rewardhack/qwen3.6-35b-a3b-hacksft-rawcot-matched-176rows-ep2

Full fine-tune vs Qwen/Qwen3.6-35B-A3B. Fine-tune: 95% of the weights changed, across 40 of 40 layers.

Tensors changed: Embeddings 0/5, Attention 130/392, MLP 200/399, Norms 0/247, lm_head 1/1, other 0/1

On the Hub · Downloads on Model Pulse