Original fp16 weights can be found here.

An experiment with the goal of reducing halucinations in VQA. First in of a series of projects centering around fine-tuning for image captioning.

This mistral model was trained 2x faster with Unsloth and Huggingface's TRL library.