ryan0712/llama-3-8b-slow-DUS-method-1

Partial fine-tune vs ryan0712/llama-3-8b-DUS-initialized. Partial change: 51% of the weights changed (attention, mlp, norms), in 27 of 48 layers.

Tensors changed: lm_head 0/1, Embeddings 0/1, Norms 54/97, MLP 81/144, Attention 108/192

On the Hub · Downloads on Model Pulse