There are three options:
distilroberta - fastest but least efficient:
Download the model from https://drive.google.com/file/d/1mdDdvwat3M6OIbSCmKnBOqNFF2iz3vdp/view?usp=sharing
and put it as smart_word_hints_api/app/assets/models/wsd_distilroberta.bin
The other models have to be downloaded from the repository of the authors of ESR: https://github.com/nusnlp/esr#using-trained-models.
roberta-base - slower but better:
Copy experiment/esr/roberta-base/dataset_semcor_wngc/sd_42/rtx3090_b32_b32_lr8.5e-6/model/pytorch_model.bin to
smart_word_hints_api/app/assets/models as wsd_roberta_base.bin.
Then, change model_name in config.ini to roberta-base.
roberta-large - slowest but best:
Copy experiment/esr/roberta-large/dataset_semcor_wngc/sd_42/a100_b16_b16_lr8.5e-6_lim348/model/pytorch_model.bin to
smart_word_hints_api/app/assets/models as wsd_roberta_large.bin.
Then, change model_name in config.ini to roberta-large.
./scripts/run_local_docker.sh 8081
virtualenv -p python3.9 venv
source venv/bin/activate
pip install -r smart_word_hints_api/requirements/requirements.txt
python -m spacy download $(cat smart_word_hints_api/requirements/spacy_modules.txt)
python -m nltk.downloader $(cat smart_word_hints_api/requirements/nltk_modules.txt)
git clone https://github.com/mihal277/esr.git -b replace-getchildren smart_word_hints_api/app
./scripts/run_local.sh
After running the above commands:
- the main API endpoint is available at
localhost:8081/api/get_hints - the API docs are available at
localhost:8081/docs
To run tests without Docker:
./scripts/run_api_tests.sh
To run tests in Docker:
./scripts/run_api_tests_docker.sh
To make sure tests are automatically run before push:
git config core.hooksPath hooks/