TokenGate: Adapting LLMs for pre-fill with learned skimming for ~11x speedup(github.com)
1 pointby anandvc3 hours ago1 comment