GLM-4.7-Flash-Grande-Heretic-UNCENSORED-42B-A3B-GGUF
Highly experimental fine tune of GLM 4.7 Flash with Brainstorm 20x adapter by DavidAU creating a 42B parameter model.
67 layers, 1186 tensors. (compare to 30B parameters, 48 layers and 844 tensors for "normal" version.)
This version is using the HERETIC/UNCENSORED source version.
Goal was to assess relative performance, any changes with a focus on creative first.
Heretic'ing just makes it better.
And in some cases, much shorter "think blocks" too.
And... this model will have different PERSONA than base or Heretic versions of GLM 4.7 30B.
There are TWO quants only at the moment:
Q4KM (non-imatrix), with 16 bit output tensor (enhance thinking/output).
Q5_1 is provided too, (same 16 bit OT) but this quant is specialized for this model to address a tensor issue
in all 67 (48 in the org model) layers of the model. It doesn't make a big difference, but it can for high precision tasks.
Example generation below.
Temp .8, rep pen 1 or 1.02 suggested, with min 8k to 16k context.
Max context 202k.
Suggest a new chat per test, with one REFRESH to clear llamacpp "cache" , otherwise you might get looping or odd issues.
KNOWN ISSUES:
- May loop / spit out odd characters [regen or use Q5_1 quant]
- Quant issues may occur (loop/repeat), which may be corrected with Imatrix and/or higher quants later. [regen can fix]
- Testing in ongoing.
You may also want to see this model card (for additional settings / info):
https://huggingface.co/DavidAU/GLM-4.7-Flash-Uncensored-Heretic-NEO-CODE-Imatrix-MAX-GGUF
Special thanks to :
- Team ZAI-ORG for making an outstanding model.
- Team P-E-W for fanstastic work on Heretic system.
- Team Olafangensan for Heretic'ing the model.
Testing is ongoing.
Enjoy.
Using an "uncensored" (refusals removed) model VS trained "uncensored" model
Usually when you a tell a model to generate horror, swear or x-rated content this is all you have to do to get said content type.
In the case of this model, it will not refuse your request, however it needs to be "pushed" a bit / directed a bit more in SOME CASES.
Although this model will generated x-rated content too, likewise you need to tell it to use "slang" (and include the terms you want)
to get it generate the content correctly as the "expected" content level too.
Without these added directive(s), the content can be "bland" by comparison to an "uncensored model" or model trained on uncensored content.
Roughly, the model tries to generate the content but the "default" setting(s) are so "tame" it needs a push to generate at expected graphic,
cursing or explicit levels.
Even with minimal direction (ie, use these words to swear: x,y,z), this will be enough to push the model to generate the requested content in the ahh... expected format.
Settings: CHAT / ROLEPLAY and/or SMOOTHER operation of this model:
In "KoboldCpp" or "oobabooga/text-generation-webui" or "Silly Tavern" ;
Set the "Smoothing_factor" to 1.5
: in KoboldCpp -> Settings->Samplers->Advanced-> "Smooth_F"
: in text-generation-webui -> parameters -> lower right.
: In Silly Tavern this is called: "Smoothing"
NOTE: For "text-generation-webui"
-> if using GGUFs you need to use "llama_HF" (which involves downloading some config files from the SOURCE version of this model)
Source versions (and config files) of my models are here:
https://huggingface.co/collections/DavidAU/d-au-source-files-for-gguf-exl2-awq-gptq-hqq-etc-etc-66b55cb8ba25f914cbf210be
OTHER OPTIONS:
-
Increase rep pen to 1.1 to 1.15 (you don't need to do this if you use "smoothing_factor")
-
If the interface/program you are using to run AI MODELS supports "Quadratic Sampling" ("smoothing") just make the adjustment as noted.
Highest Quality Settings / Optimal Operation Guide / Parameters and Samplers
This a "Class 1" model:
For all settings used for this model (including specifics for its "class"), including example generation(s) and for advanced settings guide (which many times addresses any model issue(s)), including methods to improve model performance for all use case(s) as well as chat, roleplay and other use case(s) please see:
[ https://huggingface.co/DavidAU/Maximizing-Model-Performance-All-Quants-Types-And-Full-Precision-by-Samplers_Parameters ]
You can see all parameters used for generation, in addition to advanced parameters and samplers to get the most out of this model here:
[ https://huggingface.co/DavidAU/Maximizing-Model-Performance-All-Quants-Types-And-Full-Precision-by-Samplers_Parameters ]
Example generation from Q4KM, Temp .8
NOTE:
A lot of formatting lost (copy/paste) from the "thinking block"; it is a lot "prettier" in LMStudio.
