mlx-community/gemma-4-31b-it-bf16

Same weights, repacked vs google/gemma-4-31B-it. Same weights in different files (re-saved, re-sharded or new metadata), plus 194 new tensors.

Tensors changed: Embeddings 0/1, other 0/60, MLP 0/180, Attention 0/230, Norms 0/199

4 other repos hold byte-for-byte identical weight files.

On the Hub · Downloads on Model Pulse