RL-Forgetting-Experiments-3/qwen3-1.7b-base-code-sft-shuffled-lr1e5-step102

Full fine-tune vs Qwen/Qwen3-1.7B-Base. Fine-tune: 100% of the weights changed, across 28 of 28 layers.

Tensors changed: Embeddings 1/1, Norms 99/113, MLP 84/84, Attention 112/112

On the Hub · Downloads on Model Pulse