Skip to content

tutorials.md

Latest commit

 

History

History
87 lines (56 loc) · 3.15 KB

tutorials.md

File metadata and controls

87 lines (56 loc) · 3.15 KB

ToolBrain Tutorials

Welcome to the ToolBrain tutorials! This guide will walk you through the core concepts and practical examples to help you master the library.

Part 1: Getting Started

This section covers the basics to get you up and running with ToolBrain.

  1. What is ToolBrain?

    • Understand the core problems ToolBrain solves and its fundamental concepts.
    • [Jupyter Notebook]

  2. Setting Up Your Environment

  3. Your First Training Run

Part 2: Core Concepts in Depth

Dive deeper into the key components that make ToolBrain work.

  1. Understanding Traces

    • Learn about the Trace and Turn objects, the backbone of trace-based learning.
    • [Jupyter Notebook]

  2. Crafting Custom Reward Functions

  3. DPO vs. GRPO vs. Supervised - Choosing Your Algorithm

    • Understand the differences between the available learning algorithms and when to use each.
    • [Jupyter Notebook]

Part 3: More Examples

Explore practical, real-world applications of ToolBrain.

  1. Email Search Agent

    • Build a complex agent that can search a database to answer questions.
    • [Jupyter Notebook]

  2. Hyperparameter Optimization (HPO) Training

    • Use ToolBrain in a novel way to perform HPO on a machine learning model.
    • [Jupyter Notebook]

  3. Knowledge Distillation

    • Learn how to train a small, efficient model by distilling knowledge from a larger teacher model.
    • [Jupyter Notebook]

  4. Tool Retrieval

    • Manage large sets of tools efficiently by dynamically selecting the most relevant ones for a given task.
    • [Jupyter Notebook]

  5. Automatic Task Generation (Zero-Learn)

    • Bootstrap your training process by automatically generating a dataset from a high-level description.
    • [Jupyter Notebook]

Part 4: Efficient Training

Learn how to train larger models faster and with less memory.

  • Techniques for Efficiency (FP16, LoRA, BitsAndBytes, Unsloth)
    • A guide to using LoRA, 4-bit quantization, and Unsloth to optimize your training runs.
    • [Jupyter Notebook]

Part 5: Development and Contribution

Get involved with the ToolBrain community.

  • Extending and Contributing to ToolBrain
    • Learn how to write a custom adapter for a new agent framework and how to contribute to the project.
    • [Jupyter Notebook]