evaluate_summary('litepyramid_recall') [out] Human metric: litepyramid_recall metric spearman kendalltau --- --- --- rouge1_r 0.497526 0.407974 bart_score_cnn_hypo_ref_de_id est 0.49539 0.392728 bart_score_cnn_hypo_ref_de_Videlicet 0.491011 0.388237 ... Train your custom BARTScore If ...
BARTScore就是将生成文本的评估看做是文本生成任务。 代码:https://github.com/neulab/BARTScore.git This is the Repo for the paper:BARTScore: Evaluating Generated Text as Text Generation 依赖: !pip install torch transformers 引入: frombart_scoreimportBARTScorer 例子1 # 复现英文,能不能CPU?可以的 ...
The code implementation of the paper Stumbling Blocks: Stress Testing the Robustness of Machine-Generated Text Detectors Under Attacks (ACL 2024 main) by Yichen Wang, Shangbin Feng, Abe Bohan Hou, Xiao Pu, Chao Shen, Xiaoming Liu, and Yulia Tsvetkov, and