List of steps to reproduce our re-ranking results:
- Split the training sets in 2 parts: 
	1_base_finetuning/build_training_splits.py
- Fine-tune base models: 
	1_base_finetuning/main_trainer.py
- Check the performance of the fine-tuned models: 
	1_base_finetuning/evaluate.py
- Generate summary candidates: 
	2_candidate_generation/main.py
- Score each candidate, to build the labels for the re-ranker: 
	4_supervised_reranking/main.py
- Train a re-ranker: 
	5_supervised_multitask_reranking/main_trainer.py
- Check the performance of the re-ranker: 
	5_supervised_multitask_reranking/evaluate.py