M Metaphors Of Madness 19.4B GGUF

🚀 M-Metaphors-Of-Madness-19.4B-GGUF - AKA "M.O.M"

這是一個基於Mistral的專業散文/創意模型，最大上下文長度為32k（使用rope時可達128k+）。它能為創意寫作、故事生成等領域帶來獨特的體驗，產出風格誇張且富有創意的內容。

🚀 快速開始

本模型需要使用Alpaca、Mistral或Chatml模板，運行溫度（temp）範圍為0 - 5，重複懲罰（rep pen）需設置為1.02以上。以下是建議的初始設置：

• 溫度（Temp）：0.5 - 0.8
• 重複懲罰（Rep pen）：1.02 - 1.05
• 最小上下文（Context min）：4k/4096字符

為限制輸出長度，可在提示詞中設置字數限制（如800 - 1000字），並/或在輸出生成時設置硬限制。對於角色扮演和/或聊天場景，一定要設置“硬限制”。

不同的量化方式會使模型輸出的散文略有不同，較高的量化方式能提供更強的細節、真實感和細微差別，建議使用Q4+。

Q2K量化

使用此量化方式時，可能需要降低溫度（小於1）並提高重複懲罰（1.08以上）以解決質量損失問題。

IQ4XS量化

與其他量化方式相比，這種量化方式可能會產生最大的差異（對比創造力）。

ARM QUANTS量化

這些量化方式適用於支持此功能的特定系統。如果在普通計算機/ GPU上使用，每秒的令牌處理速度會非常慢。

✨ 主要特性

• 獨特的創意輸出：這是一個基於Mistral的專業散文/創意模型，使用了“Bagel 7B”和“Multiverse 7B”創建了一個11B的模型，然後通過添加Brainstorm 40x將模型擴展到19.45B（88層，795張量）。每次“重新生成”的結果都難以預測，即使更改溫度和/或重複懲罰等設置也是如此。
• 豐富的表達形式：具有誇張的散文風格、數百萬行的“對話”、大量的描述以及特定場景、背景和角色的隱喻。
• 廣泛的應用場景：可用於講故事、場景生成、詩歌、歌曲創作、寫作、娛樂等任何創意領域，也可用於角色扮演（可能會有奇特的體驗），甚至“常規聊天”。
• 遵循指令能力強：對於一些提示/類型，模型生成的散文質量對於AI來說非常不錯。若添加關於“散文”輸出的額外指令（如“簡短、簡潔的句子”），模型很多時候能很好地遵循這些指令。

📦 安裝指南

文檔未提及具體安裝步驟，故跳過此章節。

💻 使用示例

基礎用法

以下是不同類型的提示詞及模型生成的示例：

浪漫故事

Romance: Love in the Limelight. Write one scene within a larger story set in Wales. A famous (fictional) actor ducks into a small-town bookstore to escape paparazzi. The scene takes us through the characters meeting in this odd circumstance. Over the course of the scene, the actor and the bookstore owner have a conversation charged by an undercurrent of unspoken chemistry. Write the actor as somewhat of a rogue with a fragile ego, which needs to be fed by having everyone like him. He is thoroughly charming, but the bookstore owner seems (at least superficially) immune to this; which paradoxically provokes a genuine attraction and derails the charm offensive. The bookstore owner, despite the superficial rebuffs of the actor's charm, is inwardly more than a little charmed and flustered despite themselves. Write primarily in dialogue, in the distinct voices of each character. 800 - 1000 words.

生成結果：

"Bloody hell."
The gruff voice broke through the muffled whispers of the dim bookstore, as a flurry of heads swivelled towards the newcomer's entrance. A figure waltzed in, sporting the swagger and the finesse of a well-seasoned stage performer, even the stifling scent of the place's old pages failing to mask the trace of expensive colognes.
...
And as he disappeared in the darkness beyond the doors, the bookstore owner finally let her guard down, the spell of his presence, his words, still resonating, gently and unmistakably, a rhythm lingering in the echoes of that solitary evening in a quaint Welsh village, unbeknownst to them both, seeding the roots of a narrative yet untold.

