SJJ0854/FSFR
收藏资源简介:
--- license: mit task_categories: - image-text-to-text language: - en pretty_name: OmniVL-Guard Training Data size_categories: - 10B<n<100B --- # OmniVL-Guard <div align="center"> <!-- Keep this line empty for spacing --> [](https://arxiv.org/abs/2602.10687) [](https://github.com/shen8424/OmniVL-Guard) [](https://huggingface.co/SJJ0854/OmniVL-Guard-2B) [](https://icml.cc) [](./) </div> ## FSFR — Training Data Training datasets for the OmniVL-Guard safety guard model, accepted at **ICML 2026**. ## Contents | File | Size | Description | |------|------|-------------| | `media_data_RL.zip` | 27.3 GB | Multi-modal RL training data | | `media_data.zip` | 22.1 GB | Multi-modal training data | | `Refined-SFT.jsonl` | 105 MB | Refined SFT examples | | `RL.jsonl` | 38 MB | RL training examples |
This dataset is the training data for the OmniVL-Guard safety guard model, accepted at ICML 2026. It includes multi-modal RL training data, multi-modal training data, refined SFT examples, and RL training examples, designed for image-text-to-text tasks, with a size between 10B and 100B, and in English language.




