Thank you for sharing, that was quite an interesting read. And more importantly, thank you for sharing your research openly, and for developing the custom scheduler implementation. I am looking forward to seeing how this is received over the coming months, and if it proves popular, there’s a good chance this could be made available as a built-in option in Anki in the future. I think there are Rust libraries that can be used for model training, so it may be possible to do the training directly on a device in the future, instead of having to upload the collection to a website.
@dae I thought I’d be able to port the main window to Svelte over the weekend, because I felt that’s the last big area that could really do with a design overhaul before 2.1.55, but the WIP is still too rough for a PR at this stage.
I’ll do my best to get the Tags done today to not block the Beta any further.