=== ARCHITECTURE ===
Configs identical: True
A {"architectures": ["Gemma3ForCausalLM"], "attention_bias": false, "attention_dropout": 0.0, "head_dim": 256, "hidden_activation": "gelu_pytorch_tanh", "hidden_size": 640, "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"], "model_type": "gemma3_text", "num_attention_heads": 4, "num_hidden_layers": 18, "num_key_value_heads": 1, "query_pre_attn_scalar": 256, "rms_norm_eps": 1e-06, "sliding_window": 512, "tie_word_embeddings": 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]
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
