yuyang/distil_cnndm

# Distilled CNN/DailyMail Dataset

This folder contains the distilled data and dataset loading script to build a dataset on top of it.

- `cnn_bart_pl` is downloaded from [Saved Pseudo-Labels](https://github.com/huggingface/transformers/blob/main/examples/research_projects/seq2seq-distillation/precomputed_pseudo_labels.md), which is generated by facebook/bart-large-cnn, this corresponds to version "1.0.0". It contains train/validataion/test splits.
- `pegasus_cnn_cnn_pls` is also downloaded from [Saved Pseudo-Labels](https://github.com/huggingface/transformers/blob/main/examples/research_projects/seq2seq-distillation/precomputed_pseudo_labels.md). It is generated by sshleifer/pegasus-cnn-ft-v2, and it corresponds to version "2.0.0". It only includes the train split.

## Updates
- 03/16/2023
1. Remove "(CNN)" in the beginning of articles.

原始信息汇总

Distilled CNN/DailyMail Dataset 概述

数据集组成

• cnn_bart_pl:

  • 来源: Saved Pseudo-Labels
  • 生成模型: facebook/bart-large-cnn
  • 版本: "1.0.0"
  • 包含内容: 训练集、验证集、测试集

• pegasus_cnn_cnn_pls:

  • 来源: Saved Pseudo-Labels
  • 生成模型: sshleifer/pegasus-cnn-ft-v2
  • 版本: "2.0.0"
  • 包含内容: 训练集

更新记录

• 2023年3月16日
  • 移除了文章开头的"(CNN)"字样。