Ngai Wong

Sparse prefilling can dramatically accelerate long-context inference in diffusion language models, achieving up to 28x speedup without sacrificing quality.

Attention Sink, where Transformers fixate on seemingly irrelevant tokens, is more than just a quirk – it's a fundamental challenge impacting training, inference, and even causing hallucinations, demanding a systematic approach to understanding and mitigating its effects.

Achieve better compression in low-bit quantization by considering not just numerical sensitivity, but also the structural role of each layer.