dfdx

dfdx copied to clipboard

Mean, Sum across arbitrary dimensions (especially batch dimension)

4

This might potentially depend on Reshape, but it would be good to compute mean/sum across arbitrary dimensions by letting mean/sum take `dim` or `axis` as an argument just like how...

vikigenius

coreylowman

0.9.0 release

2

Tracking issue for next release - [x] #34 - [x] #55 - [ ] #1 - [ ] #67 - [x] #123 - [ ] #115 - [x] #111 -...

coreylowman

Refactor unit tests with large arrays - move to .npy files in the repo

1

Unit tests should be easy to scan. large arrays make this almost impossible and add a lot of bloat. _Originally posted by @coreylowman in https://github.com/coreylowman/dfdx/pull/120#discussion_r929261072_

coreylowman

Closes #1 - [x] Add tests for conv2d & batch_conv2d - [x] Add tests for Conv2D layer - [ ] Add example file - [ ] Clean up implementation

coreylowman

This has been lightly discussed in some of the PRs as confusion has come up. Probably will want to link this in the README too, as I imagine this will...

coreylowman

coreylowman

coreylowman

Embedding Layers

5

In NLP, tokens are converted from tensors of `usize` to tensors of `f32`, where each `usize` is an index into a "library" of token embedding vectors. I'm pretty sure the...

jafioti

Convolution mega issue

15

coreylowman