OUTPUT of run 1 of at most 60:
```
=== ARCHITECTURE ===
A config:
{
  "_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
}
Configs identical: True

=== TENSOR INVENTORY ===
Tensor counts: 235 235
Only in A: []
Only in B: []
Layer indices: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17]

Non-layer tensors:
model.norm.weight (640,) BF16

Tensor layout of layer 0:
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

Shape mismatches: []
Nonfinite tensors: []

=== GLOBAL DIFFERENCE SUMMARY ===
Compared scalar parameters: 100326016
Exactly equal tensors: 32
Changed scalar fraction: 0.7909969234699801
Global ||B-A||_F: 2.4499819121509616
Symmetric relative difference: 0.00010808547312499057

=== DIFFERENCES BY LAYER AND PARAMETER GROUP ===
These are descriptive differences, NOT causal importance scores.
layer group          parameters changed% relative_delta delta_energy% normB/normA
   -1 norm/other            640    0.000              0        0.0000     1.000000
    0 attention         1638400   74.973     0.00591896        1.3943     1.000011
    0 mlp               3932160   76.790     0.00690131        3.6549     0.999994
    0 norm/other           3072    3.418    1.12045e-05        0.0004     1.000000
    1 attention         1638400   78.282     0.00674871        1.4018     0.999988
    1 mlp               3932160   78.781     0.00729587        3.7922     0.999999
    1 norm/other           3072    1.953    3.19596e-06        0.0002     1.000000
    2 attention         1638400   77.021     0.00683122        1.6032     0.999988
    2 mlp               3932160   78.353     0.00723523        3.7968     1.000000
    2 norm/other           3072    2.051    3.44613e-06        0.0002     1.000000
    3 attention         1638400   80.226      0.0072564        1.5419     0.999998
    3 mlp               3932160   79.356     0.00718072        3.6574     0.999990
    3 norm/other           3072    1.790    2.25045e-06        0.0002     1.000000
    4 attention         1638400   80.096     0.00726861        1.5731     1.000010
    4 mlp               3932160   80.719     0.00742734        3.8578     0.999995
    4 norm/other           3072    2.083    2.31125e-06        0.0002     1.000000
    5 attention         1638400   80.375     0.00694586        1.7078     1.000015
    5 mlp               3932160   78.260     0.00711069        3.8926     0.999994
    5 norm/other           3072    0.879    1.78275e-06        0.0001     1.000000
    6 attention         1638400   79.406     0.00718304        1.8018     1.000004
    6 mlp               3932160   79.605     0.00731966        3.8242     1.000000
    6 norm/other           3072    1.921    2.81889e-06        0.0002     1.000000
    7 attention         1638400   78.122     0.00679245        1.6157     0.999995
    7 mlp               3932160   79.016     0.00721547        3.7844     1.000000
    7 norm/other           3072    1.335    2.82725e-06        0.0002     1.000000
    8 attention         1638400   79.193     0.00716659        1.6541     1.000005
    8 mlp               3932160   80.234     0.00736057        3.8152     0.999992
    8 norm/other           3072    1.660    2.32946e-06        0.0002     1.000000
    9 attention         1638400   78.525     0.00686355        1.7421     1.000010
    9 mlp               3932160   78.855     0.00719309        3.8398     0.999995
    9 norm/other           3072    1.921    2.27697e-06        0.0002     1.000000
   10 attention         1638400   78.004     0.00682887        1.9176     1.000009
   10 mlp               3932160   80.272     0.00737493        3.8545     0.999993
   10 norm/other           3072    1.432    1.12043e-06        0.0001     1.000000
   11 attention         1638400   79.060     0.00656583        1.7607     1.000005
   11 mlp               3932160   79.115     0.00723587        3.8579     1.000004
   11 norm/other           3072    1.042    3.98316e-07        0.0001     1.000000
   12 attention         1638400   80.283     0.00716661        1.8878     1.000004
   12 mlp               3932160   79.312     0.00734666        4.0607     0.999999
   12 norm/other           3072    1.204    4.00411e-07        0.0001     1.000000
   13 attention         1638400   80.141     0.00721613        1.7109     1.000009
   13 mlp               3932160   80.026     0.00766519        4.1671     1.000009
   13 norm/other           3072    2.767    5.88078e-07        0.0003     1.000000
   14 attention         1638400   78.407     0.00659811        1.6215     1.000005
   14 mlp               3932160   80.583     0.00784049        4.3254     1.000005
   14 norm/other           3072    3.678    5.82501e-07        0.0003     1.000000
   15 attention         1638400   77.998     0.00717706        1.8393     0.999996
   15 mlp               3932160   79.964     0.00773986        4.3219     0.999996
   15 norm/other           3072    1.888    4.01052e-07        0.0003     1.000000
   16 attention         1638400   79.967     0.00742445        1.6089     1.000010
   16 mlp               3932160   78.967     0.00742196        4.0210     1.000013
   16 norm/other           3072    1.074     2.1777e-07        0.0001     1.000000
   17 attention         1638400   78.463     0.00647325        1.4416     1.000011
