aipackage/cezen-portal/tool-qlora.html
2026-06-30 10:51:41 +05:30

194 lines
8.9 KiB
HTML
Raw Permalink Blame History

This file contains ambiguous Unicode characters

This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.

<!DOCTYPE html>
<html lang="en">
<head>
<meta charset="UTF-8">
<meta name="viewport" content="width=device-width, initial-scale=1.0">
<title>QLoRA / Unsloth — Nexus One AI Portal</title>
<link rel="stylesheet" href="style.css?v=4">
</head>
<body>
<header class="topnav">
<a href="index.html" class="brand">Nexus One <span>AI</span></a>
<nav>
<a href="index.html">Home</a>
<a href="quickstart.html">Quick Start</a>
<a href="prompts.html">Prompt Library</a>
<a href="usecases.html">Use Cases</a>
<span class="nav-sep"></span>
<div class="nav-dropdown">
<button class="nav-drop-btn">Help ▾</button>
<div class="nav-drop-menu">
<span class="nav-drop-cat">LEARN /</span>
<a href="quickstart.html">Quick Start</a>
<a href="models.html">Models</a>
<span class="nav-drop-cat">SUPPORT /</span>
<a href="troubleshooting.html">Troubleshoot</a>
<a href="faq.html">FAQ</a>
<span class="nav-drop-cat">MORE /</span>
<a href="glossary.html">Glossary</a>
<a href="whats-new.html">What's New</a>
</div>
</div>
<div class="nav-dropdown">
<button class="nav-drop-btn">Admin ▾</button>
<div class="nav-drop-menu nav-drop-menu-wide">
<span class="nav-drop-cat">DOCS /</span>
<a href="security.html">Security & Privacy</a>
<a href="admin.html">Admin Guide</a>
<span class="nav-drop-cat">MONITOR /</span>
<a href="dashboard.html">Dashboard</a>
<a href="analytics.html">Usage Analytics</a>
<a href="audit.html">Audit Log</a>
<a href="feedback.html">Feedback &amp; Ratings</a>
<span class="nav-drop-cat">MANAGE /</span>
<a href="users.html">Users</a>
<a href="teams.html">Teams</a>
<a href="models-admin.html">Model Manager</a>
<a href="training.html">Training</a>
<a href="knowledge.html">Knowledge Base</a>
<span class="nav-drop-cat">TOOLS /</span>
<a href="apikeys.html">API Keys</a>
<a href="benchmark.html">Benchmarking</a>
<a href="model-compare.html">Model Compare</a>
<a href="api-playground.html">API Playground</a>
<a href="guardrails.html">Guardrails</a>
<a href="rag-quality.html">RAG Quality</a>
<a href="router.html">Model Router</a>
<a href="connectors.html">Connectors</a>
<span class="nav-drop-cat">SYSTEM /</span>
<a href="console.html">Console</a>
<a href="settings.html">Settings</a>
</div>
</div>
<div class="nav-dropdown">
<button class="nav-drop-btn">AI Tools ▾</button>
<div class="nav-drop-menu">
<span class="nav-drop-cat">INTELLIGENCE /</span>
<a href="documents.html">Document Intelligence</a>
<a href="chat-multi.html">Multimodal Chat</a>
<a href="prompt-studio.html">Prompt Studio</a>
<a href="meeting.html">Meeting Assistant</a>
<span class="nav-drop-cat">AUTOMATION /</span>
<a href="agents.html">Agent Builder</a>
<a href="schedules.html">Scheduled Jobs</a>
<a href="workflows.html">Workflow Automation</a>
<span class="nav-drop-cat">QUALITY /</span>
<a href="evals.html">AI Eval Suite</a>
<a href="chatrooms.html">Chat Rooms</a>
</div>
</div>
</nav>
<a href="notifications.html" style="position:relative">🔔</a>
<span class="badge" data-brand="tier">Basic Tier</span>
<div id="nav-org-logo" class="nav-org-logo"></div>
</header>
<div class="tool-hero">
<div class="tool-icon">🎯</div>
<div class="tool-meta">
<a href="index.html#tools" class="back-link">← All Tools</a>
<h1>QLoRA / Unsloth</h1>
<div class="tagline">Fine-tune a large AI model on your own data — efficiently, without needing massive GPU memory.</div>
<div class="hero-actions">
<a href="http://ai.local:8888" target="_blank" class="btn-primary">Open in Jupyter ↗</a>
<span class="access-pill">📓 Run fine-tuning jobs via Jupyter</span>
</div>
</div>
</div>
<div class="content">
<div class="section-title">What is fine-tuning?</div>
<div class="info-grid">
<div class="info-card">
<h4>In plain terms</h4>
<p>Fine-tuning is the process of taking a general-purpose AI model (like Llama 3.1) and continuing to train it on your own data so it learns to behave in a way specific to your organisation — using your terminology, following your formats, reflecting your policies. The result is a model that's significantly better at your specific tasks than the base model.</p>
