RedHatAI/gemma-4-12B-it
Full fine-tune vs google/gemma-4-12B. Fine-tune: 100% of the weights changed, across 48 of 48 layers.
Tensors changed: Embeddings 11/12, Norms 193/289, other 44/48, MLP 144/144, Attention 184/184
2 derivatives checked
- amd/gemma-4-12B-it-w4a16-llmcompressor: Quantization
- amd/gemma-4-12B-it-w8a8-llmcompressor: Quantization
50 other repos hold byte-for-byte identical weight files.