ptaken
Disclaimer: I am a relatively active contributor
As for the number of dimensions, in a sense they are a training variable just as the content itself. The more dimensions you utilize for your embeddings the more complex your relations can be during clustering. Too many dimensions can easily lead to over fitting however and too little dimensions can usually not accurately represent the training corpus.
The issue with wireless keyboards and QMK is that Bluetooth is such a messed up protocol and to implement it QMK would realistically have to deal with proprietary binary blobs which they choose not to. I heard there's some work on "custom" 2.4GHz transmission modes but that's not as universal or ubiquitous as Bluetooth is.
Maybe check out ZMK, basically a QMK for wireless keyboards but more opiniated. I'm about to build a wireless corne with it after years on a wired QMK Planck.
Battery life for closed source keyboards made by big companies will always be better because they have the resources to perfectly engineer the hardware and code for every model and the willingness to cut lots of corners in terms of customizability. Same with any other piece of hardware.
I do agree though that pandas can be a pain sometimes...
[1] https://beta.docs.qmk.fm/using-qmk/software-features/feature...