Rowhammer bit-flip in Triton slows or corrupts LLM inference
LLVMhammer raised single-GPU latency 4-16% on four models and could cut BERTScore to 25%.
2 stories, newest first · one card per story
LLVMhammer raised single-GPU latency 4-16% on four models and could cut BERTScore to 25%.
Ten epochs shrink the sensitive tail 5 to 13 times on nine vision models while staying within 0.5 accuracy points.
Today 2.2K readers · 2.7K views · 404 reads · all time 108K views