Tuning Workflow

Supported TensorFlow versions include 1.15 and 2.6.5.
This commissioning entry supports the following models:
- Atlas training product
- Atlas A2 training product/Atlas A2 inference product
- Atlas A3 training product/Atlas A3 inference product

There are two types of tuning in the training scenario:

Subgraph and operator tuning: Figure 1 shows the tuning procedure. For detailed operations, see Table 1.
Gradient tuning: Figure 2 shows the tuning procedure. For detailed operations, see Table 2.

Figure 1 Tuning procedure (subgraph and operator tuning)

Figure 2 Tuning procedure (gradient tuning)

**Table 1** Tuning procedure description (subgraph and operator tuning)
Operation	Reference
Set up the development environment and operating environment.	Environment Setup
Configure environment variables.	Subgraph/Operator Tuning
Run the training script.	Subgraph/Operator Tuning

**Table 2** Tuning procedure description (gradient tuning)
Operation	Reference
Set up the development environment and operating environment.	Environment Setup
Configure environment variables.	Gradient Splitting Tuning
Modify the training script.	Gradient Splitting Tuning
Run the training script.	Gradient Splitting Tuning

Parent topic: Online Tuning in TensorFlow-based Training Scenarios