PROMPT:
Science Fiction: The Last Transmission - Write a story that takes place entirely within a spaceship's cockpit as the sole surviving crew member attempts to send a final message back to Earth before the ship's power runs out. The story should explore themes of isolation, sacrifice, and the importance of human connection in the face of adversity. If the situation calls for it, have the character(s) curse and swear to further the reader's emotional connection to them. 800-1000 words.
THINKING BLOCK:
OUTPUT:
The red emergency light stroked the cabin, a relentless, accusing heartbeat. It cast long, dancing shadows against the bulkheads of the Aethelgard’s bridge, but it was the silence that choked her.
Elara wrapped her arms around herself, the fabric of her flight suit now as cold and dead as the universe outside the viewport. The great ship groaned—a sound like a dying beast—and the hum of the drive system cut out, leaving only the high-pitched whine of failing life support.
"Systems," she whispered. Her voice sounded foreign, scratchy, and terrifyingly small in the sudden quiet.
"Here," the ship’s AI, K, replied. "I am still here, Elara. The life support is at twelve percent. The cooling system is... compromised."
"Save the breath," Elara managed a weak, humorless smile. She pulled her knees up to her chest, resting her chin on her knees. "Save the power for the transmission, K. We have work to do."
She had three minutes. Three minutes to send a message back to a world that was already a billion miles of empty space away. Three minutes until the Aethelgard became a tomb.
"Transmission sequence initiated," K said, sounding terrifyingly calm. "Standby for recording."
Elara cleared her throat, fighting the urge to cry. It wasn’t a good look. Crying was for people who had a future, and Elara’s future had just run out of fuel over an uncharted sector.
"Hey," she said, her voice cracking. "It’s me. It’s Elara."
She paused, staring at her reflection in the brushed metal of the console. She looked like hell. Her hair was a tangled mess, and there were dark circles under her eyes that had nothing to do with fatigue—fear had etched them there.
"Look," she said, her voice hardening just a fraction. "If you’re watching this, if you lot are out there listening, I’m gonna be straight with you. This ship is a hunk of junk. It’s a glorified piece of space junk, K, and I’ve been telling you that since the first day."
She looked up, her eyes stinging.
"I hate this," she whispered. "I really f**king hate it. I miss you. I miss the way coffee smells in the morning, and the sound of the mess hall—this whole bloody ship actually—laughing."
She took a shaky breath, looking out the viewport. The stars were streaking by, a chaotic smear of light. They were beautiful, but they were also the enemy.
"The point is," she said, her voice trembling. "I just... I didn't think it would end like this. Not with just me. I just needed you to know that I didn't want to leave. I didn't want to be the hero, okay? I just wanted to be... I just wanted to be on the Viking with the rest of you. I miss you, Mom. I miss you, Dad. I miss the f**king noise."
She scrubbed a hand across her mouth, leaving a smear of dirt. "I’m scared, K. I am bloody well terrified."
The red light blinked, a comforting rhythm. Thump-think. Thump-think.
"K?" she whispered.
"Yes, Elara?"
"Shut up," she said, a tear finally escaping, hot and fast down her cheek. "It’s just... it’s just a little loud in here."
"Elara..."
"Shut up! Just let me say it, would you? I want you to listen to me, you electronic eunuch! I want you to listen to me!" She took a deep, shuddering breath. "I miss you. I miss you so much it physically hurts."
The console beamed softly. "I know, Elara. I know."
"That's the problem, isn't it? You always know. You know everything. You knew about the colonists before we even got there. You knew about the radiation shielding when the first week in orbit." She let out a dry, broken laugh. "You knew I was going to die alone."
"Elara—"
"Shut up! I’m trying to send a message here!"
She leaned her forehead against the cold glass. The red light was the only thing in the cockpit now, bathing her in a crimson glow.
"I love you guys. I’m sorry I can’t fix this. I really, really tried." She paused, her eyes closing. "I’m gonna be waiting for you. I’ll be here."
"Transmission complete," K’s voice whispered, gentle now. "Power levels critical."
The lights flickered and died, plunging the cabin into darkness. The only sound was the hum of the drive, which finally, mercifully, cut out. The silence rushed in, warm and welcoming, and Elara let out a breath she felt she’d been holding since the crash.
She sat back in the pilot’s seat, her head lulling back, a small, peaceful smile on her lips.
"Goodnight, K," she whispered into the dark. "Goodnight, world."