We study how training data contributes to the emergence of toxic behaviors in large language models. Most prior work on reducing model toxicity adopts *reactive* approaches, such as fine-tuning pre-trained (and potentially toxic) models to align them with human values. In contrast, we propose a *proactive* approach—IF-Guide—that leverages influence functions to identify and suppress harmful tokens in the training data. To this end, we first show that standard influence functions are ineffective at discovering harmful training records. We then present a novel adaptation that measures token-level attributions from training data to model toxicity, along with techniques for selecting toxic training documents and a learning objective that can be integrated into both pre-training and fine-tuning. Moreover, IF-Guide does not rely on human-preference data, which is typically required by existing alignment methods. In our evaluation, we demonstrate that IF-Guide substantially reduces both explicit and implicit toxicity—by up to 10$\times$ compared to uncensored models, and up to 3$\times$ compared to baseline alignment methods such as DPO and RAD—across both pre-training and fine-tuning scenarios. IF-Guide is computationally efficient: a billion-parameter model is *not necessary* for computing influence scores; a million-parameter model—with 7.5$\times$ fewer parameters—can effectively serve as a proxy for identifying harmful data.

Deep neural networks are known to be resilient to random bitwise faults in their parameters. However, this resilience has primarily been established through studies of classification models. The extent to which this claim holds for large-language models remains under-explored. In this work, we conduct an extensive measurement study on the impact of random bitwise faults in commercial-scale language model inference. We first expose that these language models are not truly resilient to random bit-flips. While aggregate metrics such as accuracy may suggest resilience, an in-depth inspection of the generated outputs shows significant degradation in text quality. Our analysis also shows that tasks requiring more complex reasoning suffer more from performance and quality degradation. Moreover, we extend our resilience analysis to models with augmented reasoning capabilities, such as Chain-of-Thought or Mixture of Experts architectures.

In this paper, we systematically evaluate the robustness of multi-exit language models against adversarial slowdown. To audit their robustness, we design a slowdown attack that generates natural adversarial text bypassing early-exit points. We use the resulting WAFFLE attack as a vehicle to conduct a comprehensive evaluation of three multi-exit mechanisms with the GLUE benchmark against adversarial slowdown. We then show our attack significantly reduces the computational savings provided by the three methods in both white-box and black-box settings. The more complex a mechanism is, the more vulnerable it is to adversarial slowdown. We also perform a linguistic analysis of the perturbed text inputs, identifying common perturbation patterns that our attack generates, and comparing them with standard adversarial text attacks. Moreover, we show that adversarial training is ineffective in defeating our slowdown attack, but input sanitization with a conversational model, e.g., ChatGPT, can remove perturbations effectively. This result suggests that future work is needed for developing efficient yet robust multi-exit models. Our code is available at: https://github.com/ztcoalson/WAFFLE