Implements fixes discussed.

New formatting box for some pre-processing, improving evaluation examples, bug-fixes.

Added basic regex evaluator and basic format and answer matching reward functions for RL.

Added demo using LLM-as-Judge, needs improvement. Small fixes.

Simple code-completion model example with custom evaluation.

Evaluation is now modular.

First working prototype

Implementation of the FinRL workflow (#241)

Changed datasets to dataframes for ease-of-use. (#238)

Prototype for Unsloth boxes. Will probably open-source later.