ProRAC: A Neuro-symbolic Method for Reasoning about Actions with LLM-based Progression and Search

This is the code repository for the EMNLP 2026 paper.

Packages

1. uv sync
2. uv pip install -e .

API

In each *_run.py file, you need to replace the API_KEY and BASE_URL with your own. Besides, BASE_URL should not contain v1/chat/completion. For example, if the base url is https://api.deepseek.com/v1/chat/completion, you should replace it with https://api.deepseek.com/v1.

File Structure

• utils.py: Utility functions for sending requests, extracting content, etc.

• general_prompt.py: General prompt for all benchmarks, including reasoning and search.

• tree.py: Implementation for tree search algorithms, including A*, BFS, and ToT.

• self_consistency_utils.py: Utility functions for self-consistency evaluation.

• heuristic.py: Implementation for local LoRA model for heuristic computation.

• heuristic_computer.py: The new heuristic computation system with robust error handling, proper configuration validation, and automatic fallback mechanisms.

Synthesied Data

https://drive.google.com/file/d/19CqEC1ULKNpC157emd111gtq5VmqfH4w/view?usp=sharing