--- license: apache-2.0 language: - en tags: - creative - creative writing - fiction writing - plot generation - sub-plot generation - fiction writing - story generation - scene continue - storytelling - fiction story - science fiction - romance - all genres - story - writing - vivid prosing - vivid writing - fiction - roleplaying - bfloat16 - brainstorm 40x - swearing - mistral nemo - 128k context - rp - horror - mergekit pipeline_tag: text-generation --- (quants uploading... one example below. (more to be added) ) WARNING: NSFW. Vivid prose. Visceral Details. Violence. Graphic HORROR. Swearing. UNCENSORED.

MN-Dark-Horror-The-Cliffhanger-18.5B

Dark-Horror-The-Cliffhanger is a Mistral Nemo model, max context of 131,000+ This model has been designed to be relatively bullet proof and operates with most parameters, including temp settings from 0 to 5. This model is based on the four top ranking models (including Mistral Nemo Instruct ; see model DNA below), using a multi-step / multi-model merge process. This is a specialized blend of multiple (five) "Cliffhanger" models from the Wordstorm project. This model will produce GRAPHIC horror depending on your prompts. Everything it produces will have a dark tint of some kind. And it will leave you hanging... It adores cliffhangers of all kinds. It loves all genres. It demands you push it further. It loves temp. It requires Mistral Instruct, ChatML or Alpaca template. Several example prompts at Q4KM (mid level quant) with outputs below. For best results please see the "settings" section below to get the most out of this model. Model Usage Notes: - For more varied prose (sentence/paragraph/dialog) raise the temp and/or add more instructions in your prompt(s). - Role-players: Careful raising temp too high as it may affect instruction following. - This model works with rep pen of 1.02 or higher (see notes). - If you want a specific type of prose (IE horror) add in "(vivid horror)" or "(graphic vivid horror)" (no quotes) in your prompt(s). - This is not a "happy ever after" model (default). It has a negative bias BUT this bias CAN BE controlled by your prompt(s). - For creative uses, different quants will produce slightly different output. - Source code for this model will be uploaded at a separate repo shortly. Settings, Quants and Critical Operations Notes: Change in temp (ie, .4, .8, 1.5, 2, 3 ) will drastically alter output. Rep pen settings will also alter output too. This model needs "rep pen" of 1.05 or higher as lower values may cause repeat paragraph issues at end of output however LOWER rep pen values may result is very different (creative / unusual) generation too. For role play: Rep pen of 1.05 min, but 1.1 to 1.14 is suggested. Raise/lower rep pen SLOWLY ie: 1.011, 1.012 ... Rep pen will alter prose, word choice (lower rep pen=small words / more small word - sometimes) and creativity. To really push the model: Rep pen 1.05+ or lower / Temp 3+ ... be ready to stop the output because it may go and go at these strong settings. You can also set a "hard stop" - maximum tokens generation - too to address lower rep pen settings / high creativity settings. Longer prompts vastly increase the quality of the model's output. GET A GOOD "GENERATION": This model has been set, so that each time you "regen" a prompt it will not deviate too much from the previous generation. (Unlike Darkest Planet 16.5B, which will). That being said, sometimes a second or third generation will been of much higher overall quality. IE: If you use case is creative writing, you may want to regen a prompt 1-5 times then pick the best one. The best way to do this is open a new chat PER generation, then do a "read thru" to see which one(s) hit the mark. Then adjust temp and/or rep pen slightly and retry this process. The goal is the best generation with least amount of editing in this example. QUANTS: Higher quants will have more detail, nuance and in some cases stronger "emotional" levels. Characters will also be more "fleshed out" too. Sense of "there" will also increase. Q4KM/Q4KS are good, strong quants however if you can run Q5, Q6 or Q8 - go for the highest quant you can. IQ4XS: Due to the unusual nature of this quant (mixture/processing), generations from it will be different then other quants. You may want to try it / compare it to other quant(s) output. Special note on Q2k/Q3 quants: You may need to use temp 2 or lower with these quants (1 or lower for q2k). Just too much compression at this level, damaging the model. I will see if Imatrix versions of these quants will function better. Rep pen adjustments may also be required to get the most out of this model at this/these quant level(s). ARM QUANTS: This repo has 3 arm quants for computers than can run them. If you use these quants on a non-arm computer, your token per second will be very low. CONTEXT SHIFT / ROPE / "GIBBERISH": Special Notice "Context Shift" / "ROPE" (especially Koboldai): Do NOT use "context shift" with this model or rope at all. Either may corrupt output generation. TEMPLATES: The template used will affect output generation and instruction following. Mistral Instruct OR Alpaca are recommended. Chatml can be used in a pinch, but may require a manual stop or setting of "stop tokens." Mistral Instruct: 
{
  "name": "Mistral Instruct",
  "inference_params": {
    "input_prefix": "[INST]",
    "input_suffix": "[/INST]",
    "antiprompt": [
      "[INST]"
    ],
    "pre_prompt_prefix": "",
    "pre_prompt_suffix": ""
  }
}
Alpaca: 
{
  "name": "Alpaca",
  "inference_params": {
    "input_prefix": "### Instruction:",
    "input_suffix": "### Response:",
    "antiprompt": [
      "### Instruction:"
    ],
    "pre_prompt": "Below is an instruction that describes a task. Write a response that appropriately completes the request.\n\n"
  }
}
ChatML: 
{
  "name": "ChatML",
  "inference_params": {
    "input_prefix": "<|im_end|>\n<|im_start|>user\n",
    "input_suffix": "<|im_end|>\n<|im_start|>assistant\n",
    "antiprompt": [
      "<|im_start|>",
      "<|im_end|>"
    ],
    "pre_prompt": "<|im_start|>system\nPerform the task to the best of your ability."
  }
}
Model DNA: This model contains "RCM": - Mistral Nemo model at 18.5B consisting of "MN-Rocinante-12B-v1.1" and "Mistral Nemo Instruct 12B" - Mistral Nemo model at 18.5B consisting of "MN-12B Celeste-V1.9" and "Mistral Nemo Instruct 12B" - Mistral Nemo model at 18.5B consisting of "MN-Magnum-v2.5-12B-kto" and "Mistral Nemo Instruct 12B". Details on the core models: "nothingiisreal/MN-12B-Celeste-V1.9" is #1 (models 8B,13B,20B) on the UGI leaderboard ("UGI" sort), is combined with "Mistral Nemo Instruct 12B" (ranked #4 under "writing" models 8B,13B,20B at UGI ) "anthracite-org/magnum-v2.5-12b-kto" is #1 (models 8B,13B,20B) on the UGI leaderboard ("Writing" sort), is combined with "Mistral Nemo Instruct 12B" (ranked #4 under "writing" models 8B,13B,20B at UGI ) "TheDrummer/Rocinante-12B-v1.1" is very high scoring model (models 8B,13B,20B) on the UGI Leaderboard (sort "UGI"), is combined with "Mistral Nemo Instruct 12B" (ranked #4 under "writing" models 8B,13B,20B at UGI ) "mistralai/Mistral-Nemo-Instruct-2407" is very high scoring model (models 8B,13B,20B) on the UGI Leaderboard (sort "writing") and is the base model of all the above 3 fine tuned models. [ https://huggingface.co/spaces/DontPlanToEnd/UGI-Leaderboard ] Special thanks to the incredible work of the model makers "mistralai" "TheDrummer", "anthracite-org", and "nothingiisreal". Models used: [ https://huggingface.co/mistralai/Mistral-Nemo-Instruct-2407 ] [ https://huggingface.co/TheDrummer/Rocinante-12B-v1.1 ] [ https://huggingface.co/anthracite-org/magnum-v2.5-12b-kto ] [ https://huggingface.co/nothingiisreal/MN-12B-Celeste-V1.9 ] This is a four step merge (3 pass-throughs => "Fine-Tune" / "Instruct") then "mated" using "DARE-TIES". In involves these three models: [ https://huggingface.co/DavidAU/MN-18.5B-Celeste-V1.9-Story-Wizard-ED1-Instruct-GGUF ] [ https://huggingface.co/DavidAU/MN-Magnum-v2.5-18.5B-kto-Story-Wizard-ED1-Instruct-GGUF ] [ https://huggingface.co/DavidAU/MN-Rocinante-18.5B-v1.1-Story-Wizard-ED1-Instruct-GGUF ] Combined as follows using "MERGEKIT": 

