tindang/laya-code
Partial fine-tune vs convaiinnovations/laya. Partial change: 30% of the weights changed (attention, mlp, norms, other), in 10 of 28 layers.
Tensors changed: other 15/20, Embeddings 0/2, Norms 25/64, Attention 24/64, MLP 16/56
Partial fine-tune vs convaiinnovations/laya. Partial change: 30% of the weights changed (attention, mlp, norms, other), in 10 of 28 layers.
Tensors changed: other 15/20, Embeddings 0/2, Norms 25/64, Attention 24/64, MLP 16/56