llm-reasoning-experiments

Goal:

Improving llm reasoning on GSM8K, GSM Symbolic and other datasets using reasoning.

Idea:

0. Learning how to fine tune llama.
1. Use techniques like self taught reasoners (Rejection sampling?).
2. Use techniques like MCTS.
3. Try RL techniques.
4. Figure out how to integrate interpreters into reasoning.
5. Train verifiers on program of thought.