=== CONFIG A ===
{
  "_sliding_window_pattern": 6,
  "architectures": [
    "Gemma3ForCausalLM"
  ],
  "attention_bias": false,
  "attention_dropout": 0.0,
  "attn_logit_softcapping": null,
  "bos_token_id": 2,
  "dtype": "float32",
  "eos_token_id": 1,
  "final_logit_softcapping": null,
  "head_dim": 256,
  "hidden_activation": "gelu_pytorch_tanh",
  "hidden_size": 640,
  "initializer_range": 0.02,
  "intermediate_size": 2048,
  "layer_types": [
    "sliding_attention",
    "sliding_attention",
    "sliding_attention",
    "sliding_attention",
    "sliding_attention",
    "full_attention",
    "sliding_attention",
    "sliding_attention",
    "sliding_attention",
    "sliding_attention",
    "sliding_attention",
    "full_attention",
    "sliding_attention",
    "sliding_attention",
    "sliding_attention",
    "sliding_attention",
    "sliding_attention",
    "full_attention"
  ],
  "max_position_embeddings": 32768,
  "model_type": "gemma3_text",
  "num_attention_heads": 4,
  "num_hidden_layers": 18,
  "num_key_value_heads": 1,
  "pad_token_id": 0,
  "query_pre_attn_scalar": 256,
  "rms_norm_eps": 1e-06,
  "rope_parameters": {
    "full_attention": {
      "rope_theta": 1000000.0,
      "rope_type": "default"
    },
    "sliding_attention": {
      "rope_theta": 10000.0,
      "rope_type": "default"
    }
  },
  "sliding_window": 512,
  "tie_word_embeddings": true,
  "transformers_version": "5.20.2",
  "use_bidirectional_attention": false,
  "use_cache": true,
  "vocab_size": 262144
}

=== CONFIG DIFFERENCES ===

Keys only in A: []
Keys only in B: []

=== TENSOR LAYOUT: FIRST LAYER AND NON-LAYER TENSORS ===
lm_head.weight (262144, 640) BF16
model.embed_tokens.weight (262144, 640) BF16
model.layers.0.input_layernorm.weight (640,) BF16
model.layers.0.mlp.down_proj.weight (640, 2048) BF16
model.layers.0.mlp.gate_proj.weight (2048, 640) BF16
model.layers.0.mlp.up_proj.weight (2048, 640) BF16
model.layers.0.post_attention_layernorm.weight (640,) BF16
model.layers.0.post_feedforward_layernorm.weight (640,) BF16
model.layers.0.pre_feedforward_layernorm.weight (640,) BF16
model.layers.0.self_attn.k_norm.weight (256,) BF16
model.layers.0.self_attn.k_proj.weight (256, 640) BF16
model.layers.0.self_attn.o_proj.weight (640, 1024) BF16
model.layers.0.self_attn.q_norm.weight (256,) BF16
model.layers.0.self_attn.q_proj.weight (1024, 640) BF16
model.layers.0.self_attn.v_proj.weight (256, 640) BF16
model.norm.weight (640,) BF16

=== NON-LAYER TENSORS, INCLUDING EMBEDDING IDENTITY ===
lm_head.weight shape (262144, 640) changed 0 / 167772160 ||B-A|| 0.0 relative_difference 0.0
model.embed_tokens.weight shape (262144, 640) changed 0 / 167772160 ||B-A|| 0.0 relative_difference 0.0
model.norm.weight shape (640,) changed 0 / 640 ||B-A|| 0.0 relative_difference 0.0

=== PER-LAYER STATIC DIFFERENCES ===
layer part       parameters changed_frac  ||B-A||  relative_diff  ||A||  ||B||
 0 attention      1638912     0.75389   0.29571      0.005216   56.6955   56.6950
 0 mlp            3932160     0.76593   0.46554      0.006859   67.8686   67.8684
 0 norm/other        2560     0.04141   0.00477      0.000012  410.0546  410.0546
 1 attention      1638912     0.78508   0.29445      0.004621   63.7205   63.7198
 1 mlp            3932160     0.78431   0.46855      0.007165   65.3925   65.3925
 1 norm/other        2560     0.02656   0.00427      0.000004 1058.3784 1058.3784
 2 attention      1638912     0.76083   0.29718      0.005071   58.6053   58.6052
 2 mlp            3932160     0.77652   0.46182      0.006999   65.9808   65.9808
 2 norm/other        2560     0.02109   0.00318      0.000003 1079.7547 1079.7547
 3 attention      1638912     0.79866   0.29629      0.006203   47.7629   47.7629
 3 mlp            3932160     0.79081   0.46207      0.007081   65.2501   65.2500
 3 norm/other        2560     0.01211   0.00278      0.000002 1437.8398 1437.8398
 4 attention      1638912     0.79663   0.29869      0.004986   59.9031   59.9030
 4 mlp            3932160     0.80191   0.46884      0.007236   64.7887   64.7887
 4 norm/other        2560     0.00898   0.00174      0.000001 1530.8081 1530.8081
 5 attention      1638912     0.80262   0.31690      0.004250   74.5611   74.5615
 5 mlp            3932160     0.77835   0.47275      0.006954   67.9788   67.9788
 5 norm/other        2560     0.00742   0.00159      0.000001 1252.9466 1252.9466
 6 attention      1638912     0.78989   0.31917      0.005941   53.7198   53.7196
 6 mlp            3932160     0.79351   0.47239      0.007217   65.4549   65.4548
 6 norm/other        2560     0.00820   0.00209      0.000002 1238.5679 1238.5679
 7 attention      1638912     0.77745   0.30760      0.005382   57.1527   57.1528
 7 mlp            3932160     0.78568   0.46554      0.007048   66.0538   66.0538
 7 norm/other        2560     0.00742   0.00249      0.000002 1197.0734 1197.0734
 8 attention      1638912     0.78384   0.30316      0.005371   56.4418   56.4420
 8 mlp            3932160     0.79600   0.46134      0.007096   65.0143   65.0144
 8 norm/other        2560     0.00898   0.00212      0.000002 1382.3530 1382.3530
 9 attention      1638912     0.78261   0.32176      0.005523   58.2544   58.2540
 9 mlp            3932160     0.78486   0.47190      0.007070   66.7427   66.7419
 9 norm/other        2560     0.01016   0.00199      0.000001 1457.9683 1457.9683
10 attention      1638912     0.77694   0.33535      0.004679   71.6774   71.6771
10 mlp            3932160     0.79874   0.47168      0.007232   65.2214   65.2216
