Belatedly, welcome to the year 2024. This is just a small holding-pattern post to make sure I don't lose touch with the reflection side of my programming process again.
- I want to do stuff with transformer models
- There's extensive and mainstream support for said models in Python, despite some movement on the JVM
- I'm going to want to put together some Android/otherwise mobile front-ends for what I've got in mind, and Python also has support for that
- So it would just complicate things to put some lisp-based server in the middle of this, just do all of it in Python
So far? No regrets. Given my current understanding of transformer models, I don't really see why Python is a necessary component on any level, but whatever, it's not awful and I'm not about to go back and re-invent this much infrastructure. Kivy is not my cup of tea UI-wise, and if the goal was just getting cross-platform, non-mobile GUI work done, I think I'd still reach for seesaw, but the difference between writing an API server with
tornado and writing one with http-kit is close enough to trivial that I don't particularly care.
Over the holidays, I've gotten a preliminary Android app running for my own minimal purposes, and I've been plugging away at the boring work of making
blogcast more and more automated. Generating the audio versions of these blog posts takes me on the order of a half hour of human interaction time once the text is online. I'm hoping to push it up to fully automated relatively soon, but that's likely going to involve taking a quality hit. The short term plan is to get a web UI running for it, and possibly a better
There's been a bit more progress in some specific libraries, but I'll blog about those in a separate piece shortly.