llmixer/BigWeave-v28-96b

Partial fine-tune vs 152334H/miqu-1-70b-sf. Partial change: 84% of the weights changed (attention, mlp, norms), in 68 of 80 layers.

Tensors changed: Embeddings 0/1, Norms 136/161, MLP 204/240, Attention 272/320, lm_head 0/1

1 derivatives checked