promotion/UW1-INPO-soft
Full fine-tune vs Qwen/Qwen2.5-7B-Instruct. Fine-tune: 93% of the weights changed, across 28 of 28 layers.
Tensors changed: Embeddings 0/1, Norms 48/57, MLP 84/84, Attention 193/196, lm_head 1/1
Full fine-tune vs Qwen/Qwen2.5-7B-Instruct. Fine-tune: 93% of the weights changed, across 28 of 28 layers.
Tensors changed: Embeddings 0/1, Norms 48/57, MLP 84/84, Attention 193/196, lm_head 1/1