cosine": 0.6694921418838832, + "relative_weight_rmse": 0.010918642035532514 + }, + { + "index": 48, + "total": 252, + "projection": "layers.14.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.0826, + "parameters": 16777216, + "reference_delta_l2": 2.1856064529455286, + "retained_delta_l2": 2.10889967923761, + "delta_retention_norm_ratio": 0.9649036661634387, + "delta_cosine": 0.7919398570664227, + "relative_weight_rmse": 0.01350813377663138 + }, + { + "index": 49, + "total": 252, + "projection": "layers.14.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0191, + "parameters": 4194304, + "reference_delta_l2": 0.9013074414020275, + "retained_delta_l2": 0.8098390973556806, + "delta_retention_norm_ratio": 0.8985159338038268, + "delta_cosine": 0.7149864158653795, + "relative_weight_rmse": 0.011479685591920221 + }, + { + "index": 50, + "total": 252, + "projection": "layers.15.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.3207, + "parameters": 50331648, + "reference_delta_l2": 2.086789207992008, + "retained_delta_l2": 1.659957545345398, + "delta_retention_norm_ratio": 0.7954600967783783, + "delta_cosine": 0.6150906974827606, + "relative_weight_rmse": 0.00935880829300255 + }, + { + "index": 51, + "total": 252, + "projection": "layers.15.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.2859, + "parameters": 50331648, + "reference_delta_l2": 2.4256639917796314, + "retained_delta_l2": 2.004613189385262, + "delta_retention_norm_ratio": 0.8264183317140072, + "delta_cosine": 0.6413103507947647, + "relative_weight_rmse": 0.010663762745827537 + }, + { + "index": 52, + "total": 252, + "projection": "layers.15.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.2769, + "parameters": 50331648, + "reference_delta_l2": 2.595704581856843, + "retained_delta_l2": 2.2266853540778584, + "delta_retention_norm_ratio": 0.8578346587056507, + "delta_cosine": 0.672538349064441, + "relative_weight_rmse": 0.010848416048602223 + }, + { + "index": 53, + "total": 252, + "projection": "layers.15.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0206, + "parameters": 4194304, + "reference_delta_l2": 0.7174737517442046, + "retained_delta_l2": 0.6089911363066031, + "delta_retention_norm_ratio": 0.8487991857905933, + "delta_cosine": 0.662850053918743, + "relative_weight_rmse": 0.010668703613887207 + }, + { + "index": 54, + "total": 252, + "projection": "layers.15.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.0837, + "parameters": 16777216, + "reference_delta_l2": 2.0240964109023434, + "retained_delta_l2": 1.9264964091557202, + "delta_retention_norm_ratio": 0.9517809521221802, + "delta_cosine": 0.7758733106480629, + "relative_weight_rmse": 0.013361216275558942 + }, + { + "index": 55, + "total": 252, + "projection": "layers.15.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.0813, + "parameters": 16777216, + "reference_delta_l2": 2.1135189974599835, + "retained_delta_l2": 2.02665696980818, + "delta_retention_norm_ratio": 0.9589017048078612, + "delta_cosine": 0.7916011745445279, + "relative_weight_rmse": 0.012965520219919589 + }, + { + "index": 56, + "total": 252, + "projection": "layers.15.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0197, + "parameters": 4194304, + "reference_delta_l2": 0.7616490923373335, + "retained_delta_l2": 0.6541228230333136, + "delta_retention_norm_ratio": 0.8588243977629574, + "delta_cosine": 0.6752154346427254, + "relative_weight_rmse": 0.010342522352985545 + }, + { + "index": 57, + "total": 252, + "projection": "layers.16.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.3305, + "parameters": 50331648, + "reference_delta_l2": 2.2303157381509173, + "retained_delta_l2": 1.8342983474627628, + "delta_retention_norm_ratio": 0.8224388664286251, + "delta_cosine": 0.639472222010357, + "relative_weight_rmse": 0.009852833359225589 + }, + { + "index": 58, + "total": 252, + "projection": "layers.16.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.2731, + "parameters": 50331648, + "reference_delta_l2": 2.4166462281703205, + "retained_delta_l2": 2.0057662695293548, + "delta_retention_norm_ratio": 0.8299792688514244, + "delta_cosine": 0.6450837084917049, + "relative_weight_rmse": 0.010391807503082114 + }, + { + "index": 59, + "total": 252, + "projection": "layers.16.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.2808, + "parameters": 50331648, + "reference_delta_l2": 2.3920914943172247, + "retained_delta_l2": 1.9746787002661033, + "delta_retention_norm_ratio": 0.8255029980906882, + "delta_cosine": 0.640919380175158, + "relative_weight_rmse": 0.010358761010515978 + }, + { + "index": 60, + "total": 252, + "projection": "layers.16.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0201, + "parameters": 4194304, + "reference_delta_l2": 0.6590984356455544, + "retained_delta_l2": 0.5477319763947867, + "delta_retention_norm_ratio": 0.8310321293029778, + "delta_cosine": 0.6469395956324974, + "relative_weight_rmse": 0.010300532914943077 + }, + { + "index": 61, + "total": 252, + "projection": "layers.16.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.0821, + "parameters": 16777216, + "reference_delta_l2": 1.5777249464740992, + "retained_delta_l2": 1.389874264125338, + "delta_retention_norm_ratio": 0.8809357215473014, + "delta_cosine": 0.6947001732960237, + "relative_weight_rmse": 0.011524895889996816 + }, + { + "index": 62, + "total": 252, + "projection": "layers.16.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.0827, + "parameters": 16777216, + "reference_delta_l2": 1.7397412979225706, + "retained_delta_l2": 1.599333859427975, + "delta_retention_norm_ratio": 0.9192940705251657, + "delta_cosine": 0.7386192162456077, + "relative_weight_rmse": 0.01216112369416335 + }, + { + "index": 63, + "total": 252, + "projection": "layers.16.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.019, + "parameters": 4194304, + "reference_delta_l2": 0.8530860626636955, + "retained_delta_l2": 0.7554267831913181, + "delta_retention_norm_ratio": 0.8855223596462897, + "delta_cosine": 0.7018652535197768, + "relative_weight_rmse": 0.011241759488540051 + }, + { + "index": 64, + "total": 252, + "projection": "layers.17.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.3266, + "parameters": 50331648, + "reference_delta_l2": 2.317107496556227, + "retained_delta_l2": 1.924559504330395, + "delta_retention_norm_ratio": 0.8305870604582432, + "delta_cosine": 0.6468727141506057, + "relative_weight_rmse": 0.010114956868444772 + }, + { + "index": 65, + "total": 252, + "projection": "layers.17.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.2722, + "parameters": 50331648, + "reference_delta_l2": 3.086873344951891, + "retained_delta_l2": 2.7992915749567815, + "delta_retention_norm_ratio": 0.9068371980776584, + "delta_cosine": 0.722672586364596, + "relative_weight_rmse": 0.012225240939740489 + }, + { + "index": 66, + "total": 252, + "projection": "layers.17.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.2746, + "parameters": 50331648, + "reference_delta_l2": 2.9036635508281687, + "retained_delta_l2": 2.579993038764108, + "delta_retention_norm_ratio": 0.8885302975367978, + "delta_cosine": 0.7026662603657728, + "relative_weight_rmse": 0.01164404960770755 + }, + { + "index": 67, + "total": 252, + "projection": "layers.17.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0196, + "parameters": 4194304, + "reference_delta_l2": 0.7066499516158611, + "retained_delta_l2": 0.6160339511399299, + "delta_retention_norm_ratio": 0.8717667774989241, + "delta_cosine": 0.689884095292728, + "relative_weight_rmse": 0.010721400388771996 + }, + { + "index": 68, + "total": 252, + "projection": "layers.17.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.0823, + "parameters": 16777216, + "reference_delta_l2": 1.6344810946602308, + "retained_delta_l2": 1.4619995818789637, + "delta_retention_norm_ratio": 0.89447322863216, + "delta_cosine": 0.7109335062500769, + "relative_weight_rmse": 0.011753573182973771 + }, + { + "index": 69, + "total": 252, + "projection": "layers.17.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.0817, + "parameters": 16777216, + "reference_delta_l2": 1.4864809414200086, + "retained_delta_l2": 1.2848595929969922, + "delta_retention_norm_ratio": 0.864363314183893, + "delta_cosine": 0.6769874056184836, + "relative_weight_rmse": 0.01113307097335904 + }, + { + "index": 70, + "total": 252, + "projection": "layers.17.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.02, + "parameters": 4194304, + "reference_delta_l2": 0.9355528627118188, + "retained_delta_l2": 0.8580613997950565, + "delta_retention_norm_ratio": 0.9171704069269334, + "delta_cosine": 0.7352963378442325, + "relative_weight_rmse": 0.012006218333944534 + }, + { + "index": 71, + "total": 252, + "projection": "layers.18.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.4035, + "parameters": 50331648, + "reference_delta_l2": 2.623621093517205, + "retained_delta_l2": 2.29715425491584, + "delta_retention_norm_ratio": 0.8755663158037403, + "delta_cosine": 0.6895082587554823, + "relative_weight_rmse": 0.01092384570817268 + }, + { + "index": 72, + "total": 252, + "projection": "layers.18.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.2952, + "parameters": 50331648, + "reference_delta_l2": 3.9516853977153295, + "retained_delta_l2": 3.814789749658205, + "delta_retention_norm_ratio": 0.9653576552080105, + "delta_cosine": 0.7899677144921917, + "relative_weight_rmse": 0.014049752567654311 + }, + { + "index": 73, + "total": 252, + "projection": "layers.18.