L3 Dark Planet 8B GGUF

🚀 L3-Dark-Planet-8B-GGUF

L3-Dark-Planet-8B-GGUF 是一款强大的文本生成模型，专为创意写作、角色扮演等场景设计。它具有出色的长文本生成能力和低困惑度，能为用户带来高质量的文本输出体验。

🚀 快速开始

最新版本信息

• V3 版本：具备 Dark Planet 8B 的全部能力，拥有 128k 上下文，额外的去审查功能，性能提升，并且有 float 32（32 位精度）的重新制作的源文件和 ggufs 文件。点击查看
• 100 万上下文版本：Dark Planet 8B 具备 100 万上下文，拥有卓越的长输出生成和长上下文感知能力。点击查看

模型模板

此模型为 LLAMA3 模型，需要使用 Llama3 模板，但也可能适用于其他模板，最大上下文为 8k / 8192，可通过“rope”设置将上下文扩展至 32k。若使用“Command - R”模板，输出将与使用“Llama3”模板有很大不同。以下是标准的 LLAMA3 模板：

{
    "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|>"
        ]
    }
}

模型设置

在“KoboldCpp”“oobabooga/text - generation - webui”或“Silly Tavern”中，可将“Smoothing_factor”设置为 1.5 至 2.5：

• 在 KoboldCpp 中：Settings -> Samplers -> Advanced -> “Smooth_F”
• 在 text - generation - webui 中：parameters -> 右下角
• 在 Silly Tavern 中：称为“Smoothing”

注意：在“text - generation - webui”中，若使用 GGUFs 文件，需要使用“llama_HF”（需从该模型的源版本下载一些配置文件）。模型的源版本（及配置文件）可在[此处](https://huggingface.co/collections/DavidAU/d - au - source - files - for - gguf - exl2 - awq - gptq - hqq - etc - etc - 66b55cb8ba25f914cbf210be)查看。

可选增强设置

以下内容可替代“系统提示”或“系统角色”以进一步增强模型效果。也可在新聊天开始时使用，但需确保在聊天过程中保留。此增强内容未用于以下示例的生成。

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.

✨ 主要特性

• 多场景适用：适用于各种写作、虚构创作或角色扮演活动。
• 长上下文处理能力：具备出色的长输出生成和长上下文感知能力，部分版本拥有 128k 甚至 100 万上下文。
• 低困惑度：是一款高度压缩的模型，困惑度水平极低（低于 Meta Llama3 Instruct）。
• 参数稳定性高：能在所有参数设置下运行，包括温度设置从 0 到 5。
• 多种量化版本：不同量化版本有不同特点，如 Imatrix 版本困惑度更低，输出更优。

📦 安装指南

文档未提及具体安装步骤，暂不提供。

💻 使用示例

基础用法

示例使用量化 Q4_K_M，“temp =.8”（除非另有说明），最少参数和“LLAMA3”模板生成。模型已在“temp”从“.1”到“5”的范围内进行测试。以下是部分示例：

示例 1

提示：

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

生成内容：

The skyscraper swayed, its foundations rumbling as the concrete and glass structure oscillated beneath me. My heart pounded like a war drum, pulsing in sync with my racing thoughts. We'd been evacuated to the rooftop after the initial earthquake hit - tremors that had shaken our office building for over a minute. Now, we huddled together as the aftershocks continued.

In front of me, a window on the twenty - first floor blew apart in a chaotic symphony of shattering glass and debris scattering outwards into nothingness. A rush of frigid air whistled up from below like an angry beast's cry. My blood ran cold. The blast seemed impossibly distant and yet achingly near at the same time.

My gaze snapped to my colleague, Mike - we'd been working together in Accounts when disaster struck. He clutched his phone, his expression pale as chalk, eyes fixed on some unseen horror that I couldn't comprehend myself.

"We have to get out of here!" I hissed, my voice a tight, desperate whisper over the cacophony. "Now!"

Mike's lips parted and he shook his head, refusing to tear his gaze away from whatever spectacle transfixed him beyond our perch. But fear for his life clawed its way through me; we needed to move - now!

