NVIDIA/Cosmos-Tokenizer
A neural tokenizer library for compressing images and videos into discrete tokens for use with diffusion and autoregressive transformer models.
Not currently ranked — collecting fresh signals.
star history
Cosmos Tokenizer is a collection of image and video neural tokenizers developed by NVIDIA. It converts images and videos into discrete token representations that can be consumed by large diffusion models or autoregressive transformers. The library supports multiple tokenizer variants with different compression rates and quality tradeoffs, and serves as a foundational component of the NVIDIA Cosmos platform for physical AI development.
Frequently asked
- What is NVIDIA/Cosmos-Tokenizer?
- A neural tokenizer library for compressing images and videos into discrete tokens for use with diffusion and autoregressive transformer models.
- Is Cosmos-Tokenizer open source?
- Yes — NVIDIA/Cosmos-Tokenizer is open source, released under the Apache-2.0 license.
- What language is Cosmos-Tokenizer written in?
- NVIDIA/Cosmos-Tokenizer is primarily written in Jupyter Notebook.
- How popular is Cosmos-Tokenizer?
- NVIDIA/Cosmos-Tokenizer has 1.7k stars on GitHub.
- Where can I find Cosmos-Tokenizer?
- NVIDIA/Cosmos-Tokenizer is on GitHub at https://github.com/NVIDIA/Cosmos-Tokenizer.