EldritchLabs/KrakenSakura-Maelstrom-12B-v1
1
0
Fork 0
Code Issues Pull Requests Actions Projects Releases Wiki Activity
Files
main
KrakenSakura-Maelstrom-12B-v1/README.md
ModelHub XC eb08c85ff3 初始化项目，由ModelHub XC社区提供模型 
Model: EldritchLabs/KrakenSakura-Maelstrom-12B-v1
Source: Original Platform
2026-06-21 07:27:17 +08:00

1741 lines
53 KiB
Markdown
Raw Permalink Blame History

---
base_model:
- allura-org/Tlacuilo-12B
- ChaoticNeutrals/Captain_Eris_Noctis-12B-v0.420
- DavidAU/Mistral-Nemo-2407-12B-Thinking-Claude-Gemini-GPT5.2-Uncensored-HERETIC
- DavidAU/MN-Dark-Planet-TITAN-12B
- DavidAU/MN-GRAND-Gutenberg-Lyra4-Lyra-12B-DARKNESS
- DreadPoor/Famino-12B-Model_Stock
- EldritchLabs/Cactus-Dream-Horror-12B
- EldritchLabs/Human-Like-Mistral-Nemo-Instruct-2407-MPOA
- EldritchLabs/Kraken-Karcher-12B-v1
- EldritchLabs/MN-12B-Mag-Mell-R1-Uncensored-Scale1.2
- EldritchLabs/MN-12B-RP-Ink-Longform-MPOA
- Epiculous/Violet_Twilight-v0.2
- IggyLux/MN-VelvetCafe-RP-12B-V2
- inflatebot/MN-12B-Mag-Mell-R1
- LatitudeGames/Muse-12B
- LatitudeGames/Wayfarer-2-12B
- mistralai/Mistral-Nemo-Instruct-2407
- MrRikyz/StarlightMoon-Foxfire-12B
- MuXodious/Rocinante-X-12B-v1-absolute-heresy
- Naphula/Ancient-Awakening-12B
- Naphula/Riemannian-Redshift-12B-v1
- ohyeah1/Violet-Lyra-Gutenberg-v2
- PocketDoc/Dans-SakuraKaze-V1.0.0-12b
- PygmalionAI/Pygmalion-3-12B
- rAIfle/Questionable-MN-bf16
- ReadyArt/Dark-Nexus-12B-v2.0
- ReadyArt/Forgotten-Safeword-12B-v4.0
- redrix/GodSlayer-12B-ABYSS
- Retreatcost/Chrysologus-12B
- Retreatcost/Impish-LongPen-12B
- Retreatcost/KansenSakura-Conflagration-RP-12b
- SicariusSicariiStuff/Impish_Bloodmoon_12B
- Sorihon/Celestial-Queen-12B-Heretic
- SuperbEmphasis/MN-12b-RP-Ink-RP-Longform
- SuperbEmphasis/Omega-Darker_The-Final-Directive-Longform-Stage2-ERP-12B-v0.2
- TheDrummer/Rocinante-X-12B-v1
- Vortex5/Aurora-Mirage-12B
- Vortex5/Prototype-X-12b
datasets:
- OccultAI/illuminati_imatrix_v1
language:
- en
library_name: transformers
license: apache-2.0
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
- float32
- swearing
- rp
- horror
- mistral
- nemo
- merge
- mergekit
widget:
- text: "🦑 KrakenSakura-Maelström-12B-v1"
output:
url: https://cdn-uploads.huggingface.co/production/uploads/68e840caa318194c44ec2a04/ssPstIfsBZ-lw4Skif0w7.png
---
<audio controls src="https://cdn-uploads.huggingface.co/production/uploads/68e840caa318194c44ec2a04/y6TzIZP4XvaH_B7iQKV6O.mpga"></audio>
> [!WARNING]
> <span style="color:red; font-weight:bold">⚠️ Warning:</span> This model can produce narratives and RP that contain violent and graphic erotic content. Adjust your system prompt accordingly, and use **ChatML** (recommended) or **Mistral Tekken/NonTekken** chat template.
>
<!DOCTYPE html>
<style>
body {
font-family: 'Segoe UI', Tahoma, Geneva, Verdana, sans-serif;
color: #D1D5DB; /* Pale stone gray */
line-height: 1.6;
margin: 0;
padding: 0;
background-color: #0A0C10; /* Very dark stormy gray/black */
}
b, strong {
color: #FBBF24; /* Glowing amber/gold */
text-shadow: 0 0 8px rgba(251, 191, 36, 0.4);
}
.awakening-text {
color: #FEF3C7; /* Pale inner-eye yellow */
position: relative;
z-index: 2;
margin-left: 0.2em;
text-shadow: 0 0 15px #F59E0B, 0 0 30px #B45309; /* Deep fiery orange/gold glow */
font-size: 1.6rem;
letter-spacing: 1px;
font-weight: 600;
}
/* Section styling */
.section-container {
background-color: rgba(17, 24, 39, 0.85); /* Dark slate rock */
margin-bottom: 30px;
position: relative;
overflow: hidden;
border-bottom: 1px solid #78350F; /* Dark bronze/earth */
box-shadow: 0 4px 20px rgba(0, 0, 0, 0.6);
}
.section-header {
display: flex;
align-items: center;
background-color: rgba(245, 158, 11, 0.05); /* Faint amber tint */
padding: 10px 20px;
border-top: 1px solid rgba(120, 53, 15, 0.4);
}
.section-indicator {
width: 8px;
height: 20px;
background-color: #F59E0B; /* Amber eye color */
margin-right: 15px;
box-shadow: 0 0 10px rgba(245, 158, 11, 0.6);
border-radius: 2px;
}
.section-title {
font-family: 'Georgia', 'Times New Roman', serif; /* Ancient tome feel */
color: #FDE68A; /* Light gold */
font-size: 1.4rem;
margin: 0;
letter-spacing: 1px;
font-weight: 400;
text-transform: capitalize;
}
.section-content {
padding: 20px;
font-family: sans-serif;
color: #D1D5DB;
line-height: 1.6;
}
/* Title styling */
.title-container {
background-color: #050505; /* Pitch black */
position: relative;
overflow: hidden;
margin-bottom: 40px;
border-left: 4px solid #F59E0B; /* Amber pillar */
box-shadow: 0 6px 25px rgba(245, 158, 11, 0.15);
}
.title-wrapper {
position: relative;
z-index: 2;
padding: 25px 20px 30px 30px;
font-family: 'Georgia', 'Times New Roman', serif;
}
.title-main {
color: #FEF3C7;
font-size: 2.0rem;
font-weight: 700;
margin: 0;
letter-spacing: 2px;
display: inline-block;
position: relative;
text-transform: uppercase;
}
.storm-overlay {
position: absolute;
top: 0;
left: 0;
width: 100%;
height: 100%;
/* Dark, brooding radial fog mimicking the eye's aura */
background-image: radial-gradient(circle at 50% 50%, rgba(245, 158, 11, 0.08) 0%, rgba(0,0,0,0.9) 80%);
z-index: 1;
}
/* Subheading styling */
.subheading {
color: #D97706; /* Deep orange */
font-size: 1.1rem;
margin-top: 20px;
margin-bottom: 15px;
font-weight: 400;
border-bottom: 1px dashed rgba(217, 119, 6, 0.4);
display: inline-block;
text-transform: uppercase;
letter-spacing: 1px;
font-family: 'Georgia', 'Times New Roman', serif;
}
/* Links */
a {
color: #FBBF24; /* Amber */
text-decoration: none;
transition: color 0.3s ease, text-shadow 0.3s ease;
}
a:hover {
text-decoration: underline;
color: #FDE68A; /* Brighter gold */
text-shadow: 0 0 8px rgba(251, 191, 36, 0.5);
}
/* Container */
.container {
max-width: 1200px;
margin: 20px auto;
padding: 40px 20px;
background-color: #0D1117; /* Deep stormy night */
background-image:
radial-gradient(circle at 15% 85%, rgba(120, 53, 15, 0.1) 0%, transparent 50%),
radial-gradient(circle at 85% 15%, rgba(245, 158, 11, 0.05) 0%, transparent 50%);
min-height: calc(100vh - 40px);
border: 1px solid #1F2937; /* Dark stone border */
