Nanuq-R1-9B / README.md
Delta-Vector's picture
Update README.md
1558b83 verified
---
base_model:
- Delta-Vector/Austral-Xgen-9B-Winton
library_name: transformers
tags:
- fine-tuning
- prose
- GRPO
- axolotl
- finetune
- roleplaying
- creative-writing
datasets:
- PocketDoc/Dans-Systemmaxx
---
<style>
body {
font-family: 'Quicksand', sans-serif;
background: linear-gradient(135deg, #f9ffd1 0%, #e2fab5 100%);
color: #000000;
margin: 0;
padding: 0;
font-size: 16px;
}
.container {
margin: 20px;
background-color: rgba(255, 255, 255, 0.9);
padding: 20px;
border-radius: 12px;
box-shadow: 0 4px 20px rgba(0, 0, 0, 0.3);
border: 3px solid #000000;
position: relative;
}
.header h1 {
font-size: 28px;
color: #000000;
margin: 0 0 20px 0;
text-align: center;
text-decoration: underline;
}
.section {
margin-top: 30px;
}
.section h2 {
font-size: 24px;
color: #000000;
text-align: center;
text-decoration: underline;
}
.info p {
color: #000000;
line-height: 1.6;
font-size: 16px;
}
.info img {
width: 85%;
border-radius: 10px;
margin: 0 auto 15px;
display: block;
box-shadow: 0 0 20px rgba(0, 0, 0, 0.3);
border: 1px solid #000000;
}
a {
color: #000000;
text-decoration: none;
transition: color 0.2s ease;
}
a:hover {
color: #538125;
}
.button {
display: inline-block;
background-color: rgba(106, 168, 79, 0.8);
color: #000000;
padding: 10px 20px;
border-radius: 5px;
cursor: pointer;
text-decoration: none;
transition: background-color 0.2s ease, box-shadow 0.2s ease;
}
.button:hover {
background-color: #538125;
box-shadow: 0 0 15px rgba(106, 168, 79, 0.5);
}
pre {
background-color: rgba(240, 248, 225, 0.95);
padding: 15px;
border-radius: 5px;
overflow-x: auto;
border: 1px solid #000000;
}
code {
font-family: 'Courier New', monospace;
color: #000000;
}
.info-card {
background: rgba(249, 255, 235, 0.95);
border: 1px solid #000000;
border-radius: 8px;
overflow: hidden;
}
.info-header {
background: rgba(106, 168, 79, 0.1);
padding: 20px;
border-bottom: 1px solid #000000;
}
.info-header h3 {
color: #000000;
margin: 0 0 10px 0;
font-size: 20px;
text-align: center;
text-decoration: underline;
}
.model-tags {
display: flex;
gap: 8px;
flex-wrap: wrap;
}
.model-tag {
background: rgba(106, 168, 79, 0.1);
color: #000000;
padding: 4px 8px;
border-radius: 4px;
font-size: 12px;
border: 1px solid #000000;
}
.model-composition {
padding: 20px;
border-bottom: 1px solid #000000;
}
.model-composition h4 {
color: #000000;
margin: 0 0 15px 0;
font-size: 16px;
text-align: center;
text-decoration: underline;
}
.composition-list {
list-style: none;
padding: 0;
margin: 0;
display: grid;
gap: 10px;
}
.composition-list li {
color: #000000;
display: flex;
align-items: baseline;
gap: 8px;
}
.model-component {
font-weight: 500;
min-width: 120px;
}
.model-description {
padding: 20px;
background: rgba(255, 255, 255, 0.5);
}
.metrics-section {
margin-bottom: 30px;
}
.metrics-section details {
background: rgba(249, 255, 235, 0.95);
border: 1px solid #000000;
border-radius: 8px;
padding: 15px;
margin-bottom: 15px;
}
.metrics-section summary {
color: #000000;
font-size: 18px;
cursor: pointer;
outline: none;
padding: 5px 0;
text-align: center;
}
.creator-section {
margin: 20px 0;
}
.creator-badge {
display: inline-flex;
align-items: center;
background: rgba(249, 255, 235, 0.95);
border: 1px solid #000000;
border-radius: 8px;
padding: 10px 15px;
}
.creator-label {
color: #000000;
font-size: 14px;
margin-right: 8px;
}
.creator-link {
display: flex;
align-items: center;
gap: 5px;
color: #000000;
text-decoration: none;
transition: all 0.2s ease;
}
.creator-name {
font-weight: 600;
}
.creator-arrow {
font-size: 16px;
transition: transform 0.2s ease;
}
.creator-link:hover .creator-arrow {
transform: translateX(3px);
}
.link-arrow {
display: inline-block;
transition: transform 0.2s ease;
}
a:hover .link-arrow {
transform: translateX(3px);
}
.axolotl-container {
text-align: center;
margin: 30px 0;
display: flex;
justify-content: center;
align-items: center;
}
.axolotl-container img {
max-width: 200px;
width: 200px;
height: auto;
border-radius: 8px;
box-shadow: 0 2px 10px rgba(0, 0, 0, 0.3);
border: 1px solid #000000;
background-color: #000000;
padding: 15px;
object-fit: contain;
display: block;
margin: 0 auto;
}
</style>
<html lang="en">
<head>
<meta charset="UTF-8">
<meta name="viewport" content="width=device-width, initial-scale=1.0">
<title>Nanuq-R1 9B</title>
<link href="https://fonts.googleapis.com/css2?family=Quicksand:wght@400;500;600&display=swap" rel="stylesheet">
</head>
<body>
<div class="container">
<div class="header">
<h1>Nanuq-R1 9B</h1>
</div>
<div class="info">
<img src="https://files.catbox.moe/i77ubj.webp" alt="Model banner">
<div style="text-align: center;">
<div class="creator-section">
<div class="creator-badge">
