HYU-NLP/OMHBench
收藏Hugging Face2026-04-17 更新2026-05-10 收录
下载链接：
https://hf-mirror.com/datasets/HYU-NLP/OMHBench
下载链接
链接失效反馈官方服务：
资源简介：
# OMHBench: Benchmarking Balanced and Grounded Omni-Modal Multi-Hop Reasoning (ACL 2026 Findings)
## Dataset Structure
| Column | Description |
|--------|-------------|
| `sample_id` | Unique sample ID |
| `question_id` | Base question ID. Each question ID has 6 reasoning paths. |
| `question_id_with_reasoning_path` | Question ID with reasoning path (e.g. `0_T-I-S`) |
| `reasoning_path` | Reasoning path (e.g. `T-I-S`, `S-T-I`) |
| `domain` | Domain: `nutrition`, `finance`, `economics`, `climate` |
| `mode` | `Connect` or `Reasoning` |
| `type` | Multi-hop reasoning type (e.g. `Lookup-Comparison-Retrieval`) |
| `answer` | Ground truth answer |
| `question` | Question |
| `input_text` | Text modality input |
| `input_image` | Image modality input |
| `input_speech` | Speech modality input |
提供机构：
HYU-NLP
