NPU_ENABLE_PERF

Description

Prints graph time consumption of the TF Adapter in TensorFlow 2.6.5 training and online inference scenarios.

• 1 or true: Prints graph time consumption.
• 0 or false: Does not print graph time consumption.

After enabling the TF Adapter graph time consumption printing function, you can search for the keywords Graph engine run and cost in logs to query the time consumption. For example:

Graph engine run 1 times for graph 0 cost 11257 ms

Example

export NPU_ENABLE_PERF=1

Constraints

• This environment variable takes effect only when the debug logging for TF Adapter is enabled. That is, set NPU_DEBUG to 1 or true.

  export NPU_DEBUG=1

• This environment variable should be set before the import npu_device operation.
• This environment variable applies only to the scenario where the TensorFlow 2.6.5 network training or online inference is performed on the Ascend platform.

Applicability

Atlas 350 Accelerator Card

Atlas A3 training product/Atlas A3 inference product

Atlas A2 training product/Atlas A2 inference product

Atlas training product