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.2975, + "parameters": 50331648, + "reference_delta_l2": 2.975283543456475, + "retained_delta_l2": 2.672285012665405, + "delta_retention_norm_ratio": 0.898161460457289, + "delta_cosine": 0.7131976175914162, + "relative_weight_rmse": 0.011765179844720721 + }, + { + "index": 74, + "total": 252, + "projection": "layers.18.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0208, + "parameters": 4194304, + "reference_delta_l2": 1.1137359991025886, + "retained_delta_l2": 1.0817885328986219, + "delta_retention_norm_ratio": 0.9713150457292312, + "delta_cosine": 0.8014797022907222, + "relative_weight_rmse": 0.013870573203487047 + }, + { + "index": 75, + "total": 252, + "projection": "layers.18.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.0929, + "parameters": 16777216, + "reference_delta_l2": 1.5162794257656762, + "retained_delta_l2": 1.328227386411706, + "delta_retention_norm_ratio": 0.8759779786242172, + "delta_cosine": 0.6919780168741544, + "relative_weight_rmse": 0.01116224109623716 + }, + { + "index": 76, + "total": 252, + "projection": "layers.18.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.0881, + "parameters": 16777216, + "reference_delta_l2": 1.8813425857431623, + "retained_delta_l2": 1.761399415184713, + "delta_retention_norm_ratio": 0.9362459705811264, + "delta_cosine": 0.7578788430641704, + "relative_weight_rmse": 0.012728323169903293 + }, + { + "index": 77, + "total": 252, + "projection": "layers.18.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0226, + "parameters": 4194304, + "reference_delta_l2": 0.8633863260109812, + "retained_delta_l2": 0.7703261769314331, + "delta_retention_norm_ratio": 0.8922149375361261, + "delta_cosine": 0.708406561753843, + "relative_weight_rmse": 0.011269247563962978 + }, + { + "index": 78, + "total": 252, + "projection": "layers.19.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.319, + "parameters": 50331648, + "reference_delta_l2": 2.5417637873727275, + "retained_delta_l2": 2.194293303255207, + "delta_retention_norm_ratio": 0.8632955250036511, + "delta_cosine": 0.6779440790016057, + "relative_weight_rmse": 0.010722249439618881 + }, + { + "index": 79, + "total": 252, + "projection": "layers.19.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.2698, + "parameters": 50331648, + "reference_delta_l2": 3.6374279097339968, + "retained_delta_l2": 3.439368567901866, + "delta_retention_norm_ratio": 0.9455496172715586, + "delta_cosine": 0.7647211066286131, + "relative_weight_rmse": 0.013621033733582259 + }, + { + "index": 80, + "total": 252, + "projection": "layers.19.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.2725, + "parameters": 50331648, + "reference_delta_l2": 2.762804194662718, + "retained_delta_l2": 2.4001113822471773, + "delta_retention_norm_ratio": 0.8687229398608113, + "delta_cosine": 0.6832749515460651, + "relative_weight_rmse": 0.011285681848780737 + }, + { + "index": 81, + "total": 252, + "projection": "layers.19.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0181, + "parameters": 4194304, + "reference_delta_l2": 0.911919532599503, + "retained_delta_l2": 0.8576706193518299, + "delta_retention_norm_ratio": 0.9405112936960216, + "delta_cosine": 0.7652673572184512, + "relative_weight_rmse": 0.012584524885586413 + }, + { + "index": 82, + "total": 252, + "projection": "layers.19.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.0853, + "parameters": 16777216, + "reference_delta_l2": 1.5551248854139976, + "retained_delta_l2": 1.3691271157560179, + "delta_retention_norm_ratio": 0.8803968919779299, + "delta_cosine": 0.695319870840718, + "relative_weight_rmse": 0.011327888696033118 + }, + { + "index": 83, + "total": 252, + "projection": "layers.19.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.0853, + "parameters": 16777216, + "reference_delta_l2": 1.755104352838205, + "retained_delta_l2": 1.6199177558526192, + "delta_retention_norm_ratio": 0.9229751799276359, + "delta_cosine": 0.7423155567664863, + "relative_weight_rmse": 0.01208754545011881 + }, + { + "index": 84, + "total": 252, + "projection": "layers.19.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0182, + "parameters": 4194304, + "reference_delta_l2": 1.0831238658546205, + "retained_delta_l2": 1.0280764178780222, + "delta_retention_norm_ratio": 0.9491771442658, + "delta_cosine": 0.7753840719743617, + "relative_weight_rmse": 0.01294070070702427 + }, + { + "index": 85, + "total": 252, + "projection": "layers.2.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.3149, + "parameters": 50331648, + "reference_delta_l2": 1.4234882918656107, + "retained_delta_l2": 1.2061008715333281, + "delta_retention_norm_ratio": 0.8472854173971627, + "delta_cosine": 0.6863852764857693, + "relative_weight_rmse": 0.009407109661443678 + }, + { + "index": 86, + "total": 252, + "projection": "layers.2.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.2926, + "parameters": 50331648, + "reference_delta_l2": 2.191820181261131, + "retained_delta_l2": 2.011655969514262, + "delta_retention_norm_ratio": 0.9178015544855482, + "delta_cosine": 0.7659305807977074, + "relative_weight_rmse": 0.011770580379080117 + }, + { + "index": 87, + "total": 252, + "projection": "layers.2.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.2919, + "parameters": 50331648, + "reference_delta_l2": 2.761679938522529, + "retained_delta_l2": 2.697117560202633, + "delta_retention_norm_ratio": 0.9766220634696589, + "delta_cosine": 0.8389325810813097, + "relative_weight_rmse": 0.01405361187597274 + }, + { + "index": 88, + "total": 252, + "projection": "layers.2.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0249, + "parameters": 4194304, + "reference_delta_l2": 0.530888361819072, + "retained_delta_l2": 0.38520761860965247, + "delta_retention_norm_ratio": 0.7255906256632768, + "delta_cosine": 0.554529133357542, + "relative_weight_rmse": 0.008682646027018926 + }, + { + "index": 89, + "total": 252, + "projection": "layers.2.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.0892, + "parameters": 16777216, + "reference_delta_l2": 1.7678602103562455, + "retained_delta_l2": 1.600839620127675, + "delta_retention_norm_ratio": 0.9055238704677256, + "delta_cosine": 0.7176664166780741, + "relative_weight_rmse": 0.012641181982004271 + }, + { + "index": 90, + "total": 252, + "projection": "layers.2.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.0892, + "parameters": 16777216, + "reference_delta_l2": 1.1507312444501292, + "retained_delta_l2": 0.8859188775360538, + "delta_retention_norm_ratio": 0.7698747051570546, + "delta_cosine": 0.5907894243743868, + "relative_weight_rmse": 0.009674117713943687 + }, + { + "index": 91, + "total": 252, + "projection": "layers.2.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0207, + "parameters": 4194304, + "reference_delta_l2": 1.0904290529480396, + "retained_delta_l2": 1.0394254308026631, + "delta_retention_norm_ratio": 0.9532260975553751, + "delta_cosine": 0.7800620905527914, + "relative_weight_rmse": 0.013025487112775229 + }, + { + "index": 92, + "total": 252, + "projection": "layers.20.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.3396, + "parameters": 50331648, + "reference_delta_l2": 2.4208288436366288, + "retained_delta_l2": 2.0116675785458877, + "delta_retention_norm_ratio": 0.8309829849531664, + "delta_cosine": 0.6451165140730315, + "relative_weight_rmse": 0.010519303702515534 + }, + { + "index": 93, + "total": 252, + "projection": "layers.20.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.2761, + "parameters": 50331648, + "reference_delta_l2": 3.4874710954715287, + "retained_delta_l2": 3.2772327189895423, + "delta_retention_norm_ratio": 0.9397160949219105, + "delta_cosine": 0.7579808679328547, + "relative_weight_rmse": 0.013336498890557155 + }, + { + "index": 94, + "total": 252, + "projection": "layers.20.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.2758, + "parameters": 50331648, + "reference_delta_l2": 2.727859600528451, + "retained_delta_l2": 2.3557936631685323, + "delta_retention_norm_ratio": 0.8636051733425574, + "delta_cosine": 0.6782957869277911, + "relative_weight_rmse": 0.011179866765553926 + }, + { + "index": 95, + "total": 252, + "projection": "layers.20.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0191, + "parameters": 4194304, + "reference_delta_l2": 0.7589017611994083, + "retained_delta_l2": 0.6787277545565447, + "delta_retention_norm_ratio": 0.8943552239012432, + "delta_cosine": 0.7128861863951973, + "relative_weight_rmse": 0.011415075810978939 + }, + { + "index": 96, + "total": 252, + "projection": "layers.20.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.0842, + "parameters": 16777216, + "reference_delta_l2": 1.5289282697500297, + "retained_delta_l2": 1.3397865955364099, + "delta_retention_norm_ratio": 0.8762913356003659, + "delta_cosine": 0.6917474952652859, + "relative_weight_rmse": 0.011402773641846947 + }, + { + "index": 97, + "total": 252, + "projection": "layers.20.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.0806, + "parameters": 16777216, + "reference_delta_l2": 1.6938623059639735, + "retained_delta_l2": 1.541284460229232, + "delta_retention_norm_ratio": 0.9099231116971402, + "delta_cosine": 0.7301091098716266, + "relative_weight_rmse": 0.011523558952480195 + }, + { + "index": 98, + "total": 252, + "projection": "layers.20.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0185, + "parameters": 4194304, + "reference_delta_l2": 0.9129813777821141, + "retained_delta_l2": 0.8319464001453759, + "delta_retention_norm_ratio": 0.9112413685440172, + "delta_cosine": 0.7293886338239729, + "relative_weight_rmse": 0.011760719644457438 + }, + { + "index": 99, + "total": 252, + "projection": "layers.21.