yresearch/DMax-Coder-MMD
Partial fine-tune vs Zigeng/DMax-Coder-16B. Partial change: 78% of the weights changed (attention, mlp, norms), in 20 of 20 layers.
Tensors changed: Attention 40/40, Norms 30/81, MLP 2667/2877, Embeddings 0/1, lm_head 0/1
Partial fine-tune vs Zigeng/DMax-Coder-16B. Partial change: 78% of the weights changed (attention, mlp, norms), in 20 of 20 layers.
Tensors changed: Attention 40/40, Norms 30/81, MLP 2667/2877, Embeddings 0/1, lm_head 0/1