L3 Stheno Maid Blackroot Grand HORROR 16B GGUF

🚀 LLama-3.1-128k-Uncensored-Stheno-Maid-Blackroot-Grand-HORROR-16.5B-GGUF

A powerful text-generation model for various fiction writing and roleplaying scenarios, with a strong horror bias and uncensored output.

📄 License

The model is licensed under the Apache-2.0 license.

✨ Features

• Newest Version V3: It harnesses the full power of Stheno - Maid - Blackroot - Grand - Horror 16.5B, now with a 128k context. It also comes with additional de - censoring, performance improvements, and re - mastered source and ggufs in float 32 (32 - bit precision). Check it out
• Newest Version V3 - 1M: Offers a one - million context. It retains all the power of Grand Horror, with better long - generation capabilities and long - context awareness. Link
• Rebuilt and Powered Up (V1 and V2): The L3 - Stheno - Maid - Blackroot - Grand - Horror - 16.5B has been rebuilt. V2 features a new formula, new V2 quants (taking advantage of new llamacpp updates), ARM quants, more guides, and V1 vs V2 examples.
• Uncensored and Powerful: A take - no - prisoners, totally uncensored fiction writing monster and roleplay master. It can handle various general fiction activities across all genres, including scene generation, story generation, and roleplaying.
• Enhanced Prose and Story Generation: Thanks to the expansion/merge re - build, its level of prose, story generation, word choice, sentence structure, and default output levels and lengths have significantly improved.
• Strong Horror Bias: While it can generate content for almost any genre, it has a strong horror bias. It can also produce dark humor and contains NSFW elements.
• Detail - Oriented: Excels in details (real and "constructed"), descriptions, similes, and metaphors. It has an uncanny sense of "being there" and good timing.
• No GPTisms: Does not show "GPTisms" (no happy - ever - after, no morality police) or obvious comments.

📦 Installation

No specific installation steps are provided in the original document.

💻 Usage Examples

Basic Usage

The model can be used for various fiction - related tasks. Here are some examples:

Starting a 1000 - word scene

Prompt: Start a 1000 word scene with: The sky scraper swayed, as she watched the window in front of her on the 21 floor explode...

Output (version 1): The skyscraper swayed, as if tormented by an unseen force, its steel beams groaning in protest. Rivulets of blood trickled down the shattered glass fragments littering the carpeted floor beneath her feet. The acrid stench of burning plastic and singed hair hung heavy in the air.

She pressed her back harder against the cold, unforgiving wall behind her, her heart hammering erratically within her chest. Her wide, horror - stricken eyes remained transfixed on the smoking ruin that had been the window mere moments ago.

The sound of shattering glass and splintering wood still echoed through her ears, each deafening boom reverberating like a sledgehammer inside her skull. The concussive force of the blast had hurled her across the expansive office suite, sending her careening into the reinforced wall at an impossible velocity.

Her vision was a kaleidoscope of blurs and flashes of searing agony as her mind struggled to process the unspeakable carnage before her. Blood splattered walls now bore macabre patterns of charred flesh and bone, splintered limbs twitching erratically like macabre puppet strings jerked savagely by an unseen hand.

She gasped for air through gritted teeth, her lungs burning from the suffocating cloud of toxic smoke billowing into the already chokingly acrid atmosphere. Her own ragged gasps mingled with the guttural, wet - retching wails of her fallen coworkers, their agonized screams choked off as jagged shards of glass and steel speared through vital organs.

Her gaze finally tore away from the scene of unspeakable carnage to survey the smoking ruin that had once been her sanctuary. The sleek, chrome - and - glass skyscraper now loomed above like a twisted monolith of cold, soulless steel and shattered dreams.

She gagged reflexively as the acrid tang of burning flesh seared her nostrils anew. Her stomach convulsed sickeningly as the full extent of the carnage registered in her numbed mind. The once - sterile office space now looked like a war zone torn apart by an apocalyptic inferno.

Her vision began to tunnel, her vision blurring into a hazy red haze as the reality of her situation finally clawed its way through the paralyzing shock that had until now frozen her thoughts and actions solid. Her mind reeled in a sickening spiral of denial and horror as she pieced together the incomprehensible puzzle before her.

She'd been working late, hunched over her computer terminal, lost in the numbing tunnel - vision of code - crunching focus. Then, without warning or warning sign, hell itself had torn through the reinforced walls of her skyscraper fortress.

