Code and datasets for a paper: Reasoning-CV: Fine-tuning Powerful Reasoning LLMs for Knowledge-Assisted Claim Verification

Arxiv: https://arxiv.org/pdf/2505.12348

This repository includes training data, testing data, training scripts, testing scripts, and testing resultss. First, unzip the zip files to obtain complete data.

Refer to \trainingset for training data.

Refer to \testset for testing data.

Refer to sft-lora.sh, sft-lora-dpo-stage3-guide.sh, sft-lora-dpo-stage3-guide2.sh for training scripts.

For evaluation, run vllm-evaluate.py first for vericities with different LLMs, then run Judge_f1.py for F1 scores.

\testset includes report results on some datasets. We will publish our LLMs on Huggingface afterward, and you can now run Judge_f1.py to get the F1 scores for these results.

See https://huggingface.co/zz1358m/Reasoning-CV to download LLMs for CoT-Verify and Decompose.