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.3265, + "parameters": 50331648, + "reference_delta_l2": 2.8313863577431513, + "retained_delta_l2": 2.4949310729952527, + "delta_retention_norm_ratio": 0.8811694194161191, + "delta_cosine": 0.6942913392598284, + "relative_weight_rmse": 0.011540670638753315 + }, + { + "index": 100, + "total": 252, + "projection": "layers.21.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.2996, + "parameters": 50331648, + "reference_delta_l2": 3.581268905576441, + "retained_delta_l2": 3.38096570423564, + "delta_retention_norm_ratio": 0.9440692093718774, + "delta_cosine": 0.762439197397351, + "relative_weight_rmse": 0.013782512921707265 + }, + { + "index": 101, + "total": 252, + "projection": "layers.21.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.2956, + "parameters": 50331648, + "reference_delta_l2": 2.940294335586719, + "retained_delta_l2": 2.60905610019324, + "delta_retention_norm_ratio": 0.887345211877442, + "delta_cosine": 0.7013493809133563, + "relative_weight_rmse": 0.011735881806793533 + }, + { + "index": 102, + "total": 252, + "projection": "layers.21.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0204, + "parameters": 4194304, + "reference_delta_l2": 0.77410035838613, + "retained_delta_l2": 0.6893916529822197, + "delta_retention_norm_ratio": 0.8905714168890017, + "delta_cosine": 0.7066130616326094, + "relative_weight_rmse": 0.01174789561808868 + }, + { + "index": 103, + "total": 252, + "projection": "layers.21.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.0915, + "parameters": 16777216, + "reference_delta_l2": 1.639992732532299, + "retained_delta_l2": 1.4793613166097839, + "delta_retention_norm_ratio": 0.9020535806433206, + "delta_cosine": 0.7189615443238899, + "relative_weight_rmse": 0.01192240297118375 + }, + { + "index": 104, + "total": 252, + "projection": "layers.21.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.091, + "parameters": 16777216, + "reference_delta_l2": 1.2205028435765317, + "retained_delta_l2": 0.9755466710777216, + "delta_retention_norm_ratio": 0.7992989743628973, + "delta_cosine": 0.6191341941737905, + "relative_weight_rmse": 0.009587733282748874 + }, + { + "index": 105, + "total": 252, + "projection": "layers.21.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0218, + "parameters": 4194304, + "reference_delta_l2": 1.1860379824198768, + "retained_delta_l2": 1.1488803099808462, + "delta_retention_norm_ratio": 0.9686707567634406, + "delta_cosine": 0.7978091745973054, + "relative_weight_rmse": 0.01366196800146398 + }, + { + "index": 106, + "total": 252, + "projection": "layers.22.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.341, + "parameters": 50331648, + "reference_delta_l2": 2.9910598559406982, + "retained_delta_l2": 2.701743807454628, + "delta_retention_norm_ratio": 0.9032730662639717, + "delta_cosine": 0.7181042575501958, + "relative_weight_rmse": 0.011729870238563058 + }, + { + "index": 107, + "total": 252, + "projection": "layers.22.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.2813, + "parameters": 50331648, + "reference_delta_l2": 3.777630486236153, + "retained_delta_l2": 3.609844765563156, + "delta_retention_norm_ratio": 0.9555844010460189, + "delta_cosine": 0.7793781225783791, + "relative_weight_rmse": 0.013768834214939765 + }, + { + "index": 108, + "total": 252, + "projection": "layers.22.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.2689, + "parameters": 50331648, + "reference_delta_l2": 3.1620192849135553, + "retained_delta_l2": 2.876731912335432, + "delta_retention_norm_ratio": 0.9097768397747383, + "delta_cosine": 0.7257917626151396, + "relative_weight_rmse": 0.011995691529653844 + }, + { + "index": 109, + "total": 252, + "projection": "layers.22.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0187, + "parameters": 4194304, + "reference_delta_l2": 0.9980371973618855, + "retained_delta_l2": 0.9581176806419454, + "delta_retention_norm_ratio": 0.960001975051171, + "delta_cosine": 0.7933572727606913, + "relative_weight_rmse": 0.01350631395620473 + }, + { + "index": 110, + "total": 252, + "projection": "layers.22.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.0855, + "parameters": 16777216, + "reference_delta_l2": 2.0063344121830644, + "retained_delta_l2": 1.893268654854589, + "delta_retention_norm_ratio": 0.9436456073115698, + "delta_cosine": 0.763043309624487, + "relative_weight_rmse": 0.01336187079091329 + }, + { + "index": 111, + "total": 252, + "projection": "layers.22.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.083, + "parameters": 16777216, + "reference_delta_l2": 1.8013330279960227, + "retained_delta_l2": 1.6650094943185272, + "delta_retention_norm_ratio": 0.9243207493790558, + "delta_cosine": 0.747278912792796, + "relative_weight_rmse": 0.012280580099934449 + }, + { + "index": 112, + "total": 252, + "projection": "layers.22.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0187, + "parameters": 4194304, + "reference_delta_l2": 1.059150396618244, + "retained_delta_l2": 1.0033485870682053, + "delta_retention_norm_ratio": 0.9473145553943916, + "delta_cosine": 0.7703772546384692, + "relative_weight_rmse": 0.012863320602021833 + }, + { + "index": 113, + "total": 252, + "projection": "layers.23.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.3371, + "parameters": 50331648, + "reference_delta_l2": 3.0230259828536648, + "retained_delta_l2": 2.7483078798014526, + "delta_retention_norm_ratio": 0.9091247959460523, + "delta_cosine": 0.7250317710698788, + "relative_weight_rmse": 0.011629094049505573 + }, + { + "index": 114, + "total": 252, + "projection": "layers.23.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.2919, + "parameters": 50331648, + "reference_delta_l2": 5.000263946159526, + "retained_delta_l2": 4.980002093083772, + "delta_retention_norm_ratio": 0.9959478432950891, + "delta_cosine": 0.831809651737954, + "relative_weight_rmse": 0.015881822890547393 + }, + { + "index": 115, + "total": 252, + "projection": "layers.23.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.2924, + "parameters": 50331648, + "reference_delta_l2": 4.454117936701333, + "retained_delta_l2": 4.354295531700759, + "delta_retention_norm_ratio": 0.9775887377884518, + "delta_cosine": 0.8063541037430465, + "relative_weight_rmse": 0.014560552787199508 + }, + { + "index": 116, + "total": 252, + "projection": "layers.23.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0207, + "parameters": 4194304, + "reference_delta_l2": 0.7266939802548371, + "retained_delta_l2": 0.630308449034524, + "delta_retention_norm_ratio": 0.8673643461495131, + "delta_cosine": 0.6826958743765825, + "relative_weight_rmse": 0.011273850499938792 + }, + { + "index": 117, + "total": 252, + "projection": "layers.23.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.0923, + "parameters": 16777216, + "reference_delta_l2": 1.8946229502748384, + "retained_delta_l2": 1.7597239440078793, + "delta_retention_norm_ratio": 0.9287990223873356, + "delta_cosine": 0.7459816596101202, + "relative_weight_rmse": 0.012561695261136123 + }, + { + "index": 118, + "total": 252, + "projection": "layers.23.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.0899, + "parameters": 16777216, + "reference_delta_l2": 1.7125996196835842, + "retained_delta_l2": 1.5415253508809772, + "delta_retention_norm_ratio": 0.9001084276579401, + "delta_cosine": 0.7177602975638901, + "relative_weight_rmse": 0.011632013912543447 + }, + { + "index": 119, + "total": 252, + "projection": "layers.23.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0217, + "parameters": 4194304, + "reference_delta_l2": 0.8868773646670587, + "retained_delta_l2": 0.8089120451245109, + "delta_retention_norm_ratio": 0.9120900784610545, + "delta_cosine": 0.7316220241406097, + "relative_weight_rmse": 0.011318348079675751 + }, + { + "index": 120, + "total": 252, + "projection": "layers.24.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.3394, + "parameters": 50331648, + "reference_delta_l2": 3.1518128058745902, + "retained_delta_l2": 2.8689717395460304, + "delta_retention_norm_ratio": 0.9102608296402062, + "delta_cosine": 0.7251792360856227, + "relative_weight_rmse": 0.012012265245717246 + }, + { + "index": 121, + "total": 252, + "projection": "layers.24.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.2724, + "parameters": 50331648, + "reference_delta_l2": 4.587213119030321, + "retained_delta_l2": 4.488477555342335, + "delta_retention_norm_ratio": 0.9784759153050082, + "delta_cosine": 0.8046703996672511, + "relative_weight_rmse": 0.015356809746409182 + }, + { + "index": 122, + "total": 252, + "projection": "layers.24.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.2736, + "parameters": 50331648, + "reference_delta_l2": 3.7073519293553274, + "retained_delta_l2": 3.482757360249292, + "delta_retention_norm_ratio": 0.9394191397564217, + "delta_cosine": 0.7562527399791866, + "relative_weight_rmse": 0.0133520489083886 + }, + { + "index": 123, + "total": 252, + "projection": "layers.24.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0213, + "parameters": 4194304, + "reference_delta_l2": 0.8845529687518748, + "retained_delta_l2": 0.8227142190408355, + "delta_retention_norm_ratio": 0.9300903937971117, + "delta_cosine": 0.7504080474234667, + "relative_weight_rmse": 0.012512043117811553 + }, + { + "index": 124, + "total": 252, + "projection": "layers.24.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.0942, + "parameters": 16777216, + "reference_delta_l2": 1.8177560952171528, + "retained_delta_l2": 1.6660417920596295, + "delta_retention_norm_ratio": 0.9165375907379922, + "delta_cosine": 0.7337509812864685, + "relative_weight_rmse": 0.012155999048714632 + }, + { + "index": 125, + "total": 252, + "projection": "layers.24.