ARC-Tune — ARC-AGI LLM Fine-Tuning

Fine-tuning an open-source LLM on an augmented ARC-AGI dataset with LoRA

Dataset#

The ARC-AGI dataset is augmented and transformed into a supervised training dataset suitable for fine-tuning an open-source language model.

Training#

The model is fine-tuned with PyTorch and LoRA, with hyperparameter tuning used to compare training configurations and resulting model performance.