Skip to content

Latest commit

 

History

History
106 lines (66 loc) · 2.86 KB

TUTORIALS.md

File metadata and controls

106 lines (66 loc) · 2.86 KB
Raw

TUTORIALS

A simple example by using provided docker image

This example demonstrates how to use Adlik to compile a serving model and deploy a serving service end to end.

Compile the model

  1. Pull the compiler docker image

    docker pull registry.cn-beijing.aliyuncs.com/adlik/model-compiler:v0.5.0_trt7.2.1.6_cuda11.0

  2. Use the example code to train a simple keras model:

    git clone https://github.com/Adlik/Adlik.git

cd Adlik/examples/keras_model

python3 train_model.py

    Then, a .h5 file will be generated in the “model” directory.

  3. Run the image

    Run the follwing command:

    docker run -it --rm -v $PWD:/home/john/model
registry.cn-beijing.aliyuncs.com/adlik/model-compiler:v0.5.0_trt7.2.1.6_cuda11.0 bash

  4. Compile the model: convert the .h5 file to TensorFlow serving model

    In the docker container:

    cd /home/john/model/

python3 compile_model.py

    After that there will be a serving model in the model_repos directory

Deploy TensorFlow serving service

  1. Pull the image for TensorFlow serving

    docker pull registry.cn-beijing.aliyuncs.com/adlik/serving-tensorflow-cpu:v0.3.0

  2. In the Adlik/examples/keras_model directory, run the command:

    docker run -d -p 8500:8500 -v $PWD/model_repos:/srv/adlik-serving
registry.cn-beijing.aliyuncs.com/adlik/serving-tensorflow-cpu:v0.3.0

Do inference

  1. Install the Adlik serving package

    wget https://github.com/Adlik/Adlik/releases/download/v0.3.0/adlik_serving_api-0.3.0-py2.py3-none-any.whl

python3 -m pip install adlik_serving_api-0.3.0-py2.py3-none-any.whl

  2. Infer with the provided client script:

    Adlik serving provides grpc and http interfaces, grpc interface mainly supports model inference and model management, and http interface mainly supports model inference. For detailed interface description, refer to adlik_serving_apis.md.

    Examples of model inference are as follows:

    Infer with the python script.

    In the Adlik/examples/keras_model directory, run the command:

    python3 mnist_client.py data/0.png

    The result will be as follows:

    Image: 'data/0.png', result: [{'idx': 0, 'score': 0.9999995231628418, 'label': ''}]

    Infer with the cpp script.

    ./cpp_client

    Currently, we provide both bazel and cmake to build cpp client. Please refer to adlik_serving/clients/cpp/README.md for instructions.

More

Refer to the above example, you can also convert PyTorch, Keras, TensorFlow models to OpenVINO, TensorFlow Lite, TVM, TF-TRT and TensorRT serving model and deploy these serving models easily with our provided images.