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.087, + "parameters": 16777216, + "reference_delta_l2": 2.9053735259013806, + "retained_delta_l2": 2.8934678342811617, + "delta_retention_norm_ratio": 0.9959021821070234, + "delta_cosine": 0.843715608256805, + "relative_weight_rmse": 0.01518422659127935 + }, + { + "index": 126, + "total": 252, + "projection": "layers.24.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0188, + "parameters": 4194304, + "reference_delta_l2": 1.0337579943004718, + "retained_delta_l2": 0.9772839036433504, + "delta_retention_norm_ratio": 0.9453701050260448, + "delta_cosine": 0.769169087266629, + "relative_weight_rmse": 0.012635955369569987 + }, + { + "index": 127, + "total": 252, + "projection": "layers.25.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.3721, + "parameters": 50331648, + "reference_delta_l2": 3.4450965978095587, + "retained_delta_l2": 3.186482018700214, + "delta_retention_norm_ratio": 0.924932560882683, + "delta_cosine": 0.7393372155827733, + "relative_weight_rmse": 0.012808033611830865 + }, + { + "index": 128, + "total": 252, + "projection": "layers.25.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.2874, + "parameters": 50331648, + "reference_delta_l2": 4.836393393002488, + "retained_delta_l2": 4.760006423474026, + "delta_retention_norm_ratio": 0.9842057989660268, + "delta_cosine": 0.8104606899433433, + "relative_weight_rmse": 0.01582758491989961 + }, + { + "index": 129, + "total": 252, + "projection": "layers.25.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.2799, + "parameters": 50331648, + "reference_delta_l2": 3.5694317531546527, + "retained_delta_l2": 3.294683142876742, + "delta_retention_norm_ratio": 0.9230273530135185, + "delta_cosine": 0.7363368990986017, + "relative_weight_rmse": 0.013230779177507392 + }, + { + "index": 130, + "total": 252, + "projection": "layers.25.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0202, + "parameters": 4194304, + "reference_delta_l2": 0.9684747142564072, + "retained_delta_l2": 0.9139151451869285, + "delta_retention_norm_ratio": 0.9436644361836857, + "delta_cosine": 0.7667776638829131, + "relative_weight_rmse": 0.013066873828265414 + }, + { + "index": 131, + "total": 252, + "projection": "layers.25.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.0853, + "parameters": 16777216, + "reference_delta_l2": 1.8723731810900412, + "retained_delta_l2": 1.7128064476755265, + "delta_retention_norm_ratio": 0.9147783491955275, + "delta_cosine": 0.728811149111852, + "relative_weight_rmse": 0.012857156453864262 + }, + { + "index": 132, + "total": 252, + "projection": "layers.25.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.0818, + "parameters": 16777216, + "reference_delta_l2": 2.2995768828199927, + "retained_delta_l2": 2.237261999690431, + "delta_retention_norm_ratio": 0.9729015874202284, + "delta_cosine": 0.804253375532493, + "relative_weight_rmse": 0.014049139264545053 + }, + { + "index": 133, + "total": 252, + "projection": "layers.25.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.019, + "parameters": 4194304, + "reference_delta_l2": 1.143475468788218, + "retained_delta_l2": 1.0940195160007935, + "delta_retention_norm_ratio": 0.9567494413851879, + "delta_cosine": 0.7792480631184641, + "relative_weight_rmse": 0.013155256927629555 + }, + { + "index": 134, + "total": 252, + "projection": "layers.26.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.3606, + "parameters": 50331648, + "reference_delta_l2": 3.100600291359713, + "retained_delta_l2": 2.78999051565817, + "delta_retention_norm_ratio": 0.8998226967316285, + "delta_cosine": 0.7122871122785357, + "relative_weight_rmse": 0.011871667884522654 + }, + { + "index": 135, + "total": 252, + "projection": "layers.26.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.2918, + "parameters": 50331648, + "reference_delta_l2": 4.893807320286038, + "retained_delta_l2": 4.8117371373593025, + "delta_retention_norm_ratio": 0.983229788678738, + "delta_cosine": 0.8084951726018041, + "relative_weight_rmse": 0.016007899478833562 + }, + { + "index": 136, + "total": 252, + "projection": "layers.26.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.2879, + "parameters": 50331648, + "reference_delta_l2": 3.346538786572717, + "retained_delta_l2": 3.0393184470170573, + "delta_retention_norm_ratio": 0.9081975858793818, + "delta_cosine": 0.7211040808517322, + "relative_weight_rmse": 0.012555611507868866 + }, + { + "index": 137, + "total": 252, + "projection": "layers.26.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0238, + "parameters": 4194304, + "reference_delta_l2": 1.029310458939815, + "retained_delta_l2": 0.9828256739356831, + "delta_retention_norm_ratio": 0.9548389073477297, + "delta_cosine": 0.7789737204154659, + "relative_weight_rmse": 0.013816204935721046 + }, + { + "index": 138, + "total": 252, + "projection": "layers.26.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.0936, + "parameters": 16777216, + "reference_delta_l2": 2.0150367336082815, + "retained_delta_l2": 1.8735956356598982, + "delta_retention_norm_ratio": 0.9298071863458749, + "delta_cosine": 0.7467704867104441, + "relative_weight_rmse": 0.013268451660151346 + }, + { + "index": 139, + "total": 252, + "projection": "layers.26.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.0958, + "parameters": 16777216, + "reference_delta_l2": 2.4415805033686278, + "retained_delta_l2": 2.3860146046965705, + "delta_retention_norm_ratio": 0.9772418322494819, + "delta_cosine": 0.8109300150687073, + "relative_weight_rmse": 0.014571772704839753 + }, + { + "index": 140, + "total": 252, + "projection": "layers.26.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0199, + "parameters": 4194304, + "reference_delta_l2": 1.2568343213357946, + "retained_delta_l2": 1.2165891361066137, + "delta_retention_norm_ratio": 0.9679789256658687, + "delta_cosine": 0.7980823330124911, + "relative_weight_rmse": 0.014054609773679536 + }, + { + "index": 141, + "total": 252, + "projection": "layers.27.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.3304, + "parameters": 50331648, + "reference_delta_l2": 2.9353829255202477, + "retained_delta_l2": 2.583391304076503, + "delta_retention_norm_ratio": 0.8800866427396827, + "delta_cosine": 0.6926504154991127, + "relative_weight_rmse": 0.011416576102284396 + }, + { + "index": 142, + "total": 252, + "projection": "layers.27.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.2745, + "parameters": 50331648, + "reference_delta_l2": 4.621100690329421, + "retained_delta_l2": 4.51226220412324, + "delta_retention_norm_ratio": 0.9764474973605427, + "delta_cosine": 0.7987724775225853, + "relative_weight_rmse": 0.015467352890931468 + }, + { + "index": 143, + "total": 252, + "projection": "layers.27.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.2955, + "parameters": 50331648, + "reference_delta_l2": 2.9433083517247884, + "retained_delta_l2": 2.5303005007309234, + "delta_retention_norm_ratio": 0.8596790408480848, + "delta_cosine": 0.6717519690332472, + "relative_weight_rmse": 0.01165486483635682 + }, + { + "index": 144, + "total": 252, + "projection": "layers.27.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0222, + "parameters": 4194304, + "reference_delta_l2": 0.9194693899536039, + "retained_delta_l2": 0.8600552773735223, + "delta_retention_norm_ratio": 0.9353821745136293, + "delta_cosine": 0.7599026923746851, + "relative_weight_rmse": 0.013110172409810411 + }, + { + "index": 145, + "total": 252, + "projection": "layers.27.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.0921, + "parameters": 16777216, + "reference_delta_l2": 1.7790627812408335, + "retained_delta_l2": 1.6067912785954226, + "delta_retention_norm_ratio": 0.9031672718568944, + "delta_cosine": 0.7163036853094985, + "relative_weight_rmse": 0.012496548305428564 + }, + { + "index": 146, + "total": 252, + "projection": "layers.27.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.0916, + "parameters": 16777216, + "reference_delta_l2": 2.4260204123321056, + "retained_delta_l2": 2.3651331226871024, + "delta_retention_norm_ratio": 0.9749024001053341, + "delta_cosine": 0.808750994766525, + "relative_weight_rmse": 0.014580906870193738 + }, + { + "index": 147, + "total": 252, + "projection": "layers.27.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0213, + "parameters": 4194304, + "reference_delta_l2": 1.242049958017081, + "retained_delta_l2": 1.211006410844666, + "delta_retention_norm_ratio": 0.9750062008600882, + "delta_cosine": 0.8028187484830746, + "relative_weight_rmse": 0.013984726316701689 + }, + { + "index": 148, + "total": 252, + "projection": "layers.28.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.3622, + "parameters": 50331648, + "reference_delta_l2": 3.6386928056409515, + "retained_delta_l2": 3.4087694478197332, + "delta_retention_norm_ratio": 0.9368115501630764, + "delta_cosine": 0.7541776733335837, + "relative_weight_rmse": 0.012819379374375835 + }, + { + "index": 149, + "total": 252, + "projection": "layers.28.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.2912, + "parameters": 50331648, + "reference_delta_l2": 6.18610943773179, + "retained_delta_l2": 6.282238923939749, + "delta_retention_norm_ratio": 1.0155395709008352, + "delta_cosine": 0.8614499190330942, + "relative_weight_rmse": 0.0176093265485177 + }, + { + "index": 150, + "total": 252, + "projection": "layers.28.