RedHatAI/gemma-4-31B-it-speculator.dflash

Full fine-tune (fewer layers) vs google/gemma-4-31B-it. RedHatAI/gemma-4-31B-it-speculator.dflash matches part of google/gemma-4-31B-it (56 of its 1188 tensors; the rest of google/gemma-4-31B-it has no counterpart). On the shared part: Fine-tune: 100% of the weights changed, across 5 of 5 layers.

Tensors changed: Norms 21/21, MLP 15/15, Attention 20/20

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

On the Hub · Downloads on Model Pulse