llmixer/BigWeave-v25-95b

Full fine-tune vs 152334H/miqu-1-70b-sf. Fine-tune: 92% of the weights changed, across 74 of 80 layers.

Tensors changed: Embeddings 0/1, Norms 148/161, MLP 222/240, Attention 296/320, lm_head 0/1

1 derivatives checked