luminal
luminal copied to clipboard
jafioti
Deep learning at the speed of light.
Related to #58, this is a temporary fix that may help to indicate how the Cuda training is going wrong. With this change this cuda training works.
Hi, I've found a runtime error and made this minimized example: ```rust #[test] fn test_add_grad_decreasing_idx() { let mut cx = Graph::new(); let a: GraphTensor = cx.tensor(); let a: GraphTensor =...
We're looking into superoptimizers, specifically adding a luminal_gpu crate with a gpu-agnostic superoptimizer pass, which will then feed into downstream gpu compilers in luminal_cuda and luminal_metal. So far, I'm specifically...