BentoML is an open-source framework for high-performance ML model serving.
What does BentoML do?
- Create API endpoint serving trained models with just a few lines of code
- Support all major machine learning training frameworks
- High-Performance online API serving with adaptive micro-batching support
- Model Registry for teams, providing Web UI dashboard and CLI/API access
- Flexible deployment orchestration with DevOps best practices baked-in, supporting Docker, Kubernetes, Kubeflow, Knative, AWS Lambda, SageMaker, Azure ML, GCP and more
๐ To follow development updates and discussion, join the Bentoml Slack community and the contributors mailing list.
Getting Machine Learning models into production is hard. Data Scientists are not experts in building production services and DevOps best practices. The trained models produced by a Data Science team are hard to test and hard to deploy. This often leads us to a time consuming and error-prone workflow, where a pickled model or weights file is handed over to a software engineering team.
BentoML is an end-to-end solution for model serving, making it possible for Data Science teams to build production-ready model serving endpoints, with common DevOps best practices and performance optimizations baked in.
Check out Frequently Asked Questions page on how does BentoML compares to Tensorflow-serving, Clipper, AWS SageMaker, MLFlow, etc.
BentoML requires python 3.6 or above, install with pip
:
pip install bentoml
A minimal prediction service in BentoML looks something like this:
# https://github.com/bentoml/BentoML/blob/master/guides/quick-start/iris_classifier.py
from bentoml import env, artifacts, api, BentoService
from bentoml.adapters import DataframeInput
from bentoml.artifact import SklearnModelArtifact
@env(auto_pip_dependencies=True)
@artifacts([SklearnModelArtifact('model')])
class IrisClassifier(BentoService):
@api(input=DataframeInput())
def predict(self, df):
# Optional pre-processing, post-processing code goes here
return self.artifacts.model.predict(df)
This code defines a prediction service that bundles a scikit-learn model and provides an
API that expects input data in the form of pandas.Dataframe
. The user-defined API
function predict
defines how the input dataframe data will be processed and used for
inference with the bundled scikit-learn model. BentoML also supports other API input
types such as ImageInput
, JsonInput
and
more.
The following code trains a scikit-learn model and packages the trained model with the
IrisClassifier
class defined above. It then saves the IrisClassifier instance to disk
in the BentoML SavedBundle format:
# https://github.com/bentoml/BentoML/blob/master/guides/quick-start/main.py
from sklearn import svm
from sklearn import datasets
from iris_classifier import IrisClassifier
if __name__ == "__main__":
# Load training data
iris = datasets.load_iris()
X, y = iris.data, iris.target
# Model Training
clf = svm.SVC(gamma='scale')
clf.fit(X, y)
# Create a iris classifier service instance
iris_classifier_service = IrisClassifier()
# Pack the newly trained model artifact
iris_classifier_service.pack('model', clf)
# Save the prediction service to disk for model serving
saved_path = iris_classifier_service.save()
By default, BentoML stores SavedBundle files under the ~/bentoml
directory. Users
can also customize BentoML to use a different directory or cloud storage like
AWS S3 and MinIO, via BentoML's
model management component YataiService,
which provides advanced model management features including a dashboard web UI:
Learn more about using YataiService for model management and try out the Web UI here.
The BentoML SavedBundle directory contains all the code, data and configs required to
deploy the model. To start a REST API model server with the IrisClassifier
SavedBundle, use
the bentoml serve
command:
bentoml serve IrisClassifier:latest
The IrisClassifier
model is now served at localhost:5000
. Use curl
command to send
a prediction request:
curl -i \
--header "Content-Type: application/json" \
--request POST \
--data '[[5.1, 3.5, 1.4, 0.2]]' \
http://localhost:5000/predict
The BentoML API server also provides a web UI for accessing predictions and debugging the server. Visit http://localhost:5000 in the browser and use the Web UI to send prediction request:
BentoML provides a convenient way to containerize the model API server with Docker:
-
Find where the SavedBundle directory is created in the file system:
- The saved path is return by the
iris_classifier_service.save()
call - The saved path is printed in the stdout when saving:
INFO - BentoService bundle 'IrisClassifier:20200121114004_360ECB' saved to: ...
