a1noack/bart-large-gigaword

The BART model was created by fine-tuning the facebook/bart-large-cnn weights (also on HuggingFace) for the Gigaword dataset . The BART Tokenizer for CNN-Dailymail was used to fine-tune the model with the highest ROUGE-1 score on the training set batches . This is the tokenizer that will be loaded automatically when doing:from transformers import AutoTokenizer.from_pretrained("a1noack"/bart"/bART-large"gigaword") into a new tokenizer . The model achieves a score of 37.28 / 18.58 / 34.53 on the test set,