</div>
<div class="info-card">
<h4>When fine-tuning makes sense</h4>
<ul>
<li>You want the AI to answer in your organisation's voice and style</li>
<li>The base model doesn't know your industry's terminology</li>
<li>You have hundreds of example question-answer pairs from your domain</li>
<li>RAG alone isn't giving accurate enough results</li>
<li>You want to teach the model a specific task format (extract fields from forms)</li>
</ul>
</div>
</div>
<div class="section-title">What QLoRA and Unsloth are</div>
<div class="info-grid">
<div class="info-card">
<h4>QLoRA</h4>
<p><strong>Quantised Low-Rank Adaptation</strong> — a technique that dramatically reduces the GPU memory needed for fine-tuning. Instead of updating all the weights in a model (which would need 48× the model's VRAM), QLoRA only trains a small set of adapter layers and keeps the rest of the model in 4-bit compressed format. This makes it possible to fine-tune a 7B or 8B model on the RTX Pro 6000.</p>
</div>
<div class="info-card">
<h4>Unsloth</h4>
<p>Unsloth is a Python library that makes QLoRA fine-tuning <strong>24× faster</strong> and uses <strong>4070% less memory</strong> than standard QLoRA implementations. It achieves this through hand-optimised GPU kernels. For Cezen Entry tier with the RTX Pro 6000, Unsloth is the recommended way to fine-tune — it makes jobs that would otherwise take 10 hours complete in 34 hours.</p>
</div>
</div>
<div class="notice">
💡 <strong>Fine-tuning is an advanced task.</strong> You'll need training data in the right format and some Python knowledge. Start by exploring <a href="tool-openwebui.html">Open WebUI</a> and <a href="tool-chromadb.html">RAG with ChromaDB</a> — for most use cases, RAG gives 80% of the benefit of fine-tuning with 10% of the effort.
</div>
<div class="section-title">How to approach a fine-tuning project</div>
<div class="steps">
<div class="step">
<div class="step-num">1</div>
<div>
<h4>Prepare your training data</h4>
<p>Fine-tuning needs examples in a structured format — typically a JSONL file where each line is a conversation: a prompt and the ideal response. A minimum of 50100 high-quality examples is needed; 5002000 is better. Quality matters more than quantity.</p>
</div>
</div>
<div class="step">
<div class="step-num">2</div>
<div>
<h4>Open Jupyter and set up the training script</h4>
<p>Open <a href="tool-jupyter.html">Jupyter</a> at <code>http://ai.local:8888</code>. Load the Cezen fine-tuning notebook template (provided by your administrator) or start from scratch with the Unsloth documentation examples.</p>
</div>
</div>
<div class="step">
<div class="step-num">3</div>
<div>
<h4>Configure and run training</h4>
<p>Set your base model, data file path, and training parameters (epochs, learning rate). A typical fine-tuning run for Llama 3.1 8B on 500 examples takes 4590 minutes on the Entry tier RTX Pro 6000.</p>
</div>
</div>
<div class="step">
<div class="step-num">4</div>
<div>
<h4>Export to Ollama format</h4>
<p>Once training is complete, export the fine-tuned model to GGUF format using Unsloth's export function. Then load it into Ollama with <code>ollama create my-model -f Modelfile</code> and use it in <a href="tool-openwebui.html">Open WebUI</a> like any other model.</p>
</div>
</div>
<div class="step">
<div class="step-num">5</div>
<div>
<h4>Test and iterate</h4>
<p>Compare your fine-tuned model against the base model on your specific tasks. If it's not where you need it, add more training examples focused on the weak areas and re-run.</p>
</div>
</div>
</div>
<div class="section-title">Works with</div>
<div class="works-with">
<a href="tool-jupyter.html">📓 Jupyter</a>
<a href="tool-ollama.html">🦙 Ollama</a>
<a href="tool-openwebui.html">💬 Open WebUI</a>
<a href="tool-dcgm.html">📡 DCGM</a>
</div>
</div>
<footer>
<p>Nexus One AI &nbsp;·&nbsp; Powered by Cezen &nbsp;·&nbsp; Basic Tier</p>
<p>Questions? <a href="mailto:support@cezentech.com">support@cezentech.com</a></p>
</footer>
<script src="auth.js"></script>
<script src="branding.js"></script>
</body>
</html>