AI GitHub
Axolotl for recipe-based language model training
Describe model fine-tuning runs in configuration files and reuse the recipes across experiments.
Key Takeaways
- Axolotl configuration ties together checkpoint, data, trainer settings and hardware options.
- Prepare a small configuration with checkpoint, dataset, adapter settings and output destination.
- Run a short adapter job and test whether it can resume from a saved checkpoint.
- Training backends and acceleration options depend on the hardware and dependencies.
How it works
Axolotl configuration ties together checkpoint, data, trainer settings and hardware options. A saved recipe makes it easier to change one variable and reproduce the rest.
What you need
Prepare a small configuration with checkpoint, dataset, adapter settings and output destination. Keep the generated preprocessing artifacts with the run if they affect the training data.
What to try first
Run a short adapter job and test whether it can resume from a saved checkpoint. Inspect the prepared dataset and compare a held-out evaluation with the base model.
Keep one failed checkpoint-resume test in the experiment notes. Changing trainer, precision or adapter settings can make state incompatible with an earlier run.
Where to look for failures
Training backends and acceleration options depend on the hardware and dependencies. Record the exact environment alongside the recipe.
Related Franklin pages
Run a tiny supervised job and check the prepared examples before investing in a large run. Save the base checkpoint, data revision and validation outputs with the recipe.
- Fine-tune supported model architectures using optimized training paths and documented recipes. Unsloth for memory-conscious model tuning
- Adapt language-model tuning recipes built around PyTorch and configurable components. torchtune for explicit PyTorch training recipes
For the distinction between training from scratch and adapting a checkpoint, see Franklin’s glossary. Fine-tuning
Project documentation
Try it for your workflow
Explore the project’s documentation for setup and usage, or save this page to a collection for later.
Read the project documentation ↗