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.2855, + "parameters": 50331648, + "reference_delta_l2": 3.1102966714154165, + "retained_delta_l2": 2.7260694584834826, + "delta_retention_norm_ratio": 0.8764660566102582, + "delta_cosine": 0.6877677824189266, + "relative_weight_rmse": 0.011968009684836535 + }, + { + "index": 151, + "total": 252, + "projection": "layers.28.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0192, + "parameters": 4194304, + "reference_delta_l2": 0.9862083874382754, + "retained_delta_l2": 0.9299796439745039, + "delta_retention_norm_ratio": 0.9429849267355873, + "delta_cosine": 0.7629343145104027, + "relative_weight_rmse": 0.013695552985927895 + }, + { + "index": 152, + "total": 252, + "projection": "layers.28.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.0866, + "parameters": 16777216, + "reference_delta_l2": 1.7954228131743857, + "retained_delta_l2": 1.6349143503376387, + "delta_retention_norm_ratio": 0.9106013014544685, + "delta_cosine": 0.7244370935225608, + "relative_weight_rmse": 0.012512480231727003 + }, + { + "index": 153, + "total": 252, + "projection": "layers.28.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.0881, + "parameters": 16777216, + "reference_delta_l2": 2.0244268892946304, + "retained_delta_l2": 1.9141484041752483, + "delta_retention_norm_ratio": 0.9455260717477398, + "delta_cosine": 0.7690629313154679, + "relative_weight_rmse": 0.012917130709178035 + }, + { + "index": 154, + "total": 252, + "projection": "layers.28.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.021, + "parameters": 4194304, + "reference_delta_l2": 1.0244263792026678, + "retained_delta_l2": 0.9516176701958848, + "delta_retention_norm_ratio": 0.9289273387674266, + "delta_cosine": 0.7461564435433649, + "relative_weight_rmse": 0.012308951982824896 + }, + { + "index": 155, + "total": 252, + "projection": "layers.29.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.365, + "parameters": 50331648, + "reference_delta_l2": 3.5565162810659685, + "retained_delta_l2": 3.305066392719017, + "delta_retention_norm_ratio": 0.9292988226468666, + "delta_cosine": 0.745816707250538, + "relative_weight_rmse": 0.012565746413120983 + }, + { + "index": 156, + "total": 252, + "projection": "layers.29.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.2977, + "parameters": 50331648, + "reference_delta_l2": 5.417113848108679, + "retained_delta_l2": 5.42151814200812, + "delta_retention_norm_ratio": 1.0008130332909615, + "delta_cosine": 0.8324762974622173, + "relative_weight_rmse": 0.016828774825352794 + }, + { + "index": 157, + "total": 252, + "projection": "layers.29.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.2877, + "parameters": 50331648, + "reference_delta_l2": 3.424315115658444, + "retained_delta_l2": 3.101676017804758, + "delta_retention_norm_ratio": 0.9057799627206192, + "delta_cosine": 0.7182981034319839, + "relative_weight_rmse": 0.012533934178832876 + }, + { + "index": 158, + "total": 252, + "projection": "layers.29.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0215, + "parameters": 4194304, + "reference_delta_l2": 0.9605145594104075, + "retained_delta_l2": 0.9092781367913338, + "delta_retention_norm_ratio": 0.9466573180832113, + "delta_cosine": 0.783507734710872, + "relative_weight_rmse": 0.013053425356086912 + }, + { + "index": 159, + "total": 252, + "projection": "layers.29.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.0845, + "parameters": 16777216, + "reference_delta_l2": 1.89051543914091, + "retained_delta_l2": 1.7320212211512802, + "delta_retention_norm_ratio": 0.9161634892219378, + "delta_cosine": 0.73173914627126, + "relative_weight_rmse": 0.012673621636119101 + }, + { + "index": 160, + "total": 252, + "projection": "layers.29.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.0837, + "parameters": 16777216, + "reference_delta_l2": 2.874744189154029, + "retained_delta_l2": 2.8536079671021843, + "delta_retention_norm_ratio": 0.9926476163925859, + "delta_cosine": 0.8440291239415598, + "relative_weight_rmse": 0.015081184443721588 + }, + { + "index": 161, + "total": 252, + "projection": "layers.29.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0203, + "parameters": 4194304, + "reference_delta_l2": 1.0001311038921359, + "retained_delta_l2": 0.9315136588648855, + "delta_retention_norm_ratio": 0.9313915498076032, + "delta_cosine": 0.7529484221709883, + "relative_weight_rmse": 0.012423154016305886 + }, + { + "index": 162, + "total": 252, + "projection": "layers.3.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.3378, + "parameters": 50331648, + "reference_delta_l2": 1.6328367078875348, + "retained_delta_l2": 1.3508772476926554, + "delta_retention_norm_ratio": 0.8273192543792934, + "delta_cosine": 0.6584088554881028, + "relative_weight_rmse": 0.009621364425890839 + }, + { + "index": 163, + "total": 252, + "projection": "layers.3.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.2827, + "parameters": 50331648, + "reference_delta_l2": 2.5053235177676094, + "retained_delta_l2": 2.2246029029802536, + "delta_retention_norm_ratio": 0.8879503533988717, + "delta_cosine": 0.7076553575863995, + "relative_weight_rmse": 0.011101486563117465 + }, + { + "index": 164, + "total": 252, + "projection": "layers.3.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.274, + "parameters": 50331648, + "reference_delta_l2": 2.7206147817330204, + "retained_delta_l2": 2.607509862283104, + "delta_retention_norm_ratio": 0.9584267055338613, + "delta_cosine": 0.8175185854428151, + "relative_weight_rmse": 0.013394276951894962 + }, + { + "index": 165, + "total": 252, + "projection": "layers.3.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0189, + "parameters": 4194304, + "reference_delta_l2": 0.5357802240764117, + "retained_delta_l2": 0.39576030078112745, + "delta_retention_norm_ratio": 0.7386616433321082, + "delta_cosine": 0.5649501893590435, + "relative_weight_rmse": 0.008909499721650315 + }, + { + "index": 166, + "total": 252, + "projection": "layers.3.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.0833, + "parameters": 16777216, + "reference_delta_l2": 1.6799251397544268, + "retained_delta_l2": 1.5044440246540813, + "delta_retention_norm_ratio": 0.8955423006968053, + "delta_cosine": 0.7088951660126287, + "relative_weight_rmse": 0.011989444728619451 + }, + { + "index": 167, + "total": 252, + "projection": "layers.3.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.0832, + "parameters": 16777216, + "reference_delta_l2": 1.4293373412915793, + "retained_delta_l2": 1.2224811667208624, + "delta_retention_norm_ratio": 0.8552782687508903, + "delta_cosine": 0.6705742277797081, + "relative_weight_rmse": 0.011213731400939248 + }, + { + "index": 168, + "total": 252, + "projection": "layers.3.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0212, + "parameters": 4194304, + "reference_delta_l2": 0.8352270904275945, + "retained_delta_l2": 0.7284911564395894, + "delta_retention_norm_ratio": 0.8722072892375154, + "delta_cosine": 0.6869123726427542, + "relative_weight_rmse": 0.01133900766299753 + }, + { + "index": 169, + "total": 252, + "projection": "layers.30.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.3595, + "parameters": 50331648, + "reference_delta_l2": 3.6032802907362416, + "retained_delta_l2": 3.3594063182533187, + "delta_retention_norm_ratio": 0.932318900333703, + "delta_cosine": 0.7483490174046199, + "relative_weight_rmse": 0.012560707972111639 + }, + { + "index": 170, + "total": 252, + "projection": "layers.30.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.2952, + "parameters": 50331648, + "reference_delta_l2": 4.974662185288301, + "retained_delta_l2": 4.915490479267123, + "delta_retention_norm_ratio": 0.9881053820707328, + "delta_cosine": 0.8134401687949316, + "relative_weight_rmse": 0.016286688583266054 + }, + { + "index": 171, + "total": 252, + "projection": "layers.30.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.2897, + "parameters": 50331648, + "reference_delta_l2": 3.425780991411554, + "retained_delta_l2": 3.0838601650457003, + "delta_retention_norm_ratio": 0.9001918607105794, + "delta_cosine": 0.7118132612098148, + "relative_weight_rmse": 0.012553027922354673 + }, + { + "index": 172, + "total": 252, + "projection": "layers.30.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0233, + "parameters": 4194304, + "reference_delta_l2": 1.454615150624287, + "retained_delta_l2": 1.4618871442415353, + "delta_retention_norm_ratio": 1.0049992560672334, + "delta_cosine": 0.8472728064664286, + "relative_weight_rmse": 0.016429652274220923 + }, + { + "index": 173, + "total": 252, + "projection": "layers.30.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.0884, + "parameters": 16777216, + "reference_delta_l2": 1.8134549324581692, + "retained_delta_l2": 1.6305396279163396, + "delta_retention_norm_ratio": 0.8991343532900017, + "delta_cosine": 0.7111653079940985, + "relative_weight_rmse": 0.012591365368045006 + }, + { + "index": 174, + "total": 252, + "projection": "layers.30.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.0891, + "parameters": 16777216, + "reference_delta_l2": 2.9316027309299133, + "retained_delta_l2": 2.9348241715076675, + "delta_retention_norm_ratio": 1.0010988666860507, + "delta_cosine": 0.8501326455787492, + "relative_weight_rmse": 0.016070274349163405 + }, + { + "index": 175, + "total": 252, + "projection": "layers.30.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0203, + "parameters": 4194304, + "reference_delta_l2": 1.2023008133072424, + "retained_delta_l2": 1.1520908526014633, + "delta_retention_norm_ratio": 0.9582384373776947, + "delta_cosine": 0.7856136728802574, + "relative_weight_rmse": 0.01320798426874773 + }, + { + "index": 176, + "total": 252, + "projection": "layers.31.