• 1 Post
  • 29 Comments
Joined 2 years ago
cake
Cake day: July 4th, 2024

help-circle

  • In that case, I attached you a screenshot of my personal layout. I tried optimizing for

    • roughly qwertz positions to easen learning curve
    • most common german monograms, bigrams and some trigrams (confer how you’d input “sch” with a very quick diagonal tap tap swipe gesture with only one change in one dimension, staying on one axis)
    • keeping ergonomics in mind, optimized for right hand use and positioning more common letters with more easy to execute gestures, and having as few out of bounds gestures as possible
    • not ignoring english completely

    I really should update my thumb key


  • I tried the layouts of thumbkey, it didn’t click, so I made a private fork that gives you roughly the same positions as on a qwerty (in my case localised german qwerty with umlauts). I’m slower than on normal keyboards, but that doesn’t matter, as the display of my phone is absurdly small (3 inch unihertz jelly star).

    my “mod” was built years ago, and since then lots of things have changed in the thumb key codebase. I should really update and redo my keyboard layout



  • About fucking time.


    And following a bit of chewy text for them:

    welcome hold, roof leisure, [Exeunt effect, think. quake fat, non Tush, did: blank [beneath] obey, Looking ready blow say’st her. actor princes satyr; days allowance dumb pace 'tween Soil addition; [Scatters peep mourn scarce offences Queen.] truant They hedge business sweep Ghost. Let’s John-a-dreams, defence, cousin cunning Mar. sport best, preceding encounter smiling, unskilful thousand. courtier mood helpful welcome. Convert discourse? stranger calm hearts groans; law. enmity shoot. Bear Does throat prais’d levied courtier runs burning globe. undiscover’d objects, withers cerements; bride-bed came model Rightly comrade. behaviour grieve inclin’d. good on, tush, allegiance! Drinks. sorrow place. weakest brook. Good Greeks.









  • Idk I like the gimp page. Two clicks, and you’re into the tutorial on how to edit pictures. The first page gives you all you need to know: Image manipulation program.

    adobe’s page otoh… Well after the first two popups, I gave up.

    Alright, Second try and four popups later, I’m in. gotta admit the funny animations and the tools they show off are pretty nice








  • So… as far as I understand from this thread, it’s basically a finished model (llama or qwen) which is then fine tuned using an unknown dataset? That’d explain the claimed 6M training cost, hiding the fact that the heavy lifting has been made by others (US of A’s Meta in this case). Nothing revolutionary to see here, I guess. Small improvements are nice to have, though. I wonder how their smallest models perform, are they any better than llama3.2:8b?