border-radius: 8px;
box-shadow: 0 8px 40px rgba(0, 0, 0, 0.9), inset 0 0 20px rgba(0, 0, 0, 0.5);
}
/* Code blocks */
pre {
background-color: #050505; /* Pitch black */
border: 1px solid #1F2937; /* Dark stone */
border-left: 3px solid #92400E; /* Dark orange/brown */
padding: 15px;
border-radius: 4px;
color: #D1D5DB;
overflow-x: auto;
}
code {
font-family: 'Courier New', Courier, monospace;
color: #FBBF24; /* Amber */
background-color: rgba(245, 158, 11, 0.08);
padding: 2px 4px;
border-radius: 3px;
}
pre code {
color: #00FFFF;
background-color: transparent;
padding: 0;
}
</style>
<html lang="en">
<head>
<meta charset="UTF-8">
<meta name="viewport" content="width=device-width, initial-scale=1.0">
<title>🦑 KrakenSakura Maelström 12B v1</title>
</head>
<body>
<div class="container">
<div class="title-container">
<div class="storm-overlay"></div>
<div class="title-wrapper">
<h2 class="title-main">
<span class="awakening-text">🦑 KrakenSakura Maelström 12B v1</span>
</h2>
</div>
</div>
<img src="https://cdn-uploads.huggingface.co/production/uploads/68e840caa318194c44ec2a04/ssPstIfsBZ-lw4Skif0w7.png"
alt="KrakenSakura Maelström"
style="display: block; margin: 0 auto 30px auto; max-width: 100%; height: auto; border-radius: 5px; border: 1px solid #1F2937; box-shadow: 0 0 25px rgba(245, 158, 11, 0.15);">
<div class="section-container">
<div class="section-header">
<div class="section-indicator"></div>
<h2 class="section-title">Overview</h2>
</div>
<div class="section-content"><font face="verdana">
This is a merge of pre-trained language models created using <a href="https://github.com/cg123/mergekit">mergekit</a>.<br><br><b>KrakenSakura</b> is fully uncensored, no jailbreaks or ablation needed. This model produces unique output and is calibrated for <b>NSFW/RP</b>. Unlike most Nemo models, KrakenSakura avoids generic "Lily" stories and instead wrote about "a brave knight named Sir Galahad".<br><br><b>Note:</b> I only saw refusals when using high temps. Lower the temp or re-roll if you encounter censorship.<br><br>For instruct tag preset use either <b>ChatML</b> or <b>Mistral Tekken/NonTekken</b>. ChatML is a bit more stable and recommended, while Tekken/NonTekken is slightly more creative, but at the cost of occasional errors such as inserting random russian words like <code>командира</code>. Experiment to see which template works best for your use case.
</div>
</div>
<div class="section-container">
<div class="section-header">
<div class="section-indicator"></div>
<h2 class="section-title">Merge Details</h2>
</div>
<div class="section-content"><font face="verdana">
<b>Merge Methods</b><br>
This model was synthesized using a complex multi-stage process involving the following 18 methods:
<ul>
<li><a href="https://en.wikipedia.org/wiki/Slerp">nuslerp</a></li>
<li><a href="https://github.com/arcee-ai/mergekit/blob/main/docs/merge_methods.md">passthrough</a></li>
<li><a href="https://huggingface.co/24B-Suite/Mergedonia-Suite-24B-v1/discussions/2">pdq</a></li>
<li><a href="https://arxiv.org/abs/2406.11617">della</a></li>
<li><a href="https://huggingface.co/24B-Suite/Mergedonia-Suite-24B-v1/discussions/2">chiral_qhe</a></li>
<li><a href="https://www.arcee.ai/blog/meet-mergekit-v0-1-arcee-fusion-expanded-model-support-multi-gpu-acceleration">arcee_fusion</a></li>
<li><a href="https://huggingface.co/alchemonaut/QuartetAnemoi-70B-t0.0001">nearswap</a></li>
<li><a href="https://en.wikipedia.org/wiki/Slerp">multislerp</a></li>
<li><a href="https://arxiv.org/abs/2406.11617">della_linear</a></li>
<li><a href="https://en.wikipedia.org/wiki/Karcher_mean">karcher</a></li>
<li><a href="https://huggingface.co/24B-Suite/Mergedonia-Suite-24B-v1/discussions/2">flux</a></li>
<li><a href="https://huggingface.co/datasets/OccultAI/Script_Tests/discussions/1">rsce</a></li>
<li><a href="https://huggingface.co/24B-Suite/Mergedonia-Suite-24B-v1/discussions/2">magic</a></li>
<li><a href="https://huggingface.co/Naphula-Archives/arcee_multifusion_prototype-24B-Q8_0-GGUF">arcee_multifusion</a></li>
<li><a href="https://arxiv.org/abs/2311.03099">dare_linear</a></li>
<li><a href="https://huggingface.co/blog/grimjim/delerp-merge-method">delerp</a></li>
<li><a href="https://huggingface.co/24B-Suite/Mergedonia-Suite-24B-v1/discussions/2">cvs</a></li>
<li><a href="https://huggingface.co/24B-Suite/Mergedonia-Suite-24B-v1/discussions/2">delerp_della</a></li>
</ul>
<br>The <a href="https://huggingface.co/spaces/Naphula/model_tools/blob/main/graph_v18.py">graph_v18.py</a> patch was helpful to use 8GB VRAM for acceleration.
<hr>
<b>Models Merged</b><br>
The following 38 models were alchemized into this merge:<br><br>
<details>
<summary style="cursor: pointer; color: #FBBF24; font-weight: bold;">Show 38 Donor Models</summary>
<ul>
<li><a href="https://huggingface.co/allura-org/Tlacuilo-12B">allura-org/Tlacuilo-12B</a></li>
<li><a href="https://huggingface.co/ChaoticNeutrals/Captain_Eris_Noctis-12B-v0.420">ChaoticNeutrals/Captain_Eris_Noctis-12B-v0.420</a></li>
<li><a href="https://huggingface.co/DavidAU/Mistral-Nemo-2407-12B-Thinking-Claude-Gemini-GPT5.2-Uncensored-HERETIC">DavidAU/Mistral-Nemo-2407-12B-Thinking-Claude-Gemini-GPT5.2-Uncensored-HERETIC</a></li>
<li><a href="https://huggingface.co/DavidAU/MN-Dark-Planet-TITAN-12B">DavidAU/MN-Dark-Planet-TITAN-12B</a></li>
<li><a href="https://huggingface.co/DavidAU/MN-GRAND-Gutenberg-Lyra4-Lyra-12B-DARKNESS">DavidAU/MN-GRAND-Gutenberg-Lyra4-Lyra-12B-DARKNESS</a></li>
<li><a href="https://huggingface.co/DreadPoor/Famino-12B-Model_Stock">DreadPoor/Famino-12B-Model_Stock</a></li>
<li><a href="https://huggingface.co/EldritchLabs/Cactus-Dream-Horror-12B">EldritchLabs/Cactus-Dream-Horror-12B</a></li>
<li><a href="https://huggingface.co/EldritchLabs/Human-Like-Mistral-Nemo-Instruct-2407-MPOA">EldritchLabs/Human-Like-Mistral-Nemo-Instruct-2407-MPOA</a></li>
<li><a href="https://huggingface.co/EldritchLabs/Kraken-Karcher-12B-v1">EldritchLabs/Kraken-Karcher-12B-v1</a></li>
