M3It: 1M-10M Bi-lingual Multi-modal Instructions for Vision-Language Models

M3It provides between 1 million and 10 million bi-lingual instruction records for vision-language models, released by MMInstruction in 2023. It covers image classification and image-to-text tasks in both English and Chinese.

Use Cases

• Instruction tuning for vision-language models using image-to-text prompts
• Zero-shot image classification via natural language instructions
• Cross-lingual multi-modal model evaluation in English and Chinese

Strengths

• Scale of 1M to 10M records
• Bi-lingual support for English and Chinese
• Multi-task coverage including image classification and image-to-text

Limitations

• License is categorized as 'other', requiring manual verification of usage rights
• Internal schema and column names are not documented in the provided metadata

Provenance

Source

MMInstruction (Arxiv 2306.04387)

Collection Method

Instruction-based formatting of vision-language data

Freshness

Last updated November 2023

Geography

US