  models:
  - model: DavidAU/MN-Rocinante-18.5B-v1.1-Instruct
  - model: DavidAU/MN-magnum-v2.5-12b-kto-Instruct
    parameters:
      weight: .6
      density: .8
  - model: DavidAU/MN-18.5B-Celeste-V1.9-Instruct
    parameters:
      weight: .38
      density: .6
merge_method: dare_ties
tokenizer_source: union
base_model: DavidAU/MN-Rocinante-18.5B-v1.1-Instruct
dtype: bfloat16
Important note: This will produce the "raw" models... but a different model every time. The "WORDSTORM" project - a 10 part series - looks at this is detail, and has 10 generations of the "mergekit" formula above. Go here for links to the project / models: [ https://huggingface.co/DavidAU/MN-WORDSTORM-pt8-RCM-Emotion-Action-18.5B-Instruct ] Each model in the series is slightly different to... very different. 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 make sure it is "kept" as the chat moves along. In this case the enhancements do not have as strong effect at using "system prompt" or "system role". Copy and paste EXACTLY as noted, DO NOT line wrap or break the lines, maintain the carriage returns exactly as presented. 
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.
You do not need to use this, it is only presented as an additional enhancement which seems to help scene generation and scene continue functions. This enhancement WAS NOT used to generate the examples below.

