In this folder, we implement a simple classifier for the skill selection task. The code is based on the HuggingFace code run_glue.py.
Firstly, we generate the few-shot datasets by running:
python generate_data.py
This creates a folder with the required *.json files for training the model.
After this step, we training RoBERTa (based/large) by running:
python runner.py
This is a meta-script for automatically running scripts in multiple GPUs.
Finally, we post-process the runs with the post_process_results.py script, which convert the output into the right format for tables and plots.