Increased LLM Vulnerabilities from Fine-tuning and Quantization

D. Kumar, A. Kumar, S. Agarwal, P. Harshangi

arXiv preprint · Apr 2024first author

abstract

Model quantization and fine-tuning have become essential techniques for deploying large language models in resource-constrained environments. However, their impact on model security has been largely overlooked. In this work, we conduct the first systematic study of how quantization and fine-tuning affect adversarial robustness in LLMs. Through comprehensive experiments across multiple model families and quantization schemes, we demonstrate that aggressive quantization can increase attack success rates significantly compared to full-precision models.

QuantizationAdversarial AttacksModel CompressionSecurityLLMs

cite

@misc{kumar2024increased,
  title     = {Increased LLM Vulnerabilities from Fine-tuning and Quantization},
  author    = {Divyanshu Kumar and Anurakt Kumar and Sahil Agarwal and Prashanth Harshangi},
  howpublished = {arXiv preprint},
  year      = {2024}
}