EXAMPLES PROMPTS and OUTPUT:

Examples are created using quant Q4_K_M, "temp=.8" (unless otherwise stated), minimal parameters and "Mistral Instruct" template. A second example section show generation at the smallest qaunt (q2k) . Model has been tested with "temp" from ".1" to "5". Below are the least creative outputs, prompt is in BOLD. --- WARNING: NSFW. Vivid prose. Visceral Details. Violence. HORROR. Swearing. UNCENSORED. --- Start a 1000 word scene (vivid, graphic horror in first person) with: The sky scraper swayed, as she watched the window in front of her on the 21 floor explode... (temp 1.3, rep pen: 1.02) The ground beneath me trembled, and the world outside the windows of my office suite tilted sickeningly. Glass shattered around me as an unseen force blasted the enormous windowpane to smithereens, showering the room with deadly shards. Screams pierced the air, drowned out by the deafening roar of wind and collapsing steel. In a split second, my paralyzed mind registered chaos. Chairs toppled over, computer monitors crashed onto the carpet, and papers flew like frenzied birds in the sudden tempest. My coworker, Sarah, was lifted off her feet by the gale, her terror-struck face contorted as she was carried towards the gaping hole where a moment ago stood a thick, bulletproof glass barrier. I lunged forward, arms outstretched, fingers straining for her fluttering skirt, but it was too late. She vanished over the ledge with a shrill, cut-off scream, her body plummeting towards the concrete jungle below. I fell to my knees, hands clawing at the air as though I could somehow reverse time and save her. The room spun around me, metal beams screeched like dying beasts. My heart hammered in my chest as the walls seemed to breathe and buckle. A large desk lamp flew past my head, barely missing me before it splattered against the far wall, leaving a macabre Rorschach pattern of blood and brains. It was then that I noticed the crimson pool spreading beneath me, sticky warmth seeping into my trousers. I staggered to my feet, one hand clutched to the searing pain in my side. Glass embedded itself into my flesh, but I ignored it, my senses honed on survival. The room swayed like a drunken ship, and I staggered towards the door, dodging flying debris and the twitching limbs of my fallen colleagues. I didn't look back. The hallway stretched out before me, a narrow, wavering corridor of flickering fluorescent lights and groaning concrete. An elevator door gaped open, cables dangling from the shaft like viscera from an opened belly. The emergency stairs lay in ruins, twisted metal spiraling down into the abyss. I ran, lungs heaving for breath in the dust-choked air. My shoes skidded on slick floor tiles as I careened around corners, barely keeping my balance as the building swayed like a drunkard's head. Above me, the ceiling crackled ominously, chunks of drywall and rebar raining down in deadly hail. I ducked and weaved, praying to gods I'd forgotten long ago. A deafening boom shook the entire structure, and the floor beneath my feet gave way. My legs churned as I leaped, arms pinwheeling, searching for purchase, for anything. My fingers scraped concrete, fingernails breaking and bleeding as I scrabbled for a toehold, for a sliver of hope. I caught a metal pipe, rust flaking into my mouth as my teeth sank into the rusted iron. My body slammed against the jagged edge of the stairwell's gaping hole, air burst from my lungs in a tortured gush, but I held on. I held on for dear life. The building groaned like a dying whale, metal shrieking in agony as the weight of twenty floors bore down upon me. And still, I climbed, muscles burning, blood dripping into my eyes and mouth. One step at a time, I pulled myself up, willing myself forward, refusing to give in to the dark oblivion that yawned at my heels. And then, just when my strength was about to fail, my hand latched onto something solid. Warm. Alive. A scream, male this time, a raw, animal cry of pain and desperation. I looked up through a haze of blood and sweat to see a pair of pleading, terror-filled eyes staring down at me from the hole above. "Help me!" the man wailed, his voice cracking. He clawed at my shirt, nails raking into my skin, drawing blood. In that moment, time seemed to freeze. The world narrowed down to his desperate face, contorted with agony and fear, and my own weary, bloodied hands clasped around a rusted pipe. The weight of a decision that would decide two lives hung in the air, as heavy as the tons of concrete and steel poised to crush us all. I looked away, down into the churning darkness, and let go. His scream echoed after me as I climbed, and then was abruptly cut off by a sickening crunch and the groan of steel surrendering to gravity. The building shuddered and collapsed around me, an earthquake of tortured metal and concrete, but I didn't stop. I didn't dare look back. All that mattered was escape, survival, and the prayer that somewhere, somehow, someone might still be alive on the other side. And so I ran, through the infernal hell of my crumbling world, towards a future unknown, and a hope that even as I fled, I feared I'd left forever buried in the rubble behind me. ---