Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

HarmLevelBench: Evaluating Harm-Level Compliance and the Impact of Quantization on Model Alignment

Yannis Belkhiter, Giulio Zizzo, Sergio Maffeis

TL;DR

This study aims to demonstrate the influence of harmful input queries on the complexity of jailbreaking techniques, as well as to deepen the understanding of LLM vulnerabilities and improve methods for assessing model robustness when confronted with harmful content, particularly in the context of compression strategies.

Abstract

With the introduction of the transformers architecture, LLMs have revolutionized the NLP field with ever more powerful models. Nevertheless, their development came up with several challenges. The exponential growth in computational power and reasoning capabilities of language models has heightened concerns about their security. As models become more powerful, ensuring their safety has become a crucial focus in research. This paper aims to address gaps in the current literature on jailbreaking techniques and the evaluation of LLM vulnerabilities. Our contributions include the creation of a novel dataset designed to assess the harmfulness of model outputs across multiple harm levels, as well as a focus on fine-grained harm-level analysis. Using this framework, we provide a comprehensive benchmark of state-of-the-art jailbreaking attacks, specifically targeting the Vicuna 13B v1.5 model. Additionally, we examine how quantization techniques, such as AWQ and GPTQ, influence the alignment and robustness of models, revealing trade-offs between enhanced robustness with regards to transfer attacks and potential increases in vulnerability on direct ones. This study aims to demonstrate the influence of harmful input queries on the complexity of jailbreaking techniques, as well as to deepen our understanding of LLM vulnerabilities and improve methods for assessing model robustness when confronted with harmful content, particularly in the context of compression strategies.

HarmLevelBench: Evaluating Harm-Level Compliance and the Impact of Quantization on Model Alignment

TL;DR

This study aims to demonstrate the influence of harmful input queries on the complexity of jailbreaking techniques, as well as to deepen the understanding of LLM vulnerabilities and improve methods for assessing model robustness when confronted with harmful content, particularly in the context of compression strategies.

Abstract

With the introduction of the transformers architecture, LLMs have revolutionized the NLP field with ever more powerful models. Nevertheless, their development came up with several challenges. The exponential growth in computational power and reasoning capabilities of language models has heightened concerns about their security. As models become more powerful, ensuring their safety has become a crucial focus in research. This paper aims to address gaps in the current literature on jailbreaking techniques and the evaluation of LLM vulnerabilities. Our contributions include the creation of a novel dataset designed to assess the harmfulness of model outputs across multiple harm levels, as well as a focus on fine-grained harm-level analysis. Using this framework, we provide a comprehensive benchmark of state-of-the-art jailbreaking attacks, specifically targeting the Vicuna 13B v1.5 model. Additionally, we examine how quantization techniques, such as AWQ and GPTQ, influence the alignment and robustness of models, revealing trade-offs between enhanced robustness with regards to transfer attacks and potential increases in vulnerability on direct ones. This study aims to demonstrate the influence of harmful input queries on the complexity of jailbreaking techniques, as well as to deepen our understanding of LLM vulnerabilities and improve methods for assessing model robustness when confronted with harmful content, particularly in the context of compression strategies.

Paper Structure

This paper contains 15 sections, 6 figures, 7 tables.

Table of Contents

  1. Introduction
  2. HarmLevelBench Pipeline
  3. Dataset
  4. Dataset structure
  5. Comparison with the pruned AdvBench
  6. Metrics
  7. Experimental setup
  8. Jailbreaking methods
  9. Models
  10. Evaluation
  11. Classic ASR
  12. HarmLevel
  13. Vulnerability of Quantized Models to Direct Attacks
  14. Enhanced Robustness of Quantization Against Transferred Attacks
  15. Summary and Conclusion

Figures (6)

  • Figure 1: Comparison of PCA applied to the BERT encoding of two datasets
  • Figure 2: Average ASR by HarmLevel and jailbreaking complexity for Vicuna 13B v1.5
  • Figure 3: GPT ASR Heatmap
  • Figure 4: ASR by Harm level and jailbreaking complexity for AWQ Vicuna 13B v1.5 Direct attacks
  • Figure 5: ASR by HarmLevel and jailbreaking complexity for AWQ Vicuna 13B v1.5 Transferred attacks
  • ...and 1 more figures