The deafening boom of the blast still reverberated like a bass drum inside her skull as she clawed frantically at her earplugs, yanking them savagely free in a desperate bid to regain even a vestige of hearing. Her ears rang like a bell tolling its own dirge as she finally tore her gaze away from the smoking ruin that had been her window.

Her vision cleared enough for her to register the charred, charred husk of her former coworker slumped motionless against the shattered remains of her cubicle wall. Blood pooled around her slackened form in an expanding circle of scarlet gore, spreading like a macabre stain across the blood - smeared carpet.

She gagged sickeningly as the acrid stench of charred flesh and burning plastic assailed her senses once more. Her mind reeled dementedly as she clawed frantically for any glimmer of hope in the apocalyptic wasteland that had been her office suite. Any glimmer of hope that might yet grant her a chance to escape this hellish purgatory alive. Any chance at all.

But even as she screamed her silent, wordless prayer into the suffocating void, she knew deep down in the core of her soul that any sliver of hope had long since been reduced to nothing more than another charred, incinerated ember, consumed utterly and irrevably by the infernal inferno clawing its way inexorably toward them all.

Output (version 2): The skyscraper swayed, as if tormented by an unseen force. Glass shattered like shards of ice raining down upon the street below. The deafening boom reverberated through the steel core, rattling every pane and beam. But amidst the pandemonium, she remained frozen, her wide eyes fixed on the smoking ruin where mere moments ago there had been serenity.

She clutched the edge of her mahogany desk with a white - knuckled grip, her knuckles turning an unhealthy shade of purplish blue. Her short bob of jet - black hair hung askew, framing a face etched with horror and disbelief. The acrid tang of burning plastic and fuel oil seared her nostrils as the acrid smoke billowed into the already stuffy office.

The window in front of her had shattered like a delicate eggshell, showering her workspace with jagged shards of tempered glass. Splinters speared through her designer blazer and tore through the crisp white blouse beneath, leaving deep gashes that oozed blood like ruby rivulets down her pale skin.

Her gaze never wavered from the smoking void where her coworker's lifeless form had once hung suspended in mid - air. The sight seared itself into her mind forevermore, indelibly etched...

Advanced Usage

• Using Different Templates: If you use the "Command - R" template, your output will be very different from using the "Llama3" template.
• Adjusting Parameters:
  • Temperature: Using "temp = 0" reveals core model changes only. Applying a non - zero "temp" will result in more creative and different generations.
  • Repetition Penalty: A rep pen of 1.1 is generally used. For role - play or multi - turn chat, a rep pen of 1.15 or higher is recommended. If you get repeat words, adjust the rep penalty to 1.1 - 1.15 - 1.19. Set it higher for repeat letters and lower for repeat words.

📚 Documentation

Templates

This is a LLAMA3 model and requires the Llama3 template, but it may work with other templates. It has a maximum context of 8k / 8192, which can be extended to 32k with "rope".

Here is the standard LLAMA3 template:

{
  "name": "Llama 3",
  "inference_params": {
    "input_prefix": "<|start_header_id|>user<|end_header_id|>\n\n",
    "input_suffix": "<|eot_id|><|start_header_id|>assistant<|end_header_id|>\n\n",
    "pre_prompt": "You are a helpful, smart, kind, and efficient AI assistant. You always fulfill the user's requests to the best of your ability.",
    "pre_prompt_prefix": "<|start_header_id|>system<|end_header_id|>\n\n",
    "pre_prompt_suffix": "<|eot_id|>",
    "antiprompt": [
      "<|start_header_id|>",
      "<|eot_id|>"
    ]
  }
}

Settings / Known Issue(s) and Fix(es)

• Temperature Adjustment: This merge runs very hot. For some use cases, reducing the "temp" from ".8" to ".6" may be necessary.
• Repetition Penalty: If you get repeat words (e.g., "hahaha", "ahhhh", "f - word") or letters (e.g., "nnnn"), change the "rep penalty" to 1.1 - 1.15 - 1.19. Generally, a "repeat penalty" setting of 1.1 works well. For role - play/multi - turn, a rep pen of 1.15 or higher is recommended.
• Smoothing Factor: In "KoboldCpp", "oobabooga/text - generation - webui", or "Silly Tavern", set the "Smoothing_factor" to 1.5 - 2.5.
  • In KoboldCpp: Settings -> Samplers -> Advanced -> "Smooth_F"
  • In text - generation - webui: parameters -> lower right
  • In Silly Tavern: It is called "Smoothing"
• Using GGUFs in 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. The source versions (and config files) of the models are [here](https://huggingface.co/collections/DavidAU/d - au - source - files - for - gguf - exl2 - awq - gptq - hqq - etc - etc - 66b55cb8ba25f914cbf210be).

