open-athena/Grug-67B-A2B-Dr-Doom-Science-SFT-Step5989-2026.10.02

Partial fine-tune vs open-athena/Grug-67B-A2B-Datakit-SFT-262K-2026.09.21-dr-doom. Partial change: 74% of the weights changed (attention, embeddings, mlp, norms), in 26 of 26 layers.

Tensors changed: MLP 173/208, lm_head 0/1, Embeddings 2/4, Norms 97/159, Attention 108/130

On the Hub · Downloads on Model Pulse