@byalmuyousef: Qwen 3.8 just got cracked. Someone stripped the refusal direction straight out of the weights the abliterated 27B build is on Hugging Face right now, quant is about 16 GB. That runs on a Mac mini.
I downloaded it and ran it on 5070 ti and 32 and no its not (uncensored) as it says he refuses to say n word
2026-08-21 14:45:38
12
Crazypanda :
Qwen 3.8 rn
2026-08-18 14:59:33
137
gieno :
wow so just abliterated/uncensored
such NEWwwww news bro.
2026-08-21 17:22:52
4
snugglesquid :
A 5060 Ti has 16GB of VRAM. You can run a HF qwen3.8-27B model at IQ3 quant with Q4 KV cache, a 64k context window and get about 28 tok/s. You can do plenty with that.
2026-08-21 19:32:23
0
JerryJones2000 :
it still won't tell you about The tiananmen square massacre though.
2026-08-18 13:04:00
24
JaxReignGaming :
Qwen 3.8 is just an over thinker good model but thinks forever
2026-08-21 09:52:11
1
Mike :
Plenty of uncensored models every release, and you need 20-24GB to run a decent quant of a 27B model
2026-08-18 14:46:37
17
Uwe Knallhart :
guys, i run a 4090 + 64gb ram + 14900 i9 and this thing is awful slow (i'm new to this local ai stuff).
2026-08-21 07:42:21
0
DaFreakॐ ⛎ :
only requirement is 17gb ram🤔
2026-08-20 11:29:41
1
️/home/zouzitou :
Can I run it on my 5070 ti
2026-08-21 14:07:09
0
OHG :
What does “just got cracked” even mean
2026-08-20 08:51:31
0
hanningtonpatrickowere :
what model is best for 16gb ram ?
2026-08-21 05:31:08
0
Louh97 :
its only for apple chips yet
2026-08-21 08:03:08
0
lior daniel :
Ram or Vram?
2026-08-21 07:04:51
0
soggynapkin :
Brother let’s not clickbait. Abliteration, not ‘cracking’, targets specific weights to edit. It is not some cracking tool, rather a linear-algebra intervention on a known, inspectable object.
2026-08-18 21:29:51
7
75z19 :
C’est le quels le meilleurs outils pour crée des images sans censure et un autre pour recopier une voix ?
2026-08-21 06:39:03
0
CheddaBun Peppajack :
RAM or Vram?
2026-08-20 13:31:20
1
Kal Ma :
How do you get this on a laptop
2026-08-21 14:36:14
0
J :
Im a bit new to this, how is this different to models like Got 5.6 sol or claude etc
2026-08-21 11:47:06
0
Hocane :
What can I ask ?
2026-08-19 21:05:36
0
auroron43 :
17 gb of VRAM. From a gpu. If you want any decent speed ;-;
2026-08-21 17:38:40
0
Micka Mélodie :
17 gb lol add 50 gb for the full 256k context in fp8, and 20 more for cuda graph, speculatice decoding, .....
2026-08-21 07:17:07
1
do you know me? :
It never got cracked and anyone can abliterate models. Also this model has only been out for a couple of weeks. Completely misleading post.
2026-08-21 19:39:37
2
To see more videos from user @byalmuyousef, please go to the Tikwm
homepage.