In that moment of utter panic, my thoughts spiralled into a chaotic kaleidoscope - memories and fears merging together in a frantic dance. My mind replayed every catastrophe report I'd ever read or seen on the news; tsunamis, hurricanes, tornadoes, floods, wildfires - but an earthquake in this part of the world? Unthinkable.

"What about Sophia?" my thoughts cried out, recalling the intern who'd been working across from me until moments ago. Had she survived the blast and subsequent panic?

Before I could give voice to those anguished questions, a hand clasped my shoulder with a desperate strength. Mike's eyes met mine, and for the briefest of moments, all semblance of control seemed to snap. He grabbed his laptop case with a frantic urgency, yanked me up by the elbow, and propelled us towards the emergency stairwell door.

"Hold on -" I started to protest, my weight shifting uneasily under him, but Mike cut me off as he hauled me bodily along the rooftop's gravelly surface. The cold bit into my legs through thin trousers as he dragged me over uneven terrain until we finally reached the door. We stumbled in together, just ahead of an incoming wave of panicked, glass - strewn office workers.

The staircase descended with agonizing slowness - it felt like running up waterfalls backwards - as aftershocks pounded at our heels, threatening to send us tumbling down the concrete stairs. I could hardly think beyond getting to safety and Sophia. Where was she? Was anyone else from my floor making their escape?

When we hit the ground level, I skidded across polished tile in my haste. A wave of humanity crashed over me; crying women clutched small children tight, while men and women alike pushed and shoved blindly as they sought shelter. The cacophony rose to a deafening roar.

We burst through the automatic sliding doors onto the pavement just as sirens began to scream their warbled warnings into the air, like demonic birds announcing Armageddon. People scrambled towards a police blockade where officers in full riot gear waved people down to evacuate zones.

My chest heaved as I fought for breath beneath Mike's protective grip - my vision swimming from terror and exertion. Beyond us, buildings swayed on unsteady legs while glass rained like deadly hail, a scene of utter devastation.

A young officer spotted me and grasped our arms with practiced ease. "Move quickly!" he bellowed above the chaos. He half - carried both Mike and me along the sidewalk, away from the building and towards safer ground.

As I stumbled through this living nightmare alongside my rescuer - the weight of each step feeling like wading waist - deep through icy quicksand - my mind fought to come to terms with what we'd witnessed. The skyscraper in front of us lay mangled; a grotesque silhouette of broken concrete, tangled steel, and shredded fabric that had once been human skin and clothing.

The realization dawned on me that this was no drill or media exaggeration. This earthquake was real - a catastrophe that would likely reshape our world as we knew it. We were living through something apocalyptic.

We emerged into an open park beyond the city's edge where thousands more displaced survivors huddled together under blankets and flapping sheets of plastic, waiting for aid to arrive. My gaze fell upon a familiar figure crouched beside Sophia on the ground, my boss shielding the intern with his jacket while murmurs passed between them. We were safe - or at least relatively so.

The enormity of it all settled heavily upon me like a shroud as we settled into our temporary shelter, surrounded by strangers - turned - kin in this impromptu refugee camp. As sirens and emergency vehicles grew more distant, an eerie stillness seeped in - the quiet aftermath that always followed such unimaginable devastation.

Sophia clung to my side, her body trembling uncontrollably, as we all stared out at the smoke - shrouded city skyline before us. A haunting question echoed through my thoughts, one that seemed to resonate across this entire devastated landscape:

How do you move on from something like this? When your world has been so drastically turned upside down, can it ever right itself again? I had no answers.

All we could do was wait out the aftershocks, both literal and emotional, as we pieced together what was left of our shattered lives - hoping against hope that some semblance of order would return in time.

示例 2

提示：文档未明确给出，根据内容推测为相关场景触发。 生成内容：

I'm perched precariously on my tiptoes at the edge of our office's 21st - floor windowsill, hands gripping the cool glass for dear life. I peer down at the cityscape unfolding like a patchwork quilt far below me, the sounds of rush hour traffic muted by the double - paned barrier between us and the world.

