# microsoft/rStar

Repository: https://github.com/microsoft/rStar
Canonical: https://ross.abutalabs.com/products/rstar
Language: Python
License: MIT
License Family: permissive
Last push: 2025-09-12T08:26:02+00:00

## Health v2 (maintenance only)
Score: 41/100 (v2, computed 2026-09-02T17:46:02.011165+00:00)
- activity 41, release rhythm 35, longevity 52
- inputs: {"age_days": 736, "days_push": 355, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: no_releases
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 1425, forks 132 (observed 2026-08-28T04:04:41.518833+00:00)

## What it is
Microsoft's research repository for rStar2-Agent, a 14B math reasoning model trained with agentic reinforcement learning that autonomously uses coding tools. It includes the GRPO-RoC RL algorithm, scalable RL training infrastructure, and model inference/evaluation code.

## Use cases
- train a small LLM with agentic reinforcement learning for math reasoning
- reproduce the rStar2-Agent 14B model training pipeline
- run a math reasoning model that calls Python tools to verify answers
- implement GRPO with resample-on-correct rollout strategy
- evaluate small LLMs on competition-level math benchmarks
- set up high-throughput RL rollout infrastructure on limited GPUs

## When to choose
- you are researching agentic RL for LLM reasoning
- you want to train or fine-tune small models to use coding tools during reasoning
- you need a reference implementation of GRPO-RoC for tool-calling agents

## When to avoid
- you need a production-ready chatbot or inference server
- you lack multi-GPU resources for RL training
- you want a general-purpose agent framework rather than a research codebase

## Facets
- artifact type: library
- maturity: active
- function: machine-learning, llm-training, agent-framework, llm-inference
- domain: large-language-models, reinforcement-learning, artificial-intelligence
- platform: python
- tags: agentic-reasoning, reinforcement-learning, grpo, math-reasoning, tool-calling, research-code, microsoft, ai-agents, gpu, linux

## Member repositories
- microsoft/rStar (main) score 41

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:04:41.518833+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-30T04:37:31.668498+00:00, confidence not recorded.
  - readme: https://github.com/microsoft/rStar (fetched 2026-08-28T04:04:41.518833+00:00, sha ab04dd4cfd10)
- Data as of 2026-08-30T08:39:29.467469+00:00.