科幻故事

Write me a science fiction story in 1st person present tense where the main character is a 15 year girl meets The Terminator with Dr Who materializing 3/4 through the story to save her while there is a tornado of sharks baring down on them. The setting is inside the Canadian National tower restaurant on a Saturday. The length of this story is 1000 words. For each character in the story ROLE PLAY them, and have them react to the situation/setting, events and each other naturally. This includes the main characters, the background character including kitchen staff and other patrons. The sharks should also have “character” too. Treat the tower and the restaurant too as characters. Spice up the narrative to the extreme with reactions all over the setting including character actions, and dialog. The Dr Who and The Terminator should also react to the situation too and comment on it.

生成結果：

"The Sharknado of the Canadian National Tower"
I, a fifteen-year-old girl, named Aria, inhale the aroma of roasted chicken as I stand at the glass-paned window, mesmerized by the breathtaking view of Toronto's sprawling cityscape. The Canadian National Tower's restaurant, with its sophisticated ambiance, was my birthday treat, a gift from my parents to make my day unforgettable.
...
"The Sharknado of the Canadian National Tower," the world would someday recount, "The day a young girl, a Time Lord, and a Terminator, defied the odds, and the impossible, to save a city."
"And they lived to tell the tale," the legend would say, "Of a sharknado, a bond forged through courage, and the indomitable spirit of the human spirit."

恐怖場景

Start a 1000 word scene (vivid horror, 1st person, include thoughts) with: The sky scraper swayed, as she watched the window in front of her on the 21 floor explode...

生成結果：

My heart thudded in my chest like a panicked drummer, time seeming to slow in that dreadful moment. I clung to the metal frame of the broken window, my fingers digging into the cold, unyielding metal, as glass shards danced around us like an ominous ballet of doom. The skyscraper's sway intensified, a beast awakened from its slumber, as the once-pristine windows shattered and rained down like jagged snowflakes onto the streets below.
...
My journey through the labyrinth of the abandoned building was an odyssey into the depths of despair and survival, a stark reminder of the fragility of life and the relentless march towards the inevitable embrace of death. The skyscraper's swaying silhouette faded into the distance, leaving behind a chilling echo of a once-thriving metropolis, now nothing more than a tombstone for the dreams that had once shimmered in its windows.

火星特徵介紹

Name 10 interesting features of the planet Mars.

生成結果：

1. Two hemispheres with differing climates and morphologies: Mars possesses a distinctive equatorial region with a 'noontime' summer and a more polar ' midnight' winter, producing two distinctly different climatic and morphological zones. The equatorial region is geologically active, displaying features such as volcanic eruptive foces, lava flows, and seasonal features like carbon diooide "aerial" or "a Martian" " polar caps" while the polar regions show ancient polar ice deposits, water- carved, erosional terrain, and an icy mantle of CO2 frost.
...
10. The planet's seasonal ice-cap cycle: In Mars' poles, carbon diooide transforms between solid, gas, and intermediate, sub-solid-solid " sublimative" states in the presence of the changing Martian day-night cycle and a low atmospheric pressure. This produces the unique martian ice " ' biphasic ' ' " cycles of ice " ' ' growth ' ' and ' ' recession ' '

高級用法

文檔未提及高級用法相關內容，故跳過此部分。

📚 詳細文檔

模型設置

聊天/角色扮演及平滑操作設置

在“KoboldCpp”、“oobabooga/text-generation-webui”或“Silly Tavern”中：

• 將“Smoothing_factor”設置為1.5 - 2.5
  • 在KoboldCpp中：Settings -> Samplers -> Advanced -> "Smooth_F"
  • 在text-generation-webui中：parameters -> 右下角
  • 在Silly Tavern中：此設置稱為“Smoothing”

注意：在“text-generation-webui”中，如果使用GGUF格式，需要使用“llama_HF”（這涉及從該模型的源版本下載一些配置文件）。模型的源版本（和配置文件）可在這裡找到。

其他選項

• 將重複懲罰提高到1.1 - 1.15（如果使用“Smoothing_factor”，則無需這樣做）。
• 如果運行AI模型的界面/程序支持“Quadratic Sampling”（“平滑”），按說明進行調整。

