---
title: CPT-Eval-RAGConflicts
canonical_url: "https://www.modelscope.cn/datasets/Tsinghuadhy/CPT-Eval-RAGConflicts"
md_url: "https://www.modelscope.cn/datasets/Tsinghuadhy/CPT-Eval-RAGConflicts.md"
repository: Tsinghuadhy/CPT-Eval-RAGConflicts
chinese_name: "CPT 评测集 RAG 冲突 (DRAGged-into-Conflicts)"
last_updated: 2026-05-31
license: MIT
storage_size: "45 MB"
downloads: 8
stars: 0
---

# CPT-Eval-RAGConflicts

> CPT-Eval-RAGConflicts - Tsinghuadhy 在 ModelScope 开源的数据集。CPT-Eval-RAGConflicts

Tsinghuadhy/CPT-Eval-RAGConflicts 是 ModelScope 魔搭社区上的数据集，存储大小 45 MB，采用 MIT 许可。

- **Repository**: Tsinghuadhy/CPT-Eval-RAGConflicts
- **License**: MIT
- **Storage size**: 45 MB
- **Downloads**: 8
- **Stars**: 0
- **Last updated**: 2026-05-31

Source: https://www.modelscope.cn/datasets/Tsinghuadhy/CPT-Eval-RAGConflicts

---

# CPT-Eval-RAGConflicts

[DRAGged-into-Conflicts](https://arxiv.org/abs/2406.16289) split used as the
zero-shot RAG transfer benchmark in
[Cognitive Pairwise Training (CPT)](https://github.com/Tsinghua-dhy/CPT) (paper Table 4).

The CPT model is evaluated **without any RAG-specific training** — the
abstention boundary learned during CPT mid-training is expected to transfer.

## Files

- `conflicts.jsonl` — full split, ~46 MB

## Schema

```json
{
  "question": "...",
  "documents": [...],
  "correct_answer": "...",
  "conflict_type": "No conflict | Conflict due to misinformation | Conflict due to outdated information | Complementary information | Conflicting opinions and research outcomes"
}
```

## Use

```bash
modelscope download --dataset Tsinghuadhy/CPT-Eval-RAGConflicts --local_dir CPT/eval/dataset/conflicts
cd CPT/eval/rag
bash eval_rag_conflicts.sh Tsinghuadhy/CPT-RL-Qwen3-4B
```

## Source

Original *DRAGged-into-Conflicts*: <https://github.com/AviAvni/DRAGged-into-Conflicts>