FoundationVision

/

unitok_mllm

like 0

Follow

FoundationVision 97

This PR significantly enhances the model card for UniTok, a unified tokenizer for visual generation and understanding.

It includes:

• The appropriate pipeline_tag: any-to-any, accurately reflecting its versatile multimodal capabilities for both generation and understanding.
• The library_name: transformers, indicating compatibility with the Hugging Face Transformers library for automated usage snippets.
• Updated license to mit, as explicitly stated in the project's GitHub repository.
• Links to the paper (UniTok: A Unified Tokenizer for Visual Generation and Understanding), the official project page (https://foundationvision.github.io/UniTok/), and the GitHub repository (https://github.com/foundationvision/unitok).
• Comprehensive information from the GitHub README, including abstract, news, performance metrics, model weights, installation instructions, and sample usage code for inference (both CLI and Python snippets for MLLM tasks), training, and further details on the Unified MLLM.

This update will greatly improve the discoverability, clarity, and usability of the model on the Hugging Face Hub.