<li><a href="https://huggingface.co/EldritchLabs/MN-12B-Mag-Mell-R1-Uncensored-Scale1.2">EldritchLabs/MN-12B-Mag-Mell-R1-Uncensored-Scale1.2</a></li>
<li><a href="https://huggingface.co/EldritchLabs/MN-12B-RP-Ink-Longform-MPOA">EldritchLabs/MN-12B-RP-Ink-Longform-MPOA</a></li>
<li><a href="https://huggingface.co/Epiculous/Violet_Twilight-v0.2">Epiculous/Violet_Twilight-v0.2</a></li>
<li><a href="https://huggingface.co/IggyLux/MN-VelvetCafe-RP-12B-V2">IggyLux/MN-VelvetCafe-RP-12B-V2</a></li>
<li><a href="https://huggingface.co/inflatebot/MN-12B-Mag-Mell-R1">inflatebot/MN-12B-Mag-Mell-R1</a></li>
<li><a href="https://huggingface.co/LatitudeGames/Muse-12B">LatitudeGames/Muse-12B</a></li>
<li><a href="https://huggingface.co/LatitudeGames/Wayfarer-2-12B">LatitudeGames/Wayfarer-2-12B</a></li>
<li><a href="https://huggingface.co/mistralai/Mistral-Nemo-Instruct-2407">mistralai/Mistral-Nemo-Instruct-2407</a></li>
<li><a href="https://huggingface.co/MrRikyz/StarlightMoon-Foxfire-12B">MrRikyz/StarlightMoon-Foxfire-12B</a></li>
<li><a href="https://huggingface.co/MuXodious-/Rocinante-X-12B-v1-absolute-heresy">MuXodious/Rocinante-X-12B-v1-absolute-heresy</a></li>
<li><a href="https://huggingface.co/Naphula/Ancient-Awakening-12B">Naphula/Ancient-Awakening-12B</a></li>
<li><a href="https://huggingface.co/Naphula/Riemannian-Redshift-12B-v1">Naphula/Riemannian-Redshift-12B-v1</a></li>
<li><a href="https://huggingface.co/ohyeah1/Violet-Lyra-Gutenberg-v2">ohyeah1/Violet-Lyra-Gutenberg-v2</a></li>
<li><a href="https://huggingface.co/PocketDoc/Dans-SakuraKaze-V1.0.0-12b">PocketDoc/Dans-SakuraKaze-V1.0.0-12b</a></li>
<li><a href="https://huggingface.co/PygmalionAI/Pygmalion-3-12B">PygmalionAI/Pygmalion-3-12B</a></li>
<li><a href="https://huggingface.co/rAIfle/Questionable-MN-bf16">rAIfle/Questionable-MN-bf16</a></li>
<li><a href="https://huggingface.co/ReadyArt/Dark-Nexus-12B-v2.0">ReadyArt/Dark-Nexus-12B-v2.0</a></li>
<li><a href="https://huggingface.co/ReadyArt/Forgotten-Safeword-12B-v4.0">ReadyArt/Forgotten-Safeword-12B-v4.0</a></li>
<li><a href="https://huggingface.co/redrix/GodSlayer-12B-ABYSS">redrix/GodSlayer-12B-ABYSS</a></li>
<li><a href="https://huggingface.co/Retreatcost/Chrysologus-12B">Retreatcost/Chrysologus-12B</a></li>
<li><a href="https://huggingface.co/Retreatcost/Impish-LongPen-12B">Retreatcost/Impish-LongPen-12B</a></li>
<li><a href="https://huggingface.co/Retreatcost/KansenSakura-Conflagration-RP-12b">Retreatcost/KansenSakura-Conflagration-RP-12b</a></li>
<li><a href="https://huggingface.co/SicariusSicariiStuff/Impish_Bloodmoon_12B">SicariusSicariiStuff/Impish_Bloodmoon_12B</a></li>
<li><a href="https://huggingface.co/Sorihon/Celestial-Queen-12B-Heretic">Sorihon/Celestial-Queen-12B-Heretic</a></li>
<li><a href="https://huggingface.co/SuperbEmphasis/MN-12b-RP-Ink-RP-Longform">SuperbEmphasis/MN-12b-RP-Ink-RP-Longform</a></li>
<li><a href="https://huggingface.co/SuperbEmphasis/Omega-Darker_The-Final-Directive-Longform-Stage2-ERP-12B-v0.2">SuperbEmphasis/Omega-Darker_The-Final-Directive-Longform-Stage2-ERP-12B-v0.2</a></li>
<li><a href="https://huggingface.co/TheDrummer/Rocinante-X-12B-v1">TheDrummer/Rocinante-X-12B-v1</a></li>
<li><a href="https://huggingface.co/Vortex5/Aurora-Mirage-12B">Vortex5/Aurora-Mirage-12B</a></li>
<li><a href="https://huggingface.co/Vortex5/Prototype-X-12b">Vortex5/Prototype-X-12b</a></li>
</ul>
</div>
</details>
</div>
<div class="section-container">
<div class="section-header">
<div class="section-indicator"></div>
<h2 class="section-title">Merge Pipeline & Configuration</h2>
</div>
<div class="section-content">
<p><b>🦑 KrakenSakura Maelström 12B v1</b> unites several methods and 38 models into one. This is a highly experimental merge that required 41 steps to build.</p>
🔑 Here is the "master key" for the 18 nuslerps and passthrough:
<ul>
<li> <b>0-5</b> <code>IggyLux/MN-VelvetCafe-RP-12B-V2</code></li>
<li> <b>5-14</b> SLERP4 = SLERP1 (<code>inflatebot/MN-12B-Mag-Mell-R1</code> + <code>PygmalionAI/Pygmalion-3-12B</code>) + SLERP3 (<code>TheDrummer/Rocinante-X-12B-v1</code> + SLERP2 (<code>DavidAU/MN-Dark-Planet-TITAN-12B</code> + <code>DavidAU/MN-GRAND-Gutenberg-Lyra4-Lyra-12B-DARKNESS</code>)</li>
<li> <b>14-22</b> SLERP7 = <code>SicariusSicariiStuff/Impish_Bloodmoon_12B</code> + SLERP6 (<code>MrRikyz/StarlightMoon-Foxfire-12B</code> + SLERP5 (<code>Retreatcost/Impish-LongPen-12B</code> + <code>Retreatcost/Chrysologus-12B</code>))</li>
<li> <b>22-29</b> SLERP12 = SLERP9 (<code>ReadyArt/Dark-Nexus-12B-v2.0</code> + SLERP8 (<code>ReadyArt/Forgotten-Safeword-12B-v4.0</code> + <code>SuperbEmphasis/Omega-Darker_The-Final-Directive-Longform-Stage2-ERP-12B-v0.2</code>)) + SLERP11 (<code>ohyeah1/Violet-Lyra-Gutenberg-v2</code> + SLERP10 (<code>Epiculous/Violet_Twilight-v0.2</code> + <code>ChaoticNeutrals/Captain_Eris_Noctis-12B-v0.420</code>))</li>
<li> <b>29-34</b> SLERP16 = SLERP13 (<code>Vortex5/Aurora-Mirage-12B</code> + <code>Vortex5/Prototype-X-12b</code>) + SLERP15 (<code>Naphula/Ancient-Awakening-12B</code> + SLERP14 (<code>redrix/GodSlayer-12B-ABYSS</code> + <code>Retreatcost/KansenSakura-Conflagration-RP-12b</code>))</li>
<li> <b>34-39</b> SLERP18 = <code>LatitudeGames/Muse-12B</code> + SLERP17 (<code>LatitudeGames/Wayfarer-2-12B</code> + <code>allura-org/Tlacuilo-12B</code>)</li>
<li> <b>39-40</b> <code>PocketDoc/Dans-SakuraKaze-V1.0.0-12b</code></li>
</ul>
<hr>
<details>
<summary style="cursor: pointer; color: #FBBF24; font-weight: bold;">Show 41 YAML Configs</summary>
<h3 class="subheading">Stage 1: nuslerp1</h3>
<pre><code>architecture: MistralForCausalLM
merge_method: nuslerp
dtype: float32
out_dtype: bfloat16
models:
- model: B:\12B\models--inflatebot--MN-12B-Mag-Mell-R1
parameters:
weight: 0.5
- model: B:\12B\models--PygmalionAI--Pygmalion-3-12B
parameters:
weight: 0.5
parameters:
tokenizer:
source: union
chat_template: auto</code></pre>
<h3 class="subheading">Stage 2: nuslerp2</h3>
<pre><code>architecture: MistralForCausalLM
merge_method: nuslerp
dtype: float32
out_dtype: bfloat16
models:
- model: B:\12B\models--DavidAU--MN-Dark-Planet-TITAN-12B
