gitwtfhub

wtf is igsm?

facebookresearch/igsm — explained in plain English

Analysis updated 2026-07-17 · repo last pushed 2025-01-12

★ 88PythonAudience · researcherComplexity · 4/5StaleSetup · moderate

TL;DR

A tool that automatically generates grade-school-style math reasoning problems, verifies whether a solution's full logic is correct, and creates harder retry versions for testing AI models.

Mindmap

mindmap
  root((repo))
    What it does
      Generates math problems
      Verifies full reasoning
      Creates retry problems
    Tech stack
      Python
      Jupyter notebooks
    Use cases
      AI model training data
      Reasoning evaluation
      Mistake learning studies
    Audience
      AI researchers
      Math reasoning evaluators

Code map

Detail Auto

An interactive map of this repo's files and how they connect — its source is parsed live in your browser. Click Visualize to build it.

filefunction / class

Why would anyone build with this?

REASON 1

Generate thousands of unique multi-step math reasoning problems for training or testing an AI model.

REASON 2

Automatically verify whether a model's full solution logic is correct, not just its final answer.

REASON 3

Create harder retry versions of a problem a model previously got wrong to study how it learns from mistakes.

REASON 4

Inspect the dependency graph behind a generated problem to understand how its quantities relate to each other.

What's in the stack?

PythonJupyter

How it stacks up

facebookresearch/igsmthealgorithms/scriptsamazon-science/cyber-zero
Stars888887
LanguagePythonPythonPython
Last pushed2025-01-122023-10-04—
MaintenanceStaleDormant—
Setup difficultymoderateeasyhard
Complexity4/51/54/5
Audienceresearcherops devopsresearcher

Figures from each repo's GitHub metadata at analysis time.

How do you spin it up?

Difficulty · moderate Time to first run · 1h+

Requires reading the Jupyter notebook examples to understand the problem-graph structure before generating your own problems.

Yoink these prompts

Prompt 1
Using iGSM, show me how to generate a batch of grade-school math reasoning problems with a set difficulty level.
Prompt 2
How do I use iGSM to verify whether a model's step-by-step solution to a generated problem is logically correct?
Prompt 3
Walk me through the Jupyter notebook examples in this repo to create a retry version of a problem a model answered wrong.
Prompt 4
Explain how iGSM represents a math problem as a dependency graph and how I can inspect that structure.

Frequently asked questions

wtf is igsm?

A tool that automatically generates grade-school-style math reasoning problems, verifies whether a solution's full logic is correct, and creates harder retry versions for testing AI models.

What language is igsm written in?

Mainly Python. The stack also includes Python, Jupyter.

Is igsm actively maintained?

Stale — no commits in 1-2 years (last push 2025-01-12).

How hard is igsm to set up?

Setup difficulty is rated moderate, with roughly 1h+ to a first successful run.

Who is igsm for?

Mainly researcher.

View the repo → Decode another repo

This repo across BitVibe Labs

Don't trust strangers blindly. Verify against the repo.

Other repos in this lane