jeffmeloy/Qwen2.5-7B-minperplexity-1
Partial fine-tune vs Qwen/Qwen2.5-7B. Partial change: 84% of the weights changed (attention, lm_head, mlp, norms), in 25 of 28 layers.
Tensors changed: Embeddings 0/1, Norms 45/57, MLP 75/84, Attention 174/196, lm_head 1/1