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.3524, + "parameters": 50331648, + "reference_delta_l2": 3.8810682441090334, + "retained_delta_l2": 3.6820452560203676, + "delta_retention_norm_ratio": 0.9487195340121222, + "delta_cosine": 0.7672569290578201, + "relative_weight_rmse": 0.01294720387724206 + }, + { + "index": 177, + "total": 252, + "projection": "layers.31.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.2803, + "parameters": 50331648, + "reference_delta_l2": 3.9223406788622537, + "retained_delta_l2": 3.7336836843750696, + "delta_retention_norm_ratio": 0.9519019356212812, + "delta_cosine": 0.7693485087735655, + "relative_weight_rmse": 0.014172196485858192 + }, + { + "index": 178, + "total": 252, + "projection": "layers.31.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.281, + "parameters": 50331648, + "reference_delta_l2": 3.623399834624329, + "retained_delta_l2": 3.3226520060300264, + "delta_retention_norm_ratio": 0.9169984428104154, + "delta_cosine": 0.7302631770450293, + "relative_weight_rmse": 0.01281987193963721 + }, + { + "index": 179, + "total": 252, + "projection": "layers.31.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0202, + "parameters": 4194304, + "reference_delta_l2": 1.058268080045612, + "retained_delta_l2": 1.0259746177602465, + "delta_retention_norm_ratio": 0.9694846108521258, + "delta_cosine": 0.7989432352726251, + "relative_weight_rmse": 0.013939990322789822 + }, + { + "index": 180, + "total": 252, + "projection": "layers.31.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.0835, + "parameters": 16777216, + "reference_delta_l2": 1.8867381401875651, + "retained_delta_l2": 1.7354123646384578, + "delta_retention_norm_ratio": 0.9197950302027267, + "delta_cosine": 0.7354900241339846, + "relative_weight_rmse": 0.012340214884699231 + }, + { + "index": 181, + "total": 252, + "projection": "layers.31.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.0823, + "parameters": 16777216, + "reference_delta_l2": 2.9886940841019882, + "retained_delta_l2": 3.005044412553113, + "delta_retention_norm_ratio": 1.0054707266756064, + "delta_cosine": 0.8543355348065301, + "relative_weight_rmse": 0.016057610254095182 + }, + { + "index": 182, + "total": 252, + "projection": "layers.31.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0186, + "parameters": 4194304, + "reference_delta_l2": 0.8857199160991623, + "retained_delta_l2": 0.77501534643497, + "delta_retention_norm_ratio": 0.8750117642699613, + "delta_cosine": 0.6921349673342734, + "relative_weight_rmse": 0.011009704772931905 + }, + { + "index": 183, + "total": 252, + "projection": "layers.32.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.3027, + "parameters": 50331648, + "reference_delta_l2": 2.896536184285432, + "retained_delta_l2": 2.5290553029548155, + "delta_retention_norm_ratio": 0.8731309198468469, + "delta_cosine": 0.686871665027865, + "relative_weight_rmse": 0.01086357178676724 + }, + { + "index": 184, + "total": 252, + "projection": "layers.32.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.2754, + "parameters": 50331648, + "reference_delta_l2": 5.160132814296684, + "retained_delta_l2": 5.147276654109073, + "delta_retention_norm_ratio": 0.9975085602153512, + "delta_cosine": 0.8283919781893335, + "relative_weight_rmse": 0.016636527728732507 + }, + { + "index": 185, + "total": 252, + "projection": "layers.32.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.2745, + "parameters": 50331648, + "reference_delta_l2": 3.9571520455669007, + "retained_delta_l2": 3.7115304062341608, + "delta_retention_norm_ratio": 0.9379296937533891, + "delta_cosine": 0.7552081979176619, + "relative_weight_rmse": 0.013379703382347076 + }, + { + "index": 186, + "total": 252, + "projection": "layers.32.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0196, + "parameters": 4194304, + "reference_delta_l2": 1.0581180770275143, + "retained_delta_l2": 1.0206194721872686, + "delta_retention_norm_ratio": 0.9645610394015878, + "delta_cosine": 0.7934770464785275, + "relative_weight_rmse": 0.014015358346925525 + }, + { + "index": 187, + "total": 252, + "projection": "layers.32.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.0848, + "parameters": 16777216, + "reference_delta_l2": 1.9059274659853354, + "retained_delta_l2": 1.7442278343465185, + "delta_retention_norm_ratio": 0.915159608891401, + "delta_cosine": 0.7290675782804628, + "relative_weight_rmse": 0.012663736807892926 + }, + { + "index": 188, + "total": 252, + "projection": "layers.32.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.0839, + "parameters": 16777216, + "reference_delta_l2": 3.386354161601392, + "retained_delta_l2": 3.437371326120008, + "delta_retention_norm_ratio": 1.0150655135535176, + "delta_cosine": 0.8787210072049341, + "relative_weight_rmse": 0.016779092018824694 + }, + { + "index": 189, + "total": 252, + "projection": "layers.32.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0196, + "parameters": 4194304, + "reference_delta_l2": 1.035264008716314, + "retained_delta_l2": 0.9559619705252885, + "delta_retention_norm_ratio": 0.923399212642042, + "delta_cosine": 0.7423517918723571, + "relative_weight_rmse": 0.012119942967608452 + }, + { + "index": 190, + "total": 252, + "projection": "layers.33.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.2976, + "parameters": 50331648, + "reference_delta_l2": 4.158056237987143, + "retained_delta_l2": 4.004873229183985, + "delta_retention_norm_ratio": 0.9631599478131848, + "delta_cosine": 0.7858069270485945, + "relative_weight_rmse": 0.013473337260780828 + }, + { + "index": 191, + "total": 252, + "projection": "layers.33.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.2775, + "parameters": 50331648, + "reference_delta_l2": 3.5037537092766597, + "retained_delta_l2": 3.244679998070068, + "delta_retention_norm_ratio": 0.9260582413311018, + "delta_cosine": 0.7390390031138095, + "relative_weight_rmse": 0.013600405401520944 + }, + { + "index": 192, + "total": 252, + "projection": "layers.33.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.2765, + "parameters": 50331648, + "reference_delta_l2": 4.153804678041435, + "retained_delta_l2": 3.9647059325031586, + "delta_retention_norm_ratio": 0.954475773370394, + "delta_cosine": 0.7760354082093615, + "relative_weight_rmse": 0.013498489161895274 + }, + { + "index": 193, + "total": 252, + "projection": "layers.33.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0208, + "parameters": 4194304, + "reference_delta_l2": 1.163209272843663, + "retained_delta_l2": 1.1531137400696825, + "delta_retention_norm_ratio": 0.9913209660465478, + "delta_cosine": 0.8259170052841797, + "relative_weight_rmse": 0.015514670628493776 + }, + { + "index": 194, + "total": 252, + "projection": "layers.33.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.0835, + "parameters": 16777216, + "reference_delta_l2": 1.917218370680728, + "retained_delta_l2": 1.7672382025223023, + "delta_retention_norm_ratio": 0.9217719950674301, + "delta_cosine": 0.7383447960236257, + "relative_weight_rmse": 0.012106365343981023 + }, + { + "index": 195, + "total": 252, + "projection": "layers.33.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.0811, + "parameters": 16777216, + "reference_delta_l2": 1.6553096736770672, + "retained_delta_l2": 1.5189603787286248, + "delta_retention_norm_ratio": 0.917629131807368, + "delta_cosine": 0.738370996876396, + "relative_weight_rmse": 0.011720228551450676 + }, + { + "index": 196, + "total": 252, + "projection": "layers.33.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0188, + "parameters": 4194304, + "reference_delta_l2": 1.185793149834154, + "retained_delta_l2": 1.129896173369846, + "delta_retention_norm_ratio": 0.9528611069543403, + "delta_cosine": 0.7771329614959223, + "relative_weight_rmse": 0.012453042571658742 + }, + { + "index": 197, + "total": 252, + "projection": "layers.34.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.3127, + "parameters": 50331648, + "reference_delta_l2": 2.821877675503641, + "retained_delta_l2": 2.523347175254057, + "delta_retention_norm_ratio": 0.8942085609021647, + "delta_cosine": 0.7087053800668754, + "relative_weight_rmse": 0.010829783070142294 + }, + { + "index": 198, + "total": 252, + "projection": "layers.34.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.2893, + "parameters": 50331648, + "reference_delta_l2": 3.238533774093452, + "retained_delta_l2": 2.9915522313131575, + "delta_retention_norm_ratio": 0.9237366166269393, + "delta_cosine": 0.7353802799688653, + "relative_weight_rmse": 0.012517406687409704 + }, + { + "index": 199, + "total": 252, + "projection": "layers.34.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.2973, + "parameters": 50331648, + "reference_delta_l2": 3.0235831105993602, + "retained_delta_l2": 2.6943013028485963, + "delta_retention_norm_ratio": 0.8910954997081291, + "delta_cosine": 0.7038879602808115, + "relative_weight_rmse": 0.011149633146402372 + }, + { + "index": 200, + "total": 252, + "projection": "layers.34.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0193, + "parameters": 4194304, + "reference_delta_l2": 0.7928188424322914, + "retained_delta_l2": 0.7317704908256062, + "delta_retention_norm_ratio": 0.9229983593485306, + "delta_cosine": 0.7535575556119187, + "relative_weight_rmse": 0.011670621769581674 + }, + { + "index": 201, + "total": 252, + "projection": "layers.34.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.091, + "parameters": 16777216, + "reference_delta_l2": 2.4645978849554875, + "retained_delta_l2": 2.4001892009258414, + "delta_retention_norm_ratio": 0.9738664532568122, + "delta_cosine": 0.7983828446517862, + "relative_weight_rmse": 0.014459432274182425 + }, + { + "index": 202, + "total": 252, + "projection": "layers.34.