Before You Start

Configure environment variables by referring to Environment Setup. Then, you can directly use the Performance Modeling function of msKPP.
- You can create an operator model based on the msKPP APIs in any directory. Pay attention to the following points during the implementation:
  - Before modeling an operator, you need to import instructions for tensor, chip, and operator implementation (all named in lowercase).
  - Use the with statement to enable the entry of the operator implementation code. The enable_trace and enable_metrics APIs can be used to enable the trace dotting diagram and instruction statistics functions. For details, see main.py in Analyzing Extreme Performance.
  - For details about the instruction API for operator modeling, see External API Usage Description.
- If you need the instruction proportion pie chart (instruction_cycle_consumption.html), install the third-party Python library plotly as it is a dependency for generating the pie chart.
```
pip3 install plotly
```
To use the Auto Tuning function, download the Ascend C template library from Link.
Ensure that the input data is reliable and secure during secondary development.

Parent topic: msKPP (Operator Design)