Corey Lowman

Yeah I'm actually working on this now! Will be able to support reducing 1 dimension at first. e.g.: ```rust let t: Tensor4D = ... let _: Tensor3D = t.sum_axis::; let...

> > * Tensor with 2-max-values in Max Last stores wrong gradients [Tensor with 2-max-values in Max Last stores wrong gradients #111](https://github.com/coreylowman/dfdx/issues/111) > > according to your last comment in...

Downside of this is that it becomes unclear what is in the npy files & how they are produced. Ofc it is already unclear how the arrays are produced, so...

I was thinking this could be an Tensor2D, where indices can be between 0 and M, and then the `usize` is just used to index into that tensor to grab...

It'd be similar to the `gather_last_dim()` function but I guess instead it would be gathering from the first dimension? ```rust let embeddings: Tensor2D = Tensor2D::zeros(); let batch_embeddings: Tensor2D = gather(embeddings,...

This should now be possible via the `.select()` method of `Select1` trait. Should be something close to: ```rust struct Embeddings { data: Tensor2D, } impl Module for Embeddings { type...

There should also be a `PaddedEmbedding` which is equivalent to using the pad_idx in pytorch

A somewhat clear implementation using im2col in darknet: https://github.com/pjreddie/darknet/blob/master/src/convolutional_layer.c#L445

Another here https://github.com/pytorch/pytorch/blob/master/aten/src/ATen/native/Im2Col.cpp

Since doing expressions with const generics isn't stable, the first thought of how to implement this won't work. tracking issue: https://github.com/rust-lang/rust/issues/76560 ```rust #[derive(Clone, Debug, Default)] pub struct Conv2D< const IN_CHANNELS:...