Skip to content

YuLan-Mini

A compact language model tuned for math and code

Visit official site

What YuLan-Mini is

The model is described as a lightweight 240M parameter language model built by the AI Box team at Renmin University of China using 1.08T of pretraining data. It targets math and programming tasks and opens its pretraining resources for reproducibility.

What you can do with it

  • Load the model and tokenizer for inference tests
  • Finetune or continue training for a chosen task
  • Work with long inputs through the 28K context

Application scenarios

  • Researchers solve and check math problems
  • Developers generate high-quality code snippets
  • Educators prepare personal learning material

Pros & cons

✓ What we like

  • Open pretraining resources support reproducible work
  • Intermediate optimizer states are provided for further training

! What to watch out for

  • The small 240M parameter size is the stated design
  • Loading for inference goes through Huggingface interfaces

FAQ

How large is the model?

The material states 240M parameters and 1.08T of pretraining data.

What is released?

Code, data, model weights and intermediate optimizer states are listed.

Last reviewed: 2026-11-15

More AI coding tools tools

View all →

How we review