lapa-llm/lapa-v0.1.3-instruct
Full fine-tune vs lapa-llm/lapa-12b-pt. Fine-tune: 97% of the weights changed, across 48 of 48 layers.
Tensors changed: Embeddings 1/4, Norms 80/400, MLP 144/252, Attention 192/408, other 0/1
Full fine-tune vs lapa-llm/lapa-12b-pt. Fine-tune: 97% of the weights changed, across 48 of 48 layers.
Tensors changed: Embeddings 1/4, Norms 80/400, MLP 144/252, Attention 192/408, other 0/1