Just saw an adaptation of Karpathy’s Autoresearch intended to chew on Triton/CUDA kernels: AutoKernel.
This may be worth experimenting with running overnight… though it may need a bit of expansion to include more modern model types.
MIT license
Just saw an adaptation of Karpathy’s Autoresearch intended to chew on Triton/CUDA kernels: AutoKernel.
This may be worth experimenting with running overnight… though it may need a bit of expansion to include more modern model types.
MIT license