Paulo Valente
Paulo Valente
Although #1424 did move eigh to defn, it's still worth looking into using a new implementation for speed and accuracy
@christianjgreen in short, you can just compare the execution time of `jax.linalg.eigh` vs `Nx.LinAlg.eigh` using a 200x200 f32 tensor using `EXLA` as the default compiler and backend. You'll notice that...
For what it's worth, IREE might be able to provide a way out. We're focusing on Metal support, but we just might get ROCm "for free"
There are many kinds of vocoders. I think the best way to approach this would be to choose a specific model we want to support and work towards porting the...
I believe append_layers would still be accepted in VegaLite, even though they're deprecating VegaLite.Data (@josevalim, can you confirm?) Having that available, how would we go about bringing that code in?...
@noplay I'm gonna need some help with the language definitions. I think that maybe the `inlineOpenToken` category is not needed anymore, but I reckon you know more of the languages...
One improvement that needs addressing before merging is that the nesting for comprehensions is not implemented properly (only counting instead of using stacks).
> Any update? :) I've been kinda busy (and sick) these last weeks, but I intend to get back to this as soon as possible!
Sorry it took so long, guys. I got sick, and then was busy with other things. Code is now ready for review (and possible merging).