(For MindSpore 1.2 Only)

Build cloud inference-serving with MindSpore API on CMC training server

Setup & Installation

The detailed instructions are all listed in the documentation below — the content listed on this page is a condensed form from the doc.

MindSpore/docs

• CMC Appulis Configurations

Step 1: Downloading .whl file directly

• Ascend 910
• Arm64 - aarch64
• Python version ****= 3.7.5
• MindsporeServing = 1.3.0
  • mindspore_serving-1.3.0-cp37-cp37m-linux_aarch64.whl from this webpage
• Mindspore-Ascend = 1.5.0

Step 2: Pip Install the wheel file

pip install mindspore_serving-1.3.0-cp37-cp37m-linux_aarch64.whl

RESTful Integration — RESTful-based MindSpore Serving Access

On the CMC Apulis platform, log into your instance and change directory to:

/home/huaweiuser/userdata/serving/example/add

Project Directory Structure

The model in this example is a simple Adder network (hence the model name is ‘add’)

1. <model_name>/servable_config.py — Config file for servable services, this is where you register the APIs (method names) that you wish to support for your model here.
2. master_with_worker.py — Configuration for Master/Worker node for lightweight deployment, this is where you configure the IP and port number for your service.

<aside> 💡 Only POST method is supported with MindSporeServing. The POST-Request format is: POST <http://$>{HOST}:${PORT}/model/${MODLE_NAME}[/version/${VERSION}]:${METHOD_NAME}

</aside>

Inside master_with_worker.py — use the master.start_restful_server API to start the RESTful service. Using the Add network as an example, the following arguments are: