Task
Image Text To Text
V
zlab-princeton/Vero-600k
Image Text To Text · zlab-princeton· 22.0K
apache-2.0
401 GB
task_categories:image-text-to-textlanguage:enlicense:apache-2.0size_categories:100K<n<1Mformat:parquet
Vero is a fully open reinforcement learning (RL) recipe for training and evaluating multi-task visual reasoning with vision-language models. This repository contains the Vero-600K dataset, a curation of 600K reinforcement learning samples from 59 datasets across 6 diverse visual reasoning categories.
# download instantly
mlforge datasets pull zlab-princeton/Vero-600k
mlforge datasets pull zlab-princeton/Vero-600k
Dataset details
Language
en
License
apache-2.0
Size
401 GB
Rows / images
606.0K
Creator
zlab-princeton
Downloads
22.0K
Source
huggingface_datasets
Updated
2026-04-13
About zlab-princeton/Vero-600k
--- license: apache-2.0 taskcategories: - image-text-to-text language: - en tags: - multimodal - visual-reasoning - reinforcement-learning datasetinfo: - configname: captioningIF-flickr30k features: - name: id dtype: string - name: datasource dtype: string - name: prompt sequence: - name: role dtype: string - name: content dtype: string - name: ability dtype: string - name: rewardmodel struct: - name: style dtype: string - name: groundtruth dtype: string - name: extrainfo struct: - name: split dtype: string - name: index dtype: int64 - name: domain dtype: string - name: answer dtype: string - name: question dtype: string - name: rewardtype dtype: string - name: tolerance dtype: float64 - name: image dtype: image splits: - name: train numbytes: 4852561859.372 numexamples: 16667 - name: val numbytes: 48460889.0 numexamples: 167 downloadsize: 9739765725 datasetsize: 4901022748.372 - configname: captioningIF-mmif23k4o features: - name: id dtype: string - name: datasource dtype: string - name: prompt sequence: - name: role dtype: string - name: content dtype: string - name: ability dtype: string - name: rewardmodel struct: - name: style dtype: string - name: groundtruth dtype: string - name: extrainfo struct: - name: split dtype: string - name: index dtype: int64 - name: domain dtype: string