   17 mlp               3932160   78.874      0.0071387        3.6488     0.999994
   17 norm/other           3072    1.302    2.65725e-07        0.0001     1.000000

=== TOP 20 TENSORS BY ABSOLUTE DIFFERENCE ENERGY ===
model.layers.14.mlp.gate_proj.weight                              energy%=   1.521 relative=0.00676324 max_abs_delta=0.00195312
model.layers.15.mlp.gate_proj.weight                              energy%=   1.498 relative=0.00670883 max_abs_delta=0.00134277
model.layers.15.mlp.up_proj.weight                                energy%=   1.496 relative=0.00670756 max_abs_delta=0.00146484
model.layers.14.mlp.up_proj.weight                                energy%=   1.478 relative=0.00667512 max_abs_delta=0.00146484
model.layers.13.mlp.gate_proj.weight                              energy%=   1.472 relative=0.00665486 max_abs_delta=0.00146484
model.layers.12.mlp.gate_proj.weight                              energy%=   1.452 relative=0.006594 max_abs_delta=0.00146484
model.layers.13.mlp.up_proj.weight                                energy%=   1.448 relative=0.00661386 max_abs_delta=0.00146484
model.layers.11.mlp.gate_proj.weight                              energy%=   1.432 relative=0.00654692 max_abs_delta=0.00195312
model.layers.6.mlp.gate_proj.weight                               energy%=   1.422 relative=0.00653479 max_abs_delta=0.00195312
model.layers.5.mlp.gate_proj.weight                               energy%=   1.414 relative=0.00650045 max_abs_delta=0.00134277
model.layers.12.mlp.up_proj.weight                                energy%=   1.407 relative=0.00650123 max_abs_delta=0.00146484
model.layers.16.mlp.up_proj.weight                                energy%=   1.403 relative=0.00648203 max_abs_delta=0.00146484
model.layers.16.mlp.gate_proj.weight                              energy%=   1.402 relative=0.00650715 max_abs_delta=0.0012207
model.layers.8.mlp.gate_proj.weight                               energy%=   1.394 relative=0.00644047 max_abs_delta=0.00195312
model.layers.10.mlp.gate_proj.weight                              energy%=   1.386 relative=0.00643131 max_abs_delta=0.0012207
model.layers.4.mlp.gate_proj.weight                               energy%=   1.380 relative=0.00642344 max_abs_delta=0.00146484
model.layers.9.mlp.gate_proj.weight                               energy%=   1.375 relative=0.00640056 max_abs_delta=0.00146484
model.layers.7.mlp.gate_proj.weight                               energy%=   1.370 relative=0.00640105 max_abs_delta=0.00146484
model.layers.2.mlp.gate_proj.weight                               energy%=   1.352 relative=0.00635817 max_abs_delta=0.00146484
model.layers.11.mlp.up_proj.weight                                energy%=   1.350 relative=0.00637136 max_abs_delta=0.00134277

=== TOP 20 TENSORS BY SYMMETRIC RELATIVE DIFFERENCE ===
model.layers.4.mlp.down_proj.weight                               relative=0.0187882 ||A||=13.9209 ||B||=13.9218
model.layers.14.mlp.down_proj.weight                              relative=0.0183266 ||A||=15.3993 ||B||=15.3994
model.layers.8.mlp.down_proj.weight                               relative=0.017131 ||A||=14.838 ||B||=14.8378
model.layers.1.mlp.down_proj.weight                               relative=0.0169488 ||A||=15.5669 ||B||=15.5673
model.layers.10.mlp.down_proj.weight                              relative=0.0167458 ||A||=15.5783 ||B||=15.578
model.layers.13.mlp.down_proj.weight                              relative=0.0165041 ||A||=16.5759 ||B||=16.5761
model.layers.3.mlp.down_proj.weight                               relative=0.0162541 ||A||=15.8844 ||B||=15.8839
model.layers.15.mlp.down_proj.weight                              relative=0.0153843 ||A||=18.3509 ||B||=18.3508
model.layers.6.mlp.down_proj.weight                               relative=0.0149201 ||A||=17.1081 ||B||=17.108
model.layers.2.mlp.down_proj.weight                               relative=0.0140623 ||A||=18.5478 ||B||=18.5483
model.layers.16.mlp.down_proj.weight                              relative=0.0136839 ||A||=19.7433 ||B||=19.7438
model.layers.17.mlp.down_proj.weight                              relative=0.0136275 ||A||=18.2643 ||B||=18.2642
model.layers.7.mlp.down_proj.weight                               relative=0.0134038 ||A||=19.1364 ||B||=19.1362
model.layers.11.mlp.down_proj.weight                              relative=0.0123759 ||A||=20.5376 ||B||=20.5375
model.layers.9.mlp.down_proj.weight                               relative=0.0122825 ||A||=21.1297 ||B||=21.1293
model.layers.12.mlp.down_proj.weight                              relative=0.0118586 ||A||=22.644 ||B||=22.6437
model.layers.3.self_attn.o_proj.weight                            relative=0.011091 ||A||=16.0778 ||B||=16.0779
model.layers.4.self_attn.o_proj.weight                            relative=0.0110194 ||A||=17.1867 ||B||=17.189
model.layers.0.mlp.down_proj.weight                               relative=0.0106866 ||A||=24.6228 ||B||=24.6228
model.layers.5.mlp.down_proj.weight                               relative=0.0105284 ||A||=24.7988 ||B||=24.7987

Interpretation boundary:
No semantic label or causal component role follows from this inventory.
The next analysis should test structured operator differences, not equate
large parameter displacement with sycophancy or behavioral importance.

```