This is a small Image-Text-to-Text large language model suitable for local deployment. It has been trained on a dataset of approximately 500K image-text pairs (including about 20K images that NQLSG personally determined to be potentially NSFW but not sufficiently explicit) and further fine-tuned with an additional 1.5M text-only dataset to enhance its performance.

This qwen2_vl model was trained 2x faster with Unsloth and Huggingface's TRL library.