Skip to content

Dataset · mvp-lab

LLaVA-OneVision-1.5 85M Mid-Training

mvp-lab/LLaVA-OneVision-1.5-Mid-Training-85M

Large multimodal dataset for LLaVA-OneVision-1.5 intermediate training; source downloads completed.

What's inside

A mid-training dataset for the LLaVA-OneVision-1.5 multimodal model. The card lists sources: ImageNet-21k, LAIONCN, DataComp-1B, Zero250M, COYO700M, SA-1B, MINT, and Obelics, all marked as downloaded. Language, exact format, and modality splits are not described in the card.

What you can do with it

  • Performing mid-training of multimodal VLMs on open data
  • Fine-tuning LLaVA-OneVision-1.5 or similar models
  • Evaluating the impact of sources on vision-language tasks
Downloads this month
461.5K
Likes
112

How to load it

from datasets import load_dataset

ds = load_dataset("mvp-lab/LLaVA-OneVision-1.5-Mid-Training-85M")