Search papers, labs, and topics across Lattice.
University of Michigan
1
0
3
Cut LLM cold starts from minutes to seconds by pre-materializing CUDA graph execution contexts, sidestepping brittle kernel patching and heavyweight checkpointing.
