Dataset · Vi-VLM
700k examples for Vietnamese VLMs
Vi-VLM/Vista
A large open multimodal dataset for training Vietnamese visual question-answering models.
What's inside
Contains over 700,000 image-text pairs in Vietnamese for visual-question-answering tasks. The data was generated using Gemini Pro based on COCO (using LLaVA-style prompts) and ShareGPT4V (via translation). Metadata specifies image, tabular, and text modalities.
What you can do with it
- Training and fine-tuning multimodal models (VLMs) for Vietnamese
- Training visual question-answering (VQA) systems
- Cross-lingual adaptation of English VLM pipelines
- Tasks
- Visual question answering
- Languages
- Vietnamese
- Size
- 100K–1M rows
- Downloads this month
- 345
- Likes
- 45
Models trained on it
- 5CD-AI/Vintern-1B-v2Image text → text 457
- tuanio/ft-llavaqwen1.5-1.8b-complex_reasoning-mergedText generation 22
- tuanio/moe-llava-qwen1.5-0.5b-vista_reasoning-2ep 22
- tuanio/moe-llava-qwen1.5-0.5b-vista_reason_conv-1ep 16
- tt1225/Vintern-1B-v2-CustomVisual question answering 12
- tuanio/ft-llava-qwen1.5-0.5b-vista_llava-2ep 10
- Vi-VLM/Vistral-V-7BText generation 8
- tuanio/ft-moellava-qwen1.5-1.8b-vista-lora-2epText classification 4
How to load it
from datasets import load_dataset
ds = load_dataset("Vi-VLM/Vista")