Phi-3 Safety Post-Training: Aligning Language Models with a "Break-Fix" Cycle

• URL: http://arxiv.org/abs/2407.13833v2
• Date: Fri, 23 Aug 2024 00:04:31 GMT
• Title: Phi-3 Safety Post-Training: Aligning Language Models with a "Break-Fix" Cycle
• Authors: Emman Haider, Daniel Perez-Becker, Thomas Portet, Piyush Madan, Amit Garg, Atabak Ashfaq, David Majercak, Wen Wen, Dongwoo Kim, Ziyi Yang, Jianwen Zhang, Hiteshi Sharma, Blake Bullwinkel, Martin Pouliot, Amanda Minnich, Shiven Chawla, Solianna Herrera, Shahed Warreth, Maggie Engler, Gary Lopez, Nina Chikanov, Raja Sekhar Rao Dheekonda, Bolor-Erdene Jagdagdorj, Roman Lutz, Richard Lundeen, Tori Westerhoff, Pete Bryan, Christian Seifert, Ram Shankar Siva Kumar, Andrew Berkley, Alex Kessler,
• Abstract summary: We present our methodology for safety aligning the Phi-3 series of language models. We utilize a "break-fix" cycle, performing multiple rounds of dataset curation, safety post-training, benchmarking, red teaming, and vulnerability identification. Results indicate that this approach iteratively improved the performance of the Phi-3 models across a wide range of responsible AI benchmarks.
• Score: 11.619942198710334
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Recent innovations in language model training have demonstrated that it is possible to create highly performant models that are small enough to run on a smartphone. As these models are deployed in an increasing number of domains, it is critical to ensure that they are aligned with human preferences and safety considerations. In this report, we present our methodology for safety aligning the Phi-3 series of language models. We utilized a "break-fix" cycle, performing multiple rounds of dataset curation, safety post-training, benchmarking, red teaming, and vulnerability identification to cover a variety of harm areas in both single and multi-turn scenarios. Our results indicate that this approach iteratively improved the performance of the Phi-3 models across a wide range of responsible AI benchmarks. Finally, we include additional red teaming strategies and evaluations that were used to test the safety behavior of Phi-3.5-mini and Phi-3.5-MoE, which were optimized for multilingual capabilities.

Related papers

• In-Context Experience Replay Facilitates Safety Red-Teaming of Text-to-Image Diffusion Models [97.82118821263825]
  Text-to-image (T2I) models have shown remarkable progress, but their potential to generate harmful content remains a critical concern in the ML community. We propose ICER, a novel red-teaming framework that generates interpretable and semantic meaningful problematic prompts. Our work provides crucial insights for developing more robust safety mechanisms in T2I systems.
  arXiv  Detail & Related papers  (2024-11-25T04:17:24Z)
• Steering Language Model Refusal with Sparse Autoencoders [16.78963326253821]
  We identify and steer features in Phi-3 Mini that mediate refusal behavior. We find that feature steering can improve Phi-3 Minis robustness to jailbreak attempts across various harms. However, feature steering can adversely affect overall performance on benchmarks.
  arXiv  Detail & Related papers  (2024-11-18T05:47:02Z)
• Prefix Guidance: A Steering Wheel for Large Language Models to Defend Against Jailbreak Attacks [27.11523234556414]
  We propose a plug-and-play and easy-to-deploy jailbreak defense framework, namely Prefix Guidance (PG) PG guides the model to identify harmful prompts by directly setting the first few tokens of the model's output. We demonstrate the effectiveness of PG across three models and five attack methods.
  arXiv  Detail & Related papers  (2024-08-15T14:51:32Z)
• SEAS: Self-Evolving Adversarial Safety Optimization for Large Language Models [19.486685336959482]
  Large language models (LLMs) continue to advance in capability and influence, ensuring their security and preventing harmful outputs has become crucial. A promising approach to address these concerns involves training models to automatically generate adversarial prompts for red teaming. We introduce the $mathbfStextelf-mathbfEtextvolving mathbfAtextdversarial mathbfStextafetyety mathbf(SEAS)$ optimization framework. SEAS operates through three iterative stages: Initialization, Attack, and Adversa
  arXiv  Detail & Related papers  (2024-08-05T16:55:06Z)
• What Makes and Breaks Safety Fine-tuning? A Mechanistic Study [64.9691741899956]
  Safety fine-tuning helps align Large Language Models (LLMs) with human preferences for their safe deployment. We design a synthetic data generation framework that captures salient aspects of an unsafe input. Using this, we investigate three well-known safety fine-tuning methods.
  arXiv  Detail & Related papers  (2024-07-14T16:12:57Z)
• SecureBERT and LLAMA 2 Empowered Control Area Network Intrusion Detection and Classification [2.824211356106516]
  We develop two distinct models for CAN intrusion detection: CAN-SecureBERT and CAN-LLAMA2. Can-LLAMA2 model surpasses the state-of-the-art models by achieving an exceptional performance 0.999993 in terms of balanced accuracy, precision detection rate, F1 score, and a remarkably low false alarm rate of 3.10e-6.
  arXiv  Detail & Related papers  (2023-11-19T23:49:08Z)
• Text generation for dataset augmentation in security classification tasks [55.70844429868403]
  This study evaluates the application of natural language text generators to fill this data gap in multiple security-related text classification tasks. We find substantial benefits for GPT-3 data augmentation strategies in situations with severe limitations on known positive-class samples.
  arXiv  Detail & Related papers  (2023-10-22T22:25:14Z)
• ASSERT: Automated Safety Scenario Red Teaming for Evaluating the Robustness of Large Language Models [65.79770974145983]
  ASSERT, Automated Safety Scenario Red Teaming, consists of three methods -- semantically aligned augmentation, target bootstrapping, and adversarial knowledge injection. We partition our prompts into four safety domains for a fine-grained analysis of how the domain affects model performance. We find statistically significant performance differences of up to 11% in absolute classification accuracy among semantically related scenarios and error rates of up to 19% absolute error in zero-shot adversarial settings.
  arXiv  Detail & Related papers  (2023-10-14T17:10:28Z)
• CodeLMSec Benchmark: Systematically Evaluating and Finding Security Vulnerabilities in Black-Box Code Language Models [58.27254444280376]
  Large language models (LLMs) for automatic code generation have achieved breakthroughs in several programming tasks. Training data for these models is usually collected from the Internet (e.g., from open-source repositories) and is likely to contain faults and security vulnerabilities. This unsanitized training data can cause the language models to learn these vulnerabilities and propagate them during the code generation procedure.
  arXiv  Detail & Related papers  (2023-02-08T11:54:07Z)
• On Adversarial Robustness of Point Cloud Semantic Segmentation [16.89469632840972]
  PCSS has been applied in many safety-critical applications like autonomous driving. This study shows how PCSS models are affected under adversarial samples. We call the attention of the research community to develop new approaches to harden PCSS models.
  arXiv  Detail & Related papers  (2021-12-11T00:10:00Z)
• Adversarial GLUE: A Multi-Task Benchmark for Robustness Evaluation of Language Models [86.02610674750345]
  Adversarial GLUE (AdvGLUE) is a new multi-task benchmark to explore and evaluate the vulnerabilities of modern large-scale language models under various types of adversarial attacks. We apply 14 adversarial attack methods to GLUE tasks to construct AdvGLUE, which is further validated by humans for reliable annotations. All the language models and robust training methods we tested perform poorly on AdvGLUE, with scores lagging far behind the benign accuracy.
  arXiv  Detail & Related papers  (2021-11-04T12:59:55Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.