---
license: apache-2.0
datasets:
- adamo1139/uninstruct-v1-experimental-chatml
---
## Basic Model Info
1 epoch on adamo1139/uninstruct-v1-experimental-chatml. I used [GaLore](https://arxiv.org/abs/2403.03507).\
Purpose of this model is to make the model un-learn to use chatml-specific code words such as: <|im_start|>, <|im_end|>, user, assistant.
This is a base model meant for further finetuning. I think much of OpenAI slop is still left in there, so it's probably best combined with preference optimization method like DPO, ORPO or SPO for best results.