InstructLab Fine-tuning Guide

A quick setup guide for fine-tuning models with InstructLab.

Prerequisites

• micromamba or conda
• CUDA-capable GPU (recommended)
• At least 16GB RAM

Environment Setup

1. Create and activate the environment (with micromamba or conda):

micromamba create -f environment_cuda.yaml
micromamba activate instructlab

# Verify installation
which ilab

2. Initialize InstructLab configuration:

ilab config init

Configuration Tips

When running ilab config init:

• Set the taxonomy path to ./instructlab/taxonomy within your project directory
• Choose the appropriate system profile based on your hardware (NVIDIA, AMD, etc.)
• Specify the path to your model file

Required Models

You'll need two models for fine-tuning:

1. Base Model (to be fine-tuned):

   • Granite 7B
   • Place in ./models/ directory

2. Training Assistant Model:

   • Merlinite 7B
   • Place in ./models/ directory

Usage

1. Test model serving:

ilab model serve --model-path ./models/granite-7b-lab.Q4_K_M.gguf

2. Validate taxonomy data:

ilab taxonomy diff

Web Interface

For a graphical interface, follow the Playground Chat setup instructions

To set it up locally:

1. Install pnpm, npm or deno

2. Clone and run the UI:

git clone https://github.com/instructlab/ui
cd ui
npm install
npm run dev

Additional Resources

• Video Tutorial
• InstructLab Documentation