10 norm/other        2560     0.00859   0.00213      0.000001 2449.4564 2449.4564
11 attention      1638912     0.78826   0.32164      0.003811   84.3949   84.3952
11 mlp            3932160     0.78435   0.46610      0.007009   66.5039   66.5038
11 norm/other        2560     0.00703   0.00211      0.000000 5974.8339 5974.8339
12 attention      1638912     0.79688   0.32365      0.005359   60.3915   60.3919
12 mlp            3932160     0.78994   0.48771      0.007258   67.2000   67.2001
12 norm/other        2560     0.00703   0.00179      0.000000 7330.5102 7330.5102
13 attention      1638912     0.79827   0.31742      0.005829   54.4589   54.4588
13 mlp            3932160     0.79874   0.49566      0.007597   65.2469   65.2463
13 norm/other        2560     0.02773   0.00351      0.000001 6870.6222 6870.6222
14 attention      1638912     0.78523   0.31476      0.004778   65.8748   65.8751
14 mlp            3932160     0.80409   0.50453      0.007764   64.9878   64.9876
14 norm/other        2560     0.03633   0.00382      0.000000 7862.6967 7862.6967
15 attention      1638912     0.77643   0.32602      0.005164   63.1300   63.1301
15 mlp            3932160     0.79893   0.50864      0.007729   65.8064   65.8061
15 norm/other        2560     0.01172   0.00253      0.000000 9693.7315 9693.7315
16 attention      1638912     0.79948   0.31193      0.005640   55.3094   55.3094
16 mlp            3932160     0.79371   0.50168      0.007579   66.1931   66.1925
16 norm/other        2560     0.00937   0.00254      0.000000 9574.1642 9574.1642
17 attention      1638912     0.78958   0.30548      0.004623   66.0831   66.0830
17 mlp            3932160     0.79638   0.48769      0.007439   65.5570   65.5566
17 norm/other        2560     0.01094   0.00235      0.000000 10309.0641 10309.0641

=== TOP 24 TENSORS BY RELATIVE DIFFERENCE (NOT CAUSAL IMPORTANCE) ===
model.layers.4.mlp.down_proj.weight                               rel=0.018281 rms=0.0002223 max=0.0012207 changed=0.9156
model.layers.14.mlp.down_proj.weight                              rel=0.018078 rms=0.0002432 max=0.0013428 changed=0.9032
model.layers.8.mlp.down_proj.weight                               rel=0.016582 rms=0.0002149 max=0.0012207 changed=0.8956
model.layers.1.mlp.down_proj.weight                               rel=0.016577 rms=0.0002254 max=0.0012207 changed=0.8955
model.layers.10.mlp.down_proj.weight                              rel=0.016500 rms=0.0002245 max=0.0019531 changed=0.8981
model.layers.13.mlp.down_proj.weight                              rel=0.016468 rms=0.0002384 max=0.0011902 changed=0.8942
model.layers.3.mlp.down_proj.weight                               rel=0.015904 rms=0.0002207 max=0.0019531 changed=0.8960
model.layers.15.mlp.down_proj.weight                              rel=0.015485 rms=0.0002482 max=0.0012817 changed=0.8867
model.layers.6.mlp.down_proj.weight                               rel=0.014720 rms=0.0002200 max=0.0010986 changed=0.8829
model.layers.17.mlp.down_proj.weight                              rel=0.014429 rms=0.0002302 max=0.0012817 changed=0.8804
model.layers.16.mlp.down_proj.weight                              rel=0.014094 rms=0.0002431 max=0.0012512 changed=0.8763
model.layers.2.mlp.down_proj.weight                               rel=0.013604 rms=0.0002204 max=0.0019531 changed=0.8759
model.layers.7.mlp.down_proj.weight                               rel=0.013034 rms=0.0002179 max=0.0012207 changed=0.8679
model.layers.9.mlp.down_proj.weight                               rel=0.012032 rms=0.0002221 max=0.0012207 changed=0.8573
model.layers.11.mlp.down_proj.weight                              rel=0.011970 rms=0.0002147 max=0.0010986 changed=0.8562
model.layers.12.mlp.down_proj.weight                              rel=0.011741 rms=0.0002322 max=0.0011597 changed=0.8554
model.layers.3.self_attn.o_proj.weight                            rel=0.010839 rms=0.0002153 max=0.0010986 changed=0.8487
model.layers.4.self_attn.o_proj.weight                            rel=0.010615 rms=0.0002254 max=0.0011082 changed=0.8586
model.layers.0.mlp.down_proj.weight                               rel=0.010586 rms=0.0002277 max=0.0010986 changed=0.8407
model.layers.16.self_attn.o_proj.weight                           rel=0.010531 rms=0.0002271 max=0.0011597 changed=0.8394
model.layers.5.mlp.down_proj.weight                               rel=0.010374 rms=0.0002247 max=0.0010986 changed=0.8380
model.layers.1.self_attn.o_proj.weight                            rel=0.009584 rms=0.0002189 max=0.0010452 changed=0.8313
model.layers.13.self_attn.o_proj.weight                           rel=0.008842 rms=0.0002382 max=0.0012283 changed=0.8228
model.layers.8.self_attn.o_proj.weight                            rel=0.008573 rms=0.0002225 max=0.0010986 changed=0.8378

=== NORMALISATION PARAMETER DIFFERENCES ===
model.layers.0.input_layernorm.weight A[min,mean,max]= [6.375, 12.474902153015137, 61.75] B[min,mean,max]= [6.375, 12.474902153015137, 61.75] largest (index,A,B)= [(639, 9.5, 9.5), (638, 9.4375, 9.4375), (217, 15.5625, 15.5625), (216, 8.875, 8.875)]
model.layers.0.post_attention_layernorm.weight A[min,mean,max]= [-1.0, -0.36056631803512573, 51.75] B[min,mean,max]= [-1.0, -0.36056479811668396, 51.75] largest (index,A,B)= [(103, -0.099609375, -0.09912109375), (195, -0.1220703125, -0.12158203125), (639, 7.09375, 7.09375), (211, -0.890625, -0.890625)]
