While Replimat itself will provide some built-in grammars for Base and Flux types, I expect to be able to also add grammars from other packages. This will require figuring out how to merge grammars first.
Try taking common models and breaking them down into isolated sub-layers within each layer, or across multiple layers.
Implement a simple recurrent model which can make use ofrotationandaffine transformationlayers to iteratively preprocess the MNIST inputs into something more suitable for recognition.
REPORTED RESOLVED FIXED
Currently, if no accounts are specified in accounts.conf, aerc will crash with a runtime error. We could instead print an informative error indicating to add an account.
Excellent, will do.
I've gotten a simple patch in place for auto-creating these files (and the enclosing parent directories) if necessary, as well as binds.conf. I was wondering if just creating empty files is OK, or if it's preferred to copy-pasta the config/*.conf contents into the appropriate locations instead? Or I can just write a nice error telling the user to create them manually, whatever works.
To fulfill the requirements of the AGPL, we should add a link to each service's source repo in the page source as a comment (ideally at the top of the page where it's most obvious). This will use the links in the sr.ht config, which should be created if they don't exist for each service (since the code might be in different locations for each service?). A short blurb could also be added directly in the page source next to the source link indicating what the license is.