trotsky1997/MathBlackBox

Health v2 · maintenance only

Flags: no_releases no_license

Adoption not part of the score

1033 stars · 104 forks observed · 2026-08-28

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-30, confidence not recorded

MathBlackBox (MCTSr) is a Python research implementation of Monte Carlo Tree Self-Refine, an algorithm that combines MCTS with LLM self-refinement to solve challenging math problems using models like LLaMA-3 8B. It runs against vLLM or OpenAI-compatible servers and benchmarks on datasets such as GSM8K, MATH, and OlympiadBench.

Use cases

• solve olympiad math problems with an open-source LLM
• run MCTS self-refine search over gsm8k with llama-3 8b
• benchmark LLM mathematical reasoning on MATH and AIME datasets
• reproduce the MCTSr paper results
• serve a local vllm model and evaluate it on math benchmarks

When to choose

• you want to reproduce or extend the MCTSr / LLaMA-Berry math reasoning research
• you have GPU resources and a vLLM or OpenAI-compatible inference server
• you need MCTS-style search with early stopping on math benchmarks

When to avoid

• you need a production-ready math solver or polished application
• you lack GPU infrastructure or a Slurm cluster
• you want a maintained library with a stable API or license

Facets

library · maturity experimental

llm-inference agent-framework machine-learning benchmarking large-language-models mathematics artificial-intelligence python mcts self-refine mathematical-reasoning vllm slurm research-code olympiad-math natural-language-processing linux gpu

Member repositories

For agents

markdown · JSON · MCP: product_card(name="trotsky1997/MathBlackBox")