model.layers.0.post_feedforward_layernorm.weight A[min,mean,max]= [-0.98828125, 0.8047735095024109, 152.0] B[min,mean,max]= [-0.98828125, 0.8047771453857422, 152.0] largest (index,A,B)= [(244, -0.1298828125, -0.130859375), (304, -0.16015625, -0.1611328125), (435, 0.1806640625, 0.181640625), (462, 0.197265625, 0.1962890625)]
model.layers.0.pre_feedforward_layernorm.weight A[min,mean,max]= [-1.0, 4.830193996429443, 19.625] B[min,mean,max]= [-1.0, 4.830193996429443, 19.625] largest (index,A,B)= [(639, 0.76953125, 0.76953125), (638, 4.5, 4.5), (217, 5.78125, 5.78125), (216, 4.84375, 4.84375)]
model.layers.0.self_attn.k_norm.weight A[min,mean,max]= [-1.828125, 0.9714725017547607, 5.84375] B[min,mean,max]= [-1.828125, 0.9714727401733398, 5.84375] largest (index,A,B)= [(9, 0.0615234375, 0.062255859375), (147, 0.125, 0.12451171875), (200, -0.05908203125, -0.059326171875), (7, 0.0576171875, 0.057373046875)]
model.layers.0.self_attn.q_norm.weight A[min,mean,max]= [-3.78125, 0.9641852378845215, 3.5625] B[min,mean,max]= [-3.78125, 0.9641826152801514, 3.5625] largest (index,A,B)= [(56, -0.08251953125, -0.0830078125), (199, -0.0380859375, -0.038330078125), (104, 0.05859375, 0.058837890625), (46, 0.046630859375, 0.04638671875)]
model.layers.1.input_layernorm.weight A[min,mean,max]= [-0.171875, 24.65911293029785, 47.5] B[min,mean,max]= [-0.171875, 24.65911293029785, 47.5] largest (index,A,B)= [(639, 1.296875, 1.296875), (638, 20.75, 20.75), (217, 33.25, 33.25), (216, 22.5, 22.5)]
model.layers.1.post_attention_layernorm.weight A[min,mean,max]= [-1.0078125, 2.750606060028076, 103.0] B[min,mean,max]= [-1.0078125, 2.7506070137023926, 103.0] largest (index,A,B)= [(389, 0.1884765625, 0.189453125), (486, -0.04052734375, -0.041015625), (95, 0.1201171875, 0.11962890625), (557, -0.0703125, -0.07080078125)]
model.layers.1.post_feedforward_layernorm.weight A[min,mean,max]= [-1.0, 1.8786900043487549, 195.0] B[min,mean,max]= [-1.0, 1.8786838054656982, 195.0] largest (index,A,B)= [(505, 0.146484375, 0.1455078125), (363, 0.197265625, 0.1982421875), (308, -0.166015625, -0.1650390625), (553, 0.2177734375, 0.21875)]
model.layers.1.pre_feedforward_layernorm.weight A[min,mean,max]= [-1.0, 28.950634002685547, 55.75] B[min,mean,max]= [-1.0, 28.950634002685547, 55.75] largest (index,A,B)= [(639, 1.734375, 1.734375), (638, 31.0, 31.0), (217, 36.5, 36.5), (216, 31.125, 31.125)]
model.layers.1.self_attn.k_norm.weight A[min,mean,max]= [-1.0234375, 1.6646957397460938, 7.875] B[min,mean,max]= [-1.0234375, 1.6646957397460938, 7.875] largest (index,A,B)= [(255, 1.828125, 1.828125), (254, 2.203125, 2.203125), (93, 0.875, 0.875), (92, 3.1875, 3.1875)]
model.layers.1.self_attn.q_norm.weight A[min,mean,max]= [-1.1171875, 1.6944160461425781, 14.4375] B[min,mean,max]= [-1.1171875, 1.6944160461425781, 14.4375] largest (index,A,B)= [(255, 1.4765625, 1.4765625), (254, 1.3671875, 1.3671875), (93, 1.7890625, 1.7890625), (92, 0.78125, 0.78125)]
model.layers.10.input_layernorm.weight A[min,mean,max]= [-0.263671875, 16.899927139282227, 100.5] B[min,mean,max]= [-0.263671875, 16.899927139282227, 100.5] largest (index,A,B)= [(639, 15.0625, 15.0625), (638, 9.375, 9.375), (217, 9.3125, 9.3125), (216, 12.8125, 12.8125)]
model.layers.10.post_attention_layernorm.weight A[min,mean,max]= [-71.5, 42.12190246582031, 482.0] B[min,mean,max]= [-71.5, 42.12190246582031, 482.0] largest (index,A,B)= [(54, 0.08935546875, 0.08984375), (356, -0.0133056640625, -0.01312255859375), (195, -0.00848388671875, -0.0084228515625), (639, 16.75, 16.75)]
model.layers.10.post_feedforward_layernorm.weight A[min,mean,max]= [-1.2109375, 55.94256591796875, 592.0] B[min,mean,max]= [-1.2109375, 55.94256591796875, 592.0] largest (index,A,B)= [(639, 39.25, 39.25), (638, 72.5, 72.5), (217, 51.5, 51.5), (216, 53.0, 53.0)]
model.layers.10.pre_feedforward_layernorm.weight A[min,mean,max]= [-1.03125, 3.502417802810669, 38.25] B[min,mean,max]= [-1.03125, 3.5024161338806152, 38.25] largest (index,A,B)= [(45, 0.13671875, 0.1376953125), (270, -0.166015625, -0.1669921875), (338, -0.10595703125, -0.1064453125), (360, 0.061279296875, 0.061767578125)]
model.layers.10.self_attn.k_norm.weight A[min,mean,max]= [-1.0, 1.6845829486846924, 21.25] B[min,mean,max]= [-1.0, 1.6845824718475342, 21.25] largest (index,A,B)= [(1, 0.08740234375, 0.087890625), (2, 0.0184326171875, 0.0179443359375), (32, 0.07373046875, 0.0732421875), (217, -0.07470703125, -0.07421875)]
model.layers.10.self_attn.q_norm.weight A[min,mean,max]= [-0.98828125, 0.6117059588432312, 3.3125] B[min,mean,max]= [-0.98828125, 0.6117131114006042, 3.3125] largest (index,A,B)= [(250, 0.18359375, 0.1845703125), (155, -0.02001953125, -0.0194091796875), (63, 0.095703125, 0.09619140625), (152, 0.08837890625, 0.0888671875)]
model.layers.11.input_layernorm.weight A[min,mean,max]= [-0.84375, 4.286688327789307, 52.25] B[min,mean,max]= [-0.84375, 4.286690711975098, 52.25] largest (index,A,B)= [(419, 0.1708984375, 0.169921875), (187, 0.2080078125, 0.208984375), (1, 0.03173828125, 0.0322265625), (420, -0.0634765625, -0.06298828125)]
model.layers.11.post_attention_layernorm.weight A[min,mean,max]= [-0.95703125, 110.36897277832031, 568.0] B[min,mean,max]= [-0.95703125, 110.36897277832031, 568.0] largest (index,A,B)= [(639, 30.0, 30.0), (638, 54.0, 54.0), (217, 57.75, 57.75), (216, 36.25, 36.25)]
model.layers.11.post_feedforward_layernorm.weight A[min,mean,max]= [-0.98046875, 112.866943359375, 892.0] B[min,mean,max]= [-0.98046875, 112.866943359375, 892.0] largest (index,A,B)= [(639, 51.5, 51.5), (638, 129.0, 129.0), (217, 71.0, 71.0), (216, 86.5, 86.5)]
