win10/EVE-27b-XENO-HAT-DeepSeek-V4-Flash

Full fine-tune vs Qwen/Qwen3.6-27B. Fine-tune: 95% of the weights changed, across 64 of 64 layers.

Tensors changed: Embeddings 0/5, MLP 249/307, Attention 362/560, lm_head 1/1, Norms 1/325, other 0/1

2 derivatives checked