ProGamerGov

Optim-wip: Add the CLIP ResNet 50x4 model

1

* Added the CLIP ResNet 50x4 model used in the 'Multimodal Neurons in Artificial Neural Networks' research paper: https://distill.pub/2021/multimodal-neurons/ * To make it easier to conduct research, the CLIP ResNet...

* Wrap all remaining `torch.__version__` calls in `version.parse`. * Remove unused version check in `typing.py`. * Expose `MaxPool2dRelaxed` to users so that tutorials using it work. * Expose `dataset` module...

This PR does the following: * Ensures loss testing coverage is as high as possible. * Simplified loss composition code with new `rmodule_op` function. * Removed the NumPy import from...

* Improved the `LaplacianImage` code as part of removing the ToDo comment. * Added `"same"` padding option to `GaussianSmoothing`. * Improved the `NaturalImage` docs. * Also removed the built-in clamp...

I discovered what appears to be a forgotten class, called `captum.attr.InputBaselineXGradient`

2

## 🐛 Bug I found a module that was listed in the code and on the API docs site, but there's not public path to use it and no tests...

I helped the TorchText team with developing their CLIP tokenizer to ensure that I could use it like this for Captum.

This tutorial covers how to create the facet weights used in the `FacetLoss` objective.

Not sure why the line count is so high in this PR as it's rather short and simple.