SpERT torchserve
Spert_torchserve is the Relation Extraction model (SpERT)Span-based Entity and Relation Transformer API deployed with pytorch/serve.
Install Requirements
pip install -r requirements.txt
Get your pretrained model
Train your own model with SpERT or use model trained by CoNLL04 from Huggingface. Only pytorch_model.bin is required to be archived. Download it here and save it into model directory.
You can also save your own model at Huggingface with large file storage.
Archive your model
Before serving model, pack the model and other configs into a .mar file.
torch-model-archiver --model-name spert --version 1.0 --model-file models.py --serialized-file model/pytorch_model.bin --export-path model_store --extra-files entities.py,input_reader.py,loss.py,prediction.py,sampling.py,conll04_types.json --handler spert_handler
This command pack all the needed files into a spert.mar file in model_store fold.
Serve the model
Start the service using the spert.mar with command:
torchserve --start --model-store model_store --models my_tc=spert.mar --ncs
Test
Post the text example "In 1822, the 18th president of the United States, Ulysses S. Grant, was born in Point Pleasant, Ohio" to:
http://127.0.0.1:8080/predictions/my_tc
A prediction would be returned:
{ "tokens": [ "In", "1822", ",", "the", "18th", "president", "of", "the", "United", "States", ",", "Ulysses", "S.", "Grant", ",", "was", "born", "in", "Point", "Pleasant", ",", "Ohio" ], "entities": [ { "type": "Loc", "start": 8, "end": 10 }, { "type": "Peop", "start": 11, "end": 14 }, { "type": "Loc", "start": 18, "end": 22 } ], "relations": [ { "type": "Live_In", "head": 1, "tail": 2 } ] }
References
Markus Eberts, Adrian Ulges. Span-based Joint Entity and Relation Extraction with Transformer Pre-training. 24th European Conference on Artificial Intelligence, 2020.
6 May 22, 2022
11 Nov 17, 2022
1 Oct 28, 2021
48 Dec 06, 2022
586 Dec 24, 2022
858 Dec 27, 2022
293 Jan 02, 2023
1.3k Jan 03, 2023
3 Jan 24, 2022
4 Feb 27, 2022
1 Jan 16, 2022
45 Nov 21, 2022
54 Dec 03, 2022
3 Jan 14, 2022
4 Jan 20, 2022
1 Oct 18, 2021
1 Feb 02, 2022
10 Oct 21, 2022
42 Jul 09, 2022
8.4k Dec 30, 2022