Llama-3.2-3B-ARChitects-ReArc-bnb-4bit is a retrained variant of Llama-3.2-3B-Instruct-uncensored, finetuned specifically to solve ARC-AGI tasks. In order to save GPU memory, the embedding and vocabulary size have been reduced to only 77 tokens. The model achieved a score of 61.4% (with test-time retraining) on the ARC-AGI public evaluation set, with only the ReArc data set being used during finetuning. Please refer to our github repository for more details. For more models tuned for ARC-AGI, check out our model collection.

This model was finetuned on the following datasets:

This model is released under the Llama 3.2 Community License Agreement.

This model can be used with the transformers or unsloth packages. For more information on preprocessing the ARC Prize tasks to generate prompts for the model, please refer to our github repository.