DJL Serving is a high performance universal stand-alone model serving solution powered by DJL. It takes a deep learning model, several models, or workflows and makes them available through an HTTP endpoint. It can serve the following model types out of the box:
You can install extra extensions to enable the following models:
For macOS
brew install djl-serving
# Start djl-serving as service:
brew services start djl-serving
# Stop djl-serving service
brew services stop djl-serving
For Ubuntu
curl -O https://publish.djl.ai/djl-serving/djl-serving_0.19.0-1_all.deb
sudo dpkg -i djl-serving_0.19.0-1_all.deb
For Windows
We are considering to create a chocolatey
package for Windows. For the time being, you can
download djl-serving zip file from here.
curl -O https://publish.djl.ai/djl-serving/serving-0.19.0.zip
unzip serving-0.19.0.zip
# start djl-serving
serving-0.19.0\bin\serving.bat
You can also use docker to run DJL Serving:
docker run -itd -p 8080:8080 deepjavalibrary/djl-serving
Use the following command to start model server locally:
djl-serving
The model server will be listening on port 8080. You can also load a model for serving on start up:
djl-serving -m "https://resources.djl.ai/demo/mxnet/resnet18_v1.zip"
Open another terminal, and type the following command to test the inference REST API:
curl -O https://resources.djl.ai/images/kitten.jpg
curl -X POST http://localhost:8080/predictions/resnet18_v1 -T kitten.jpg
or:
curl -X POST http://localhost:8080/predictions/resnet18_v1 -F "data=@kitten.jpg"
[
{
"className": "n02123045 tabby, tabby cat",
"probability": 0.4838452935218811
},
{
"className": "n02123159 tiger cat",
"probability": 0.20599420368671417
},
{
"className": "n02124075 Egyptian cat",
"probability": 0.18810515105724335
},
{
"className": "n02123394 Persian cat",
"probability": 0.06411745399236679
},
{
"className": "n02127052 lynx, catamount",
"probability": 0.010215568356215954
}
]
# Load models from the DJL model zoo on startup
djl-serving -m "djl://ai.djl.pytorch/resnet"
# Load version v1 of a PyTorch model on GPU(0) from the local file system
djl-serving -m "resnet:v1:PyTorch:0=file:$HOME/models/pytorch/resnet18/"
# Load a TensorFlow model from TFHub
djl-serving -m "resnet=https://tfhub.dev/tensorflow/resnet_50/classification/1"
# Use the default data processing for a well-known application
djl-serving -m "file:/resnet?application=CV/image_classification"
# Specify a custom data processing with a Translator
djl-serving -m "file:/resnet?translatorFactory=MyFactory"
## Pass parameters for data processing
djl-serving -m "djl://ai.djl.pytorch/resnet?applySoftmax=false"
# Load a model from an AWS S3 Bucket
djl-serving -m "s3://djl-ai/demo/resnet/resnet18.zip"
# Load a model from HDFS
djl-serving -m "hdfs://localhost:50070/models/pytorch/resnet18/"
# Use a HuggingFace tokenizer
djl-serving -m "file:/resnet?transaltorFactory=ai.djl.huggingface.BertQATranslator"
djl-serving --help
usage: djl-serving [OPTIONS]
-f,--config-file <CONFIG-FILE> Path to the configuration properties file.
-h,--help Print this help.
-m,--models <MODELS> Models to be loaded at startup.
-s,--model-store <MODELS-STORE> Model store location where models can be loaded.
-w,--workflows <WORKFLOWS> Workflows to be loaded at startup.
See configuration for more details about defining models, model-store, and workflows.
DJL Serving uses a RESTful API for both inference and management calls.
When DJL Serving starts up, it has two web services:
By default, DJL Serving listens on port 8080 and is only accessible from localhost. Please see DJL Serving Configuration for how to enable access from a remote host.
DJL serving is built on top of Deep Java Library. You can visit the DJL github repository to learn more.
It is also possible to leverage only the worker thread pool using the separate WorkLoadManager module. The separate WorkLoadManager can be used to take advantage of DJL serving's model batching and threading and integrate it into a custom Java service.
DJL Serving supports plugins, user can implement their own plugins to enrich DJL Serving features. See DJL Plugin Management for how to install plugins to DJL Serving.
you can set the logging level on the command-line adding a parameter for the JVM
-Dai.djl.logging.level={FATAL|ERROR|WARN|INFO|DEBUG|TRACE}
此处可能存在不合适展示的内容,页面不予展示。您可通过相关编辑功能自查并修改。
如您确认内容无涉及 不当用语 / 纯广告导流 / 暴力 / 低俗色情 / 侵权 / 盗版 / 虚假 / 无价值内容或违法国家有关法律法规的内容,可点击提交进行申诉,我们将尽快为您处理。