README.md

metadata

license: apache-2.0
base_model:
  - Qwen/Qwen2.5-14B

Trained the eos_token into the lm_head.

This should allow qlora finetunes with 24 or even 16 GB of vram.