PRISM-CoT-new: Expanded Supervised Fine-Tuning Corpus for Vision-Language Model Safety

PRISM-CoT-new is an expanded supervised fine-tuning corpus for the PRISM Vision-Language Model safety alignment framework. It supersedes the original PRISM-CoT dataset for SFT use cases and was created by andyc03, with contributions from sources like prism-cot-orig and holisafe-bedrock. The dataset was last updated on May 14, 2026.

Use Cases

• Supervised fine-tuning of vision-language models based on the PRISM safety alignment framework.
• Training models on integrated safety reasoning for multimodal inputs, as described in the associated paper.
• Benchmarking safety alignment methods using the expanded corpus from multiple sources.

Strengths

• Supersedes a previous version (andyc03/PRISM-CoT), indicating iterative improvement.
• Composition includes at least 5,644 records from listed sources like prism-cot-orig (4,818) and holisafe-bedrock (826).
• Associated with a published paper and open-source code repository, providing methodological context.

Limitations

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

Provenance

Source

huggingface, andyc03

Collection Method

Curated from multiple sources including prism-cot-orig and holisafe-bedrock, with roles like 'Original PRISM-CoT' and 'Curator'.

Time Range

null

Freshness

Last updated 2026-05-14 12:38:38; freshness should be verified.

Geography

null