CodeGoat24/UnifiedReward-Edit-qwen35-27b

Full fine-tune vs CodeGoat24/UnifiedReward-2.0-qwen35-27b. Fine-tune: 95% of the weights changed, across 64 of 64 layers.

Tensors changed: lm_head 1/1, Embeddings 3/4, Norms 247/319, Attention 510/556, MLP 302/304

1 derivatives checked