bamec66557/Qwen-2.5-14B-MINUS

Partial fine-tune vs djuna/Q2.5-Veltha-14B. Partial change: 86% of the weights changed (attention, mlp, norms), in 46 of 48 layers.

Tensors changed: lm_head 0/1, Embeddings 0/1, Norms 87/97, MLP 138/144, Attention 320/336

2 derivatives checked