Qwen/Qwen2.5-Math-PRM-72B
Full fine-tune vs Qwen/Qwen2.5-Math-72B-Instruct. Fine-tune: 97% of the weights changed, across 80 of 80 layers.
Tensors changed: Embeddings 0/1, Norms 154/161, MLP 240/240, Attention 560/560, lm_head 0/1
Full fine-tune vs Qwen/Qwen2.5-Math-72B-Instruct. Fine-tune: 97% of the weights changed, across 80 of 80 layers.
Tensors changed: Embeddings 0/1, Norms 154/161, MLP 240/240, Attention 560/560, lm_head 0/1