HOTFIX: tqdm version for datasets

Merge pull request #222 from thunlp/dev

Dev

Version 2.1.0

• Fix a series of issues and add some functions (#122, #132, #150, #153, #211, #212, #214, #216)
• Re-organize README.md
• Re-design the document, and add more examples
• Add some new attack methods

Version 2.0.1

• Add more evaluation metrics
• Fix a bug in PSO and accelerate the attack process
• Fix a bug in BERT-Attck

Version 2.0.1

• Add more evaluation metrics
• Fix a bug in PSO and accelerate the attack process
• Fix a bug in BERT-Attck

Version 2.0.0

• Added support for multi-processing attacks. #113
• Fixed a potential bug in wordnet substitute. #114
• Fixed a problem caused by Python3.8(or above) urllib package when download resources. #123
• Fixed an encoding problem when loading counter-fit-word2vec on Windows. #124
• Added support for Chinese datasets and attacks. #131, #147
• Added CDN to accelerate global download speeds. #150
• Added support for transformers 4. #151
• Fixed a problem when loading Universal Sentence Encoder. #178

Version 1.1.1

Fix the bug when running on Windows in #116.

• Add models including bert, xlnet, albert and roberta on datasets including sst-2, ag-news, mnli, snli and imdb.

Version 1.1.0

Version 1.1.0

• Added Dataset and DataInstance to manage dataset and sentences instead of a list of tuples.
• Modified "get_grad" interface definition.
• Added an example of adversarial training
• Updated documents, added PyPI installation.
• Added GloveSubstitute and Word2VecSubstitute