Prem Studio Tutorials

Hands-on tutorials showing how to build, fine-tune, evaluate, and automate workflows in Prem Studio—featuring code examples in Python and TypeScript.

TUTORIALS' OVERVIEW

• Continuous Fine-Tuning - Automate model improvement using traces and evaluation feedback to continuously refine fine-tuned models.

• Guarding BYOE - Build and deploy custom evaluation servers for safety guardrail models with flexible scoring strategies.

• Nemotron Safety Dataset Adaptation - Convert the Nemotron Safety Guard Dataset to the messages format required for fine-tuning.

• Dataset Quality Labeling - Automatically label datapoints in your dataset based on quality criteria using predefined quality levels.

• PDF Synthetic Dataset - Generate synthetic training datasets from PDF documents for structured data extraction tasks.

• Two-Step Fine-Tuning - Perform full fine-tuning followed by LoRA fine-tuning to progressively improve model performance on specific tasks.

• Web Synthetic Safety Dataset - Generate safety classification datasets from web sources using synthetic Q&A generation.

• YouTube Synthetic Dataset - Generate synthetic training datasets from YouTube video transcripts for structured information extraction.

Tutorial Format

Each tutorial README includes:

1. Prerequisites - Required knowledge or setup
2. Setup Environment - Environment setup instructions for Python and TypeScript
3. Outcome - What you'll achieve
4. Steps - Detailed walkthrough
5. Code Snippets - TypeScript and Python examples with instructions on how to run the experiments
6. Resources - Sample files and datasets
7. Next Steps - Related tutorials and learning paths

Tags

Tutorials are tagged with:

• Platform Sections: dataset, finetuning, evaluation, inference
• Complexity: beginner, intermediate, advanced
• Domain: safety, finance, medicine, education, etc.

Tags appear at the top of each tutorial README.

Repository Structure

Each tutorial is organized in its own directory under /tutorials/, following this structure:

/tutorials/
    /<tutorial-name>/
        README.md          # Tutorial documentation
        python/            # Python implementation
            script.py
            requirements.txt
        typescript/        # TypeScript implementation
            script.ts
            package.json
        resources/         # Shared resources
            dataset.jsonl  # Optional: sample dataset in JSONL format
            qa_templates.json
            ...

Note: Not all tutorials provide both Python and TypeScript implementations.

Contributing

We welcome contributions! To add a new tutorial:

1. Copy the template: Use /tutorials/_template/ as a starting point
2. Follow the structure: Ensure all required files are included in the appropriate python or typescript subfolders
3. Add tags: Include platform sections, complexity, and domain tags
4. Submit a PR: Open a pull request with your tutorial

See /tutorials/_template/README.md for detailed guidelines.