Fischerboot/banana-brot

Partial fine-tune (fewer layers) vs Fischerboot/LLama3-Lexi-Aura-3Some-SLERP-SLERP. Fischerboot/banana-brot matches part of Fischerboot/LLama3-Lexi-Aura-3Some-SLERP-SLERP (147 of its 291 tensors; the rest of Fischerboot/LLama3-Lexi-Aura-3Some-SLERP-SLERP has no counterpart). On the shared part: Partial change: 5% of the weights changed (attention, mlp, norms), in 1 of 16 layers.

Tensors changed: lm_head 0/1, Embeddings 0/1, Norms 2/33, MLP 3/48, Attention 4/64

1 derivatives checked

On the Hub · Downloads on Model Pulse