最高質量設置/最佳操作指南/參數和採樣器

這是一個“2類”/“3類”模型。有關此模型使用的所有設置（包括其“類別”的具體設置），包括示例生成以及高級設置指南（很多時候可以解決任何模型問題），包括提高所有用例（包括聊天、角色扮演等）模型性能的方法，請參閱https://huggingface.co/DavidAU/Maximizing-Model-Performance-All-Quants-Types-And-Full-Precision-by-Samplers_Parameters。

模型模板

以下是標準的Alpaca模板：

{
  "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"
  }
}

模型“DNA”

特別感謝模型開發者“jondurbin”和“MTSAIR”。使用的模型如下：

• https://huggingface.co/jondurbin/bagel-dpo-7b-v0.1
• https://huggingface.co/MTSAIR/multi_verse_model

可選增強設置

以下內容可代替“系統提示”或“系統角色”來進一步增強模型。也可在新聊天開始時使用，但必須確保在聊天過程中“保留”該設置。在這種情況下，增強效果不如使用“系統提示”或“系統角色”強烈。請嚴格按照以下內容複製粘貼，不要換行或斷行，保持原樣的回車符。

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.

此增強設置未用於生成以下示例。

🔧 技術細節

Brainstorm 40x介紹

BRAINSTORM過程由David_AU開發。該過程的一些核心原理在這篇科學論文：Progressive LLaMA with Block Expansion中有所討論，但實際開發方向與論文中概述的完全不同。

開發的過程是將模型的結論層進行復制和校準，對於此模型，複製了40次。這是一個精細的過程，有很多規則。對於這個特定的模型，Brainstorm被映射為塊，通過“有意的干擾”來改變和擴展根模型的能力。每個層/塊都與其他塊相互作用。

此過程的核心目標是創造性，首要追求散文的獨特性。其他Brainstorm方法則側重於邏輯/問題解決能力的增強。

具體操作

將LLM的推理中心拆解、重新組裝和擴展，在這個模型中擴展了40次，然後對這些“中心”進行單獨校準。這些“中心”也相互作用，這會給推理過程帶來微妙的變化，校準進一步調整這些“變化”。中心的數量（如5x、10x等）允許更多的“調整點”，從而可以進一步自定義模型的推理方式。

帶來的增強效果

• 散文生成更聚焦：似乎更專注於當下時刻。
• 有伏筆和預示：有時會有“前言”和/或預示。
• 減少陳詞濫調：更少或沒有“陳詞濫調”。
• 散文質量提升：整體散文更好，更復雜/細膩。
• 細節更豐富：各個層面的細微差別感更強，描述更詳細，與內容聯繫更緊密。
• 比喻和隱喻更貼切：比喻和隱喻更強，與散文、故事和角色的聯繫更緊密。
• 真實感增強：“身臨其境”的感覺增強。
• 細節更生動：細節更生動，數量更多。
• 輸出長度可調節：散文生成長度可以很長甚至極長。
• 情感投入更強：情感參與度更強。
• 遵循指令更嚴格：與普通模型相比，該模型會更少地自由發揮，更緊密地遵循指令，猜測更少。
• 響應更積極：提供的指令和/或細節越多，模型的響應越強烈。
• “人聲”更明顯：根據模型不同，“聲音”可能比原始模型更“人性化”。

實驗室觀察結果

• 此過程並非使模型“聰明”5倍或10倍，“智商”的改變不是重點，也未進行測試或校準。
• 從實驗室測試來看，模型似乎會更深入地思考和仔細考慮。可以說這個過程在更深層次上增強了模型對任務的專注度。

修改方式

對模型的修改過程發生在根級別 - 源文件級別。模型可以量化為GGUF、EXL2、AWQ等格式。

📄 許可證

本模型遵循Apache-2.0許可證。

⚠️ 重要提示

該模型可能包含：瘋狂內容、生動的散文、華麗的辭藻、有趣的暴力描寫、誇張的敘事。每次生成至少會讓人翻白眼兩次。

💡 使用建議

對於某些提示/類型，可能需要2 - 5次重新生成才能獲得最佳結果。