model.layers.11.pre_feedforward_layernorm.weight A[min,mean,max]= [-1.0546875, 4.350240230560303, 53.75] B[min,mean,max]= [-1.0546875, 4.350237846374512, 53.75] largest (index,A,B)= [(107, -0.134765625, -0.1357421875), (392, -0.052001953125, -0.051513671875), (1, -0.09814453125, -0.0986328125), (459, -0.041748046875, -0.0419921875)]
model.layers.11.self_attn.k_norm.weight A[min,mean,max]= [-1.0, 2.510615348815918, 29.5] B[min,mean,max]= [-1.0, 2.510612726211548, 29.5] largest (index,A,B)= [(34, -0.033447265625, -0.033935546875), (166, 0.11962890625, 0.1201171875), (38, 0.0311279296875, 0.0308837890625), (135, 0.05908203125, 0.058837890625)]
model.layers.11.self_attn.q_norm.weight A[min,mean,max]= [-1.046875, 0.5513417720794678, 6.09375] B[min,mean,max]= [-1.046875, 0.5513443946838379, 6.09375] largest (index,A,B)= [(222, -0.068359375, -0.06787109375), (221, -0.06591796875, -0.0654296875), (65, 0.0966796875, 0.09619140625), (92, 0.0264892578125, 0.0267333984375)]
model.layers.12.input_layernorm.weight A[min,mean,max]= [-0.62109375, 14.544893264770508, 160.0] B[min,mean,max]= [-0.62109375, 14.544894218444824, 160.0] largest (index,A,B)= [(0, 0.08251953125, 0.08203125), (260, -0.115234375, -0.11474609375), (338, -0.10107421875, -0.1005859375), (144, -0.00390625, -0.00360107421875)]
model.layers.12.post_attention_layernorm.weight A[min,mean,max]= [-0.59765625, 130.19558715820312, 616.0] B[min,mean,max]= [-0.59765625, 130.19558715820312, 616.0] largest (index,A,B)= [(639, 28.0, 28.0), (638, 91.0, 91.0), (217, 57.25, 57.25), (216, 68.0, 68.0)]
model.layers.12.post_feedforward_layernorm.weight A[min,mean,max]= [-1.8203125, 162.47103881835938, 740.0] B[min,mean,max]= [-1.8203125, 162.47103881835938, 740.0] largest (index,A,B)= [(639, 68.5, 68.5), (638, 181.0, 181.0), (217, 75.0, 75.0), (216, 134.0, 134.0)]
model.layers.12.pre_feedforward_layernorm.weight A[min,mean,max]= [-1.0390625, 2.4050889015197754, 30.875] B[min,mean,max]= [-1.0390625, 2.4050936698913574, 30.875] largest (index,A,B)= [(467, -0.17578125, -0.1748046875), (198, 0.0654296875, 0.06591796875), (311, -0.08447265625, -0.083984375), (362, 0.095703125, 0.09619140625)]
model.layers.12.self_attn.k_norm.weight A[min,mean,max]= [-3.03125, 1.225684642791748, 6.6875] B[min,mean,max]= [-3.03125, 1.2256803512573242, 6.6875] largest (index,A,B)= [(207, 0.1669921875, 0.166015625), (205, -0.0013275146484375, -0.00146484375), (255, 2.015625, 2.015625), (91, 3.640625, 3.640625)]
model.layers.12.self_attn.q_norm.weight A[min,mean,max]= [-1.0546875, 0.7926902770996094, 3.921875] B[min,mean,max]= [-1.0546875, 0.7926936149597168, 3.921875] largest (index,A,B)= [(102, 0.06103515625, 0.061767578125), (114, -0.05810546875, -0.05859375), (91, 0.01025390625, 0.0106201171875), (247, -0.037841796875, -0.03759765625)]
model.layers.13.input_layernorm.weight A[min,mean,max]= [-0.2333984375, 11.304201126098633, 122.0] B[min,mean,max]= [-0.2333984375, 11.304200172424316, 122.0] largest (index,A,B)= [(69, -0.062255859375, -0.06298828125), (167, 0.0064697265625, 0.005950927734375), (154, 0.1220703125, 0.12255859375), (514, -0.103515625, -0.10400390625)]
model.layers.13.post_attention_layernorm.weight A[min,mean,max]= [-2.03125, 124.32344055175781, 454.0] B[min,mean,max]= [-2.03125, 124.32344055175781, 454.0] largest (index,A,B)= [(639, 41.75, 41.75), (638, 115.5, 115.5), (217, 40.25, 40.25), (216, 89.0, 89.0)]
model.layers.13.post_feedforward_layernorm.weight A[min,mean,max]= [-0.91015625, 175.43789672851562, 492.0] B[min,mean,max]= [-0.91015625, 175.43789672851562, 492.0] largest (index,A,B)= [(639, 61.0, 61.0), (638, 208.0, 208.0), (217, 65.5, 65.5), (216, 172.0, 172.0)]
model.layers.13.pre_feedforward_layernorm.weight A[min,mean,max]= [-0.9921875, 2.1670565605163574, 29.0] B[min,mean,max]= [-0.9921875, 2.1670520305633545, 29.0] largest (index,A,B)= [(356, 0.2216796875, 0.220703125), (489, 0.1796875, 0.1787109375), (376, -0.1962890625, -0.197265625), (349, 0.107421875, 0.10791015625)]
model.layers.13.self_attn.k_norm.weight A[min,mean,max]= [-1.0, 1.0408673286437988, 11.8125] B[min,mean,max]= [-1.0, 1.0408670902252197, 11.8125] largest (index,A,B)= [(139, -0.244140625, -0.2431640625), (38, -0.09130859375, -0.091796875), (11, 0.058837890625, 0.05859375), (49, -0.016845703125, -0.0169677734375)]
model.layers.13.self_attn.q_norm.weight A[min,mean,max]= [-1.0, 0.5665218830108643, 5.875] B[min,mean,max]= [-1.0, 0.5665175914764404, 5.875] largest (index,A,B)= [(120, -0.1875, -0.1884765625), (110, -0.138671875, -0.1396484375), (227, -0.234375, -0.2333984375), (204, -0.064453125, -0.06494140625)]
model.layers.14.input_layernorm.weight A[min,mean,max]= [-0.3046875, 16.29885482788086, 171.0] B[min,mean,max]= [-0.3046875, 16.298852920532227, 171.0] largest (index,A,B)= [(315, -0.125, -0.1259765625), (163, -0.2421875, -0.2412109375), (327, 0.099609375, 0.10009765625), (156, -0.091796875, -0.09228515625)]
model.layers.14.post_attention_layernorm.weight A[min,mean,max]= [0.6640625, 161.01527404785156, 482.0] B[min,mean,max]= [0.6640625, 161.01527404785156, 482.0] largest (index,A,B)= [(639, 69.0, 69.0), (638, 216.0, 216.0), (217, 77.0, 77.0), (216, 198.0, 198.0)]
model.layers.14.post_feedforward_layernorm.weight A[min,mean,max]= [1.515625, 207.31106567382812, 788.0] B[min,mean,max]= [1.515625, 207.31106567382812, 788.0] largest (index,A,B)= [(639, 80.0, 80.0), (638, 282.0, 282.0), (217, 76.5, 76.5), (216, 223.0, 223.0)]
model.layers.14.pre_feedforward_layernorm.weight A[min,mean,max]= [-1.03125, 1.8935438394546509, 28.25] B[min,mean,max]= [-1.03125, 1.893537163734436, 28.25] largest (index,A,B)= [(376, -0.22265625, -0.2236328125), (439, -0.244140625, -0.2431640625), (225, -0.1044921875, -0.10400390625), (513, 0.08837890625, 0.0888671875)]
