Compilethings

AI, Datasets, Fine tuning small llm's, ML student, Glassblower

[Compilethings @Compilethings]

@mathfax I sent you a DM with a full run down so far. I will update you on my findings as we test other areas in our workflow. I see huge potential. I build synthetic domain specific datasets.

[Compilethings @Compilethings]

@MiaAI_lab I get 10-12000$ in API usage for 200$ I would say the usage is great.

[Compilethings @Compilethings]

@virtualcity69 @jun_song DGX spark is over 6k in Norway.

[Compilethings @Compilethings]

@virtualcity69 @jun_song Sure, but OEM's are sold out.

[Compilethings @Compilethings]

@virtualcity69 @jun_song I agree, but DGX spark is 7000 now.

[Compilethings @Compilethings]

@itsPaulAi I'm building the dataset now for my workflow as I work. GLiNER2.5 looks interesting for a finetune.

[Compilethings @Compilethings]

@AMD We need this in a prosumer setup. The market is there. They are to hard to get our hands on. Put a blower on it and let consumers run with it.

[AMD @AMD]

Bringing AI into the enterprise doesn’t have to mean starting from scratch.

AMD Instinct MI350P is designed to bring LLM-scale AI into existing environments with an air-cooled form factor that helps organizations deploy without rethinking their entire infrastructure.

[Compilethings @Compilethings]

@sudoingX 1 up 1 down fiber. Fastest we can get in residential.

[Compilethings @Compilethings]

If you are complaining about token usage and not using something like @_codeburn , then you are not even trying to understand why, understand before you complain.

[Compilethings @Compilethings]

Do you keep your frontier agents doing frontier work?

Do you waste frontier review on things local tools should catch?

If so you should be adding tooling like, Ruff, Hypothesis, schemas, Pyright, Semgrep and coverage behind one fail closed quality gate.

Keep frontier doing frontier work!

[Compilethings @Compilethings]

@DavidKPiano I have started working it in slowly.

[Compilethings @Compilethings]

@DavidKPiano codify it all, each "skill" a codified subagent.

[Compilethings @Compilethings]

@TTrimoreau I go touch grass, breaks are needed.

[Compilethings @Compilethings]

@Im_IrushiK Don't chase hype, chase stability.

[Compilethings @Compilethings]

@Im_IrushiK No problems here, everything is as smooth as butter

[Compilethings @Compilethings]

@dev_maims yes, I use them together all day everyday for months.

[Compilethings @Compilethings]

@stefanopineda Nice! I build datasets and fine tune models to test the datasets: here is some of my work that I published.
https://t.co/VeQC4hFBJk