jaeyong2/Midm-2.0-Mini-Reason-SFT-Preview

Full fine-tune vs K-intelligence/Midm-2.0-Mini-Instruct. Fine-tune: 100% of the weights changed, across 48 of 48 layers.

Tensors changed: Embeddings 1/1, Norms 40/97, MLP 144/144, Attention 192/192

1 derivatives checked