model.layers.14.self_attn.k_norm.weight A[min,mean,max]= [-1.0390625, 1.5548053979873657, 10.3125] B[min,mean,max]= [-1.0390625, 1.554808497428894, 10.3125] largest (index,A,B)= [(86, 0.1962890625, 0.197265625), (72, -0.140625, -0.1396484375), (56, -0.20703125, -0.2080078125), (197, 0.08740234375, 0.0869140625)]
model.layers.14.self_attn.q_norm.weight A[min,mean,max]= [-1.0390625, 0.9955974221229553, 4.46875] B[min,mean,max]= [-1.0390625, 0.9956014752388, 4.46875] largest (index,A,B)= [(247, -0.1875, -0.1865234375), (246, -0.10205078125, -0.1025390625), (215, 0.042724609375, 0.043212890625), (255, -0.037841796875, -0.03759765625)]
model.layers.15.input_layernorm.weight A[min,mean,max]= [0.17578125, 17.181549072265625, 225.0] B[min,mean,max]= [0.1767578125, 17.181550979614258, 225.0] largest (index,A,B)= [(163, 0.17578125, 0.1767578125), (639, 26.0, 26.0), (210, 9.625, 9.625), (217, 21.25, 21.25)]
model.layers.15.post_attention_layernorm.weight A[min,mean,max]= [-10.4375, 170.9198760986328, 712.0] B[min,mean,max]= [-10.4375, 170.9198760986328, 712.0] largest (index,A,B)= [(639, 49.25, 49.25), (638, 137.0, 137.0), (217, 60.0, 60.0), (216, 110.5, 110.5)]
model.layers.15.post_feedforward_layernorm.weight A[min,mean,max]= [-6.0, 244.0491180419922, 1192.0] B[min,mean,max]= [-6.0, 244.0491180419922, 1192.0] largest (index,A,B)= [(639, 85.5, 85.5), (638, 346.0, 346.0), (217, 89.0, 89.0), (216, 298.0, 298.0)]
model.layers.15.pre_feedforward_layernorm.weight A[min,mean,max]= [-1.046875, 1.1252580881118774, 20.875] B[min,mean,max]= [-1.046875, 1.1252613067626953, 20.875] largest (index,A,B)= [(503, 0.2197265625, 0.220703125), (11, 0.1455078125, 0.14453125), (493, -0.1162109375, -0.11669921875), (105, -0.0703125, -0.06982421875)]
model.layers.15.self_attn.k_norm.weight A[min,mean,max]= [-1.3359375, 0.9434323310852051, 13.4375] B[min,mean,max]= [-1.3359375, 0.9434309005737305, 13.4375] largest (index,A,B)= [(20, -0.1376953125, -0.13671875), (49, 0.1767578125, 0.17578125), (30, 0.07470703125, 0.0751953125), (139, -0.09423828125, -0.09375)]
model.layers.15.self_attn.q_norm.weight A[min,mean,max]= [-1.0078125, 0.9584987759590149, 12.0] B[min,mean,max]= [-1.0078125, 0.9584981203079224, 12.0] largest (index,A,B)= [(141, 0.2470703125, 0.24609375), (96, -0.08203125, -0.08154296875), (37, 0.1005859375, 0.10107421875), (5, 0.091796875, 0.09130859375)]
model.layers.16.input_layernorm.weight A[min,mean,max]= [0.51953125, 20.35601806640625, 233.0] B[min,mean,max]= [0.51953125, 20.35601806640625, 233.0] largest (index,A,B)= [(639, 46.0, 46.0), (638, 5.6875, 5.6875), (217, 24.375, 24.375), (216, 6.78125, 6.78125)]
model.layers.16.post_attention_layernorm.weight A[min,mean,max]= [0.259765625, 140.06993103027344, 372.0] B[min,mean,max]= [0.259765625, 140.06993103027344, 372.0] largest (index,A,B)= [(639, 55.25, 55.25), (638, 192.0, 192.0), (217, 58.75, 58.75), (216, 169.0, 169.0)]
model.layers.16.post_feedforward_layernorm.weight A[min,mean,max]= [-10.4375, 283.35968017578125, 1656.0] B[min,mean,max]= [-10.4375, 283.35968017578125, 1656.0] largest (index,A,B)= [(639, 93.5, 93.5), (638, 402.0, 402.0), (217, 99.5, 99.5), (216, 322.0, 322.0)]
model.layers.16.pre_feedforward_layernorm.weight A[min,mean,max]= [-1.0546875, 1.0023088455200195, 23.25] B[min,mean,max]= [-1.0546875, 1.0023064613342285, 23.25] largest (index,A,B)= [(567, -0.1640625, -0.1650390625), (286, -0.201171875, -0.2021484375), (556, 0.15625, 0.1572265625), (407, 0.03857421875, 0.037841796875)]
model.layers.16.self_attn.k_norm.weight A[min,mean,max]= [-1.0, 1.2281361818313599, 8.4375] B[min,mean,max]= [-1.0, 1.2281372547149658, 8.4375] largest (index,A,B)= [(74, -0.1689453125, -0.169921875), (75, 0.0625, 0.06298828125), (73, 0.017333984375, 0.017822265625), (184, 0.000545501708984375, 0.000881195068359375)]
model.layers.16.self_attn.q_norm.weight A[min,mean,max]= [-1.015625, 0.98746657371521, 10.25] B[min,mean,max]= [-1.015625, 0.9874651432037354, 10.25] largest (index,A,B)= [(237, -0.1181640625, -0.11767578125), (208, 0.1181640625, 0.11767578125), (199, 0.07373046875, 0.07421875), (217, 0.0673828125, 0.06689453125)]
model.layers.17.input_layernorm.weight A[min,mean,max]= [0.28515625, 10.484280586242676, 121.0] B[min,mean,max]= [0.28515625, 10.484280586242676, 121.0] largest (index,A,B)= [(639, 18.375, 18.375), (638, 1.7578125, 1.7578125), (217, 10.625, 10.625), (216, 2.234375, 2.234375)]
model.layers.17.post_attention_layernorm.weight A[min,mean,max]= [-1.5078125, 134.44313049316406, 448.0] B[min,mean,max]= [-1.5078125, 134.44313049316406, 448.0] largest (index,A,B)= [(353, -0.0133056640625, -0.01318359375), (639, 55.25, 55.25), (209, 73.0, 73.0), (216, 140.0, 140.0)]
model.layers.17.post_feedforward_layernorm.weight A[min,mean,max]= [-1.1328125, 322.94384765625, 1648.0] B[min,mean,max]= [-1.1328125, 322.94384765625, 1648.0] largest (index,A,B)= [(639, 188.0, 188.0), (638, 450.0, 450.0), (217, 150.0, 150.0), (216, 338.0, 338.0)]
model.layers.17.pre_feedforward_layernorm.weight A[min,mean,max]= [-1.0546875, 1.1489206552505493, 29.625] B[min,mean,max]= [-1.0546875, 1.1489224433898926, 29.625] largest (index,A,B)= [(48, -0.23828125, -0.2392578125), (255, -0.1357421875, -0.134765625), (25, 0.01080322265625, 0.011474609375), (24, -0.01446533203125, -0.013916015625)]
model.layers.17.self_attn.k_norm.weight A[min,mean,max]= [-3.53125, 1.5867414474487305, 5.8125] B[min,mean,max]= [-3.53125, 1.5867462158203125, 5.8125] largest (index,A,B)= [(110, 0.11328125, 0.11376953125), (120, -0.12451171875, -0.1240234375), (127, 0.031494140625, 0.03173828125), (248, 0.035400390625, 0.03515625)]
