Search papers, labs, and topics across Lattice.
Duke University
1
0
3
11
LLM-based GPU kernel generators ace synthetic benchmarks, but choke when faced with real-world production constraints, achieving at best 0.94x speedup.
