Damn, was hoping this would be an upgrade over (the seemingly abandoned) UVR5

The stem models were getting better and better, but stopped dead right around when GPTs started to work.

There is a robust community around it but it is not very user friendly; most info is found in discord and a Google spreadsheet and custom patches/builds of uvr

My only interest is if there are any newer, cleaner algorithms/models. htdemucs (as well as everything else in UVR) are good but not great and leave a lot of artifacts. The acapellas it stamps are 'good enough' til you get to final mixdown and then the flaws stand out real heavy.

I was running one of the older llamas (3.1 I think?) at slow-ish (10-20 tok/sec at Q4?) but OK speeds on 12 year old DDR3 ECC Xeon machine

I find 10 to be very usable. It’s not (that) interactive but it chews through tasks. I let Kimi churn away at 4 overnight and it gives good results that are ready for me in the morning.

Cool! I played the original with a joystick back in the day. I tried the previous port to XWA with Tie Fighter more recently, and it had been so long that I couldnt even complete the tutorial with the joystick! Proper gamepad support would be clutch

Thrustmaster FCS & WCS MK1 for me. And as much as I'd love to play it again, it will never be the same.

You can incentivize cooperation without having to compel action with a warrant. This sort of corrupt quid pro quo is quite common, in other countries at least...

I am working this out with Fable right now, for getting this running on my DGX spark homelab; it mentioned that there might be issues with the 'optimized LM-head restrictions' that unsloth NVFP4 ships with. Have you had any issues here?

Are you trying this with vLLM? Or a different engine? I am getting about 15 tok/s on my spark on my current setup using the 0.26 nvidia vLLM image and MTP.

If you are using the Unsloth nvfp4 checkpoint, you need to patch vLLM+DFlash 2 to accept the quant's FP8 `lm_head`.

Dr Bronners is local to me and they are very active in the burner scene (I believe the founder is an OG burner). I have a few friends who work there and Bronners often has a presence at local burner events (and Burning Man) -- notable because these are environments where commercial expression is generally forbidden. Yet they sometimes get a pass, I think because they embody the burner spirit and often supply a needed service (showers) totally free and in a way that fits in with the culture. It is the most guerilla marketing you ever saw and I would not be suprised if that wasn't their intent -- rather its just the company gifting itself and embodying the spirit of the ten principles.

They remind me a lot of Valve in that they are able to chart their own path and find success.

Bronners certainly isn't without its problems as a company. But I find it to be very inspirational that companies can find success without having to devolve into the bland milquetoast mediocrity that feels like 95% of US business culture. We need more people being visibly, authentically, and unashamedly weird.

Oh man I cannot tell you how many times I've gotten into arguments on here and other technical forums years ago about self-hosting and how 'stupid' I was to not want to rent cloud servers

The actual infection rates for some STIs are also similarly low, per-encounter. But they are high enough that if you are promiscuous and have a large number of at-risk encounters then you will eventually get a bad roll of the dice.

XP? What with its Fischer-Price primary color scheme? You have got to be kidding me.

Finally, a throwback lib that seems to be mostly pixel-accurate! Some of the element spacing seems to be off but it does actually look like Windows 98.

The lack of accuracy and attention to detail in a lot of these efforts is incredibly irksome and annoying, just about as obnoxious as pixel-art games with inconsistent pixel sizes (like how hard is it to render to a smaller buffer and do a simple linear upscale?!?)