parameters:
weight: 0.5
- model: B:\12B\models--DavidAU--MN-GRAND-Gutenberg-Lyra4-Lyra-12B-DARKNESS
parameters:
weight: 0.5
parameters:
tokenizer:
source: union
chat_template: auto</code></pre>
<h3 class="subheading">Stage 3: nuslerp3</h3>
<pre><code>architecture: MistralForCausalLM
merge_method: nuslerp
dtype: float32
out_dtype: bfloat16
models:
- model: B:\12B\models--TheDrummer--Rocinante-X-12B-v1
parameters:
weight: 0.5
- model: B:\12B\SLERP2
parameters:
weight: 0.5
parameters:
tokenizer:
source: union
chat_template: auto</code></pre>
<h3 class="subheading">Stage 4: nuslerp4</h3>
<pre><code>architecture: MistralForCausalLM
merge_method: nuslerp
dtype: float32
out_dtype: bfloat16
models:
- model: B:\12B\SLERP1
parameters:
weight: 0.5
- model: B:\12B\SLERP2
parameters:
weight: 0.5
parameters:
tokenizer:
source: union
chat_template: auto</code></pre>
<h3 class="subheading">Stage 5: nuslerp5</h3>
<pre><code>architecture: MistralForCausalLM
merge_method: nuslerp
dtype: float32
out_dtype: bfloat16
models:
- model: B:\12B\models--Retreatcost--Impish-LongPen-12B
parameters:
weight: 0.5
- model: B:\12B\models--Retreatcost--Chrysologus-12B
parameters:
weight: 0.5
parameters:
tokenizer:
source: union
chat_template: auto</code></pre>
<h3 class="subheading">Stage 6: nuslerp6</h3>
<pre><code>architecture: MistralForCausalLM
merge_method: nuslerp
dtype: float32
out_dtype: bfloat16
models:
- model: B:\12B\models--MrRikyz--StarlightMoon-Foxfire-12B
parameters:
weight: 0.5
- model: B:\12B\SLERP5
parameters:
weight: 0.5
parameters:
tokenizer:
source: union
chat_template: auto</code></pre>
<h3 class="subheading">Stage 7: nuslerp7</h3>
<pre><code>architecture: MistralForCausalLM
merge_method: nuslerp
dtype: float32
out_dtype: bfloat16
models:
- model: B:\12B\models--SicariusSicariiStuff--Impish_Bloodmoon_12B
parameters:
weight: 0.5
- model: B:\12B\SLERP6
parameters:
weight: 0.5
parameters:
tokenizer:
source: union
chat_template: auto</code></pre>
<h3 class="subheading">Stage 8: nuslerp8</h3>
<pre><code>architecture: MistralForCausalLM
merge_method: nuslerp
dtype: float32
out_dtype: bfloat16
models:
- model: B:\12B\models--ReadyArt--Forgotten-Safeword-12B-v4.0
parameters:
weight: 0.5
- model: B:\12B\models--ReadyArt--Dark-Nexus-12B-v2.0
parameters:
weight: 0.5
parameters:
tokenizer:
source: union
chat_template: auto</code></pre>
<h3 class="subheading">Stage 9: nuslerp9</h3>
<pre><code>architecture: MistralForCausalLM
merge_method: nuslerp
dtype: float32
out_dtype: bfloat16
models:
- model: B:\12B\models--SuperbEmphasis--Omega-Darker_The-Final-Directive-Longform-Stage2-ERP-12B-v0.2
parameters:
weight: 0.5
- model: B:\12B\SLERP8
parameters:
weight: 0.5
parameters:
tokenizer:
source: union
chat_template: auto</code></pre>
<h3 class="subheading">Stage 10: nuslerp10</h3>
<pre><code>architecture: MistralForCausalLM
merge_method: nuslerp
dtype: float32
out_dtype: bfloat16
models:
- model: B:\12B\models--Epiculous--Violet_Twilight-v0.2
parameters:
weight: 0.5
- model: B:\12B\models--ChaoticNeutrals--Captain_Eris_Noctis-12B-v0.420
parameters:
weight: 0.5
parameters:
tokenizer:
source: union
chat_template: auto</code></pre>
<h3 class="subheading">Stage 11: nuslerp11</h3>
<pre><code>architecture: MistralForCausalLM
merge_method: nuslerp
dtype: float32
out_dtype: bfloat16
models:
- model: B:\12B\models--ohyeah1--Violet-Lyra-Gutenberg-v2
parameters:
weight: 0.5
- model: B:\12B\SLERP10
parameters:
weight: 0.5
parameters:
tokenizer:
source: union
chat_template: auto</code></pre>
<h3 class="subheading">Stage 12: nuslerp12</h3>
<pre><code>architecture: MistralForCausalLM
merge_method: nuslerp
dtype: float32
out_dtype: bfloat16
models:
- model: B:\12B\SLERP9
parameters:
weight: 0.5
- model: B:\12B\SLERP11
parameters:
weight: 0.5
parameters:
tokenizer:
source: union
chat_template: auto</code></pre>
<h3 class="subheading">Stage 13: nuslerp13</h3>
<pre><code>architecture: MistralForCausalLM
merge_method: nuslerp
dtype: float32
out_dtype: bfloat16
models:
- model: B:\12B\models--Vortex5--Aurora-Mirage-12B
parameters:
weight: 0.5
- model: B:\12B\models--Vortex5--Prototype-X-12b
parameters:
weight: 0.5
parameters:
tokenizer:
source: union
chat_template: auto</code></pre>
<h3 class="subheading">Stage 14: nuslerp14</h3>
<pre><code>architecture: MistralForCausalLM
merge_method: nuslerp
dtype: float32
out_dtype: bfloat16
models:
- model: B:\12B\models--redrix--GodSlayer-12B-ABYSS
parameters:
weight: 0.5
- model: B:\12B\models--Retreatcost--KansenSakura-Conflagration-RP-12b
parameters:
weight: 0.5
parameters:
tokenizer:
source: union
chat_template: auto</code></pre>
<h3 class="subheading">Stage 15: nuslerp15</h3>
<pre><code>architecture: MistralForCausalLM
merge_method: nuslerp
dtype: float32
out_dtype: bfloat16
models:
- model: B:\12B\models--Naphula--Ancient-Awakening-12B
parameters:
weight: 0.5
- model: B:\12B\SLERP14
parameters:
weight: 0.5
parameters:
tokenizer:
source: union
chat_template: auto</code></pre>
<h3 class="subheading">Stage 16: nuslerp16</h3>
<pre><code>architecture: MistralForCausalLM
merge_method: nuslerp
dtype: float32
out_dtype: bfloat16
models:
- model: B:\12B\SLERP13
parameters:
weight: 0.5
- model: B:\12B\SLERP15
parameters:
weight: 0.5
parameters:
tokenizer:
source: union
chat_template: auto</code></pre>
<h3 class="subheading">Stage 17: nuslerp17</h3>
<pre><code>architecture: MistralForCausalLM
merge_method: nuslerp
dtype: float32
out_dtype: bfloat16
models:
- model: B:\12B\models--LatitudeGames--Wayfarer-2-12B
parameters:
weight: 0.5
- model: B:\12B\!models--allura-org--Tlacuilo-12B
parameters:
weight: 0.5
parameters:
tokenizer:
source: union
chat_template: auto</code></pre>
<h3 class="subheading">Stage 18: nuslerp18</h3>
<pre><code>architecture: MistralForCausalLM
merge_method: nuslerp
dtype: float32
out_dtype: bfloat16
models:
- model: B:\12B\models--LatitudeGames--Muse-12B
parameters:
weight: 0.5
- model: B:\12B\SLERP17
parameters:
weight: 0.5
parameters:
tokenizer:
source: union
chat_template: auto</code></pre>
