For detailed README file, please see anthoer repository TorchServe Deployment (simplified Chinese only).

https://pytorch.org/serve/getting_started.html

1. Put script model under root folder of project, then the project structure will be like:

.  
├── script_model.pt     <--- torch script model file  
├── config.properties  
├── curl.sh  
├── data.json  
├── handler.py  
├── index_to_name.json  
├── restart.sh  
└── stop.sh  

1. Edit restart.sh, modify the project root path in line 1 and modify the conda activate path in line 2;
2. Edit curl.sh, modify the service request url and choose the right method to send request;
3. Edit data.json;
4. Edit handler.py to process request data and prediction result;
5. Edit index_to_name.json, mapping index and class name;

Give execute permission to the .sh files:

chmod +x restart.sh curl.sh stop.sh

1. Execute restart.sh to start torchserve service;
2. Execute curl.sh to test torchserve service;
3. Execute stop.sh to stop torchserve service;