dfdx

dfdx copied to clipboard

Add complex numbers

3

Closes #821. I had to use a proxy object to implement cuda traits. It built with `cargo build -F complex,cuda,nightly,f16`

opfromthestart

Consider providing interim instructions for Linux "happy path" using Docker

2

Hello, I spent part of this afternoon banging my head against a wall with getting `dfdx` with the `cuda` feature enabled up and running on my computer. It turns a...

quietlychris

Add support for complex tensors

1

Pytorch has support for complex tensors and can also do Autograd for them. https://pytorch.org/docs/stable/complex_numbers.html https://pytorch.org/docs/stable/notes/autograd.html#autograd-for-complex-numbers We can gate this behind a feature flag if we want. But it would be...

vikigenius

Operator fusion

16

Let's discuss how operator fusion might work in dfdx. I suspect it will require a lot of work. On cuda side of things it will at least require jit compiling...

coreylowman

The merge semantics for `Arc<Mutex<OwnedTape<_, _>>>` seem a bit unintuitive

2

The merge semantics for `Arc` seem a bit unintuitive. In particular, one would hope that when these tapes are merged, they would essentially be replaced with one tape which is...

emchristiansen

Bug: Incompatible uniqueness assumption causes crashes

1

Code to reproduce for `OwnedTape`: ```rust let x = dev.tensor(1.0); let y = x .clone() .put_tape(OwnedTape::::default()) + x.with_empty_tape(); // Crash is here. y.backward(); ``` Code to reproduce for `Arc`: ```rust...

emchristiansen

Using no_std on thumbv6m-none-eabi

16

I'm trying to use this awesome library for an embedded project (just inference) but I'm having problems compiling it. ```bash error[E0432]: unresolved import `__alloc::sync` --> ~/.cargo/registry/src/index.crates.io-6f17d22bba15001f/no-std-compat-0.4.1/src/generated.rs:173:48 | 173 | #[cfg(feature...

spadarian

Tensor `Display` Implementation?

4

Hey there! I'm a fairly new to machine learning, but I'm wanting to use `dydx` instead of PyTorch as I go through pieces of the d2l course. One thing I...

zicklag

Add Gradient scaler

1

With the addition of `AMP` dtype, we also need to add gradient scaling, which is commonly used with AMP training. I think the frontend interface could look something like: ```rust...

coreylowman

impl Default for Tensor?

3

_Thanks for working on this! I once had a project where we regularly worked with 6 dimensional tensors and it was such a pain to keep track of the axes...

emchristiansen