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.0945, + "parameters": 16777216, + "reference_delta_l2": 2.695511834193117, + "retained_delta_l2": 2.699277898890852, + "delta_retention_norm_ratio": 1.0013971612552248, + "delta_cosine": 0.8541317814560682, + "relative_weight_rmse": 0.015023564746444064 + }, + { + "index": 203, + "total": 252, + "projection": "layers.34.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0209, + "parameters": 4194304, + "reference_delta_l2": 1.0207525569136708, + "retained_delta_l2": 0.9269824099360373, + "delta_retention_norm_ratio": 0.9081362605045485, + "delta_cosine": 0.7288602731730222, + "relative_weight_rmse": 0.01164327693543864 + }, + { + "index": 204, + "total": 252, + "projection": "layers.35.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.3233, + "parameters": 50331648, + "reference_delta_l2": 1.9823095213349184, + "retained_delta_l2": 1.6833332043325306, + "delta_retention_norm_ratio": 0.8491777828918198, + "delta_cosine": 0.6694214090375227, + "relative_weight_rmse": 0.008826801286047611 + }, + { + "index": 205, + "total": 252, + "projection": "layers.35.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.2822, + "parameters": 50331648, + "reference_delta_l2": 2.4543541404296376, + "retained_delta_l2": 2.095939159235461, + "delta_retention_norm_ratio": 0.8539676995710831, + "delta_cosine": 0.6683413065179096, + "relative_weight_rmse": 0.010069223796027413 + }, + { + "index": 206, + "total": 252, + "projection": "layers.35.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.286, + "parameters": 50331648, + "reference_delta_l2": 2.499227317560434, + "retained_delta_l2": 2.1108465532485687, + "delta_retention_norm_ratio": 0.8445996642310335, + "delta_cosine": 0.6601572203554984, + "relative_weight_rmse": 0.009871612295636545 + }, + { + "index": 207, + "total": 252, + "projection": "layers.35.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0211, + "parameters": 4194304, + "reference_delta_l2": 0.7804828082506937, + "retained_delta_l2": 0.7046762074304084, + "delta_retention_norm_ratio": 0.9028721708935631, + "delta_cosine": 0.7212488537699842, + "relative_weight_rmse": 0.011677823781018356 + }, + { + "index": 208, + "total": 252, + "projection": "layers.35.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.0937, + "parameters": 16777216, + "reference_delta_l2": 1.9358674579050277, + "retained_delta_l2": 1.826476966807879, + "delta_retention_norm_ratio": 0.9434927785730074, + "delta_cosine": 0.7613448331168875, + "relative_weight_rmse": 0.012846772474266906 + }, + { + "index": 209, + "total": 252, + "projection": "layers.35.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.0884, + "parameters": 16777216, + "reference_delta_l2": 1.789558247625012, + "retained_delta_l2": 1.6807697827683608, + "delta_retention_norm_ratio": 0.9392093188355123, + "delta_cosine": 0.7543470277448979, + "relative_weight_rmse": 0.012638227777340855 + }, + { + "index": 210, + "total": 252, + "projection": "layers.35.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0213, + "parameters": 4194304, + "reference_delta_l2": 0.8107990141519036, + "retained_delta_l2": 0.6923380793625497, + "delta_retention_norm_ratio": 0.8538960547290944, + "delta_cosine": 0.6701267360411922, + "relative_weight_rmse": 0.010610346900305287 + }, + { + "index": 211, + "total": 252, + "projection": "layers.4.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.3217, + "parameters": 50331648, + "reference_delta_l2": 2.043343408116534, + "retained_delta_l2": 1.7174886693961613, + "delta_retention_norm_ratio": 0.8405286466161204, + "delta_cosine": 0.6572729898384961, + "relative_weight_rmse": 0.01012551010778355 + }, + { + "index": 212, + "total": 252, + "projection": "layers.4.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.2898, + "parameters": 50331648, + "reference_delta_l2": 3.0372171514712547, + "retained_delta_l2": 2.61912948708616, + "delta_retention_norm_ratio": 0.8623451523107035, + "delta_cosine": 0.6878635130036593, + "relative_weight_rmse": 0.01020563041830835 + }, + { + "index": 213, + "total": 252, + "projection": "layers.4.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.2857, + "parameters": 50331648, + "reference_delta_l2": 3.1333670723021627, + "retained_delta_l2": 2.991480228291325, + "delta_retention_norm_ratio": 0.9547174522688178, + "delta_cosine": 0.7852368918844295, + "relative_weight_rmse": 0.013581849235775443 + }, + { + "index": 214, + "total": 252, + "projection": "layers.4.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0198, + "parameters": 4194304, + "reference_delta_l2": 0.5244352595642214, + "retained_delta_l2": 0.37542370971843775, + "delta_retention_norm_ratio": 0.7158628312489808, + "delta_cosine": 0.5453996124130298, + "relative_weight_rmse": 0.008572015911100742 + }, + { + "index": 215, + "total": 252, + "projection": "layers.4.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.0856, + "parameters": 16777216, + "reference_delta_l2": 1.4702355284702313, + "retained_delta_l2": 1.2361517781877487, + "delta_retention_norm_ratio": 0.8407848635476487, + "delta_cosine": 0.6543602580925502, + "relative_weight_rmse": 0.010979184466776731 + }, + { + "index": 216, + "total": 252, + "projection": "layers.4.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.0827, + "parameters": 16777216, + "reference_delta_l2": 1.1330369850333857, + "retained_delta_l2": 0.8538898964738426, + "delta_retention_norm_ratio": 0.753629323449386, + "delta_cosine": 0.5765922595156237, + "relative_weight_rmse": 0.009442583482949792 + }, + { + "index": 217, + "total": 252, + "projection": "layers.4.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0186, + "parameters": 4194304, + "reference_delta_l2": 0.9219868617247438, + "retained_delta_l2": 0.8307413349834236, + "delta_retention_norm_ratio": 0.9010338102100188, + "delta_cosine": 0.7170312757451748, + "relative_weight_rmse": 0.011933906834989327 + }, + { + "index": 218, + "total": 252, + "projection": "layers.5.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.294, + "parameters": 50331648, + "reference_delta_l2": 1.60459054270334, + "retained_delta_l2": 1.1396238620246248, + "delta_retention_norm_ratio": 0.7102272085591625, + "delta_cosine": 0.5413391457434584, + "relative_weight_rmse": 0.0082082789800158 + }, + { + "index": 219, + "total": 252, + "projection": "layers.5.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.2787, + "parameters": 50331648, + "reference_delta_l2": 2.7167476668565116, + "retained_delta_l2": 2.226031133887409, + "delta_retention_norm_ratio": 0.8193735329358358, + "delta_cosine": 0.6379289789806207, + "relative_weight_rmse": 0.009413097609736907 + }, + { + "index": 220, + "total": 252, + "projection": "layers.5.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.2949, + "parameters": 50331648, + "reference_delta_l2": 2.9329646658700654, + "retained_delta_l2": 2.687159996374674, + "delta_retention_norm_ratio": 0.9161924204693843, + "delta_cosine": 0.7311448007579043, + "relative_weight_rmse": 0.012468887222045644 + }, + { + "index": 221, + "total": 252, + "projection": "layers.5.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0185, + "parameters": 4194304, + "reference_delta_l2": 0.5437987065128471, + "retained_delta_l2": 0.4047357132618228, + "delta_retention_norm_ratio": 0.7442748730632757, + "delta_cosine": 0.5708455477077047, + "relative_weight_rmse": 0.008915155872740532 + }, + { + "index": 222, + "total": 252, + "projection": "layers.5.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.0898, + "parameters": 16777216, + "reference_delta_l2": 1.352433111519197, + "retained_delta_l2": 1.1069321443420719, + "delta_retention_norm_ratio": 0.8184745958331704, + "delta_cosine": 0.632998832522086, + "relative_weight_rmse": 0.010657983726530096 + }, + { + "index": 223, + "total": 252, + "projection": "layers.5.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.093, + "parameters": 16777216, + "reference_delta_l2": 1.264568274116871, + "retained_delta_l2": 1.0160588001837498, + "delta_retention_norm_ratio": 0.8034827545340159, + "delta_cosine": 0.6220893589407358, + "relative_weight_rmse": 0.010130382111453774 + }, + { + "index": 224, + "total": 252, + "projection": "layers.5.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.02, + "parameters": 4194304, + "reference_delta_l2": 0.9997299186419492, + "retained_delta_l2": 0.9316298816862428, + "delta_retention_norm_ratio": 0.9318815655250023, + "delta_cosine": 0.7544580375427529, + "relative_weight_rmse": 0.012434136650850247 + }, + { + "index": 225, + "total": 252, + "projection": "layers.6.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.3196, + "parameters": 50331648, + "reference_delta_l2": 4.376686711230453, + "retained_delta_l2": 4.350106184769021, + "delta_retention_norm_ratio": 0.9939267925224744, + "delta_cosine": 0.8288901954627105, + "relative_weight_rmse": 0.014787761817631922 + }, + { + "index": 226, + "total": 252, + "projection": "layers.6.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3052, + "parameters": 50331648, + "reference_delta_l2": 2.9074557058505683, + "retained_delta_l2": 2.5005355483480867, + "delta_retention_norm_ratio": 0.860042525606271, + "delta_cosine": 0.6776434227254817, + "relative_weight_rmse": 0.01017618471205037 + }, + { + "index": 227, + "total": 252, + "projection": "layers.6.