Practical-LLM-Implementation

Practical code examples and tutorials for fine-tuning Large Language Models and building AI agent systems. Companion repo for YouTube guides.

Hi! Welcome to the companion repository for the LLM Implementation YouTube channel www.youtube.com/@LLMImplementation. Here, I share code, notebooks, and resources related to my learnings and practical experiences in LLM implementation, including tutorials and project explorations focused on fine-tuning and AI agent development.

This repository covers two main areas:

1. Fine-tuning: Adapt powerful pre-trained LLMs like Google Gemini, Llama, and GPT models to your specific tasks, domains, or desired output styles
2. AI Agent Systems: Build sophisticated multi-agent systems using frameworks like Google's Agent Development Kit (ADK), LangGraph, and AutoGen

Repository Structure

🤖 AI Agent Systems

Google Agent Development Kit (ADK)

• Folder: adk/
• Description: Multi-agent systems built with Google's ADK framework
• Examples:
  • Google Search Agent: Simple agent using Google Search tool with Gemini 2.5 Flash
  • YouTube Metadata Refinement Agent: Complex multi-LLM workflow with criticism → refinement → voting → aggregation phases
  • ADK Deploy Notebook: Deployment examples for ADK agents

Agent Frameworks

• Folder: agents_frameworks/
• Description: Examples using different agent frameworks
• Examples:
  • LangGraph Basic: Fundamental LangGraph implementations
  • AutoGen (ag2) with OpenAI: Multi-agent conversations and workflows

🔧 Fine-Tuning Examples

Google Gemini

• Folder: gemini/
• Description: Examples focusing on fine-tuning Google's Gemini family of models
• Examples:
  • Supervised Fine-Tuning to Humanize Prompts: Data preparation (JSONL) and supervised tuning on Vertex AI (Companion to: Practical Gemini Fine-Tuning: Step-by-Step Guide with Vertex AI)
  • Gemini 2.5 Pro Video Analysis: Video content analysis demonstrations
  • Prompt Rewriting with Gemini 2.5 Pro Exp: Advanced prompt engineering techniques

Llama

• Folder: llama/
• Description: Llama model fine-tuning examples
• Examples:
  • Llama 3 Fine-tuning: Practical fine-tuning implementations for Llama 3

GPT Open Source

• Folder: gpt-oss-20b/
• Description: Fine-tuning examples for open-source GPT models
• Examples:
  • GPT OSS 20B Fine-tuning: Large-scale model fine-tuning techniques

📝 Context Engineering

• Folder: context-engineering/
• Description: Framework for building AI features using context engineering principles
• Key Concept: 2-step process - setup examples, then use the universal magic prompt template

Getting Started

For ADK Agent Development

1. Setup Environment:

   cd adk/
   python3 -m venv .venv
   source .venv/bin/activate  # On macOS/Linux

2. Install Dependencies:

   pip install google-adk -q
   pip install python-dotenv -q
   pip install litellm -q

3. Configure API Keys: Create a .env file in the adk/ directory:

   GOOGLE_API_KEY=your_google_api_key
   OPENAI_API_KEY=your_openai_api_key
   ANTHROPIC_API_KEY=your_anthropic_api_key

4. Run ADK Web UI:

   adk web

   Access at http://localhost:8000

For Jupyter Notebooks

1. Install Dependencies:

   pip install langgraph langchain langsmith
   pip install jupyter notebook

2. Launch Notebooks:

   jupyter notebook

Contributing

I'm sharing my learning journey here! If you find bugs, have suggestions for improvements, spot errors, or want to discuss ideas related to fine-tuning or agent development, feel free to open an issue. Pull requests that fix issues or add value are also welcome.

License

This project is licensed under the MIT License. (Assumes you chose the MIT License based on our discussion)

---

Let's learn and explore LLM fine-tuning and AI agent development together! Don't forget to check out the www.youtube.com/@LLMImplementation for video walkthroughs and discussions.