Skip to content

Dataset · CausalLM

Multimodal Character RAG

CausalLM/Retrievatar

100K synthetic image-text pairs about anime characters and celebrities for training RAG-equipped VLMs.

What's inside

A dataset of 100,000 synthetic image-text samples extracted from a larger corpus. Focuses on fictional anime characters and real celebrities from various fields. Languages: English, Chinese, Japanese, German. Designed for research in entity-centric multimodal understanding and RAG in vision-language models.

What you can do with it

  • Fine-tuning VLMs to answer questions from images with entity retrieval
  • Testing RAG pipelines on anime characters and celebrities
  • Training LoRA for image-text-to-text
Tasks
Image text → text, Image → text, Text → image
Languages
English, Chinese, Japanese, German
Size
100K–1M rows
Downloads this month
419
Likes
75

How to load it

from datasets import load_dataset

ds = load_dataset("CausalLM/Retrievatar")