What is MUSE?
MUSE stands for Multilingual Universal Sentence Encoder - multilingual extension (16 languages) of Universal Sentence Encoder (USE).
MUSE/USE models encode sentences into embedding vectors of fixed size.
MUSE paper: link.
USE paper: link.
USE Visually Explainer article: link.
What is MUSE as Service?
MUSE as Service is REST API for sentence tokenization and embedding using MUSE.
It is written on flask + gunicorn.
You can configure gunicorn with gunicorn.conf.py file.
Installation
# clone repo
git clone https://github.com/dayyass/muse_as_service.git
# install dependencies
cd muse_as_service
pip install -r requirements.txt
Run Service
To launch a service use a docker container (either locally or on a server):
docker build -t muse_as_service .
docker run -d -p 5000:5000 --name muse_as_service muse_as_service
NOTE: you can launch a service without docker using gunicorn: sh ./gunicorn.sh, or flask: python app.py, but it is preferable to launch the service inside the docker container.
NOTE: instead of building a docker image, you can pull it from Docker Hub:
docker pull dayyass/muse_as_service
Usage
After you launch the service, you can tokenize and embed any {sentence} using GET requests ({ip} is the address where the service was launched):
http://{ip}:5000/tokenize?sentence={sentence}
http://{ip}:5000/embed?sentence={sentence}
You can use python requests library to work with GET requests (example notebook):
import numpy as np
import requests
ip = "localhost"
port = 5000
sentence = "This is sentence example."
# tokenizer
response = requests.get(
url=f"http://{ip}:{port}/tokenize",
params={"sentence": f"{sentence}"},
)
tokenized_sentence = response.json()["content"]
# embedder
response = requests.get(
url=f"http://{ip}:{port}/embed",
params={"sentence": f"{sentence}"},
)
embedding = np.array(response.json()["content"][0])
# results
print(tokenized_sentence) # ['▁This', '▁is', '▁sentence', '▁example', '.']
print(embedding.shape) # (512,)
But it is better to use the built-in client MUSEClient for sentence tokenization and embedding, that wraps the functionality of the requests library and provides the user with a simpler interface (example notebook):
from muse_as_service import MUSEClient
ip = "localhost"
port = 5000
sentence = "This is sentence example."
# init client
client = MUSEClient(
ip=ip,
port=port,
)
# tokenizer
tokenized_sentence = client.tokenize(sentence)
# embedder
embedding = client.embed(sentence)
# results
print(tokenized_sentence) # ['▁This', '▁is', '▁sentence', '▁example', '.']
print(embedding.shape) # (512,)
Citation
If you use muse_as_service in a scientific publication, we would appreciate references to the following BibTex entry:
@misc{dayyass_muse_as_service,
author = {El-Ayyass, Dani},
title = {Multilingual Universal Sentence Encoder REST API},
howpublished = {\url{https://github.com/dayyass/muse_as_service}},
year = {2021},
}
35 Sep 17, 2022
13 Dec 13, 2022
2 Feb 10, 2022
189 Jan 2, 2023
74 Oct 7, 2022
2.1k Jan 7, 2023
652 Jan 6, 2023
213 Jan 01, 2023
12 Dec 14, 2022
21 Dec 26, 2022
65 Sep 21, 2022
98 Dec 09, 2022
17 Dec 23, 2022
1 Jan 29, 2022
395 Nov 21, 2022
5.9k Jan 03, 2023
4.9k Jan 04, 2023
52 Jan 05, 2023
192 Dec 23, 2022
71 Jan 04, 2023
12.8k Jan 03, 2023
2 Feb 17, 2022
11 Sep 25, 2022
211 Dec 28, 2022
29 Nov 30, 2022
1 Nov 18, 2021
11 Jun 25, 2022