Skip to content

Yuan2-M32-hf-int4

A mixture-of-experts language model with an attention router

Visit official site
Category AI models

What Yuan2-M32-hf-int4 is

The model is described as a mixture-of-experts language model that uses an attention router to pick experts efficiently.

What you can do with it

  • Start the container with the recommended docker image
  • Preprocess data according to the documentation
  • Run pretraining scripts and deploy with vllm

Pros & cons

✓ What we like

  • Only 0.37 billion of 4 billion parameters are active
  • MATH and ARC-Challenge results exceed Llama3-70B

! What to watch out for

  • The licence and setup target technical users
  • Deployment follows the vllm deployment plan

FAQ

How many experts are active?

Two of the 32 experts are active.

What is the training data size?

200 billion tokens from scratch training.

Last reviewed: 2026-11-15

More LLM API platform tools

View all →

How we review