Diogo
Diogo
duplicate of #40
Will also need a tile op to support einops
After #278 and #240 get merged, that should be all thats needed to create a backend for einops
Did some more digging into this, looks like pytorch only supports exporting to onnx even though importing/running was a requested feature. I don't have anything concrete on the decision from...
Also would the implementation be in the core C++ lib or in python, the onnx python package provides the backend helper to create the runtime. If C++ implementation is desired...
switched to the getenv function you added in the latest commit I used False to match the previous value types, but didnt see any cases where we cant just use...
I see, lmk if [this](https://github.com/dc-dc-dc/tinygrad/tree/lazy_cast) is more on track with what you envisioned
seems like you are trying to install metal on windows, given that the metal specific libraries are objective-c wrappers and target macos frameworks I doubt they would work on windows.
To run on Windows you can use the following backends - OpenCL - CUDA(if you have a NVDIA card) - CPU - LLVM