Share your thoughts, 1 month free Claude Pro on usSee more
WorkDL logo mark

Aligning Large Multimodal Models with Factually Augmented RLHF

About

Large Multimodal Models (LMM) are built across modalities and the misalignment between two modalities can result in "hallucination", generating textual outputs that are not grounded by the multimodal information in context. To address the multimodal misalignment issue, we adapt the Reinforcement Learning from Human Feedback (RLHF) from the text domain to the task of vision-language alignment, where human annotators are asked to compare two responses and pinpoint the more hallucinated one, and the vision-language model is trained to maximize the simulated human rewards. We propose a new alignment algorithm called Factually Augmented RLHF that augments the reward model with additional factual information such as image captions and ground-truth multi-choice options, which alleviates the reward hacking phenomenon in RLHF and further improves the performance. We also enhance the GPT-4-generated training data (for vision instruction tuning) with previously available human-written image-text pairs to improve the general capabilities of our model. To evaluate the proposed approach in real-world scenarios, we develop a new evaluation benchmark MMHAL-BENCH with a special focus on penalizing hallucinations. As the first LMM trained with RLHF, our approach achieves remarkable improvement on the LLaVA-Bench dataset with the 94% performance level of the text-only GPT-4 (while previous best methods can only achieve the 87% level), and an improvement by 60% on MMHAL-BENCH over other baselines. We opensource our code, model, data at https://llava-rlhf.github.io.

Zhiqing Sun, Sheng Shen, Shengcao Cao, Haotian Liu, Chunyuan Li, Yikang Shen, Chuang Gan, Liang-Yan Gui, Yu-Xiong Wang, Yiming Yang, Kurt Keutzer, Trevor Darrell• 2023

Related benchmarks

TaskDatasetResultRank
Visual Question AnsweringVizWiz
Accuracy51.7
1525
Object Hallucination EvaluationPOPE--
1455
Visual Question AnsweringGQA--
1249
Multimodal UnderstandingMMBench
Accuracy59.6
637
Multimodal UnderstandingMM-Vet--
531
Multimodal ReasoningMM-Vet
MM-Vet Score31.1
431
Instruction FollowingAlpacaEval
Win Rate81
227
Hallucination EvaluationMMHal-Bench
MMHal Score3.06
216
Instruction FollowingMT-Bench
MT-Bench Score6.18
215
Hallucination EvaluationAMBER
CHAIR8.3
172
Showing 10 of 48 rows

Other info

Code

Follow for update