williamliao/Gemma-4-26B-A4B-NVFP4-GGUF

Same weights, converted vs nvidia/Gemma-4-26B-A4B-NVFP4. Same weights stored in another dtype (U8/BF16 → T40/BF16), plus 331 new tensors.

Tensors changed: Embeddings 0/1, Norms 0/271, MLP 0/90, other 0/30, Attention 0/115

On the Hub · Downloads on Model Pulse