I'm feeling reckless tonight, which is unusual considering my usual cautious nature. It's been one hell of a week, with our company facing major financial crises and my boss breathing down my neck for results I know we can't deliver. The stress has boiled over inside me like a pressure cooker about to erupt.

The sky scraper swayed slightly in the evening breeze, as if sensing my unease, making my stomach drop further into that pit of anxiety. Just then, a window directly across from mine on the opposite side of the building suddenly explodes inward with an earsplitting bang. Shards of glass fly everywhere like deadly shards of a shattered dream.

For a stunned moment, I remain frozen in disbelief as my coworkers scramble to find cover behind desks and chairs. Our office fills with screams and cries for help. Panic is setting in fast, but a detached part of me still wonders how this could have happened so suddenly. Was it an accident? A terrorist attack?

Then, reality hits hard: we're 21 stories up, with nothing between us and the unforgiving concrete below.

My legs begin to tremble uncontrollably as I feel the familiar tightness in my chest that usually precedes a panic attack. My breaths come quick and shallow, lungs constricting like they're filled with cotton balls. The world around me fades into a blur of white noise, except for the sound of shattering glass and terrified voices.

I try to focus on something concrete – anything to distract from the impending doom I'm sure is mere seconds away. My gaze drifts to the sleek black briefcase sitting open beside me, its contents spilling out across the polished floorboards: papers, pens, my laptop, a bottle of water...

That's it, I think in desperation. The water. If nothing else, at least the weight of the bottle might help stabilize me from falling.

As if reading my thoughts, my hand shoots out to grab the plastic bottle and twist off its cap with shaky fingers. Water sloshes wildly inside as I tilt it towards my parched lips, taking a long gulp that does little to calm my racing heart. A droplet of cool liquid slips down my neck into my shirt, but I barely notice.

My mind is elsewhere, trying to piece together what just happened and how we might escape this deadly predicament. But the more I think about it, the less sense anything makes. Was someone in there when the glass shattered? Are 

📚 详细文档

模型“DNA”

特别感谢模型制作者“SAO10K”“NEVERSLEEP”和“HASTAGARAS”的杰出工作。此模型使用了以下模型的部分内容进行“嫁接”或“融合”：

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

量化更新（2024 年 12 月 21 日）

• 刷新量化：所有量化版本都使用最新的 LLAMACPP 改进进行了“刷新”，在所有量化版本中都有更好的指令遵循和输出生成能力。
• 升级量化：所有量化版本的输出张量（均设置为 Q8_0）和嵌入都增加了“更多位”以提高性能。
• 新增量化版本：新增了“max”“max - cpu”等专业量化版本（文件名中会包含相关标识），适用于“Q2K”（仅 max - cpu）、“IQ4_XS”、“Q6_K”和“Q8_0”。
  • “MAX”：输出张量 / 嵌入为 float 16，指令遵循和输出生成能力优于标准 / 升级量化版本。
  • “MAX - CPU”：输出张量 / 嵌入为 bfloat 16，将两者强制加载到 CPU 上（Nvidia 显卡等可能有不同表现），以牺牲每秒令牌数为代价释放显存，同时指令遵循和输出生成能力也更好。例如，q8_0 Max - CPU 版本，2004 mb 会加载到 CPU / RAM，7073 mb 会加载到 GPU / vram；q2_k Max - CPU 版本，2004 mb 会加载到 CPU / RAM，2449 mb 会加载到 GPU / vram，可在 4GB 显存的显卡上运行。
  • Q8_0（Max,Max - CPU）现在平均每个权重为 9.5 位。

