How to use this library??

[gauravkoradiya]

How to use this library??

[mayurbhangale]

@beekbin added a script to do all that you have mentioned above in
this pull request.

[beekbin]

Hi @gauravkoradiya

How to use this library

• Step0: Download the pre-trained BERT models for pytorch from huggingface /pytorch_transformers/modeling_bert.py. In the simplest case, you have to download three things:
  (1) a pre-trained model, for example:
  bert-base-uncased': "https://s3.amazonaws.com/models.huggingface.co/bert/bert-base-uncased-pytorch_model.bin",
  (2) the config file:
  'bert-base-uncased': "https://s3.amazonaws.com/models.huggingface.co/bert/bert-base-uncased-config.json",
  (3) the vocal file in tokenization_bert.py:
  'bert-base-uncased': "https://s3.amazonaws.com/models.huggingface.co/bert/bert-base-uncased-vocab.txt",

• Step1: Fine-tune the BertPairSim on your dataset or on the STS-B dataset;
  The run.sh script provides an example to do the fine-tuning on the STS-B dataset.

• Step2: After fine-tuned, the BertPairSim model can be used to generate sentence embeddings, and the embedding can be used to compute the Cosine similarity between sentences.

Dataset
In the example of run.sh, I am using the STS-B dataset.

The STS-B dataset can be downloaded from the GLUE GitHub:
https://github.com/nyu-mll/jiant/blob/master/scripts/download_glue_data.py