Research map / Large language models / Agents, code and math
LeanDojo: Theorem Proving with Retrieval-Augmented Language Models
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add retrieval-augmented premise selection to the tactic generator (ablation) | ReProver · LeanDojo Benchmark random test | Pass@1 (%): 47.6 -> 51.2 | Helped |
| add retrieval-augmented premise selection to the tactic generator (ablation) | ReProver · LeanDojo Benchmark novel_premises test | Pass@1 (%): 23.2 -> 26.3 | Helped |
| restrict retrieval candidates to accessible premises (ablation) | ReProver retriever · LeanDojo Benchmark random premise-selection test | R@1: 11.7 -> 13.5 | Helped |
| restrict retrieval candidates to accessible premises (ablation) | ReProver retriever · LeanDojo Benchmark novel_premises premise-selection test | R@1: 7.1 -> 9.1 | Helped |
| add in-file negative examples during retriever training (ablation) | ReProver retriever · LeanDojo Benchmark random premise-selection test | R@1: 10.8 -> 13.5 | Helped |
| add in-file negative examples during retriever training (ablation) | ReProver retriever · LeanDojo Benchmark novel_premises premise-selection test | R@1: 7.9 -> 9.1 | Helped |
| use learned retrieval rather than BM25 | premise selection · LeanDojo Benchmark random premise-selection test | R@1: 6.7 -> 13.5 | Helped |
| use learned retrieval rather than BM25 | premise selection · LeanDojo Benchmark novel_premises premise-selection test | R@1: 5.9 -> 9.1 | Helped |
| use ReProver rather than GPT-4 for theorem proving | theorem proving · LeanDojo Benchmark random test | Pass@1 (%): 29.0 -> 51.2 | Helped |
| use ReProver rather than GPT-4 for theorem proving | theorem proving · LeanDojo Benchmark novel_premises test | Pass@1 (%): 7.4 -> 26.3 | Helped |
Closest papers by meaning
- FaithSieve: Fine-Grained Evaluation of Math Proofs with Faithful Formal Evidence (NeurIPS 2026 · similarity 0.92)
- MA-LoT: Model-Collaboration Lean-based Long Chain-of-Thought Reasoning enhances Formal Theorem Proving (ICML 2025 · similarity 0.92)
- Lean-STaR: Learning to Interleave Thinking and Proving (ICLR 2025 · similarity 0.91)
- LEGO-Prover: Neural Theorem Proving with Growing Libraries (ICLR 2024 · similarity 0.91)
- APOLLO: Automated LLM and Lean Collaboration for Advanced Formal Reasoning (NeurIPS 2025 · similarity 0.91)
- MPS-Prover: Advancing Stepwise Theorem Proving by Multi-Perspective Search and Data Curation (NeurIPS 2025 · similarity 0.91)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.