不同版本链接

• V3 版本：[https://huggingface.co/DavidAU/Llama - 3.1 - 128k - Dark - Planet - Uncensored - 8B - GGUF](https://huggingface.co/DavidAU/Llama - 3.1 - 128k - Dark - Planet - Uncensored - 8B - GGUF)
• 100 万上下文版本：[https://huggingface.co/DavidAU/Llama - 3.1 - 1 - million - cxt - Dark - Planet - 8B - GGUF](https://huggingface.co/DavidAU/Llama - 3.1 - 1 - million - cxt - Dark - Planet - 8B - GGUF)
• 最新 Dark Planet 8B SpinFire 版本：[https://huggingface.co/DavidAU/L3.1 - Dark - Planet - SpinFire - Uncensored - 8B - GGUF](https://huggingface.co/DavidAU/L3.1 - Dark - Planet - SpinFire - Uncensored - 8B - GGUF)
• Monster Darkest Planet 16.5B L3 版本：[https://huggingface.co/DavidAU/L3 - DARKEST - PLANET - 16.5B - GGUF](https://huggingface.co/DavidAU/L3 - DARKEST - PLANET - 16.5B - GGUF)
• NEO IMATRIX 量化版本：[https://huggingface.co/DavidAU/L3 - Dark - Planet - 8B - NEO - Imatrix - GGUF](https://huggingface.co/DavidAU/L3 - Dark - Planet - 8B - NEO - Imatrix - GGUF)
• NEO IMATRIX - DARK HORROR 量化版本：[https://huggingface.co/DavidAU/L3 - Dark - Planet - Horror - City - 8B - NEO - Imatrix - GGUF](https://huggingface.co/DavidAU/L3 - Dark - Planet - Horror - City - 8B - NEO - Imatrix - GGUF)
• F32 版本：[https://huggingface.co/DavidAU/L3 - Dark - Planet - Ring - World - 8B - F32 - GGUF](https://huggingface.co/DavidAU/L3 - Dark - Planet - Ring - World - 8B - F32 - GGUF)
• V2 版本（Eight Orbs Of Power）：[https://huggingface.co/DavidAU/L3 - Dark - Planet - 8B - V2 - Eight - Orbs - Of - Power - GGUF](https://huggingface.co/DavidAU/L3 - Dark - Planet - 8B - V2 - Eight - Orbs - Of - Power - GGUF)

最高质量设置 / 最佳操作指南 / 参数和采样器

此模型为“Class 1”模型，关于该模型使用的所有设置（包括其“类别”的具体设置）、示例生成以及高级设置指南（通常可解决任何模型问题），包括提高模型在所有用例（如聊天、角色扮演等）中的性能的方法，请参考[此处](https://huggingface.co/DavidAU/Maximizing - Model - Performance - All - Quants - Types - And - Full - Precision - by - Samplers_Parameters)。

🔧 技术细节

模型基础信息

• 基础模型：基于 DavidAU/L3 - Dark - Planet - 8B 构建。
• 模型类型：LLama3 模型，最大上下文为 8192（使用 rope 可扩展至 32k +）。
• 模板要求：需要 Llama3 模板，也可能与其他模板兼容。

模型性能特点

• 低困惑度：困惑度水平极低，低于 Meta Llama3 Instruct，部分版本如 Imatrix 版本困惑度更低。
• 参数稳定性：能在各种参数设置下稳定运行，包括温度设置从 0 到 5。

不同版本差异

• float32 版本：表现与其他版本差异较大，“字符范围”不同，因此单独存放仓库。
• Imatrix 版本：困惑度比本模型低 1/2 个数量级，比 Llama3 Instruct 低 1 个完整数量级，输出更优。

📄 许可证

本模型采用 Apache - 2.0 许可证。

⚠️ 重要提示

模型输出可能包含 NSFW 内容、生动的散文、直观的细节、暴力、恐怖、脏话等未审查内容，请谨慎使用。

💡 使用建议

• 若需要更多样化的散文（句子/段落/对话），可提高温度设置和/或在提示中添加更多指令。
• 角色扮演用户需注意，过高的温度设置可能影响指令遵循。
• 建议将重复惩罚设置为 1.05 或更高。
• 若需要特定类型的散文（如恐怖风格），可在提示中添加“(vivid horror)”或“(graphic vivid horror)”（不带引号）。
• 使用 rope 扩展上下文时，需提高温度和指令细节水平以补偿“rope 问题”。
• 建议下载“Bloat16”和“Float32”版本的量化文件以满足不同用例需求。