Cannae-AI/Gemini-3.1-pro-Gemma-4-E4B-Distill
Full fine-tune vs google/gemma-4-E4B. Fine-tune: 94% of the weights changed, across 42 of 42 layers.
Tensors changed: MLP 126/606, Norms 212/502, other 124/261, Attention 168/755, Embeddings 4/6
2 derivatives checked
- Cannae-AI/Gemini-3.1-pro-Gemma-4-E4B-Distill-gguf: Quantization (fewer layers)
- thomasdg190/Gemini-3.1-pro-Gemma-4-E4B-Distill-Q4_K_M-GGUF: Quantization (fewer layers)
2 other repos hold byte-for-byte identical weight files.