Paulo Valente
Paulo Valente
@emilsoman this solution is good, but I think we should convert the atom to string only after "instantiating" the atom, so we fill the Atom Table properly. Otherwise, it might...
This seems like an issue with the upstream `protobuf` library. Although perhaps @sleipnir or @wingyplus can help answer the question
> Extensions cannot be provided at runtime (at least to my knowledge) only those defined in elixir-protobuf already. This seems like an upstream issue that could be tackled in `:protobuf`....
I think we're gonna need @tony612's input here. The fork is based on a really outdated version of `grpc`, and what fails (at least for me) is some part of...
Thanks for the feedback! This is an issue with the precision of the float itself, not a problem with Nx. https://www.h-schmidt.net/FloatConverter/IEEE754.html Here you can check that the same representation for...
https://github.com/tensorflow/tensorflow/pull/45475 https://github.com/tensorflow/tensorflow/issues/6504 This issue and Pull Request are related to QR grad. Might be relevant for who tackles this later. In the PR comments there's a paper related to backpropagation...
I'm starting work on triangular solve gradient
So in this construct subtensor would always be an iteration over the first dimension of the tensor?
Does this also handle `*_rest`-style globs in the end of the template?
> Good idea! There's just a merge conflict, but after that's fixed I think this is a good addition. @devonestes conflicts resolved! I guess most of them were due to...