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.2796, + "parameters": 50331648, + "reference_delta_l2": 3.1183011367968954, + "retained_delta_l2": 2.886978332990382, + "delta_retention_norm_ratio": 0.9258176828796826, + "delta_cosine": 0.7412652873621648, + "relative_weight_rmse": 0.012624356834163077 + }, + { + "index": 228, + "total": 252, + "projection": "layers.6.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0184, + "parameters": 4194304, + "reference_delta_l2": 0.5506121946790155, + "retained_delta_l2": 0.4057511541629519, + "delta_retention_norm_ratio": 0.7369091314795313, + "delta_cosine": 0.5636920049824873, + "relative_weight_rmse": 0.008997415385937207 + }, + { + "index": 229, + "total": 252, + "projection": "layers.6.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.0842, + "parameters": 16777216, + "reference_delta_l2": 1.2370675869345922, + "retained_delta_l2": 0.9687512486064946, + "delta_retention_norm_ratio": 0.7831029273081389, + "delta_cosine": 0.6026394837415747, + "relative_weight_rmse": 0.00962555464796606 + }, + { + "index": 230, + "total": 252, + "projection": "layers.6.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.0803, + "parameters": 16777216, + "reference_delta_l2": 1.2723583985596465, + "retained_delta_l2": 1.0190259907621784, + "delta_retention_norm_ratio": 0.8008954017325234, + "delta_cosine": 0.6186974562229032, + "relative_weight_rmse": 0.009970888524534274 + }, + { + "index": 231, + "total": 252, + "projection": "layers.6.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0183, + "parameters": 4194304, + "reference_delta_l2": 0.9237469929926623, + "retained_delta_l2": 0.8381914238174702, + "delta_retention_norm_ratio": 0.9073820322835175, + "delta_cosine": 0.722158367761246, + "relative_weight_rmse": 0.01192094386606325 + }, + { + "index": 232, + "total": 252, + "projection": "layers.7.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.2991, + "parameters": 50331648, + "reference_delta_l2": 1.6011354845798527, + "retained_delta_l2": 1.114551863096439, + "delta_retention_norm_ratio": 0.696100907031552, + "delta_cosine": 0.5296439743814517, + "relative_weight_rmse": 0.00797270964724083 + }, + { + "index": 233, + "total": 252, + "projection": "layers.7.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.2888, + "parameters": 50331648, + "reference_delta_l2": 2.604621866268664, + "retained_delta_l2": 2.156010126213745, + "delta_retention_norm_ratio": 0.8277631982343785, + "delta_cosine": 0.6468744511766974, + "relative_weight_rmse": 0.009582833654521025 + }, + { + "index": 234, + "total": 252, + "projection": "layers.7.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.2917, + "parameters": 50331648, + "reference_delta_l2": 2.7148627968615884, + "retained_delta_l2": 2.3940188386459096, + "delta_retention_norm_ratio": 0.8818194574743969, + "delta_cosine": 0.6931481733288808, + "relative_weight_rmse": 0.011652355285720086 + }, + { + "index": 235, + "total": 252, + "projection": "layers.7.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0213, + "parameters": 4194304, + "reference_delta_l2": 0.6375207609376344, + "retained_delta_l2": 0.5155370261913551, + "delta_retention_norm_ratio": 0.8086591963422939, + "delta_cosine": 0.6255068307050061, + "relative_weight_rmse": 0.00986875980171051 + }, + { + "index": 236, + "total": 252, + "projection": "layers.7.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.0902, + "parameters": 16777216, + "reference_delta_l2": 1.2552230316150197, + "retained_delta_l2": 0.982903345138241, + "delta_retention_norm_ratio": 0.7830507570225178, + "delta_cosine": 0.6021611665254775, + "relative_weight_rmse": 0.009895264289633535 + }, + { + "index": 237, + "total": 252, + "projection": "layers.7.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.0896, + "parameters": 16777216, + "reference_delta_l2": 1.5359069696221528, + "retained_delta_l2": 1.3581859768630977, + "delta_retention_norm_ratio": 0.8842892204579448, + "delta_cosine": 0.6978932431003803, + "relative_weight_rmse": 0.011338741686984611 + }, + { + "index": 238, + "total": 252, + "projection": "layers.7.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0199, + "parameters": 4194304, + "reference_delta_l2": 0.7051042187633481, + "retained_delta_l2": 0.5652459705552851, + "delta_retention_norm_ratio": 0.8016488279514846, + "delta_cosine": 0.6220085894138233, + "relative_weight_rmse": 0.009523562848348033 + }, + { + "index": 239, + "total": 252, + "projection": "layers.8.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.3306, + "parameters": 50331648, + "reference_delta_l2": 2.241443869715866, + "retained_delta_l2": 1.8272421294903742, + "delta_retention_norm_ratio": 0.8152076231656886, + "delta_cosine": 0.6324387939860261, + "relative_weight_rmse": 0.009842683384753201 + }, + { + "index": 240, + "total": 252, + "projection": "layers.8.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.2887, + "parameters": 50331648, + "reference_delta_l2": 2.0020562075991806, + "retained_delta_l2": 1.4535536434014302, + "delta_retention_norm_ratio": 0.7260303871011183, + "delta_cosine": 0.55432970546666, + "relative_weight_rmse": 0.008751578472771991 + }, + { + "index": 241, + "total": 252, + "projection": "layers.8.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.2765, + "parameters": 50331648, + "reference_delta_l2": 2.3050198953747403, + "retained_delta_l2": 1.8581929482664734, + "delta_retention_norm_ratio": 0.8061505030803113, + "delta_cosine": 0.6234078341037631, + "relative_weight_rmse": 0.010262403171030401 + }, + { + "index": 242, + "total": 252, + "projection": "layers.8.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0202, + "parameters": 4194304, + "reference_delta_l2": 0.8142274162935644, + "retained_delta_l2": 0.7196864411222547, + "delta_retention_norm_ratio": 0.8838887351623841, + "delta_cosine": 0.6959786563698767, + "relative_weight_rmse": 0.01141390426469825 + }, + { + "index": 243, + "total": 252, + "projection": "layers.8.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.0837, + "parameters": 16777216, + "reference_delta_l2": 1.376716956989815, + "retained_delta_l2": 1.125127602026786, + "delta_retention_norm_ratio": 0.8172541177141249, + "delta_cosine": 0.6336929939668533, + "relative_weight_rmse": 0.01029017423529436 + }, + { + "index": 244, + "total": 252, + "projection": "layers.8.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.0837, + "parameters": 16777216, + "reference_delta_l2": 1.4916791552785456, + "retained_delta_l2": 1.2646706845971023, + "delta_retention_norm_ratio": 0.8478168244973208, + "delta_cosine": 0.6609397318833818, + "relative_weight_rmse": 0.011302634113662864 + }, + { + "index": 245, + "total": 252, + "projection": "layers.8.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0208, + "parameters": 4194304, + "reference_delta_l2": 0.7376420382648309, + "retained_delta_l2": 0.5951106031452117, + "delta_retention_norm_ratio": 0.8067742512955219, + "delta_cosine": 0.6237091618733237, + "relative_weight_rmse": 0.01003165832577633 + }, + { + "index": 246, + "total": 252, + "projection": "layers.9.mlp.down_proj", + "shape": [ + 4096, + 12288 + ], + "seconds": 0.335, + "parameters": 50331648, + "reference_delta_l2": 2.198025661201102, + "retained_delta_l2": 1.7953132726162053, + "delta_retention_norm_ratio": 0.8167844917857617, + "delta_cosine": 0.6336958049607184, + "relative_weight_rmse": 0.009999010507091002 + }, + { + "index": 247, + "total": 252, + "projection": "layers.9.mlp.gate_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.3086, + "parameters": 50331648, + "reference_delta_l2": 2.367346213681673, + "retained_delta_l2": 1.8376129596715194, + "delta_retention_norm_ratio": 0.7762332983031163, + "delta_cosine": 0.597617687706116, + "relative_weight_rmse": 0.009410607398687657 + }, + { + "index": 248, + "total": 252, + "projection": "layers.9.mlp.up_proj", + "shape": [ + 12288, + 4096 + ], + "seconds": 0.2924, + "parameters": 50331648, + "reference_delta_l2": 2.7911405036465617, + "retained_delta_l2": 2.4785721061854744, + "delta_retention_norm_ratio": 0.8880140942196483, + "delta_cosine": 0.7027169076037411, + "relative_weight_rmse": 0.011728165449823365 + }, + { + "index": 249, + "total": 252, + "projection": "layers.9.self_attn.k_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0186, + "parameters": 4194304, + "reference_delta_l2": 0.7479720000175084, + "retained_delta_l2": 0.6531119478001653, + "delta_retention_norm_ratio": 0.8731770009905151, + "delta_cosine": 0.6897070446896674, + "relative_weight_rmse": 0.010911995385218833 + }, + { + "index": 250, + "total": 252, + "projection": "layers.9.self_attn.o_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.0847, + "parameters": 16777216, + "reference_delta_l2": 1.3714384751859114, + "retained_delta_l2": 1.127018845066739, + "delta_retention_norm_ratio": 0.8217786400618234, + "delta_cosine": 0.6383344346502468, + "relative_weight_rmse": 0.010335860800388192 + }, + { + "index": 251, + "total": 252, + "projection": "layers.9.self_attn.q_proj", + "shape": [ + 4096, + 4096 + ], + "seconds": 0.0825, + "parameters": 16777216, + "reference_delta_l2": 1.6757371128988487, + "retained_delta_l2": 1.5104975488994226, + "delta_retention_norm_ratio": 0.9013929077971071, + "delta_cosine": 0.7160380408555145, + "relative_weight_rmse": 0.012124181065717473 + }, + { + "index": 252, + "total": 252, + "projection": "layers.9.self_attn.v_proj", + "shape": [ + 1024, + 4096 + ], + "seconds": 0.0175, + "parameters": 4194304, + "reference_delta_l2": 0.672382652149282, + "retained_delta_l2": 0.5284218432859524, + "delta_retention_norm_ratio": 0.7858945224075065, + "delta_cosine": 0.6078980628877405, + "relative_weight_rmse": 0.009142790598844487 + } + ] +}