https://github.com/cdathuraliya/bert-inference

An attempt to make Google BERT closer to production before Hugging Face Transformers etc.
https://github.com/cdathuraliya/bert-inference

Last synced: about 1 month ago
JSON representation

An attempt to make Google BERT closer to production before Hugging Face Transformers etc.

• Host: GitHub
• URL: https://github.com/cdathuraliya/bert-inference
• Owner: cdathuraliya
• License: apache-2.0
• Created: 2019-03-10T17:11:01.000Z (almost 6 years ago)
• Default Branch: master
• Last Pushed: 2020-09-10T07:30:30.000Z (over 4 years ago)
• Last Synced: 2024-11-02T22:32:55.761Z (about 1 month ago)
• Language: Jupyter Notebook
• Homepage: https://medium.com/@cdathuraliya/online-prediction-and-model-pipelining-for-google-bert-8ace3c85b4f6
• Size: 113 KB
• Stars: 28
• Watchers: 3
• Forks: 3
• Open Issues: 1
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

• awesome-bert - cdathuraliya/bert-inference

README

        
A helper class for Google BERT (Devlin et al., 2018) to support online prediction and model pipelining. Check Jupyter Notebook for the demo. Before running, clone [BERT repo](https://github.com/google-research/bert) into the root directory. Check the pull request [#327](https://github.com/google-research/bert/pull/327) for `Sst2Processor` class. Place actual model files (`ckpt`) under `models` directory.

Based on [this](https://github.com/google-research/bert/issues/94#issuecomment-437883736) issue comment by [@brightmart](https://github.com/brightmart).