Collection of nitpicks

See below for a small collection of nitpicks; I suppose those will be adressed before the first release, but wanted to write them down somewhere:

• Documentation seems to be in a different format so will need to be refactored to work with the doc-builder
• The logging class still has mentions to 🤗 Transformers
• You have some tqdm instances in schedulers, but they’re not using the utility defined in logging in order to toggle bars on and off
• Some copyrights have wrong attribution (e.g., configuration_utils which mentions NVIDIA)
• In some places you define a global variable for configuratio name (e.g., SCHEDULER_CONFIG_NAME = "scheduler_config.json"), in other places you define them as a string (config_name = "model_index.json", in DiffusionPipeline)
• We have some conventions in transformers that IMO make sense for this library as well, wondering if it is an oversight or a decision to do differently. Those that I can see are: single letter variables, message-less asserts
• The register_modules doesn’t seem to play well with code completion tools
• The CLIP model is entirely copy/pasted from transformers, is the goal to upstream the change at some point or to keep it separate in that repo?

[vvvm23]

Hi yes, would be interested, feel free to tag me when that happens~ I don’t have much experience with continuous diffusion and contributing to big projects, but fairly good at pytorch in general.

[vvvm23]

Made a small pull request to fix the logging class issues. Not a huge contribution, just wanted to get my feet wet with contributing 🙂

It does change the environmental variables too, but I think this shouldn’t have any effects elsewhere - I checked for other occurrences of the variables using ripgrep