What is a LoRA config?
What is it?
LoRA adapts an existing model to a task without fully retraining it: only small adapters are trained. A LoRA page gathers all the settings of that training.
When to use it
- Reproduce a fine-tuning that worked.
- Start your own training from proven settings.
Page fields
| Method | LoRA, or QLoRA (4-bit compressed model, less memory). |
|---|---|
| Base model | The model to adapt, e.g. Qwen/Qwen2.5-7B-Instruct. |
| Rank, alpha, dropout | Size and strength of the adapters. |
| Learning rate, epochs, batch | Training settings. |
| Linked dataset | The catalog dataset used for training. |
How to use it
“Use” tab: download the YAML config for Axolotl, or copy the PEFT Python example.