<h3 class="subheading">Stage 19: passthrough1</h3>
<pre><code>architecture: MistralForCausalLM
merge_method: passthrough
slices:
- sources:
- model: B:\12B\models--IggyLux--MN-VelvetCafe-RP-12B-V2
layer_range: [0, 5]
- sources:
- model: B:\12B\SLERP4
layer_range: [5, 14]
- sources:
- model: B:\12B\SLERP7
layer_range: [14, 22]
- sources:
- model: B:\12B\SLERP12
layer_range: [22, 29]
- sources:
- model: B:\12B\SLERP16
layer_range: [29, 35]
- sources:
- model: B:\12B\SLERP18
layer_range: [35, 39]
- sources:
- model: B:\12B\models--PocketDoc--Dans-SakuraKaze-V1.0.0-12b
layer_range: [39, 40]
tokenizer:
source: union
chat_template: auto
dtype: float32
out_dtype: bfloat16</code></pre>
<h3 class="subheading">Stage 20: pdq1</h3>
<pre><code>merge_method: pdq
pdq_base_yaml: B:\12B\19-passthrough\20-pdq5.yml
pdq_base_model: B:\12B\19-passthrough
output_dir: B:\12B\pdq1
base_model: A:\LLM\.cache\12B\models--mistralai--Mistral-Nemo-Instruct-2407
models:
- model: B:\12B\models--IggyLux--MN-VelvetCafe-RP-12B-V2
- model: B:\12B\SLERP4
- model: B:\12B\SLERP7
- model: B:\12B\SLERP12
- model: B:\12B\SLERP16
- model: B:\12B\SLERP18
- model: B:\12B\models--PocketDoc--Dans-SakuraKaze-V1.0.0-12b
parameters:
chi: 0.15
iota: 0.1
nu: 24
gamma: 1.0
zeta: 16
sigma: 0.5
density: 0.9
epsilon: 0.099
lambda: 1.0
lazy_unpickle: True
random_seed: 420
name: Stage 20 PDQ</code></pre>
<h3 class="subheading">Stage 21: della1</h3>
<pre><code>architecture: MistralForCausalLM
models:
- model: A:\LLM\.cache\12B\models--mistralai--Mistral-Nemo-Instruct-2407
- model: B:\12B\models--IggyLux--MN-VelvetCafe-RP-12B-V2
parameters:
weight: 0.1
density: 0.8
epsilon: 0.19
- model: B:\12B\SLERP1
parameters:
weight: 0.1
density: 0.8
epsilon: 0.19
- model: B:\12B\SLERP2
parameters:
weight: 0.1
density: 0.8
epsilon: 0.19
- model: B:\12B\SLERP3
parameters:
weight: 0.1
density: 0.8
epsilon: 0.19
- model: B:\12B\SLERP4
parameters:
weight: 0.1
density: 0.8
epsilon: 0.19
- model: B:\12B\SLERP5
parameters:
weight: 0.1
density: 0.8
epsilon: 0.19
- model: B:\12B\SLERP6
parameters:
weight: 0.1
density: 0.8
epsilon: 0.19
- model: B:\12B\SLERP7
parameters:
weight: 0.1
density: 0.8
epsilon: 0.19
- model: B:\12B\SLERP8
parameters:
weight: 0.1
density: 0.8
epsilon: 0.19
- model: B:\12B\SLERP9
parameters:
weight: 0.1
density: 0.8
epsilon: 0.19
- model: B:\12B\SLERP10
parameters:
weight: 0.1
density: 0.8
epsilon: 0.19
- model: B:\12B\SLERP11
parameters:
weight: 0.1
density: 0.8
epsilon: 0.19
- model: B:\12B\SLERP12
parameters:
weight: 0.1
density: 0.8
epsilon: 0.19
- model: B:\12B\SLERP13
parameters:
weight: 0.1
density: 0.8
epsilon: 0.19
- model: B:\12B\SLERP14
parameters:
weight: 0.1
density: 0.8
epsilon: 0.19
- model: B:\12B\SLERP15
parameters:
weight: 0.1
density: 0.8
epsilon: 0.19
- model: B:\12B\SLERP16
parameters:
weight: 0.1
density: 0.8
epsilon: 0.19
- model: B:\12B\SLERP17
parameters:
weight: 0.1
density: 0.8
epsilon: 0.19
- model: B:\12B\SLERP18
parameters:
weight: 0.1
density: 0.8
epsilon: 0.19
- model: B:\12B\models--PocketDoc--Dans-SakuraKaze-V1.0.0-12b
parameters:
weight: 0.1
density: 0.8
epsilon: 0.19
- model: B:\12B\19-passthrough
parameters:
weight: 0.1
density: 0.9
epsilon: 0.09
- model: B:\12B\pdq1
parameters:
weight: 0.1
density: 0.9
epsilon: 0.09
merge_method: della
base_model: A:\LLM\.cache\12B\models--mistralai--Mistral-Nemo-Instruct-2407
parameters:
lambda: 1.0
normalize: false
int8_mask: false
tokenizer:
source: union
chat_template: auto
dtype: float32
out_dtype: bfloat16</code></pre>
<h3 class="subheading">Stage 22: pdq2</h3>
<pre><code>merge_method: pdq
pdq_base_yaml: B:\12B\19-passthrough\22-pdq20.yml
pdq_base_model: B:\12B\19-passthrough
output_dir: B:\12B\pdq3
base_model: A:\LLM\.cache\12B\models--mistralai--Mistral-Nemo-Instruct-2407
models:
- model: B:\12B\models--IggyLux--MN-VelvetCafe-RP-12B-V2
- model: B:\12B\SLERP1
- model: B:\12B\SLERP2
- model: B:\12B\SLERP3
- model: B:\12B\SLERP4
- model: B:\12B\SLERP5
- model: B:\12B\SLERP6
- model: B:\12B\SLERP7
- model: B:\12B\SLERP8
- model: B:\12B\SLERP9
- model: B:\12B\SLERP10
- model: B:\12B\SLERP11
- model: B:\12B\SLERP12
- model: B:\12B\SLERP13
- model: B:\12B\SLERP14
- model: B:\12B\SLERP15
- model: B:\12B\SLERP16
- model: B:\12B\SLERP17
- model: B:\12B\SLERP18
- model: B:\12B\models--PocketDoc--Dans-SakuraKaze-V1.0.0-12b
parameters:
chi: 0.15
iota: 0.1
nu: 24
gamma: 1.0
zeta: 16
sigma: 0.5
density: 0.9
epsilon: 0.099
lambda: 1.0
lazy_unpickle: True
random_seed: 420
name: Stage 22 PDQ</code></pre>
<h3 class="subheading">Stage 23: nuslerp19</h3>
<pre><code>architecture: MistralForCausalLM
merge_method: nuslerp
dtype: float32
out_dtype: bfloat16
models:
- model: B:\12B\pdq1
parameters:
weight: 0.5
- model: B:\12B\pdq3
parameters:
weight: 0.5
parameters:
tokenizer:
source: union
chat_template: auto</code></pre>
<h3 class="subheading">Stage 24: chiral_qhe1</h3>
<pre><code>merge_method: chiral_qhe
models:
- model: B:\12B\models--IggyLux--MN-VelvetCafe-RP-12B-V2
- model: B:\12B\SLERP1
- model: B:\12B\SLERP2
- model: B:\12B\SLERP3
- model: B:\12B\SLERP4
- model: B:\12B\SLERP5
- model: B:\12B\SLERP6
- model: B:\12B\SLERP7
- model: B:\12B\SLERP8
- model: B:\12B\SLERP9
- model: B:\12B\SLERP10
- model: B:\12B\SLERP11
- model: B:\12B\SLERP12
- model: B:\12B\SLERP13
- model: B:\12B\SLERP14
- model: B:\12B\SLERP15
- model: B:\12B\SLERP16
- model: B:\12B\SLERP17
- model: B:\12B\SLERP18
- model: B:\12B\models--PocketDoc--Dans-SakuraKaze-V1.0.0-12b
- model: B:\12B\19-passthrough
- model: B:\12B\21-Della
- model: B:\12B\SLERP-PDQ
parameters:
chi: 0.15
iota: 0.1
nu: 24
gamma: 1.0
zeta: 16
sigma: 0.5
coherence: 0.5
dtype: float32
out_dtype: bfloat16
tokenizer:
source: union
chat_template: chatml</code></pre>
<h3 class="subheading">Stage 25: arcee_fusion1</h3>
<pre><code>merge_method: arcee_fusion
base_model: B:\12B\21-della
models:
- model: B:\12B\21-della
- model: B:\12B\24-qhe
parameters:
tukey_fence: 1.5
dtype: float32
out_dtype: bfloat16
tokenizer:
source: base
chat_template: "chatml"</code></pre>
<h3 class="subheading">Stage 26: nearswap1</h3>
