---
title: SimpleSearch-VL-SFT-5K
canonical_url: "https://www.modelscope.cn/datasets/dmmm997/SimpleSearch-VL-SFT-5K"
md_url: "https://www.modelscope.cn/datasets/dmmm997/SimpleSearch-VL-SFT-5K.md"
repository: dmmm997/SimpleSearch-VL-SFT-5K
last_updated: 2026-11-12
license: "Apache License 2.0"
storage_size: "1.5 GB"
downloads: 11
stars: 0
---

# SimpleSearch-VL-SFT-5K

> SimpleSearch-VL-SFT-5K - dmmm997 在 ModelScope 开源的数据集。SimpleSearch-VL-SFT-5K

dmmm997/SimpleSearch-VL-SFT-5K 是 ModelScope 魔搭社区上的数据集，存储大小 1.5 GB，采用 Apache License 2.0 许可。

- **Repository**: dmmm997/SimpleSearch-VL-SFT-5K
- **License**: Apache License 2.0
- **Storage size**: 1.5 GB
- **Downloads**: 11
- **Stars**: 0
- **Last updated**: 2026-11-12

Source: https://www.modelscope.cn/datasets/dmmm997/SimpleSearch-VL-SFT-5K

---

# SimpleSearch-VL-SFT-5K

SFT training set for SimpleSearch-VL. 5,193 evidence-verification trajectories in ms-swift chat format.

## Statistics

| Source | Samples |
|--------|---------|
| Wiki | 1,726 |
| LiveVQA | 1,539 |
| FVQA | 689 |
| Palace | 259 |
| WebQA | 154 |
| **Total** | **5,193** |

- Single-image samples: 706
- Multi-image samples: 4,487
- Total image references: 27,614

## Files

- `sft_ms_swift.jsonl` — each line has `messages` (chat turns) and `images` (absolute paths).
- `samples/` — image files referenced by `sft_ms_swift.jsonl`, one subdirectory per sample.

## Usage

Point the ms-swift SFT script to this JSONL:

```bash
--dataset /path/to/SimpleSearch-VL-SFT-5K/sft_ms_swift.jsonl
```