Skip to main content

Week 10: Fine Tuning with axolotl and modal

Required Reading​

Chapter 13 of Fine Tuning Modal Examples

Prerequisites​

Before attending class, please complete the following setup steps to ensure you can participate in the fine-tuning exercises:

1. HuggingFace Setup​

  • Create a secret named huggingface in your Modal workspace
  • Use the HuggingFace template in your Modal dashboard
  • Get your API token from HuggingFace (settings → API tokens)
  • Set HF_TOKEN with your key
  • Rename from my-huggingface-secret to huggingface

2. Model Access​

  • For LLaMA models, visit the Hugging Face model page
  • Accept Terms and Conditions for model access (instant approval)
  • This is required for LLaMA 3 8B and similar models
  • Create a secret named wandb in your workspace for Weights & Biases logging
  • Can be created from template
  • Alternative: Consider using axolotl's MLFlow integration
  • Proper logging is essential for tracking training progress

Additional Resources​

Axolotl Resources​

Note​

The instructor will not be able to debug setup issues during class time. Please ensure you have completed all setup steps before the session begins.