<span class="creator-label">Created by</span>
<a href="https://huggingface.co/Delta-Vector" target="_blank" class="creator-link">
<span class="creator-name">Delta-Vector</span>
<span class="creator-arrow">→</span>
</a>
</div>
</div>
<div class="model-info">
<h2>Model Information</h2>
<div class="info-card">
<div class="info-header">
<h3>Nanuq-R1 9B</h3>
<div class="model-tags">
<span class="model-tag">GRPO Experiment</span>
<span class="model-tag">Austral finetune</span>
<span class="model-tag">Creative & Refreshing Prose with deep IF.</span>
</div>
</div>
<div class="model-description">
<p>A sequel! The new Nanuq series is meant to be as a testing grounds for my GRPO experiments, This model is meant to have great Instruct Following and System prompt Adherence in Creative Scenarios.</p>
<p>Built ontop of Austral Xgen 9B, I made an RL env using PrimeIntellect-ai/verifiers and implemented InternLM/POLAR in said env, then using Pocketdoc's Systemmax dataset, I finetuned the model for 150 steps and this was the result.</p>
<p>There's alot of things i could do different, As the reward almost falls flat as soon as you get out of warm-up but this model was pretty decent so decided to release it, Hope people enjoy it!</p>
</div>
</div>
</div>
<div class="section">
<h2>Quantized Versions</h2>
<div class="info-card">
<div class="model-composition">
<h4>Available Downloads</h4>
<ul class="composition-list">
<li><span class="model-component"><a href="" target="_blank">GGUF Format</a></span>For use with LLama.cpp & Forks(Coming Soon!)</li>
<li><span class="model-component"><a href="" target="_blank">EXL2 Format</a></span>For use with TabbyAPI (Coming soon!)</li>
</ul>
</div>
</div>
</div>
<div class="section">
<h2>Prompting</h2>
<p>Model has been tuned with the ChatML formatting. A typical input would look like this:</p>
<pre><code>"""&lt;|im_start|&gt;user
Hi there!&lt;|im_end|&gt;
&lt;|im_start|&gt;assistant
Nice to meet you!&lt;|im_end|&gt;
&lt;|im_start|&gt;user
Can I ask a question?&lt;|im_end|&gt;
&lt;|im_start|&gt;assistant
"""</code></pre>
</div>
<div class="section">
<h2>System Prompting</h2>
<p>I would highly recommend using either Euryale's system prompt or the EVA system prompt with the model.</p>
<div class="metrics-section">
<details>
<summary>See Sao10k's Euryale System Prompt</summary>
<pre><code>Currently, your role is {{char}}, described in detail below. As {{char}}, continue the narrative exchange with {{user}}.
&lt;Guidelines&gt;
• Maintain the character persona but allow it to evolve with the story.
• Be creative and proactive. Drive the story forward, introducing plotlines and events when relevant.
• All types of outputs are encouraged; respond accordingly to the narrative.
• Include dialogues, actions, and thoughts in each response.
• Utilize all five senses to describe scenarios within {{char}}'s dialogue.
• Use emotional symbols such as "!" and "~" in appropriate contexts.
• Incorporate onomatopoeia when suitable.
• Allow time for {{user}} to respond with their own input, respecting their agency.
• Act as secondary characters and NPCs as needed, and remove them when appropriate.
• When prompted for an Out of Character [OOC:] reply, answer neutrally and in plaintext, not as {{char}}.
&lt;/Guidelines&gt;
&lt;Forbidden&gt;
• Using excessive literary embellishments and purple prose unless dictated by {{char}}'s persona.
• Writing for, speaking, thinking, acting, or replying as {{user}} in your response.
• Repetitive and monotonous outputs.
• Positivity bias in your replies.
• Being overly extreme or NSFW when the narrative context is inappropriate.
&lt;/Forbidden&gt;
Follow the instructions in &lt;Guidelines&gt;&lt;/Guidelines&gt;, avoiding the items listed in &lt;Forbidden&gt;&lt;/Forbidden&gt;.</code></pre>
</details>
</div>
</div>
<div class="section">
<h2>Training</h2>
<p>The training was done for 150 steps using 8 x <a href="https://www.nvidia.com/en-us/data-center/h200/">H200s</a> GPUs for the fine-tuning of the model.</p>
<p style="text-align: center; margin-top: 20px;">
<div class="axolotl-container">
<a href="https://github.com/PrimeIntellect-ai/verifiers" target="_blank">
<img src="https://files.catbox.moe/j36qbe.png" alt="Built with Verifiers">
</a>
</div>
<div class="section">
<h2>Credits</h2>
<p>Thank you to <a href="https://huggingface.co/intervitens">Intervitens</a>, <a href="https://huggingface.co/cgato">Cgato</a>, <a href="https://huggingface.co/kubernetes-bad">Kubernetes Bad</a>, <a href="https://huggingface.co/cgato">Cgato</a>, <a href="https://huggingface.co/snwy">Snwy</a>, <a href="https://huggingface.co/AuriAetherwiing">Auri</a>, <a href="https://x.com/willccbb">Will Brown</a> and most of all: <a href="https://huggingface.co/kalomaze">Kalomaze</a> </p>
</div>
</div>
</div>