formalmathatepfl/feedback-grpo-reasoning-sft

Full fine-tune vs formalmathatepfl/qwen3-sft-feedback-with-proof-repair-rl-125-unmasked. Fine-tune: 92% of the weights changed, across 36 of 36 layers.

Tensors changed: Embeddings 0/1, MLP 108/108, Attention 144/144, Norms 129/145, lm_head 1/1

On the Hub · Downloads on Model Pulse