MM-RLHF: A Multimodal LLM Alignment Dataset and Reward Model

MM-RLHF is a project for aligning Multimodal Large Language Models with human preferences. The release includes a high-quality alignment dataset and a strong critique-based reward model. The project was open-sourced by yifanzhang114 in February 2025.

Use Cases

• Fine-tuning MLLMs for alignment based on human preferences.
• Training or evaluating reward models for multimodal tasks.
• Benchmarking MLLM safety and performance using the included evaluation suites.

Strengths

• Includes a critique-based reward model and its training algorithm.
• Released as part of a comprehensive project for MLLM alignment.
• Associated with benchmark suites MM-RewardBench and MM-SafetyBench.

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

yifanzhang114 on Hugging Face

Freshness

Last updated 2025-04-21 02:43:14; freshness should be verified.