DavidAU

/

Qwen3.6-27B-Fable-Fusion-711-Uncensored-Heretic-NM-DAU-NEO-MAX-MTP-GGUF

like 2.2k

Testing the prototypes today.
There are a lot of additional steps required to get these to 700+; as there is a drop off (known issue) at the expansion step.
This issue must be addressed post expansion.

IT begins:

https://huggingface.co/DavidAU/Qwen3.6-40B-Grand-Intelligence-One

Composed of TWO 700+ ARC-C Monsters.
In testing.

Full example generation on the repo test page.
This is a take no prisoners version.

Sample:

"Listen up, you trembling little scribe with the sticky fingers and the big dreams. You’re standing there, staring at a blank page like a deer caught in the headlights of a freight train made of your own inadequacy, wondering who you’re going to beg to help you craft your masterpiece? You’re thinking about writing it yourself? You’re thinking about hiring some pretentious hack who charges by the word and cries when you ask him to cut his precious exposition? Or worse—hell, worse than worse—you’re thinking about using one of those soulless, sanitized, corporate-approved AI bots that write like they’re afraid to sneeze on a nun?"

...

Update:
Alpha "tune" part one complete; proceeding on testing other alphas / 2nd/3rd level tuning.

Due to issues caused by expansion, the model requires re-tuning to regain "711" type performance/benches.

Benches of expanded [non tuned] and [light / stage tune[s]] are here:

https://huggingface.co/DavidAU/Qwen3.6-40B-Grand-Intelligence-One-IQ-tune-alpha

NOTE: THIS IS A WORK IN PROGRESS ; it will take time to get it right.
NOTE: Name of the REPO for "alpha" will CHANGE as the project proceeds.

Update;
One alpha 40B has reached "700 club" status. Work continues.

this is super cool! i applied to access your gated model, have you released it in gguf form yet? regardless, i am curious as to how the performance is on my 128gb vram system i just finished building out!

@cpui686
Thank you ;
40B model build is in progress, however this build has more steps than 27B due in part to expansion and extended testing.
Benching 40Bs take twice a long too, per test.

As of this writing there will likely be at least 2 versions, maybe more - a general version plus specific/specialized use case(s).

aaah i see, yea that makes a lot of sense! keep up the good work!

Update;

Alpha testing finishing up ; Beta testing/tuning about to start.
Onward.

Hello,
Will it be a MOE?
Otherwise, something planned like "Qwen3.6-35B-A3B Fable-Fusion-711"?

I'm really hoping the 40B model is like, significantly better at coding than the 27B; do you think it will be? Just trying to get general vibes 😃

Thank you so much for all your hard work!

Amazing, you never fail to get people excited. Keep up the good work, but also take care of yourself :)

There’s some good news: the new 27B (3.8) version is reportedly set for release next week.
DavidAU, perhaps the new 40B model—which promises higher scores, faster speeds, and greater intelligence—will prove superior for your training; August is shaping up to be a truly wonderful month.

We are waiting for 3.8 too ; the pipeline is ready ;

40B version has entered BETA testing, in the "FF711" pipeline.

I would like to throw my observations -> the 27B model is really amazing at digesting a 350k text draft of a novel, distilling the magic system into a markdown, then finding reader confusion within the system. Truly remarkable in a model I never thought possible to run on my local machine. Thank you so much. I have one question: does the model do any token watermarking (not from you, but from Qwen)? I read that Google's SynthID can be marked on text output tokens this week, so now I'm just suspicious of all of these models that they can "fingerprint" your text!

@kylebamboo

Thank you for the detailed feedback ; excellent.

RE: Watermark;
I don't put any water marks per say in my fine tunes.
I don't know if Qwen does ;

Other issues: The AI app(s) you use ; I don't know any that do so , so far.

TWO 40B models now in limited release ; GGUFS uploading first (in progress - regular and MTP GGUFS; followed by full source release.
Both 40B models have ggufs available right now.

Fable Fusion Family

The Fable Fusion family consists of (in order):

Qwen3.5-9B-The-Defiant-Fable-Uncensored-Heretic
This is a project "test pilot" for building the Fable Fusion 27B/40B models.

