Get an adapter configuration ready for training.
What it does
Produces a validated LoRA or QLoRA adapter configuration with task-appropriate hyperparameters for supervised fine-tuning.
When to use it
Use when writing or reviewing LoRA/QLoRA training configurations, choosing adapter parameters, or comparing LoRA, QLoRA, and full fine-tuning.
How to use it
Give it an SFT routing decision and target size class; it returns validated adapter configuration values for downstream script generation.
What you provide
- an SFT routing decision
- a target size class
What you need · 4
The default worked training configuration imports Unsloth's FastLanguageModel.
The worked training configuration imports TRL's SFTConfig and SFTTrainer.
The instructed hardware check uses torch to test CUDA BF16 support before selecting the training dtype.
The worked training configuration requires a GPU with BF16 support and raises an error instead of falling back to fp16.
Access · 0
No additional actions listed.