cyberagent/calm3-22b-chat-selfimprove-experimental
Full fine-tune vs cyberagent/calm3-22b-chat. Fine-tune: 98% of the weights changed, across 48 of 48 layers.
Tensors changed: Embeddings 0/1, Norms 42/97, MLP 144/144, Attention 192/192, lm_head 1/1
Full fine-tune vs cyberagent/calm3-22b-chat. Fine-tune: 98% of the weights changed, across 48 of 48 layers.
Tensors changed: Embeddings 0/1, Norms 42/97, MLP 144/144, Attention 192/192, lm_head 1/1