Unakar/Logic-RL

Health v2 · maintenance only

Flags: no_releases

Adoption not part of the score

2448 stars · 163 forks observed · 2026-08-28

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

Logic-RL is a research framework that reproduces R1-Zero-style LLM reasoning via rule-based reinforcement learning (GRPO) on logic puzzles, built on verl and vLLM. It includes data preprocessing, reward modeling, and training scripts for 4×A100 setups.

Use cases

• train an LLM to reason with rule-based reinforcement learning
• reproduce R1 Zero emergent reasoning behavior
• run GRPO training on logic puzzle datasets
• benchmark small models on multi-person logic puzzles
• generate and preprocess knight-and-knave puzzle data

When to choose

• you want to replicate or extend rule-based RL for LLM reasoning research
• you have multi-GPU (A100) resources and want a working verl/vLLM training pipeline
• you need a reproducible baseline for reasoning benchmarks

When to avoid

• you need a production LLM serving or inference system
• you lack GPU cluster resources for RL training
• you want a general-purpose RL library rather than a research reproduction

Facets

library · maturity active

llm-training reinforcement-learning machine-learning benchmarking large-language-models reinforcement-learning machine-learning artificial-intelligence python grpo verl vllm rule-based-reward logic-puzzles reasoning r1-zero research-code gpu linux docker

Member repositories

For agents

markdown · JSON · MCP: product_card(name="Unakar/Logic-RL")