W1/6 D5/5

Created 2024-03-29 / Edited 2024-03-29

Llamas, Pythons, Coffee, and Sway

Part I - Morning

  • Morning Status:
    • I feel pretty great today, but going to be Virtual anyway. Got some errands to run
    • I've got llama-cpp-python up and running, so I'm going to work on a back-end for it
    • I'll likely solicit pairing once I get it a bit further, especially for some visuals

Part II - Evening

  • General Status:
    • TIL: Drop this in your python code to get a pretty close binding.pry equivalent from IPython import embed # For debugging; put `embed()` anywhere
    • Coffee Chat with Paolo!
    • Volitional Muscles Workshop helped me re-affirm that I like the plan that I pre-gamed
      • Core: Alloy+LLM+Rails (using Human Essentials app as a dataset)
        • Explore possibilities, mash together things, dev-tooling, do some Good
      • Cake: Pair, Follow Whims, Dance With Serendipity
        • Create artifacts, play, learn, share
    • Side Note: I'm typing this in vim with copilot and it is upsetting when it suggests stiff language and then it is upsetting when it suggests exactly what I was about to write because then I'm like ... is this as bad as what you just suggested dear robot?
    • Attended applied machine learning group, but really just watched
    • .... then got a Work Call that I hopped on to help with for the end of the day
  • Alloy+LLM+Rails Project:
    • Slight progress on understanding the llama-cpp-python setup
    • Did some brushing up on python iterators and debugging tools


  • Weekend Updates:
    • Nerd Sniped on displaying the current desktop grid pop-up to the right monitor. Got it working :)
      • For reference, screen=$(swaymsg -t get_outputs | jq 'map(.focused) | index(true)') which I then pass on to aosd_cat
    • I dreamed about generating tokens from an LLM, but I think they were physical tokens and also maybe gangsters
    • I don't really need it now, but I keep wondering a cost-effective but natural-for-me way to rent GPU time to make my model run faster. I can spin up a whole linode GPU vm, but will need to make sure to delete it in-between uses. I don't really want to do hosted notebook blah blah. Not sure what I want. I guess I wish I could run a command and have a GPU on my existing linode and then run another to remove it, but that's maybe not how it works
    • We went on a some shopping and walking around and out to Thai and got groceries. Making good progress on getting the sense-of-place around here