Research map / Large language models / Agents, code and math
A Solver-in-the-Loop Framework for Improving LLMs on Answer Set Programming for Logic Puzzle Solving
Timo Pierre Schrader, Lukas Lange, Tobias Kaminski, Simon Razniewski, Annemarie Friedrich
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 |
|---|---|---|---|
| SFT on solver-filtered chosen ASP encodings | Llama-3.3 70B · LogicPuzzles (two-shot) | accuracy (%): 31.0 -> 55.4 | Improved |
| add best-of-5 reward-based sampling (no fallbacks) (ablation) | Llama-3.3 70B +SFT · LogicPuzzles (two-shot) | accuracy (%): 55.4 -> 62.7 | Improved |
| add regeneration fallback to best-of-5 (ablation) | Llama-3.3 70B +SFT +Seq · LogicPuzzles (two-shot) | accuracy (%): 62.7 -> 64.6 | Improved |
| add backtracking fallback to best-of-5 (ablation) | Llama-3.3 70B +SFT +Seq · LogicPuzzles (two-shot) | accuracy (%): 62.7 -> 62.7 | No clear effect |
| combine regeneration and backtracking (ablation) | Llama-3.3 70B +SFT +Seq · LogicPuzzles (two-shot) | accuracy (%): 62.7 -> 66.8 | Improved |
| increase best-of-N from 5 to 25 (ablation) | Llama-3.3 70B +SFT +TT · LogicPuzzles (two-shot) | accuracy (%): 66.8 -> 69.2 | Improved |
| apply solver-reward best-of-5 to closed model without fine-tuning | GPT-4.1-mini · LogicPuzzles (two-shot) | accuracy (%): 39.2 -> 55.4 | Improved |
| plug SFT-tuned models into prompt pipeline | Llama-3.3 70B · LogicPuzzles (prompt pipeline) | accuracy (%): 56.8 -> 78.4 (with TT) | Improved |
| SFT on LogicPuzzles-derived data evaluated on harder GridPuzzles | 8B models · GridPuzzles (two-shot) | accuracy (%): approx. +10pp on average | Improved |
| use DPO instead of SFT (ablation) | ASP instruction tuning · initial experiments | accuracy | Hurt |
Closest papers by meaning
- Logic.py: Bridging the Gap between LLMs and Constraint Solvers (NeurIPS 2025 · similarity 0.90)
- ZebraLogic: On the Scaling Limits of LLMs for Logical Reasoning (ICML 2025 · similarity 0.88)
- Reasoning-as-Logic-Units: Scaling Test-Time Reasoning in Large Language Models Through Logic Unit Alignment (ICML 2025 · similarity 0.88)
- Enhancing Reasoning Capabilities of LLMs via Principled Synthetic Logic Corpus (NeurIPS 2024 · similarity 0.87)
- Planning in Natural Language Improves LLM Search for Code Generation (ICLR 2025 · similarity 0.87)
- Leveraging Online Olympiad-Level Math Problems for LLMs Training and Contamination-Resistant Evaluation (ICML 2025 · similarity 0.87)
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.