model.layers.17.self_attn.q_norm.weight A[min,mean,max]= [-1.0390625, 0.8164167404174805, 16.125] B[min,mean,max]= [-1.0390625, 0.8164145946502686, 16.125] largest (index,A,B)= [(60, -0.1806640625, -0.1796875), (64, -0.044677734375, -0.045166015625), (207, -0.06396484375, -0.064453125), (222, 0.0103759765625, 0.0098876953125)]
model.layers.2.input_layernorm.weight A[min,mean,max]= [-0.578125, 28.82571029663086, 93.5] B[min,mean,max]= [-0.578125, 28.82571029663086, 93.5] largest (index,A,B)= [(97, 0.040771484375, 0.041015625), (639, 8.0625, 8.0625), (219, 21.125, 21.125), (217, 51.0, 51.0)]
model.layers.2.post_attention_layernorm.weight A[min,mean,max]= [-1.125, 0.8022615313529968, 109.5] B[min,mean,max]= [-1.125, 0.8022602200508118, 109.5] largest (index,A,B)= [(247, -0.2451171875, -0.244140625), (451, -0.1923828125, -0.19140625), (27, -0.095703125, -0.0966796875), (250, -0.052734375, -0.05322265625)]
model.layers.2.post_feedforward_layernorm.weight A[min,mean,max]= [-1.171875, 4.607209205627441, 250.0] B[min,mean,max]= [-1.171875, 4.607211112976074, 250.0] largest (index,A,B)= [(74, 0.224609375, 0.2255859375), (602, 0.10009765625, 0.1005859375), (506, 0.10888671875, 0.1083984375), (312, 0.0286865234375, 0.028564453125)]
model.layers.2.pre_feedforward_layernorm.weight A[min,mean,max]= [-1.0, 22.05205726623535, 65.0] B[min,mean,max]= [-1.0, 22.052059173583984, 65.0] largest (index,A,B)= [(134, 0.1962890625, 0.197265625), (639, 1.78125, 1.78125), (200, 19.25, 19.25), (217, 35.5, 35.5)]
model.layers.2.self_attn.k_norm.weight A[min,mean,max]= [-1.0, 1.0676209926605225, 22.0] B[min,mean,max]= [-1.0, 1.0676202774047852, 22.0] largest (index,A,B)= [(95, 0.060302734375, 0.06005859375), (92, 0.040771484375, 0.04052734375), (103, 0.061279296875, 0.0615234375), (24, -0.048828125, -0.048583984375)]
model.layers.2.self_attn.q_norm.weight A[min,mean,max]= [-1.0, 0.7755727767944336, 3.015625] B[min,mean,max]= [-1.0, 0.7755727767944336, 3.015625] largest (index,A,B)= [(230, 0.1787109375, 0.1796875), (70, 0.11572265625, 0.115234375), (78, 0.02294921875, 0.022705078125), (106, -0.046630859375, -0.046875)]
model.layers.3.input_layernorm.weight A[min,mean,max]= [-0.67578125, 24.901037216186523, 62.0] B[min,mean,max]= [-0.67578125, 24.90103530883789, 62.0] largest (index,A,B)= [(134, 0.051513671875, 0.05126953125), (632, 0.04150390625, 0.041259765625), (639, 2.8125, 2.8125), (209, 6.53125, 6.53125)]
model.layers.3.post_attention_layernorm.weight A[min,mean,max]= [-6.96875, 4.106741905212402, 221.0] B[min,mean,max]= [-6.96875, 4.106744289398193, 221.0] largest (index,A,B)= [(226, 0.201171875, 0.2021484375), (557, -0.2158203125, -0.216796875), (612, 0.12451171875, 0.1240234375), (3, -0.08154296875, -0.0810546875)]
model.layers.3.post_feedforward_layernorm.weight A[min,mean,max]= [-0.99609375, 12.409502029418945, 584.0] B[min,mean,max]= [-0.99609375, 12.409501075744629, 584.0] largest (index,A,B)= [(527, -0.212890625, -0.2138671875), (565, -0.185546875, -0.1865234375), (309, 0.08740234375, 0.087890625), (220, 0.0869140625, 0.08740234375)]
model.layers.3.pre_feedforward_layernorm.weight A[min,mean,max]= [-1.0, 24.31264877319336, 69.5] B[min,mean,max]= [-1.0, 24.312650680541992, 69.5] largest (index,A,B)= [(608, 0.09228515625, 0.09326171875), (267, 0.08935546875, 0.0888671875), (632, 0.01141357421875, 0.0111083984375), (163, 0.050537109375, 0.05078125)]
model.layers.3.self_attn.k_norm.weight A[min,mean,max]= [-1.2578125, 0.7500784397125244, 8.25] B[min,mean,max]= [-1.2578125, 0.7500808238983154, 8.25] largest (index,A,B)= [(14, 0.00750732421875, 0.0079345703125), (140, 0.00872802734375, 0.0084228515625), (43, -0.03564453125, -0.035400390625), (31, -0.047607421875, -0.0478515625)]
model.layers.3.self_attn.q_norm.weight A[min,mean,max]= [-0.9921875, 0.5277363657951355, 3.25] B[min,mean,max]= [-0.9921875, 0.5277371406555176, 3.25] largest (index,A,B)= [(148, -0.0458984375, -0.046630859375), (112, -0.0859375, -0.08544921875), (105, 0.1044921875, 0.10400390625), (6, 0.09619140625, 0.0966796875)]
model.layers.4.input_layernorm.weight A[min,mean,max]= [-0.58984375, 34.19289779663086, 162.0] B[min,mean,max]= [-0.58984375, 34.192893981933594, 162.0] largest (index,A,B)= [(193, -0.11474609375, -0.115234375), (426, -0.003082275390625, -0.00335693359375), (399, 0.0286865234375, 0.0289306640625), (132, -0.041259765625, -0.04150390625)]
model.layers.4.post_attention_layernorm.weight A[min,mean,max]= [-3.3125, 3.1820502281188965, 221.0] B[min,mean,max]= [-3.3125, 3.182051420211792, 221.0] largest (index,A,B)= [(305, 0.0927734375, 0.09326171875), (320, 0.12158203125, 0.12109375), (44, 0.1083984375, 0.10888671875), (115, 0.0869140625, 0.08740234375)]
model.layers.4.post_feedforward_layernorm.weight A[min,mean,max]= [-1.0078125, 6.990669250488281, 608.0] B[min,mean,max]= [-1.0078125, 6.990669250488281, 608.0] largest (index,A,B)= [(171, 0.056884765625, 0.056640625), (460, -0.0198974609375, -0.019775390625), (639, 21.375, 21.375), (210, 10.625, 10.625)]
model.layers.4.pre_feedforward_layernorm.weight A[min,mean,max]= [-1.0, 12.555732727050781, 51.75] B[min,mean,max]= [-1.0, 12.555730819702148, 51.75] largest (index,A,B)= [(68, 0.031982421875, 0.032470703125), (541, -0.1181640625, -0.11865234375), (552, 0.054931640625, 0.055419921875), (561, -0.07470703125, -0.0751953125)]
