Framework-based Adaptation

Context

This mode is recommended for the scenario where the needed operators have been implemented in the CANN operator library but not been adapted to a third-party framework. You can look up operators in Operator Acceleration Library API Reference.

For example, if an operator that has been adapted to the TensorFlow framework but not the ONNX framework exists in the operator library, you can directly adapt the operator to the ONNX framework.

Development Workflow

The following figure shows the development workflow of framework-based adaptation.

Figure 1 Framework-based operator adaptation

**Table 1** Operator development workflow
Action	Description	See Also
Environment setup	Set up the development and operating environments required for operator development, execution, and verification.	Environment Setup
Project creation	You are advised to develop an operator plugin based on the existing operator project. If no operator project exists, you can directly download the operator sample from the open-source Ascend community (see the See Also column). NOTICE: You can skip project creation if you are adapting a PyTorch operator.	Using Operator Samples
Operator adaptation	To develop a custom operator in a third-party framework (such as TensorFlow/Caffe/ONNX), develop a plugin capable of mapping an operator developed based on a third-party framework to one supported by Ascend AI Processor.	Operator Adaptation
Operator verification on network	Load the custom operator to a network model and execute it for verification.	Operator Verification on Network

Parent topic: Development Scenario Selection