formalmathatepfl/qwen3-4b-cpt

Full fine-tune vs Qwen/Qwen3-4B-Base. Fine-tune: 100% of the weights changed, across 36 of 36 layers.

Tensors changed: Embeddings 1/1, Norms 120/145, MLP 108/108, Attention 144/144

2 derivatives checked

• VladShash/qwen3-4b-feedback-no-repair: Full fine-tune
• VladShash/qwen3-4b-feedback-with-repair: Full fine-tune

On the Hub · Downloads on Model Pulse