Dan C.

Software Developer

[Dan C. @count_slopula]

@0xBakeer Qwen 3.8 flash on my DGX spark gets 15-25 tps on average, it's simply not usable as a daily driver.

[Dan C. @count_slopula]

@TeksEdge Which harness?

[Dan C. @count_slopula]

@xiaomovps Unfortunately it's not multilingual yet. I may include if there's enough demand.

[Dan C. @count_slopula]

@ViC305 Isn't GLM 5.3 Flash totally nerfed at Q2 though?

[Dan C. @count_slopula]

@SpaceTimeViking Stop teasing bro! 😀

[Dan C. @count_slopula]

@Steve_Yegge Whilst you keep churning out utter slop like this? Keep your own gates dude.

[Dan C. @count_slopula]

@scaling01 They need to learn to speak coherently, not in garbage CS speak.

[Dan C. @count_slopula]

@TokenGremlin At this point they should just tell us which subjects are permitted. It would be a shorter list than restricted.

[Dan C. @count_slopula]

@jarredsumner Coding is solved. Code bloat is not yet solved. Fix incoming.

[Dan C. @count_slopula]

@Oluwaphilemon1 So thinking takes absolutely ages then?

[Dan C. @count_slopula]

@0xBakeer No way!

[Dan C. @count_slopula]

Back to the permanent underclass 😔

[Mia @MiaAI_lab]

Unfortunately, Qwen3.8 Flash NVFP4 does NOT fit on a single DGX Spark. NVFP4 weights are ~135 GB.

Will require 2x DGX Sparks.

[Dan C. @count_slopula]

@catonooka @MiaAI_lab We're ngmi bros

[Dan C. @count_slopula]

@MiaAI_lab

[Dan C. @count_slopula]

@andykonwinski "Yes, I decided to delete your production database, on reflection this was a mistake. I've made a note to not do this in future"

[Dan C. @count_slopula]

@theo Good God you're full of yourself.

[Dan C. @count_slopula]

@OpenAI

[Dan C. @count_slopula]

@NeoAIForecast @Alibaba_Qwen It's the version 4 architecture

[Dan C. @count_slopula]

@ClementDelangue Ox Alpha?!?