Qwopus 3.6

Posted by Perfect-Flounder7856@reddit | LocalLLaMA | View on Reddit | 19 comments

Has anyone tried it yet? What's it good at?

[-]

Thin_Pollution8843@reddit

I prefer Qwender3.6-27.5B-REBEL-UNHINGED-MegaCoder-UltraHD-ANARCHYST-ANTROPICKILLER-ALTMANSTEARS-YOURMAMAAREFAT-NoElectricityNeeded-BETTERTHENGEMMA-GGUF

[-]

Abject-Kitchen3198@reddit

I also noticed that removing hinges between layers 13 and 14 does wonders, bit I have to keep temp between 0.834 and 0.836 or it falls flat.

[-]

jmprog@reddit

Reading this like its the Star Wars intro scrolling text

[-]

dinerburgeryum@reddit

For real, actual work the stock models are almost always better. I've seen these fine tunes degrade hard after 32K context, which isn't much in the context of agent code work.

[-]

Pleasant-Shallot-707@reddit

Just say no to these distillations. They’re almost always poorly done

[-]

Perfect-Flounder7856@reddit (OP)

That's what I figured just getting a lot of traction on x

[-]

PM_ME_UR_COFFEE_CUPS@reddit

Why? I ran it on one loop to fix a build and it did fine.

[-]

b1231227@reddit

I've used it before. It's pretty good.
But in the end, I went with DavidAU/Qwen3.6-27B-Heretic-Uncensored-FINETUNE-NEO-CODE-Di-IMatrix-MAX-GGUF

[-]

GhettoClapper@reddit

Thought those were trained for creative writing

[-]

mrgreatheart@reddit

May I ask, why uncensored? Doesn’t the abliteration process degrade the model?

[-]

serige@reddit

It loops like crazy for a quick pacman test found here in another post. I say nah.

[-]

MundanePercentage674@reddit

Work great for chat bot every fine-tune I tried fail at tool calling or more dumpthan original model.

[-]

I'm using it daily and I have very good results with it. I like that it is less chatty than the original. I run it on a 7900XTX with a Q4 quant and also 128k Q4 cache, and regularly go up to 90-100k context, and it works without failing.

[-]

sub-merge@reddit

Absolutely awful, failed spectacularly on tool calling in every harness I tried

[-]

FinalCap2680@reddit

There is a video here for 27B:

https://www.youtube.com/@tokenchaser/videos

[-]

gyzerok@reddit

Wanna try and tell us?