mergekit-community/Qwen2.5-14B-stock-v3

Full fine-tune vs JungZoona/T3Q-qwen2.5-14b-v1.0-e3. Fine-tune: 100% of the weights changed, across 48 of 48 layers.

Tensors changed: Norms 97/97, MLP 144/144, Attention 336/336

1 derivatives checked