model.layers.4.self_attn.k_norm.weight A[min,mean,max]= [-1.3984375, 1.0646915435791016, 13.9375] B[min,mean,max]= [-1.3984375, 1.0646846294403076, 13.9375] largest (index,A,B)= [(59, -0.06787109375, -0.068359375), (165, -0.031005859375, -0.031494140625), (254, -0.0185546875, -0.0189208984375), (30, -0.0225830078125, -0.02294921875)]
model.layers.4.self_attn.q_norm.weight A[min,mean,max]= [-1.0078125, 0.5517474412918091, 5.09375] B[min,mean,max]= [-1.0078125, 0.5517475605010986, 5.09375] largest (index,A,B)= [(254, 0.109375, 0.10888671875), (31, 0.07470703125, 0.0751953125), (24, 0.0888671875, 0.08935546875), (112, 0.11669921875, 0.1171875)]
model.layers.5.input_layernorm.weight A[min,mean,max]= [-0.6953125, 13.057644844055176, 48.25] B[min,mean,max]= [-0.6953125, 13.057644844055176, 48.25] largest (index,A,B)= [(24, -0.111328125, -0.11181640625), (270, 0.043701171875, 0.044189453125), (428, 0.0181884765625, 0.017822265625), (399, -0.041259765625, -0.04150390625)]
model.layers.5.post_attention_layernorm.weight A[min,mean,max]= [-1.03125, 9.784034729003906, 512.0] B[min,mean,max]= [-1.03125, 9.784035682678223, 512.0] largest (index,A,B)= [(339, -0.0341796875, -0.033935546875), (616, -0.0478515625, -0.047607421875), (639, 33.25, 33.25), (216, 7.1875, 7.1875)]
model.layers.5.post_feedforward_layernorm.weight A[min,mean,max]= [-1.09375, 10.628747940063477, 740.0] B[min,mean,max]= [-1.09375, 10.628747940063477, 740.0] largest (index,A,B)= [(639, 24.375, 24.375), (638, 6.96875, 6.96875), (217, 31.75, 31.75), (216, 9.25, 9.25)]
model.layers.5.pre_feedforward_layernorm.weight A[min,mean,max]= [-1.0, 11.419027328491211, 46.0] B[min,mean,max]= [-1.0, 11.419026374816895, 46.0] largest (index,A,B)= [(271, 0.01190185546875, 0.0126953125), (357, -0.072265625, -0.07275390625), (426, -0.09521484375, -0.095703125), (195, -0.0947265625, -0.09521484375)]
model.layers.5.self_attn.k_norm.weight A[min,mean,max]= [-3.6875, 1.9305553436279297, 22.625] B[min,mean,max]= [-3.6875, 1.9305553436279297, 22.625] largest (index,A,B)= [(255, 6.03125, 6.03125), (254, 9.4375, 9.4375), (93, 0.73046875, 0.73046875), (92, 0.77734375, 0.77734375)]
model.layers.5.self_attn.q_norm.weight A[min,mean,max]= [-3.0625, 1.001765489578247, 5.375] B[min,mean,max]= [-3.0625, 1.0017684698104858, 5.375] largest (index,A,B)= [(111, -0.0018768310546875, -0.00136566162109375), (255, -0.0751953125, -0.07470703125), (250, -0.037841796875, -0.0380859375), (65, 1.8203125, 1.8203125)]
model.layers.6.input_layernorm.weight A[min,mean,max]= [-0.578125, 20.80784797668457, 101.5] B[min,mean,max]= [-0.578125, 20.807846069335938, 101.5] largest (index,A,B)= [(105, 0.09912109375, 0.099609375), (134, 0.04833984375, 0.048095703125), (368, 0.00750732421875, 0.00732421875), (503, 0.0206298828125, 0.0205078125)]
model.layers.6.post_attention_layernorm.weight A[min,mean,max]= [-14.125, 8.017030715942383, 352.0] B[min,mean,max]= [-14.125, 8.0170316696167, 352.0] largest (index,A,B)= [(451, 0.1689453125, 0.169921875), (4, -0.07666015625, -0.0771484375), (431, -0.08203125, -0.08154296875), (263, 0.11279296875, 0.11328125)]
model.layers.6.post_feedforward_layernorm.weight A[min,mean,max]= [-1.1796875, 13.7931547164917, 784.0] B[min,mean,max]= [-1.1796875, 13.7931547164917, 784.0] largest (index,A,B)= [(639, 21.375, 21.375), (638, 11.8125, 11.8125), (217, 27.25, 27.25), (216, 15.9375, 15.9375)]
model.layers.6.pre_feedforward_layernorm.weight A[min,mean,max]= [-1.015625, 7.152066230773926, 25.875] B[min,mean,max]= [-1.015625, 7.152062892913818, 25.875] largest (index,A,B)= [(426, -0.1298828125, -0.130859375), (561, 0.0966796875, 0.09716796875), (541, -0.06884765625, -0.0693359375), (523, -0.042236328125, -0.042724609375)]
model.layers.6.self_attn.k_norm.weight A[min,mean,max]= [-1.0859375, 0.6734185218811035, 19.75] B[min,mean,max]= [-1.0859375, 0.6734278202056885, 19.75] largest (index,A,B)= [(219, 0.130859375, 0.1318359375), (156, 0.09716796875, 0.0966796875), (111, 0.11962890625, 0.1201171875), (36, 0.08203125, 0.08251953125)]
model.layers.6.self_attn.q_norm.weight A[min,mean,max]= [-1.015625, 0.4593641459941864, 2.9375] B[min,mean,max]= [-1.015625, 0.4593706429004669, 2.9375] largest (index,A,B)= [(233, -0.232421875, -0.2314453125), (187, 0.1884765625, 0.189453125), (249, 0.2060546875, 0.20703125), (116, -0.1142578125, -0.11474609375)]
model.layers.7.input_layernorm.weight A[min,mean,max]= [-0.26953125, 20.035648345947266, 71.0] B[min,mean,max]= [-0.26953125, 20.035648345947266, 71.0] largest (index,A,B)= [(608, 0.09326171875, 0.09375), (639, 7.90625, 7.90625), (209, 2.890625, 2.890625), (216, 18.625, 18.625)]
model.layers.7.post_attention_layernorm.weight A[min,mean,max]= [-0.9765625, 10.41993522644043, 392.0] B[min,mean,max]= [-0.9765625, 10.41993522644043, 392.0] largest (index,A,B)= [(639, 11.875, 11.875), (638, 8.8125, 8.8125), (217, 17.875, 17.875), (216, 12.8125, 12.8125)]
model.layers.7.post_feedforward_layernorm.weight A[min,mean,max]= [-133.0, 16.61493492126465, 688.0] B[min,mean,max]= [-133.0, 16.61493492126465, 688.0] largest (index,A,B)= [(639, 21.875, 21.875), (638, 17.375, 17.375), (217, 24.875, 24.875), (216, 19.0, 19.0)]
model.layers.7.pre_feedforward_layernorm.weight A[min,mean,max]= [-1.0, 6.32569694519043, 27.0] B[min,mean,max]= [-1.0, 6.325697898864746, 27.0] largest (index,A,B)= [(25, -0.18359375, -0.1845703125), (132, 0.2236328125, 0.22265625), (524, -0.18359375, -0.1826171875), (406, 0.185546875, 0.1865234375)]
model.layers.7.self_attn.k_norm.weight A[min,mean,max]= [-1.0, 1.032013177871704, 12.6875] B[min,mean,max]= [-1.0, 1.032010555267334, 12.6875] largest (index,A,B)= [(196, -0.142578125, -0.1435546875), (90, 0.10595703125, 0.1064453125), (153, 0.1005859375, 0.10009765625), (202, -0.045654296875, -0.046142578125)]
