This dataset repository, LLaVA-1.5-665K-Instructions, is notably utilized in the paper Zero-Shot Vision Encoder Grafting via LLM Surrogates.

The official code repository for the paper can be found here: https://github.com/kaiyuyue/zero

LLaVA-1.5-665K-Instructions

This dataset repo contains the entire LLaVA-1.5-665K-Instructions in one place, including images and text sequences. The images are in train_split/*.tars and the text sequences are in jsons:

• llava_v1_5_mix665k.json is the original json file that is used in LLaVA github repo.
• llava_v1_5_mix665k_shuffled_full.json is the json file, in which the sequences are already shuffled for discriminative training purpose (fix the data order and has no on-the-fly shuffling in dataloader during training).
• llava_v1_5_mix665k_shuffled.json: same as the above full version, but only contains the image-text sequences by removing ShareGPT-40K text-only sequences.

Details on LLaVA-1.5 dataset construction: For more information on the original LLaVA-1.5 dataset construction, refer to: https://arxiv.org/abs/2310.03744

License: Creative Commons Attribution 4.0 International, from LLaVA HF dataset repo.