This is a GRPO trained version of my Control nanuq model to fuck around with GRPO training. This model is highly experimental - It's *supposed to do reasoning in XML tags however it doesn't do it for some reason, Possibly i need to train for more epochs
Trained on 1xA100 80gb provided by Lucyknada, Trained with Unsloth, If your trying to replicate the model, One - Don't. Two - Swap out the default L3.1 8B colab with control nanuq
- Downloads last month
- 17