KernelLLM

Kernel-writing models and methods, which either generate triton or CUDA kernels, are becoming increasingly popular. KernelLLM is a relatively small model, based on Llama 3.1 8B, capable of converting PyTorch code to triton kernels.