Search papers, labs, and topics across Lattice.
1
0
3
13
LLM-powered agents can now optimize PyTorch inference by 2.88x on H100 GPUs, but only if you carefully balance exploit-heavy search with agents that fix errors.