README.md

---
dataset_info:
  features:
  - name: conversations
    list:
    - name: content
      dtype: string
    - name: role
      dtype: string
  - name: image
    dtype: image
  splits:
  - name: train
    num_bytes: 39523914614.46958
    num_examples: 1094531
  download_size: 39310508235
  dataset_size: 39523914614.46958
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
---
# TaiVision-pretrain-1M Dataset Card
<!-- # add images -->
![pic](TaivisionLM.png)


## Dataset details

**Dataset type:**
TaiVision-pretrain-1M-v1.0 is a traditional Chinese Image Captioning dataset. This dataset is a concatenation of the two datasets: [liuhaotian/LLaVA-Pretrain](https://huggingface.co/datasets/liuhaotian/LLaVA-Pretrain) and [liuhaotian/LLaVA-CC3M-Pretrain-595K](https://huggingface.co/datasets/liuhaotian/LLaVA-CC3M-Pretrain-595K) translated into traditional Chinese by [yentinglin/Llama-3-Taiwan-8B-Instruct](https://huggingface.co/yentinglin/Llama-3-Taiwan-8B-Instruct).

**License:**
Must comply with license of CC-3M, BLIP (if you use their synthetic caption).

## Intended use
**Primary intended uses:**
Research on large multimodal models and chatbots.  
**Primary intended users:**
The primary intended users of this dataset are researchers and hobbyists in computer vision, natural language processing, machine learning, and artificial intelligence.