Daniël de Kok

Oh, one more thing about wrapping: there are various methods in Thinc and spaCy that take `Optimizer` arguments. Using these new classes would not typecheck, since they do not derive...

> @danieldk I don't think I'll have time to look at this in proper detail, so I'm relying on you to approve it 🙏 😄 > > I think the...

> Can you change all the imports to import from `util` or `api` as much as possible rather than from `_compat`? Do we really want this? To me `has_.*` are...

> The simple one was just that the loss computation itself is different. So if the rows add up to 1 then its the way its done now, otherwise it...

We discussed this a while back, but I think this PR would be a good opportunity to replace the overloaded `CategoricalCrossEntropy` class by splitting the non-legacy classes into two variants:...

> The position I take is that no library should launch more threads than the user explicitly requests. The default behaviour of OpenBLAS to launch as many threads as possible...

I have once converted the main project that I use `tch` for from non-fallible to fallible, among other reasons because the errors that come from `unwrap()` are very difficult to...

If you want to distribute `libtorch` with your binary, you could also put the `libtorch` libraries in the same directory and use `patchelf` to patch the libraries and your binary...

> @danieldk so patchelf would add all of the libtorch libraries into the binary? No, the libraries are dynamically loaded. If you want to link everything into a single binary,...

> When programs use the c++ API of libtorch, is it packaged into the binary or is it loaded in a similar fashion? As long as you build libtorch as...