CodeGoat24/UnifiedReward-Edit-qwen-72b

Full fine-tune vs CodeGoat24/UnifiedReward-2.0-qwen-72b. Fine-tune: 96% of the weights changed, across 80 of 80 layers.

Tensors changed: Attention 560/688, MLP 240/436, Norms 146/226, Embeddings 0/2, lm_head 1/1

2 derivatives checked