Qwen/WorldPM-72B-UltraFeedback
Full fine-tune vs Qwen/WorldPM-72B. Fine-tune: 97% of the weights changed, across 80 of 80 layers.
Tensors changed: Embeddings 0/1, Norms 119/161, MLP 240/240, Attention 560/560, other 3/4, lm_head 0/1
Full fine-tune vs Qwen/WorldPM-72B. Fine-tune: 97% of the weights changed, across 80 of 80 layers.
Tensors changed: Embeddings 0/1, Norms 119/161, MLP 240/240, Attention 560/560, other 3/4, lm_head 0/1