GGUFS:
https://huggingface.co/DavidAU/Qwen3.5-9B-The-Defiant-Fable-Uncensored-Heretic-NEO-IMATRIX-MAX-MTP-GGUF
SOURCE:
https://huggingface.co/DavidAU/Qwen3.5-9B-The-Defiant-Fable-Uncensored-Heretic-NEO-IMATRIX-MAX-MTP

Qwen3.6-27B-Fable-Fusion-711-Uncensored-Heretic
1700+ likes, 2.3 million+ downloads, universal acclaim and 3rd party verications of performance.

GGUFS and many other quant types:
https://huggingface.co/DavidAU/Qwen3.6-27B-Fable-Fusion-711-Uncensored-Heretic-NM-DAU-NEO-MAX-MTP-GGUF
SOURCE:
https://huggingface.co/DavidAU/Qwen3.6-27B-Fable-Fusion-711-Uncensored-Heretic-NM-DAU-MTP

Qwen3.6-40B-Grand-Intelligence-Fable-Fusion-Uncensored-Heretic
Built from multiple versions of Qwen3.6-27B-Fable-Fusion-711-Uncensored-Heretic (and related 717), expanded and tuned.

GGUFS:
https://huggingface.co/DavidAU/Qwen3.6-40B-Grand-Intelligence-Fable-Fusion-Uncensored-Heretic-NM-DAU-NEO-MAX-MTP-GGUF
SOURCE:
https://huggingface.co/DavidAU/Qwen3.6-40B-Fable-Fusion-6-Core-Deckard-Eleanor-Heretic-Uncensored

Qwen3.6-40B-Fable-Fusion-6-Core-Deckard-Eleanor-Heretic-Uncensored
Built from multiple versions of Qwen3.6-27B-Fable-Fusion-711-Uncensored-Heretic (and related 717), expanded and tuned,
and Qwen3.6-40B-Grand-Intelligence-Fable-Fusion-Uncensored-Heretic then fused with THE DECKARD 40B.

GGUFS:
https://huggingface.co/DavidAU/Qwen3.6-40B-Fable-Fusion-6-Core-Deckard-Eleanor-Heretic-Uncensored-NM-DAU-NEO-MAX-MTP-GGUF
SOURCE:
https://huggingface.co/DavidAU/Qwen3.6-40B-Fable-Fusion-6-Core-Deckard-Eleanor-Heretic-Uncensored

Update: (repeating across all 3.8 threads || Primary "Update" thread.)

Currently we are testing Qwen 3.8 27B to get core metrics and primary "human testing" to get a sense of model's
character /stats / issue(s) as well as how it compares to base Qwen 3.6 27B, Qwen 3.6 27B Fable Fusion 711 and other lab versions of "711".

As of this writing we know the pipeline is compatible with this model, as well as the new GAIN training method (AKA "COLD FUSION")
which was invented during the build of Qwen 3.6 27B Fable Fusion 711.

Fable Fusion 711 used a primitive (but effective) version of GAIN during the build process. COLD FUSION takes this to the next level
with the ability to maintain BF16 metrics across 8 bit and 4 bit at 99% (of BF16/full precision) as of this writing.

The first test model (built 14 days ago) has been released here:
https://huggingface.co/DavidAU/Qwen3.5-9B-Cold-Fusion-GAIN-v1.0-Uncensored-Heretic-NEO-MAX-Imatrix-GGUF

With Qwen 3.6 Fable Fusion "Darker Hero" (built 3 days ago) also using COLD FUSION here (ggufs/source releasing shortly):
https://huggingface.co/DavidAU/Qwen3.6-27B-V1.1-FF711-Darker-Hero-GAIN-H2.0

NOTE:

• This version has been "re-heretic'ed", and trained to reduce thinking down to 1/10 to 1/2 "qwen size".
• 3 example generations.

We published both "reg training" and "GAIN training" benches (at both 9B and 27B repos above) so you can see the differences.
They are stark.

With Qwen 3.8 27B we will shortly be testing it in the pipeline and/or using COLD FUSION on it.

Conversely GAIN will also be used on current Qwen 3.6 models, and other arch/model types.

This is a lengthy, intense and detailed process so it may take some time before we publish.

Thank you ;

This August has been non-stop—from the 27B-711 to the new 40B, and now the new 3.8 27B. David, please make sure to get some rest; I’m finding myself sleep-deprived just from using your models! Yet, I’m also eagerly looking forward to your next update—it’s quite a dilemma. Keep up the great work!

@basiteis
Thank you so much ; trying to get (some) sleep ... a lot to do!

take care of yourself