FSFR: Training Data for the OmniVL-Guard Safety Model

Training datasets for the OmniVL-Guard safety guard model, accepted at ICML 2026. The collection includes multi-modal training data (22.1 GB), multi-modal reinforcement learning data (27.3 GB), and refined supervised fine-tuning examples (105 MB). The dataset was uploaded by SJJ0854 and last updated on May 11, 2026.

Use Cases

• Training multimodal safety guard models based on the provided multi-modal training data.
• Conducting reinforcement learning for AI safety based on the RL training examples.
• Performing supervised fine-tuning of multimodal models based on the refined SFT examples.

Strengths

• Multi-modal training data file is 22.1 GB, suggesting substantial visual or audio content.
• Multi-modal RL training data file is 27.3 GB, indicating a large-scale reinforcement learning corpus.
• Refined SFT examples file is 105 MB, implying curated, high-quality fine-tuning data.

Limitations

• Description metadata is limited; actual data quality requires manual inspection after download.
• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.

Provenance

Source

SJJ0854

Freshness

Last updated 2026-05-11 05:56:35; freshness should be verified.