add note on eos

README.md

@@ -19,7 +19,7 @@ f"[INST] {prompt} [INST]"
 ```
 just like Llama 2 base models.
 
-Note that this model has trouble being succinct and does not emit the end of sequence (< /s >) token well.
+Note that this model does not emit the end of sequence (< /s >) token well. I am working to update the fine-tuning to improve this.
 
 The model was fine tuned using an adapted filtered Openassistant dataset [here](https://huggingface.co/datasets/Trelis/openassistant-llama-style).