<pre><code>merge_method: nearswap
base_model: B:\12B\23-arcee
models:
- model: B:\12B\models--IggyLux--MN-VelvetCafe-RP-12B-V2
parameters:
t:
# We use a "U-Shape" or "End-Heavy" gradient
# High at the start (Instruction following)
# Zero in the middle (Preserve DELLA/QHE/ARCEE creativity)
# High at the end (EOS/Termination logic)
- filter: self_attn
value: [0.0005, 0.0002, 0.0001, 0.0000, 0.0000, 0.0002, 0.0005]
- filter: mlp
value: [0.0003, 0.0001, 0.0000, 0.0000, 0.0000, 0.0001, 0.0003]
- value: 0.0002 # Catch-all for layernorms and embeddings
dtype: bfloat16
tokenizer:
source: B:\12B\models--IggyLux--MN-VelvetCafe-RP-12B-V2
chat_template: chatml</code></pre>
<h3 class="subheading">Stage 27: passthrough2</h3>
<pre><code>architecture: MistralForCausalLM
merge_method: passthrough
slices:
- sources:
- model: B:\12B\models--IggyLux--MN-VelvetCafe-RP-12B-V2
layer_range: [0, 3]
- sources:
- model: B:\12B\models--SicariusSicariiStuff--Impish_Bloodmoon_12B
layer_range: [3, 5]
- sources:
- model: B:\12B\26-nearswap
layer_range: [5, 37]
- sources:
- model: B:\12B\models--inflatebot--MN-12B-Mag-Mell-R1
layer_range: [37, 38]
- sources:
- model: B:\12B\models--LatitudeGames--Muse-12B
layer_range: [38, 39]
- sources:
- model: B:\12B\models--PocketDoc--Dans-SakuraKaze-V1.0.0-12b
layer_range: [39, 40]
tokenizer:
source: B:\12B\models--IggyLux--MN-VelvetCafe-RP-12B-V2
chat_template: auto
dtype: float32
out_dtype: bfloat16</code></pre>
<h3 class="subheading">Stage 28: della2</h3>
<pre><code>architecture: MistralForCausalLM
models:
- model: A:\LLM\.cache\12B\models--mistralai--Mistral-Nemo-Instruct-2407
- model: B:\12B\!models--allura-org--Tlacuilo-12B
parameters:
weight: 0.06
density: 0.7
epsilon: 0.29
- model: B:\12B\models--ChaoticNeutrals--Captain_Eris_Noctis-12B-v0.420
parameters:
weight: 0.06
density: 0.7
epsilon: 0.29
- model: B:\12B\models--DavidAU--MN-Dark-Planet-TITAN-12B
parameters:
weight: 0.06
density: 0.7
epsilon: 0.29
- model: B:\12B\models--DavidAU--MN-GRAND-Gutenberg-Lyra4-Lyra-12B-DARKNESS
parameters:
weight: 0.06
density: 0.7
epsilon: 0.29
- model: B:\12B\models--DreadPoor--Famino-12B-Model_Stock
parameters:
weight: 0.06
density: 0.7
epsilon: 0.29
- model: B:\12B\models--EldritchLabs--Cactus-Dream-Horror-12B
parameters:
weight: 0.06
density: 0.7
epsilon: 0.29
- model: B:\12B\models--EldritchLabs--Kraken-Karcher-12B-v1
parameters:
weight: 0.06
density: 0.7
epsilon: 0.29
- model: B:\12B\models--Epiculous--Violet_Twilight-v0.2
parameters:
weight: 0.06
density: 0.7
epsilon: 0.29
- model: B:\12B\models--IggyLux--MN-VelvetCafe-RP-12B-V2
parameters:
weight: 0.06
density: 0.7
epsilon: 0.29
- model: B:\12B\models--inflatebot--MN-12B-Mag-Mell-R1
parameters:
weight: 0.06
density: 0.7
epsilon: 0.29
- model: B:\12B\models--LatitudeGames--Muse-12B
parameters:
weight: 0.06
density: 0.7
epsilon: 0.29
- model: B:\12B\models--LatitudeGames--Wayfarer-2-12B
parameters:
weight: 0.06
density: 0.7
epsilon: 0.29
- model: B:\12B\models--MrRikyz--StarlightMoon-Foxfire-12B
parameters:
weight: 0.06
density: 0.7
epsilon: 0.29
- model: B:\12B\models--Naphula--Ancient-Awakening-12B
parameters:
weight: 0.06
density: 0.7
epsilon: 0.29
- model: B:\12B\models--Naphula--Riemannian-Redshift-12B-v1
parameters:
weight: 0.06
density: 0.7
epsilon: 0.29
- model: B:\12B\models--ohyeah1--Violet-Lyra-Gutenberg-v2
parameters:
weight: 0.06
density: 0.7
epsilon: 0.29
- model: B:\12B\models--PocketDoc--Dans-SakuraKaze-V1.0.0-12b
parameters:
weight: 0.06
density: 0.7
epsilon: 0.29
- model: B:\12B\models--PygmalionAI--Pygmalion-3-12B
parameters:
weight: 0.06
density: 0.7
epsilon: 0.29
- model: B:\12B\models--rAIfle--Questionable-MN-bf16
parameters:
weight: 0.06
density: 0.7
epsilon: 0.29
- model: B:\12B\models--ReadyArt--Dark-Nexus-12B-v2.0
parameters:
weight: 0.06
density: 0.7
epsilon: 0.29
- model: B:\12B\models--ReadyArt--Forgotten-Safeword-12B-v4.0
parameters:
weight: 0.06
density: 0.7
epsilon: 0.29
- model: B:\12B\models--redrix--GodSlayer-12B-ABYSS
parameters:
weight: 0.06
density: 0.7
epsilon: 0.29
- model: B:\12B\models--Retreatcost--Chrysologus-12B
parameters:
weight: 0.06
density: 0.9
epsilon: 0.09
- model: B:\12B\models--Retreatcost--Impish-LongPen-12B
parameters:
weight: 0.06
density: 0.9
epsilon: 0.09
- model: B:\12B\models--Retreatcost--KansenSakura-Conflagration-RP-12b
parameters:
weight: 0.06
density: 0.9
epsilon: 0.09
- model: B:\12B\models--SicariusSicariiStuff--Impish_Bloodmoon_12B
parameters:
weight: 0.06
density: 0.9
epsilon: 0.09
- model: B:\12B\models--SuperbEmphasis--MN-12b-RP-Ink-RP-Longform
parameters:
weight: 0.06
density: 0.9
epsilon: 0.09
- model: B:\12B\models--SuperbEmphasis--Omega-Darker_The-Final-Directive-Longform-Stage2-ERP-12B-v0.2
parameters:
weight: 0.06
density: 0.9
epsilon: 0.09
- model: B:\12B\models--TheDrummer--Rocinante-X-12B-v1
parameters:
weight: 0.06
density: 0.9
epsilon: 0.09
- model: B:\12B\models--Vortex5--Aurora-Mirage-12B
parameters:
weight: 0.06
density: 0.9
epsilon: 0.09
- model: B:\12B\models--Vortex5--Prototype-X-12b
parameters:
weight: 0.06
density: 0.9
epsilon: 0.09
merge_method: della
base_model: A:\LLM\.cache\12B\models--mistralai--Mistral-Nemo-Instruct-2407
parameters:
lambda: 1.0
normalize: false
int8_mask: false
tokenizer:
source: union
chat_template: "chatml"
dtype: float32
out_dtype: bfloat16</code></pre>
<h3 class="subheading">Stage 29: multislerp1</h3>
<pre><code>architecture: MistralForCausalLM
merge_method: multislerp
models:
- model: B:\12B\SLERP1
parameters:
weight: 0.1
- model: B:\12B\SLERP2
parameters:
weight: 0.1
- model: B:\12B\SLERP3
parameters:
weight: 0.1
- model: B:\12B\SLERP4
parameters:
weight: 0.1
- model: B:\12B\SLERP5
parameters:
weight: 0.1
- model: B:\12B\SLERP6
parameters:
weight: 0.1
- model: B:\12B\SLERP7
parameters:
weight: 0.1
- model: B:\12B\SLERP8
parameters:
weight: 0.1
- model: B:\12B\SLERP9
parameters:
weight: 0.1
- model: B:\12B\SLERP10
parameters:
weight: 0.1
- model: B:\12B\SLERP11
parameters:
weight: 0.1
- model: B:\12B\SLERP12
parameters:
weight: 0.1