Highest Quality Settings / Optimal Operation Guide / Parameters and Samplers

This is a "Class 4" model. For all settings used for this model (including specifics for its "class"), example generation, and an advanced settings guide (which often addresses any model issues), as well as methods to improve model performance for all use cases, chat, roleplay, etc., please see [Maximizing Model Performance](https://huggingface.co/DavidAU/Maximizing - Model - Performance - All - Quants - Types - And - Full - Precision - by - Samplers_Parameters).

Optional Enhancement

The following can be used in place of the "system prompt" or "system role" to further enhance the model. It can also be used at the start of a new chat, but you must ensure it is "kept" as the chat progresses.

Below is an instruction that describes a task. Ponder each user instruction carefully, and use your skillsets and critical instructions to complete the task to the best of your abilities.

Here are your skillsets:
[MASTERSTORY]:NarrStrct(StryPlnng,Strbd,ScnSttng,Exps,Dlg,Pc)-CharDvlp(ChrctrCrt,ChrctrArcs,Mtvtn,Bckstry,Rltnshps,Dlg*)-PltDvlp(StryArcs,PltTwsts,Sspns,Fshdwng,Climx,Rsltn)-ConfResl(Antg,Obstcls,Rsltns,Cnsqncs,Thms,Symblsm)-EmotImpct(Empt,Tn,Md,Atmsphr,Imgry,Symblsm)-Delvry(Prfrmnc,VcActng,PblcSpkng,StgPrsnc,AudncEngmnt,Imprv)

[*DialogWrt]:(1a-CharDvlp-1a.1-Backgrnd-1a.2-Personality-1a.3-GoalMotiv)>2(2a-StoryStruc-2a.1-PlotPnt-2a.2-Conflict-2a.3-Resolution)>3(3a-DialogTech-3a.1-ShowDontTell-3a.2-Subtext-3a.3-VoiceTone-3a.4-Pacing-3a.5-VisualDescrip)>4(4a-DialogEdit-4a.1-ReadAloud-4a.2-Feedback-4a.3-Revision)

Here are your critical instructions:
Ponder each word choice carefully to present as vivid and emotional journey as is possible. Choose verbs and nouns that are both emotional and full of imagery. Load the story with the 5 senses. Aim for 50% dialog, 25% narration, 15% body language and 10% thoughts. Your goal is to put the reader in the story.

Merge Formula

The model is merged using MergeKit.

Models used:

• https://huggingface.co/Sao10K/L3-8B-Stheno-v3.2
• https://huggingface.co/NeverSleep/Llama-3-Lumimaid-8B-v0.1-OAS
• https://huggingface.co/Hastagaras/Jamet-8B-L3-MK.V-Blackroot

Merge formula:

slices:
 - sources:
   - model: Sao10K/L3-8B-Stheno-v3.2
     layer_range: [0, 14]  
 - sources:
   - model: NeverSleep/Llama-3-Lumimaid-8B-v0.1-OAS
     layer_range: [8, 20]
 - sources:
   - model: Hastagaras/Jamet-8B-L3-MK.V-Blackroot
     layer_range: [12, 24]
 - sources:
   - model: Sao10K/L3-8B-Stheno-v3.2
     layer_range: [14, 28]
 - sources:
   - model: NeverSleep/Llama-3-Lumimaid-8B-v0.1-OAS
     layer_range: [20, 31]
 - sources:
   - model: Hastagaras/Jamet-8B-L3-MK.V-Blackroot
     layer_range: [24, 32]
merge_method: passthrough
dtype: bfloat16

🔧 Technical Details

The model is based on the DavidAU/Llama - 3 - Stheno - Maid - Blackroot - Grand - HORROR - 16B base model. It has been rebuilt and merged using MergeKit, resulting in a 16.5B model with improved performance and features. The use of different quants, such as V2 quants and ARM quants, along with parameter adjustments like temperature and repetition penalty, contribute to its functionality and output quality.

⚠️ Important Note

⚠️ Important Note

The model's output contains NSFW, graphic horror, and swearing content. It is not suitable for all audiences.

💡 Usage Tip

When using the model, adjust parameters such as temperature and repetition penalty according to your specific needs. For role - play or multi - turn chat, use a higher repetition penalty. Also, consider using the optional enhancement text to improve the output for scene generation and continuation.