Hey something feels unnecessary
Hello @NuosResearch .. at first look on hf I though it would be uncensored version of model or finetuned but when I opened card i realised that it's just removing chinese political things .mmmm Question is Why do you need to use a model ? Don't use Qwen3.8 27B leave it.. use some american stuff like gemma?
BTW my home country is an enemy of china
Hello @NuosResearch .. at first look on hf I though it would be uncensored version of model or finetuned but when I opened card i realised that it's just removing chinese political things .mmmm Question is Why do you need to use a model ? Don't use Qwen3.8 27B leave it.. use some american stuff like gemma?
You'd do that when you like and want to keep 99.9% of the model, but there's just 0.1% that might give you some sort of ick. Gemma does not match Qwen technical capabilities in many areas.
Ok different opinions
Ok different opinions
It's not a matter of opinion. It's factual information that Gemma is better than Qwen at certain subjects, ie humanities, creative writing, languages, translations, while being much worse at many more technical subjects, like coding, math, etc, and that difference is not tiny. It needs only small fraction of resource to "de-ccp" Qwen, than what it would take to "qwentify" Gemma.
Ok different opinions
So now benchmark are opinion now ...
State sponsored troll ?
Ok different opinions
So now benchmark are opinion now ...
State sponsored troll ?
Different opinions on Whether this model is usefull or not
Ho i see we don't understand it the same way. your are debating utility when i think capabilities.
i agree with you, both are usefull.
But American are behind Chinese model on open weight models front in capabilities front.
(i was using gemma 31b before discovering DavidAu Qwen 3.6)
I host a qwne 3.8 next i would love to have other Moe model tu run and compare, i love these king of tamagoshi.
Europe isn't even in the course.
Same, India too
I am from India and china is our enemy
But In AI their is no country
Hey if you wanna run Qwen3.8 flash next go with gyro s or gyro m they are really accurate while being runnable
Ho i see we don't understand it the same way. your are debating utility when i think capabilities.
i agree with you, both are usefull.
But American are behind Chinese model on open weight models front in capabilities front.
(i was using gemma 31b before discovering DavidAu Qwen 3.6)
I host a qwne 3.8 next i would love to have other Moe model tu run and compare, i love these king of tamagoshi.
Europe isn't even in the course.
Hmmm Qwen4 flash is coming it may be big as Qwen3.8 Flash Next or otherwise if you have Mac style something try Mimo V2.6 flash
I am excited for Qwen 4 27B & 35B A3B
I use DavidAU/Qwen3.8-Flash-Next-Abliterated-Uncensored-Turbo-Brilliance-50x-Reasoning-Instruct-GSQ-RCO-GGUF.
I benchmarked there 27b vs the vanilla one, it is better, i didn't event try the vanilla 3.8 next XD
Same can't wait. (did you know you can increase qwen dev quality by pretending you are a chinese company ... i even made a false company card with a normal version in chinese and a an translated version in english, which let in my projects folder ...)
My dev pipeline went a little faster.
If you try the david one, know that the thinking reduction make it more "lazy" it have tendance to conclude without read facts
I will try a Gyro
Just try it you will experience opus 4.7 under 30GB
will do i need to wait for the pipeline to finish ... selfhosting is slower, i think it will work 1 to 3 days again.