OmniAlign-V: Towards Enhanced Alignment of MLLMs with Human Preference
This work addresses the problem of aligning MLLMs with human preferences for developers and researchers, though it is incremental as it builds on existing finetuning methods.
The paper tackles the gap in human preference alignment for multi-modal large language models (MLLMs) by introducing OmniAlign-V, a dataset of 200K training samples, and MM-AlignBench, a benchmark for evaluation, showing that finetuning with these resources significantly enhances alignment while maintaining performance on standard VQA benchmarks.
Recent advancements in open-source multi-modal large language models (MLLMs) have primarily focused on enhancing foundational capabilities, leaving a significant gap in human preference alignment. This paper introduces OmniAlign-V, a comprehensive dataset of 200K high-quality training samples featuring diverse images, complex questions, and varied response formats to improve MLLMs' alignment with human preferences. We also present MM-AlignBench, a human-annotated benchmark specifically designed to evaluate MLLMs' alignment with human values. Experimental results show that finetuning MLLMs with OmniAlign-V, using Supervised Fine-Tuning (SFT) or Direct Preference Optimization (DPO), significantly enhances human preference alignment while maintaining or enhancing performance on standard VQA benchmarks, preserving their fundamental capabilities. Our datasets, benchmark, code and checkpoints have been released at https://github.com/PhoenixZ810/OmniAlign-V.