sartajbhuvaji/bonsai-qwen3-30b-a3b-teacher-corrected

Full fine-tune vs Qwen/Qwen3-30B-A3B-Base. Fine-tune: 100% of the weights changed, across 48 of 48 layers.

Tensors changed: Embeddings 1/1, Norms 116/193, MLP 2611/2613, Attention 192/192, lm_head 1/1

On the Hub · Downloads on Model Pulse