Documentation

HCCL API Overview

HCCL provides C and Python development APIs to implement distributed capabilities.
  • The C APIs are used to implement framework adaptation in single-operator mode.

    For the PyTorch network, the HCCL single-operator API has been embedded in the backend code of Ascend Extension for PyTorch. PyTorch users can directly use the native PyTorch collective communication APIs to implement distributed capabilities.

  • Python APIs are used to implement framework adaptation in graph mode. Currently, they are only used to implement distributed optimization of the TensorFlow network in the Ascend AI Processor.
Figure 1 Collective communication API overview