L3 MOE 8X8B Dark Planet 8D Mirrored Chaos 47B GGUF

🚀 L3-MOE-8X8B-Dark-Planet-8D-Mirrored-Chaos-47B

This is a powerful LLama3 model using a mixture of experts, excelling in creative writing, prose, fiction, and role - play with a large context window and high - quality output generation.

✨ Features

• Model Architecture: It is a LLama3 model with a maximum context of 8192 (or 32k+ with rope). It combines eight unreleased 8B "Dark Planet" models into one 47B - parameter powerhouse using a mixture of experts.
• Exceptional Output: Its instruction - following and output generation for creative writing, prose, fiction, and role - play are outstanding. It shows great skills in description, dialog, imagery, metaphors, and prose, with diverse sentence and paragraph structures.
• Genre Versatility: It can handle various genres, including horror, romance, and all forms of fiction. It has a sense of humor and can generate "un - AI" like dialog.
• Stable Performance: It is a relatively stable and compressed model with a low perplexity level (lower than Meta Llama3 Instruct). It can operate with all parameters, including temp settings from 0 to 5.

📦 Installation

The model can be used in different applications. Here are the ways to set the number of experts in various platforms:

• LMStudio (https://lmstudio.ai): Set the number of experts at the "load" screen.
• Text - Generation - Webui (https://github.com/oobabooga/text - generation - webui): Set the number of experts at the loading screen page.
• KolboldCPP (https://github.com/LostRuins/koboldcpp) Version 1.8+: On the load screen, click on "TOKENS" and set the experts on this page, then launch the model.
• server.exe / Llama - server.exe (Llamacpp - https://github.com/ggerganov/llama.cpp/blob/master/examples/server/README.md): Add "--override - kv llama.expert_used_count=int:X" (where "X" is the number of experts to use) to the command line to start the "llamacpp server".
• API: Set the "num_experts_used" in the JSON payload (this may vary for different back - ends).

📚 Documentation

Model Details

Model Notes

• Writing Abilities: Its detail, prose, and fiction writing abilities are far superior to all combined Dark Planet 8B models.
• Temperature Settings: For more varied prose, raise the temp and/or add more instructions in your prompt. But be careful when raising the temp too high for role - playing as it may affect instruction following.
• Repetition Penalty: This model works well with a rep pen of 1 or higher, 1.02+ is recommended.
• Genre - Specific Prompting: To get a specific type of prose (e.g., horror), add "(vivid horror)" or "(graphic vivid horror)" (no quotes) in your prompt.
• GPTisms: A lot of GPTisms have been removed, but a few still remain.
• Output Length: The output length varies, and the model prefers long outputs unless specified otherwise.
• Quantization: Different quants will produce slightly different output. Due to its high stability and compressed nature, all quants will operate at above - average levels.
• Rope Extension: If using rope to extend context, increase temp and instruction detail levels to compensate for "rope issues".

Special Operations Notes

• Expert Selection: Even though the default is 4 experts, the "selected" 4 will vary during generation, resulting in different output for each prompt generation. It may take 2 - 4 regens of the same prompt to get the highest quality.
• Samplers: This model responds well to Dry, Dynamic Temp, and Smooth/Quadratic samplers. Using these can improve output quality. Higher temps (above 1) can aid in generation, especially in word choice and sentence generation. Increasing the number of experts will improve output quality at the cost of speed. Adjusting temp, samplers, and advanced samplers may be necessary when changing the number of experts. The choice of quant also impacts instruction following and output generation.

Template

This is a LLAMA3 model that 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 using "rope" settings. Using the "Command - R" template will result in different output compared to the "Llama3" template. The standard LLAMA3 template is as follows:

{
  "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 for Chat/Role - play and Smoother Operation

• Smoothing Factor: In "KoboldCpp", "oobabooga/text - generation - webui", or "Silly Tavern", set the "Smoothing_factor" to 1.5. 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).
• Other Options: Increase the rep pen to 1.1 - 1.15 (not necessary if using "smoothing_factor"). If the interface supports "Quadratic Sampling" ("smoothing"), make the adjustment as noted.

Highest Quality Settings

For all settings used for this model, including advanced settings and methods to improve performance for all use cases, please refer to [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 enhance the model. Copy and paste exactly as presented, without line - wrapping or breaking the lines:

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(

What Can I Use This Model For?

This model can be used for various activities, including:

• Scene generation
• Scene continuation
• Creative writing
• Fiction writing
• Plot generation
• Sub - plot generation
• Story generation
• Storytelling
• Role - playing
• Graphic horror
• Horror
• Dark humor
• NSFW content
• And all genres of general fiction.

Credits

• Special thanks to all the model makers/creators listed in the relevant repos.
• Special credit goes to MERGEKIT [https://github.com/arcee - ai/mergekit].
• Special thanks to Team "Mradermacher" for uploading the quants and creating IMATRIX quants. IMATRIX GGUFS can be found at [https://huggingface.co/mradermacher/L3 - MOE - 8X8B - Dark - Planet - 8D - Mirrored - Chaos - 47B - i1 - GGUF].

Warning

⚠️ Important Note

This model contains NSFW content. It has vivid prose, intense and visceral details, including violence, horror, gore, and swearing. It also includes uncensored humor, romance, and fun.