Eric Christiansen

BTW if you give me a pointer on how to do this I'd be happy to give it a go.

> Yeah split can be used for this. You'd have to clone the tensor since as you state it takes ownership of input. You mean `select`, right? I'd be happy...

I took a stab at `split_0` here, which I'm calling `unstack` for symmetry: https://github.com/coreylowman/dfdx/pull/831

Agreed. I'd actually go a step further and say you could dramatically simplify your architecture, while improving runtime speed, by using XLA as the only backend. (You probably know that...

I've never used IREE but that also seems like a good option. AOT compilation would be great, too. Perhaps the critique should be "don't manually support various backends, instead choose...

Hey, is this something you're interested in? FYI I've been using this fn and it appears to work.

FWIW you can get a pretty good string by using `.array()`, e.g. `println("{#?}", my_tensor.array())`. The array will have the same structure as the tensor (it's not a 1D array). Also,...

FYI I hacked something together using `Arc` 🤯 The basic idea is to provide another layer of indirection so we can mutate the pointer to the actual `OwnedTape`. Here's the...

As a workaround, assuming I'm doing everything on a single device (say the CPU for now), could I just define something like this and use it for my device everywhere,...

Hey, just to be clear, you mean something like [RustaCUDA's LockedBuffer](https://github.com/bheisler/RustaCUDA/blob/c74efcb14fa24d53ac5ca1ce57e6f5794b62ce25/src/memory/locked.rs#L14), right? FYI I need this to switch from RustaCUDA to cudarc.