I have just published the source to all the board games I've built on https://t.co/xiigjhpUBK.
Thirty-one games.
Chess, checkers, go, backgammon, dominoes, and many other less well known games. Play free single-player mode on the site.
Link to the repos in the replies.
@OlympicDrankz GLM-5.3-Flash suggested it as the best small model to start with- smart enough and small enough.
Fine tuning on a partial dataset gave it the ability to build a small app in the codebase using the same conventions. Waiting to see how a bigger dataset does.
Just fine tuned Qwen 2.5-4B on my 3060 as a test.
Very good results.
Went well, now setting up a fine tuning run with qwen3.5-4b with part of my codebase and docs as a training corpus. If that goes well I may run the whole thing as fp16 on my Spark cluster.
Fine tuning using qwen3.5-4B as a base, a few thousand examples pulled from my codebase + docs as prompts. Should be done in ~6 hrs, will be testing its ability to build and debug apps in my codebase while running on an RTX 3060.
If it passes the test, I may set it to monitor the task list for my projects. Full automation of the code pipeline from a model running on 12 GB VRAM.
@IceSolst I'm in the middle of tuning qwen 3.5-4B on my code base , with the docs as prompts. I'm trying to figure out the smallest model I can use as a base to get to competence on the work I need done. Initial run looks pretty good.
@honestpollster She torpedoed the second term. I was saying in July last year that she needed to go. Miller should have been the guy in that spot all along.
@Stafu10@epperson_b42@VladTheInflator Incorrect.
We're the descendants of the people who had the balls to leave in search of a better place, and when they found it, they built a country from scratch out of the wilderness.
@VladTheInflator There is a way out of it, we all see what it ends up being, but the Left is so utterly retarded they will keep pushing until they get what they asked for in spades.