Use run_experiment.py <config.yaml> to train models.
To evaluate models on other benchmarks (e.g., zero-shot transfer), use run_eval.py <config.yaml>
We adapt the framework from our ACL17 demo. Documentation and details on how our YAML configuration files are structured can be found there. We provide several basic configurations for training models in ./configs.
Documented example configs are example_config.yaml and example_eval_config.yaml.