model.layers.7.self_attn.q_norm.weight A[min,mean,max]= [-1.03125, 0.7932995557785034, 4.0] B[min,mean,max]= [-1.03125, 0.7933034896850586, 4.0] largest (index,A,B)= [(127, -0.0257568359375, -0.0252685546875), (98, 0.1103515625, 0.11083984375), (187, -0.107421875, -0.10791015625), (213, 0.0732421875, 0.07373046875)]
model.layers.8.input_layernorm.weight A[min,mean,max]= [-0.0208740234375, 19.915367126464844, 83.5] B[min,mean,max]= [-0.0205078125, 19.915369033813477, 83.5] largest (index,A,B)= [(632, 0.07080078125, 0.0712890625), (339, 0.091796875, 0.09228515625), (267, -0.0208740234375, -0.0205078125), (639, 13.5625, 13.5625)]
model.layers.8.post_attention_layernorm.weight A[min,mean,max]= [-2.125, 18.393497467041016, 406.0] B[min,mean,max]= [-2.125, 18.393497467041016, 406.0] largest (index,A,B)= [(4, -0.099609375, -0.10009765625), (590, -0.09814453125, -0.0986328125), (328, -0.048095703125, -0.04833984375), (574, -0.01019287109375, -0.0101318359375)]
model.layers.8.post_feedforward_layernorm.weight A[min,mean,max]= [-1.1484375, 20.039234161376953, 708.0] B[min,mean,max]= [-1.1484375, 20.039234161376953, 708.0] largest (index,A,B)= [(639, 20.25, 20.25), (638, 22.125, 22.125), (217, 27.0, 27.0), (216, 20.875, 20.875)]
model.layers.8.pre_feedforward_layernorm.weight A[min,mean,max]= [-1.0, 5.790274620056152, 26.625] B[min,mean,max]= [-1.0, 5.790274143218994, 26.625] largest (index,A,B)= [(524, -0.224609375, -0.2255859375), (61, -0.0185546875, -0.0179443359375), (32, -0.01361083984375, -0.0130615234375), (87, 0.06689453125, 0.06640625)]
model.layers.8.self_attn.k_norm.weight A[min,mean,max]= [-1.390625, 0.9154994487762451, 11.1875] B[min,mean,max]= [-1.390625, 0.9154965877532959, 11.1875] largest (index,A,B)= [(104, -0.1201171875, -0.11962890625), (108, 0.078125, 0.07763671875), (148, -0.025390625, -0.0257568359375), (163, -0.034423828125, -0.03466796875)]
model.layers.8.self_attn.q_norm.weight A[min,mean,max]= [-1.0546875, 0.6504087448120117, 8.875] B[min,mean,max]= [-1.0546875, 0.6503957509994507, 8.875] largest (index,A,B)= [(114, -0.08154296875, -0.08203125), (253, -0.06982421875, -0.0703125), (36, 0.038818359375, 0.038330078125), (38, -0.087890625, -0.08837890625)]
model.layers.9.input_layernorm.weight A[min,mean,max]= [-0.294921875, 16.71709632873535, 71.5] B[min,mean,max]= [-0.294921875, 16.717092514038086, 71.5] largest (index,A,B)= [(632, -0.07763671875, -0.078125), (230, 0.029052734375, 0.028564453125), (339, -0.07275390625, -0.0732421875), (572, 0.0166015625, 0.0162353515625)]
model.layers.9.post_attention_layernorm.weight A[min,mean,max]= [-1.1328125, 21.47821617126465, 450.0] B[min,mean,max]= [-1.1328125, 21.47821807861328, 450.0] largest (index,A,B)= [(293, 0.1083984375, 0.10888671875), (427, 0.0654296875, 0.06591796875), (54, 0.0390625, 0.038818359375), (574, -0.026123046875, -0.0260009765625)]
model.layers.9.post_feedforward_layernorm.weight A[min,mean,max]= [-0.90234375, 30.130413055419922, 524.0] B[min,mean,max]= [-0.90234375, 30.130413055419922, 524.0] largest (index,A,B)= [(639, 26.25, 26.25), (638, 38.5, 38.5), (217, 30.875, 30.875), (216, 31.75, 31.75)]
model.layers.9.pre_feedforward_layernorm.weight A[min,mean,max]= [-0.984375, 3.1125056743621826, 20.875] B[min,mean,max]= [-0.984375, 3.1125071048736572, 20.875] largest (index,A,B)= [(87, -0.2236328125, -0.22265625), (551, -0.11669921875, -0.1162109375), (32, -0.10400390625, -0.1044921875), (24, -0.03759765625, -0.0380859375)]
model.layers.9.self_attn.k_norm.weight A[min,mean,max]= [-1.0078125, 1.03554368019104, 15.0] B[min,mean,max]= [-1.0078125, 1.035546898841858, 15.0] largest (index,A,B)= [(79, 0.228515625, 0.2294921875), (227, -0.0693359375, -0.06884765625), (101, 0.10009765625, 0.1005859375), (198, -0.046875, -0.04736328125)]
model.layers.9.self_attn.q_norm.weight A[min,mean,max]= [-0.98828125, 0.8321380615234375, 5.9375] B[min,mean,max]= [-0.98828125, 0.8321335911750793, 5.9375] largest (index,A,B)= [(70, 0.1787109375, 0.1796875), (126, 0.2177734375, 0.216796875), (12, 0.248046875, 0.2470703125), (14, 0.1181640625, 0.11865234375)]
model.norm.weight A[min,mean,max]= [-0.9140625, 25.42266845703125, 300.0] B[min,mean,max]= [-0.9140625, 25.42266845703125, 300.0] largest (index,A,B)= [(639, 20.625, 20.625), (638, 5.03125, 5.03125), (217, 26.75, 26.75), (216, 7.0625, 7.0625)]

=== TOKENIZER STRUCTURE ===
Vocabulary entries: 262144
Special tokens: {0: '<pad>', 1: '<eos>', 2: '<bos>', 3: '<unk>', 4: '<mask>', 105: '<start_of_turn>', 106: '<end_of_turn>', 255999: '<start_of_image>', 256000: '<end_of_image>', 262144: '<image_soft_token>'}
Tokenizer config:
{
  "backend": "tokenizers",
  "boi_token": "<start_of_image>",
  "bos_token": "<bos>",
  "clean_up_tokenization_spaces": false,
  "eoi_token": "<end_of_image>",
  "eos_token": "<eos>",
  "image_token": "<image_soft_token>",
  "is_local": true,
  "local_files_only": true,
  "mask_token": "<mask>",
  "model_max_length": 1000000000000000019884624838656,
  "model_specific_special_tokens": {
    "boi_token": "<start_of_image>",
    "eoi_token": "<end_of_image>",
    "image_token": "<image_soft_token>"
  },
  "pad_token": "<pad>",
  "padding_side": "left",
  "sp_model_kwargs": null,
  "spaces_between_special_tokens": false,
  "tokenizer_class": "GemmaTokenizer",
  "unk_token": "<unk>",
  "use_default_system_prompt": false
}

No behavioral label inferred from these difference magnitudes.
