Tuning Workflow

For the Atlas A2 training product/Atlas A2 inference product, tuning in offline inference is not supported.

For the Atlas A3 training product/Atlas A3 inference product, tuning in offline inference is not supported.

Figure 1 shows the AOE tuning procedure. For detailed operations, see Table 1.

Figure 1 Tuning procedure (using AOE)

You are advised to perform subgraph tuning and then operator tuning. The reason is that performing subgraph tuning first can generate the graph partition mode. After subgraph tuning is complete, the operators are partitioned into the final shapes. Operator tuning can then be performed based on the final shapes. If operator tuning is performed first, the shapes of the tuned operators are not the final shapes after operator partitioning. This does not meet the actual application scenarios.

**Table 1** Tuning procedure description (using AOE)
Operation	Reference
Set up the development environment and operating environment.	Environment Setup
Configure environment variables.	Environment Variable Configuration
Perform tuning.	Tuning Procedure

Parent topic: Tuning in Offline Inference Scenarios