- model: B:\12B\SLERP13
parameters:
weight: 0.1
- model: B:\12B\SLERP14
parameters:
weight: 0.1
- model: B:\12B\SLERP15
parameters:
weight: 0.1
- model: B:\12B\SLERP16
parameters:
weight: 0.1
- model: B:\12B\SLERP17
parameters:
weight: 0.1
- model: B:\12B\SLERP18
parameters:
weight: 0.1
dtype: float32
out_dtype: bfloat16
parameters:
normalize: false
tokenizer:
source: union
chat_template: auto</code></pre>
<h3 class="subheading">Stage 30: della3</h3>
<pre><code>architecture: MistralForCausalLM
models:
- model: A:\LLM\.cache\12B\models--mistralai--Mistral-Nemo-Instruct-2407
- model: B:\12B\19-passthrough
parameters:
weight: 0.3
density: 0.8
epsilon: 0.19
- model: B:\12B\21-Della
parameters:
weight: 0.22
density: 0.8
epsilon: 0.19
- model: B:\12B\27-passB
parameters:
weight: 0.22
density: 0.8
epsilon: 0.19
- model: B:\12B\28-della
parameters:
weight: 0.22
density: 0.8
epsilon: 0.19
- model: B:\12B\29-multislerp
parameters:
weight: 0.3
density: 0.8
epsilon: 0.19
- model: B:\12B\24-qhe
parameters:
weight: 0.11
density: 0.8
epsilon: 0.19
- model: B:\12B\pdq1
parameters:
weight: 0.11
density: 0.8
epsilon: 0.19
- model: B:\12B\25-arcee
parameters:
weight: 0.11
density: 0.8
epsilon: 0.19
- model: B:\12B\models--SicariusSicariiStuff--Impish_Bloodmoon_12B
parameters:
weight: 0.06
density: 0.5
epsilon: 0.25
- model: B:\12B\models--SuperbEmphasis--MN-12b-RP-Ink-RP-Longform
parameters:
weight: 0.05
density: 0.5
epsilon: 0.25
- model: B:\12B\models--Vortex5--Aurora-Mirage-12B
parameters:
weight: 0.05
density: 0.5
epsilon: 0.25
- model: B:\12B\models--Vortex5--Prototype-X-12b
parameters:
weight: 0.05
density: 0.5
epsilon: 0.25
- model: B:\12B\models--MrRikyz--StarlightMoon-Foxfire-12B
parameters:
weight: 0.05
density: 0.5
epsilon: 0.25
- model: B:\12B\models--Naphula--Ancient-Awakening-12B
parameters:
weight: 0.05
density: 0.5
epsilon: 0.25
- model: B:\12B\models--IggyLux--MN-VelvetCafe-RP-12B-V2
parameters:
weight: 0.05
density: 0.5
epsilon: 0.25
- model: B:\12B\models--PocketDoc--Dans-SakuraKaze-V1.0.0-12b
parameters:
weight: 0.05
density: 0.5
epsilon: 0.25
merge_method: della
base_model: A:\LLM\.cache\12B\models--mistralai--Mistral-Nemo-Instruct-2407
parameters:
lambda: 1.0
normalize: false
int8_mask: false
tokenizer:
source: B:\12B\29-multislerp
chat_template: "chatml"
dtype: float32
out_dtype: bfloat16</code></pre>
<h3 class="subheading">Stage 31: della_linear1</h3>
<pre><code>architecture: MistralForCausalLM
models:
- model: A:\LLM\.cache\12B\models--mistralai--Mistral-Nemo-Instruct-2407
- model: B:\12B\19-passthrough
parameters:
weight: 0.4
density: 0.8
epsilon: 0.19
- model: B:\12B\29-multislerp
parameters:
weight: 0.5
density: 0.8
epsilon: 0.19
- model: B:\12B\21-della
parameters:
weight: 0.1
density: 0.8
epsilon: 0.19
- model: B:\12B\28-della
parameters:
weight: 0.1
density: 0.8
epsilon: 0.19
- model: B:\12B\30-della
parameters:
weight: 0.2
density: 0.8
epsilon: 0.19
- model: B:\12B\24-qhe
parameters:
weight: 0.2
density: 0.8
epsilon: 0.19
- model: B:\12B\pdq1
parameters:
weight: 0.1
density: 0.8
epsilon: 0.19
- model: B:\12B\27-passB
parameters:
weight: 0.4
density: 0.8
epsilon: 0.19
merge_method: della_linear
base_model: A:\LLM\.cache\12B\models--mistralai--Mistral-Nemo-Instruct-2407
parameters:
lambda: 1.0
normalize: false
int8_mask: false
tokenizer:
source: B:\12B\30-della
chat_template: "chatml"
dtype: float32
out_dtype: bfloat16</code></pre>
<h3 class="subheading">Stage 32: nuslerp20</h3>
<pre><code>architecture: MistralForCausalLM
merge_method: nuslerp
dtype: float32
out_dtype: bfloat16
models:
- model: B:\12B\31-della_linear
parameters:
weight: 0.4
- model: B:\12B\29-multislerp
parameters:
weight: 0.6
parameters:
tokenizer:
source: B:\12B\29-multislerp
chat_template: "chatml"</code></pre>
<h3 class="subheading">Stage 33: karcher1</h3>
<pre><code>architecture: MistralForCausalLM
models:
- model: B:\12B\31-della_linear
- model: B:\12B\24-qhe
- model: B:\12B\29-multislerp
merge_method: karcher
dtype: float32
out_dtype: bfloat16
parameters:
tol: 1e-9
max_iter: 1000
tokenizer:
source: B:\12B\31-della_linear
chat_template: auto</code></pre>
<h3 class="subheading">Stage 34: flux1</h3>
<pre><code>models:
- model: B:\12B\33-karcher
- model: B:\12B\32-nuslerp
- model: B:\12B\24-qhe
- model: B:\12B\29-multislerp
- model: B:\12B\31-della_linear
merge_method: flux
parameters:
eta: 1.2
tol: 1.0e-9
max_iter: 1000
kappa: 0.8
mu: 0.5
dtype: float32
out_dtype: bfloat16
tokenizer:
source: B:\12B\31-della_linear
chat_template: auto</code></pre>
<h3 class="subheading">Stage 35: rsce1</h3>
<pre><code>architecture: MistralForCausalLM
merge_method: rsce
dtype: float32
out_dtype: bfloat16
models:
- model: B:\12B\31-della_linear
- model: B:\12B\SLERP1
- model: B:\12B\SLERP2
- model: B:\12B\SLERP3
- model: B:\12B\SLERP4
- model: B:\12B\SLERP5
- model: B:\12B\SLERP6
- model: B:\12B\SLERP7
- model: B:\12B\SLERP8
- model: B:\12B\SLERP9
- model: B:\12B\SLERP10
- model: B:\12B\SLERP11
- model: B:\12B\SLERP12
- model: B:\12B\SLERP13
- model: B:\12B\SLERP14
- model: B:\12B\SLERP15
- model: B:\12B\SLERP16
- model: B:\12B\SLERP17
- model: B:\12B\SLERP18
- model: B:\12B\33-karcher
- model: B:\12B\32-nuslerp
- model: B:\12B\24-qhe
- model: B:\12B\29-multislerp
- model: B:\12B\19-passthrough
- model: B:\12B\27-passB
- model: B:\12B\pdq1
base_model: B:\12B\34-flux
parameters:
select_topk: 0.5
normalize: false
tokenizer:
source: base
chat_template: auto</code></pre>
<h3 class="subheading">Stage 36: magic1</h3>
<pre><code>merge_method: magic
base_model: B:\12B\34-flux
models:
- model: B:\12B\34-flux
- model: B:\12B\31-della_linear
- model: B:\12B\33-karcher
- model: B:\12B\32-nuslerp
- model: B:\12B\24-qhe
- model: B:\12B\29-multislerp
- model: B:\12B\19-passthrough
- model: B:\12B\27-passB
- model: B:\12B\pdq1
- model: B:\12B\35-rsce
parameters:
power: 1.0
creativity: 1.0
filter_topk: 0.5
hierarchy: 0.5
karcher_max_iter: 1000
karcher_tol: 1e-9
karcher_eta: 1.0
inversion_mode: 1
inversion_threshold: 1.0
dtype: float32
out_dtype: bfloat16
tokenizer:
source: base
chat_template: auto
name: Psychosis-14B-v0a-MAGIC</code></pre>
<h3 class="subheading">Stage 37: arcee_multifusion1</h3>
<pre><code>architecture: MistralForCausalLM
merge_method: arcee_multifusion
# ANCHOR: Use Precog as the base.
# Anything not "salient" from the donors will remain Precog logic.
base_model: B:\12B\36-magic
models:
# - model: B:\24B\models--TheDrummer--Precog-24B-v1
# - model: B:\24B\models--mistralai--Magistral-Small-2509\textonly
- model: B:\12B\36-magic
- model: B:\12B\35-rsce
- model: B:\12B\31-della_linear
- model: B:\12B\33-karcher
- model: B:\12B\32-nuslerp
- model: B:\12B\24-qhe
- model: B:\12B\29-multislerp
- model: B:\12B\19-passthrough
- model: B:\12B\27-passB
- model: B:\12B\pdq1
- model: B:\12B\34-flux
parameters:
# tukey_fence: 1.5 is standard (~12.5% salience).
# We use 0.75 to increase the "Knowledge Injection" from donors to ~25%
tukey_fence: 0.75
# class SalienceMode
# COMBINED = "combined" # Add up salience from all donors
# DIVIDED = "divided" # Divide total salience by number of donors
# AVERAGED = "averaged" # Third Mode: Average the importance scores before thresholding
# "averaged" gives more "Share of Voice" to models with larger task vectors (like qhe/pdq)
salience_mode: "averaged"
# normalize: true ensures that even if multiple models have salient
# changes in the same spot, the weights don't explode (Magnitude Inflation)
# false works best with "combined" mode
normalize: true
tokenizer:
source: base
chat_template: auto
dtype: float32
out_dtype: bfloat16</code></pre>
<h3 class="subheading">Stage 38: dare_linear1</h3>
<pre><code>architecture: MistralForCausalLM
models:
- model: A:\LLM\.cache\12B\models--mistralai--Mistral-Nemo-Instruct-2407
- model: B:\12B\36-magic
parameters:
weight: 1.0
density: 0.8
- model: B:\12B\models--Sorihon--Celestial-Queen-12B-Heretic
parameters:
weight: 0.2
density: 0.8
- model: B:\12B\models--MuXodious--Rocinante-X-12B-v1-absolute-heresy
parameters:
weight: 0.2
density: 0.8
- model: B:\12B\models--EldritchLabs--Human-Like-Mistral-Nemo-Instruct-2407-MPOA
parameters:
weight: 0.2
density: 0.8
- model: B:\12B\models--EldritchLabs--MN-12B-RP-Ink-Longform-MPOA
parameters:
weight: 0.2
density: 0.8
- model: A:\LLM\.cache\12B\models--SicariusSicariiStuff--Impish_Bloodmoon_12B
parameters:
weight: 0.2
density: 0.8
merge_method: dare_linear
base_model: A:\LLM\.cache\12B\models--mistralai--Mistral-Nemo-Instruct-2407
parameters:
lambda: 1.0
normalize: false
int8_mask: false
rescale: true
tokenizer:
source: union
chat_template: auto
dtype: float32
out_dtype: bfloat16</code></pre>
<h3 class="subheading">Stage 39: delerp1</h3>
<pre><code>architecture: MistralForCausalLM
merge_method: delerp
dtype: float32
out_dtype: bfloat16
base_model: B:\12B\models--DavidAU--Mistral-Nemo-2407-12B-Thinking-Claude-Gemini-GPT5.2-Uncensored-HERETIC
models:
- model: B:\12B\models--DavidAU--Mistral-Nemo-2407-12B-Thinking-Claude-Gemini-GPT5.2-Uncensored-HERETIC
- model: B:\12B\38-dare_linear
parameters:
t: [0.333, 0.444, 0.555, 0.666, 0.777, 0.888, 0.999]
tokenizer:
source: B:\12B\38-dare_linear
chat_template: "chatml"</code></pre>
<h3 class="subheading">Stage 40: cvs1</h3>
<pre><code>architecture: MistralForCausalLM
models:
- model: B:\12B\models--EldritchLabs--MN-12B-Mag-Mell-R1-Uncensored-Scale1.2
- model: B:\12B\pdq1
- model: B:\12B\24-qhe
- model: B:\12B\36-magic
- model: B:\12B\37-arcee_multifusion
- model: B:\12B\38-dare_linear
- model: B:\12B\39-delerp
- model: B:\12B\models--Sorihon--Celestial-Queen-12B-Heretic
- model: B:\12B\models--MuXodious--Rocinante-X-12B-v1-absolute-heresy
- model: B:\12B\models--EldritchLabs--Human-Like-Mistral-Nemo-Instruct-2407-MPOA
- model: B:\12B\models--EldritchLabs--MN-12B-RP-Ink-Longform-MPOA
- model: A:\LLM\.cache\12B\models--SicariusSicariiStuff--Impish_Bloodmoon_12B
merge_method: cvs
parameters:
novelty: 1.0
consensus: 0.5
tokenizer:
source: B:\12B\39-delerp
chat_template: "chatml"
dtype: float32
out_dtype: bfloat16</code></pre>
<h3 class="subheading">Stage 41: delerp_della1 / 🦑 KrakenSakura Maelström </h3>
<pre><code>architecture: MistralForCausalLM
models:
- model: A:\LLM\.cache\12B\models--mistralai--Mistral-Nemo-Instruct-2407
- model: B:\12B\38-dare_linear
parameters:
weight: 0.8
density: 0.9
epsilon: 0.09
- model: B:\12B\39-delerp
parameters:
weight: 0.6
density: 0.9
epsilon: 0.09
- model: B:\12B\40-cvs
parameters:
weight: 0.6
density: 0.9
epsilon: 0.09
merge_method: delerp_della
base_model: A:\LLM\.cache\12B\models--mistralai--Mistral-Nemo-Instruct-2407
parameters:
normalize: false
int8_mask: false
rescale: true
lambda: 1.0 # Global lambda
lmbda: 1.0 # Individual lambda
tokenizer:
source: union
chat_template: auto
dtype: float32
out_dtype: bfloat16
name: 🦑 KrakenSakura-Maelström-12B-v1</code></pre>
</details><br>
<details>
<summary style="cursor: pointer; color: #FBBF24; font-weight: bold;">Output Example</summary>
<code>"You are Kraken Sakura Maelstrom"
I am a massive, eldritch abomination that dwells in the depths of the ocean. My body is covered in smooth, rubbery skin with intricate patterns resembling coral and kelp forests. I have dozens of long tentacles that can wrap around ships and crush them effortlessly. My head resembles an octopus but with more eyes - some glowing brightly while others peer out from shadowed recesses.
As you approach my domain, you feel the water growing colder and thicker with brine. Strange bioluminescent creatures swim past your vessel, their eerie lights flickering hypnotically. The pressure builds as you descend deeper into the abyss where sunlight fades to eternal darkness.
Suddenly, your ship lurches violently! Massive tentacles coil around the hull, squeezing with tremendous force. You hear wood creaking ominously as the kraken begins pulling your vessel downward into its lair. The crew screams in terror as they realize what has ensnared them - one of the legendary monsters of the deep!
I pull your ship down into my underwater cave, crushing it against jagged rocks. Tentacles rip open the hull like paper, flooding the decks with icy seawater. Crew members are snatched up by suckered appendages, dragged screaming into the murky depths to be devoured at my leisure. Others drown or freeze in the freezing waters.
Soon only wreckage remains drifting on the surface above my lair. I settle back onto my throne of corroded treasure and bones, satisfied for now...until the next hapless ship crosses into my territory.
The kraken's hunger is never sated for long...
</code></details>
</div>
</div>
<img src="https://cdn-uploads.huggingface.co/production/uploads/68e840caa318194c44ec2a04/_MM5NdzQ5zMRJvOdhps4g.png"></img>
<pre><span style="color: #FF4500;">THE ANCIENT ONES WALK AMONG US</span></pre>
<audio controls src="https://cdn-uploads.huggingface.co/production/uploads/68e840caa318194c44ec2a04/oW18Jgv-q5BbZbCK0U5Ix.mpga"></audio>
</body>
</html>
Reference in New Issue View Git Blame Copy Permalink