llm-adapter

Train

python adapter_class_2.py --train_set baseline --max_epochs 20

explaination of the arguments:

• --train_set: the dataset to train on. The corresponding dataset cofinguration is in configs/dataset-{train_set}.py
• --max_epochs: the number of epochs to train the model for

Inference

python inference.py --model_dir "training_outputs/data_100000/saved_adapters/yelp_adapter_class_2_baseline/" --test_data "data_100000/test/baseline.csv" --test_output_dir "test_results/"

explaination of the arguments:

• --model_dir: the directory where the trained adapter (with head) is saved
• --test_data: the path to the test data file
• --test_output_dir: the directory where the test results will be saved

Datasets

data_100000

Train

• baseline: 5000 samples in all. Sample numbers for each category are listed below:

	Restaurant	Drinks	Shopping	Entertainment	Housing	Beauty
Amount	1000	1000	1000	1000	500	500

• restaurant: 5000 samples in Restaurant category.
• drinks: 5000 samples in Drinks category.
• shopping: 5000 samples in Shopping category.
• entertainment: 5000 samples in Entertainment category.
• housing: 2500 samples in Housing category.
• beauty: 2500 samples in Beauty category.

Test

1500 samples in all. 250 samples for each category.