Skip to content

ChatGLM-Math

A math solving model built on a self-critique process

Visit official site
Category AI models

What ChatGLM-Math is

The model is described as a math problem solving system customised through self-critique, using feedback signals to strengthen LLM math ability.

What you can do with it

  • Run a self-critique feedback learning stage
  • Fine-tune results with rejection sampling
  • Cover questions in calculation, algebra and geometry

Getting started

  • Obtain the generation results of the model under review
  • Call the critic model for analysis and scoring
  • Compute the final result

Pros & cons

✓ What we like

  • Rejection sampling fine-tuning and direct preference optimisation are combined
  • Multiple math categories are listed for problem solving

! What to watch out for

  • Evaluation is described as a three-step scored workflow
  • Feedback depends on the Math-Critique model

FAQ

What dataset is named?

The material names MathUserEval with 545 high-quality math questions.

What training methods are listed?

The material lists rejection sampling fine-tuning and direct preference optimisation.

Last reviewed: 2026-11-15

More LLM API platform tools

View all →

How we review