- Use the
bentoml get IrisClassifier:latest
command to view all the metadata including saved path
- The saved path is return by the
-
Run
docker build
under the SavedBundle directory which contains a generated Dockerfile, which will build a docker image containing the IrisClassifier API server
# If jq command not found, install jq (the command-line JSON processor) here: https://stedolan.github.io/jq/download/
saved_path=$(bentoml get IrisClassifier:latest -q | jq -r ".uri.uri")
# Build the docker image
docker build -t iris-classifier $saved_path
# Start a container with the image build above
docker run -p 5000:5000 iris-classifier
This docker images makes it possible to deploy BentoML saved bundle to container orchestration platforms such as Kubeflow, Knative, Kubernetes, which provides advanced model deployment features such as auto-scaling, A/B testing, scale-to-zero, canary rollout and multi-armed bandit.
BentoML can also deploy SavedBundle directly to cloud services such as AWS Lambda or AWS SageMaker, with the bentoml CLI command. For a list of all deployment options with BentoML, check out the BentoML deployment guides.
BentoML full documentation: https://docs.bentoml.org/
- Quick Start Guide: https://docs.bentoml.org/en/latest/quickstart.html
- Core Concepts: https://docs.bentoml.org/en/latest/concepts.html
- Deployment Guides: https://docs.bentoml.org/en/latest/deployment/index.html
- API References: https://docs.bentoml.org/en/latest/api/index.html
- Frequently Asked Questions: https://docs.bentoml.org/en/latest/faq.html
Visit bentoml/gallery repository for more examples and tutorials.
- Pet Image Classification - Google Colab | nbviewer | source
- Salary Range Prediction - Google Colab | nbviewer | source
- Sentiment Analysis - Google Colab | nbviewer | source
- Fashion MNIST - Google Colab | nbviewer | source
- CIFAR-10 Image Classification - Google Colab | nbviewer | source
- Fashion MNIST - Google Colab | nbviewer | source
- Text Classification - Google Colab | nbviewer | source
- Toxic Comment Classifier - Google Colab | nbviewer | source
- tf.Function model - Google Colab | nbviewer | source
- Fashion MNIST - Google Colab | nbviewer | source
- Movie Review Sentiment with BERT - Google Colab | nbviewer | source
- Titanic Survival Prediction - Google Colab | nbviewer | source
- League of Legend win Prediction - Google Colab | nbviewer | source
- Titanic Survival Prediction - Google Colab | nbviewer | source
- Loan Default Prediction - Google Colab | nbviewer | source
- Prostate Cancer Prediction - Google Colab | nbviewer | source
- Text Classification - Google Colab | nbviewer | source
- Scikit-Learn Iris Classifier - Google Colab | nbviewer | source
- ResNet50 Image recognition (ONNX model zoo) - Google Colab | nbviewer | source
-
End-to-end deployment management with BentoML
-
Deployment guides for open-source platforms:
-
Deployment guides for Cloud service providers:
Have questions or feedback? Post a new github issue
or discuss in our Slack channel:
Want to help build BentoML? Check out our contributing guide and the development guide.
BentoML is under active development and is evolving rapidly. Currently it is a Beta release, we may change APIs in future releases.
Read more about the latest features and changes in BentoML from the releases page.
BentoML by default collects anonymous usage data using Amplitude. It only collects BentoML library's own actions and parameters, no user or model data will be collected. Here is the code that does it.
This helps BentoML team to understand how the community is using this tool and what to build next. You can easily opt-out of usage tracking by running the following command:
# From terminal:
bentoml config set usage_tracking=false
# From python:
import bentoml
bentoml.config().set('core', 'usage_tracking', 'False')