TokenGate: Adapting LLMs for pre-fill with learned skimming for ~11x speedup

1 comments

• anandvc 7 hours ago

  Full paper: https://arxiv.org/html/2610.26754v1