TensorFlow Commercial Version
Ascend provides the TF Adapter (TensorFlow adapter) plugin, which is compatible with mainstream TensorFlow deep-learning frameworks and accelerates the execution of TensorFlow networks on the Ascend platform. Currently, TF Adapter supports only TensorFlow 1.15 and TensorFlow 2.6.5.
Model Porting
- TensorFlow 1.15 Model Porting Guide
Describes how to port training scripts developed based on the TensorFlow 1.15 Python APIs to the Ascend platform and tune the accuracy and performance.
- TensorFlow 2.6.5 Model Porting Guide
Describes how to port training scripts developed based on the TensorFlow 2.6.5 Python APIs to the Ascend platform and tune the accuracy and performance.
Scope Fusion Pattern
- TensorFlow Parser Scope Fusion Pattern Development
Describes how to develop scope fusion patterns of the TensorFlow framework to improve the execution efficiency of TensorFlow computational graphs on Ascend AI Processors.
- TensorFlow Parser Scope Fusion Patterns
Describes built-in scope fusion patterns of Ascend AI Processors. Scope fusion is a scope-based fusion capability, which is used to improve efficiency.
API
- TF Adapter API (1.x)
Describes APIs and configurations provided by TF Adapter to adapt to the TensorFlow 1.15 framework.
- TF Adapter API (2.x)
Describes APIs and configurations provided by TF Adapter to adapt to the TensorFlow 2.6.5 framework.
- TensorFlow Parser Scope Fusion Pattern Development API
Describes APIs provided by TensorFlow Parser to develop scope fusion patterns of the TensorFlow framework.