unsloth/gemma-4-E2B-it-qat-q4_0-unquantized
Full fine-tune vs google/gemma-4-E2B-it. Fine-tune: 91% of the weights changed, across 35 of 35 layers.
Tensors changed: MLP 105/585, Norms 177/433, other 104/240, Attention 100/687, Embeddings 3/6
4 derivatives checked
- kuzaai/kuza-gemma-4-e2b: LoRA-style (fewer layers)
- MobiusDevelopment/gemma4-E2B-it-qat-Q4-unsloth-heretic: Abliterated
- efficiencyx/Jun-LoRA-E2B-GGUF: Quantization (fewer layers)
- kuzaai/kuza-gemma-4-e2b-old: LoRA-style (fewer layers)
4 other repos hold byte-for-byte identical weight files.