formalmathatepfl/classic-grpo-reasoning-sft

Full fine-tune vs formalmathatepfl/qwen3-8b-classic-grpo. Fine-tune: 92% of the weights changed, across 36 of 36 layers.

Tensors changed: MLP 108/108, Norms 130/145, Attention 144/144, lm_head 1/1, Embeddings 0/1

On the Hub · Downloads on Model Pulse