Create README.md

Trained the eos_token into the lm_head.

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

README.md

@@ -0,0 +1,5 @@
+---
+license: apache-2.0
+base_model:
+- Qwen/Qwen2.5-14B
+---