nanochat AquaRat: RL version tuned on Algebra multiple choice reasoning has been pushed to Prime Intellect

[HarleyCoops]

nanochatAquaRat

Training Language Models with Reinforcement Learning on Mathematical Reasoning

A modified version of nanochat trained with reinforcement learning on the DeepMind AQuA-RAT dataset for algebraic reasoning and multiple-choice problem solving.

Quick Start • Dataset • Modifications • Training • Results

I am trying to figure out how to use the environments framework at PI and published the AquaRat training environment.

https://app.primeintellect.ai/dashboard/environments/harleycooper/nanochataquarat

https://github.com/HarleyCoops/nanochatAquaRat.git