lora-peft

Steps to Train Alpaca-Based Instruction LoRAs on the LOLA Model

1. Install Dependencies

Modify the setup script as needed, and then run:

bash setup_peft_env.sh

2. Set Up Weights & Biases (Wandb) (Optional)

If you wish to use Wandb for experiment tracking, set it up as follows:

VENV_PATH=./venv-lola-peft
source $VENV_PATH/bin/activate
wandb login

3. Download Multilingual Alpaca Data

To obtain multilingual Alpaca data, refer to the following script: ../instruction-ft/generate_multilingual_alpaca_json.py

4. Run the Training Script

Modify the training arguments to suit your setup, and then run the script:

bash run-lora-peft-train.sh

5. Inference

To test your model's output in inference mode, you can use the following python notebook: lora-inference.ipynb.