What model ?
Hey thx for your work I follow u for a while now and your model are always good.
Just a little question what model do you recommend using for agentic coding on a 24gb macbook m5, what is the maximum quality (how many parameters) can I use in an agentic coding workflow with local LLM ?
I saw I can run this one in quant 2 bit but I think it will not be good and I tried 3.8 27B and it was not great using nearly all my ram and I guess (but I am not sure) 3.5 9B is not the best coding model to use.
thx in advance for your response !
HEy;
This is the one you need:
https://huggingface.co/DavidAU/Qwen3.6-27B-Fable-Fusion-711-Uncensored-Heretic-NM-DAU-NEO-MAX-MTP-GGUF
It performs above 3.8, and all other 3.6s (see testing/community section).
Okay thx what quant go u recommend ? Bc I tested it but didn't manage to have a good working workflow using it bc it used all my ram (24gb is not that much), is a 3 bit quant good enough compared to a 8bit 9B for ex ? or a 6/4 bit of another model that is 21B for ex ?
(thx for your answer your are f*cking fast)
Hey:
For tools/agentic min quant is Q4KM ; better is Q5s, and recommended is Q6/Q8 (this is general, all models/all quants suggestions).
For lower quants do not use caching, as this may cause too much degradation.
In a pinch you MIGHT get away IQ4XS LOW // Q6 LOW.
Decision "making" may degrade too much at IQ3_M.
NOTE:
V2 of Turbo will be out in a few days, with improvements.