How‐to

How-To

python main.py --model <model_name> --dataset <dataset_name> [--dbpath <database_path>]

--model (required): Specify the name or identifier of the model you want to use for activation analysis. This should be a valid model identifier from the Hugging Face Models Hub. For example:

--model google/gemma-2b

--dataset (required): Specify the name or identifier of the dataset you want to use for activation analysis. This should be a valid dataset identifier from the Hugging Face Datasets Hub. For example:

--dataset wikimedia/wikipedia

--dbpath (optional): Specify the path where you want to store the SQLite database file for the activation analysis results. If not provided, it defaults to ./activations.db in the current directory. For example:

--dbpath /path/to/activations.db