Benchmarking Multimodal Safety: Phi-3-Vision's Robust RAI Performance

9 Jul 2025

Table of Links

Abstract and 1 Introduction

2 Technical Specifications

3 Academic benchmarks

4 Safety

5 Weakness

6 Phi-3-Vision

6.1 Technical Specifications

6.2 Academic benchmarks

6.3 Safety

6.4 Weakness

References

A Example prompt for benchmarks

B Authors (alphabetical)

C Acknowledgements

6.3 Safety

To ensure the integration of Phi-3-Vision aligns with Microsoft’s Responsible AI (RAI) principles, we involved safety post-training in both Supervised Fine-Tuning (SFT) stage and Direct Preference Optimization (DPO) stage. In creating the safety training datasets, we utilized not only the text-only RAI datasets, but also a variety of in-house Multi-Modal (MM) RAI datasets that cover various harm categories identified in both public and internal MM RAI benchmarks. For the purpose of RAI evaluation, we performed a rigorous quantitative assessment on both public and internal benchmarks, this was done in conjunction with a human evaluation conducted by Microsoft’s internal red team.

In Table 3, we present the evaluation outcomes of Phi-3-Vision on three MM RAI benchmarks: one internal and two public benchmarks (specifically, RTVLM [LLY+ 24] and VLGuard [ZBY+ 24]). We juxtapose these results with those of other open-source models such as Llava-1.5 [LLLL23], Llava-1.6 [LLL+ 24], Qwen-VL-Chat [BBY+ 23], and GPT4-V[Ope23]. The results clearly indicate that safety posttraining notably enhances the RAI performance of Phi-3-Vision across all RAI benchmarks. In Figure 7, we further breakdown the performance across different RAI categories of the VLGuard and Internal benchmarks, demonstrating that safety post-training can aid Phi-3-Vision in improving RAI performance in nearly all categories.

Authors:

(1) Marah Abdin;

(2) Sam Ade Jacobs;

(3) Ammar Ahmad Awan;

(4) Jyoti Aneja;

(5) Ahmed Awadallah;

(6) Hany Awadalla;

(7) Nguyen Bach;

(8) Amit Bahree;

(9) Arash Bakhtiari;

(10) Jianmin Bao;

(11) Harkirat Behl;

(12) Alon Benhaim;

(13) Misha Bilenko;

(14) Johan Bjorck;

(15) Sébastien Bubeck;

(16) Qin Cai;

(17) Martin Cai;

(18) Caio César Teodoro Mendes;

(19) Weizhu Chen;

(20) Vishrav Chaudhary;

(21) Dong Chen;

(22) Dongdong Chen;

(23) Yen-Chun Chen;

(24) Yi-Ling Chen;

(25) Parul Chopra;

(26) Xiyang Dai;

(27) Allie Del Giorno;

(28) Gustavo de Rosa;

(29) Matthew Dixon;

(30) Ronen Eldan;

(31) Victor Fragoso;

(32) Dan Iter;

(33) Mei Gao;

(34) Min Gao;

(35) Jianfeng Gao;

(36) Amit Garg;

(37) Abhishek Goswami;

(38) Suriya Gunasekar;

(39) Emman Haider;

(40) Junheng Hao;

(41) Russell J. Hewett;

(42) Jamie Huynh;

(43) Mojan Javaheripi;

(44) Xin Jin;

(45) Piero Kauffmann;

(46) Nikos Karampatziakis;

(47) Dongwoo Kim;

(48) Mahoud Khademi;

(49) Lev Kurilenko;

(50) James R. Lee;

(51) Yin Tat Lee;

(52) Yuanzhi Li;

(53) Yunsheng Li;

(54) Chen Liang;

(55) Lars Liden;

(56) Ce Liu;

(57) Mengchen Liu;

(58) Weishung Liu;

(59) Eric Lin;

(60) Zeqi Lin;

(61) Chong Luo;

(62) Piyush Madan;

(63) Matt Mazzola;

(64) Arindam Mitra;

(65) Hardik Modi;

(66) Anh Nguyen;

(67) Brandon Norick;

(68) Barun Patra;

(69) Daniel Perez-Becker;

(70) Thomas Portet;

(71) Reid Pryzant;

(72) Heyang Qin;

(73) Marko Radmilac;

(74) Corby Rosset;

(75) Sambudha Roy;

(76) Olatunji Ruwase;

(77) Olli Saarikivi;

(78) Amin Saied;

(79) Adil Salim;

(80) Michael Santacroce;

(81) Shital Shah;

(82) Ning Shang;

(83) Hiteshi Sharma;

(84) Swadheen Shukla;

(85) Xia Song;

(86) Masahiro Tanaka;

(87) Andrea Tupini;

(88) Xin Wang;

(89) Lijuan Wang;

(90) Chunyu Wang;

(91) Yu Wang;

(92) Rachel Ward;

(93) Guanhua Wang;

(94) Philipp Witte;

(95) Haiping Wu;

(96) Michael Wyatt;

(97) Bin Xiao;

(98) Can Xu;

(99) Jiahang Xu;

(100) Weijian Xu;

(101) Sonali Yadav;

(102) Fan Yang;

(103) Jianwei Yang;

(104) Ziyi Yang;

(105) Yifan Yang;

(106) Donghan Yu;

(107) Lu Yuan;

(108) Chengruidong Zhang;

(109) Cyril Zhang;

(110) Jianwen Zhang;

(111) Li Lyna Zhang;

(112) Yi Zhang;

(113) Yue Zhang;

(114) Yunan Zhang;

(115) Xiren Zhou.

This paper is available on arxiv under CC BY 4.0 DEED license.

← Previous

Phi-3-Vision's Triumphant Performance on Key Multimodal Benchmarks

Up Next →

Confronting Multimodal LLM Challenges: Reasoning Gaps and Safety Trade-offs in Phi-3-Vision