Skip to content

Latest commit

 

History

History
79 lines (58 loc) · 2.46 KB

demo.md

File metadata and controls

79 lines (58 loc) · 2.46 KB
Raw

#HOW TO BUILD?

  • You are supposed to have a Conda and pip. If not, then please download:

Install conda Install pip

  • After installing conda, you need to create a new virtual environment to make a balance within dependencies. Please run this commands on a shell in order:
conda create -n simpletransformers python pandas tqdm python=3.7
conda activate simpletransformers
conda install pytorch cudatoolkit=11.0 -c pytorch
pip install simpletransformers
conda install -c conda-forge keras
conda install -c conda-forge h5py=2.10.0
conda install -c anaconda flask

  • Now, you are done with dependencies. The github repo that you have installed contains only for seq2seq LSTM checkpoints. Not T5. We used seq2seq LSTM as a baseline model. T5 results state-of-the-art.

  • Please download all files in this drive link

  • Extract the compressed file into the main folder of project.

  • You must have now a folder called "checkpoint-15982-epoch-1". The file tree structure is then:

  • checkpoint-15982-epoch-1

    • training_args.bin
    • tokenizer_config.json
    • spiece.model
    • special_tokens_map.json
    • scheduler.pt
    • pytorch_model.bin
    • optimizer.pt
    • model_args.json
    • eval_results.txt
    • config.json

  • This "checkpoint-15982-epoch-1" file must be in same directory with app.py, generate.py, request.py

  • After running $ls command the output must be look like this

    • app.py
    • crawler
    • generate_lstm.py
    • generate.py
    • helpers
    • lstm_seq2seq
    • models
    • prep_data.py
    • process
    • request.py
    • requirements.sh
    • static
    • T5
    • templates
    • train_lstm.py
    • utils
    • readme.txt
    • docs

  • Now you are ready to play! Run this commands to start the demo web page on local:

conda activate simpletransformers
python3 app.py

  • Open up the browser and go to http://127.0.0.1:5000/ (i think this url is constant, however it might be different in other machines. After running the command you can see your local url address, just open this url)

  • We provide you to some abstracts to get a recommendation based on those. This abstracts in docs directory.

  • Copy a paper abstract and paste it to T5 input section or lstm input section (T5 is SotA).

  • Press the button! (T5 is relatively slow to predict, about 10 seconds).