Robustness of Llama-3-70B Block-Sparse FlashAttention Under Adversarial Perturbations

This report synthesises findings from 4 peer-reviewed papers addressing the following research question: How does the robustness of Llama-3-70b with Block-Sparse FlashAttention compare to full attention under adversarial perturbations in the ConvAITest dataset, measured by accuracy drop under noise. 8 claims were extracted from source literature; 8 were independently verified against retrieved documents. An automated multi-reviewer quality assessment produced a score of 7.5/10. This report is a machine-generated literature synthesis and does not constitute original research.

Research goal: How does the robustness of Llama-3-70b with Block-Sparse FlashAttention compare to full attention under adversarial perturbations in the ConvAITest dataset, measured by accuracy drop under noise injection or token shuffling?

Autonomous literature synthesis. Automated review score: 7.5/10. Full text and citation available at Assignee Research.