wzhouad / question-answer-consolidation

Answer Consolidation

Code for NAACL 2022 paper Answer Consolidation: Formulation and Benchmarking.

Requirements

• PyTorch
• Transformers
• wandb
• tqdm
• scikit-learn

Training from Scratch

Finetune the sentence embedding models with the following command:

>> python main.py --model_name_or_path $MODEL --format sentence_embedding

where $MODEL can be ['roberta-large', 'sentence-transformers/all-roberta-large-v1', 'princeton-nlp/sup-simcse-roberta-large'].

Finetune the NLI models with the following command:

>> python main.py --model_name_or_path $MODEL --format nli

where $MODEL can be ['roberta-large', 'roberta-large-mnli'].

Evaluation results on the development and test sets are synced to the wandb darshboard every epoch.