#KernelLLM

Meta introduces KernelLLM, an 8-billion-parameter model that automates converting PyTorch modules into efficient Triton GPU kernels, outperforming larger models in kernel generation benchmarks.