SFT and PEFT Examples

We offer many examples that show how to run supervised fine-tuning (SFT) and parameter-efficient fine-tuning(PEFT) methods across a variety of models presented in the form of playbooks or NeMo Framework Launcher commands.

Most of the PEFT examples use LoRA since it is the most common PEFT method. However, you can easily switch to other PEFT methods by modifying model.peft.peft_scheme to `ptuning, ia3, or adapter. You can also switch to SFT by setting model.peft.peft_scheme to null (along with other changes such as learning rate, if applicable).

Nemotron

Gemma and CodeGemma

Starcoder2

Mistral

Mixtral

Llama

Falcon

Baichuan2

T5