CANN Commercial Version

Describes how to install, upgrade, and uninstall CANN in different operating systems and service scenarios.

Describes how to develop AI applications using the C&C++ and Python APIs to implement functions such as target recognition and image classification. For details about the APIs, see Application Development APIs.

Describes how to tune algorithms and functions related to the image signal processing (ISP). For details about the APIs, see ISP Image Tuning APIs.

Describes how to develop operators based on the Ascend C operator programming language. For details about the APIs, see Ascend C Operator Development APIs.

Describes how to develop TBE and AI CPU custom operators based on TBE and AI CPU APIs. For details about the APIs, see TBE & AI CPU Operator Development APIs.

Describes how to use the BiSheng compiler to compile operator code into binary executable files and dynamic libraries.

Describes the CCE Intrinsic–based heterogeneous programming and multi-pipeline parallel programming. For details about the APIs, see CCE Intrinsic Development APIs.

Describes the intermediate representation (IR) built based on MLIR, which is used for building Ascend affinity operators.

Describes how to build a graph using the GE APIs. For details about the APIs, see GE Graph Engine APIs.

Describes how to build a graph (FlowGraph) based on DataFlow C++ and Python APIs. For details about the APIs, see DataFlow Graph Build APIs.

Provides a high-performance collective communication library based on the Ascend AI processors, which supports data-parallel and model-parallel collective communication between single-server multi-device and multi-server multi-device, and supports custom development of communication operators. For details about the APIs, see HCCL APIs.

Provides guidance on developing unilateral communication libraries. It is used to guide developers on how to use the unilateral communication library APIs to implement data transmission between clusters and build a separated framework for foundation model inference. For details about related APIs, see HIXL Unilateral Communication Library APIs.

Describes how to use the Ascend Transformer Boost acceleration library to improve the efficiency of Transformer model training and inference development. For details about the APIs, see ATB Acceleration Library APIs.

Describes how to use high-performance operators in the signal processing domain. For details about related APIs, see SiP Acceleration Library APIs.

Deploys the inference of the foundation model separately using the LLM DataDist APIs to improve the throughput performance of the foundation model inference. For details about the APIs, see LLM DataDist APIs.

Provides C&C++ and Python APIs for system configuration, runtime management, single-operator execution, model execution, and media data preprocessing.

Provides image tuning APIs of the ISP.

Provides basic and high-level Ascend C APIs.

Provides APIs required for TBE & AI CPU operator development.

Provides Ascend hardware APIs extended based on the C language, and implements fine-grained control over memory allocation, data synchronization, and double buffer through the CCE Intrinsic APIs.

Constructs graphs that run directly on the Ascend platform through GE APIs.

Constructs, modifies, compiles, and executes computational graphs through DataFlow C++ and Python APIs, and provides UDF APIs for users to write custom processing functions through FuncProcessPoint and GraphProcessPoint.

Provides communication operator APIs and communicator management APIs to implement distributed capabilities. In addition, HCCL provides communication operator development APIs for developers to customize communication operators.

Provides APIs in C++ and Python to provide simple, reliable, and efficient point-to-point data transmission capabilities in cluster scenarios.

Provides a wide array of high-performance operators with deep optimization and hardware affinity.

Provides APIs required for using the ATB acceleration library, including the public class definitions such as the Operation class, single-operator class, and graph operator class.

Provides APIs required for using the SiP acceleration library, including high-performance operators related to signal processing.

Provides APIs to manage KV data in a cluster, supporting separate deployment of full graphs and incremental graphs.

Provides APIs for automatic tuning, allowing you to query previously generated repository files through the APIs and obtain the tiling result.

Describes the basic data structures and APIs on which operator development and graph development depend.

Describes APIs on which the CANN open code depends, including the error reporting APIs and log APIs.

Provides quick start for using the PyTorch training scenario development tool, foundation model inference development tool, and operator development tool.

Describes how to use operator development tools (such as msKPP, msOpGen, msOpST, msSanitizer, msDebug, and msProf).

Compiles operators to generate operator binary files.

Converts a network model into an offline model (.om) supported by Ascend AI processors.

Implements automatic optimization to make full use of hardware resources and improve network performance.

Migrates the PyTorch training script to the Ascend NPU in one-click mode.

Allows you to compare accuracy and locate model accuracy problems.

Collects and analyzes profile data in the training and inference phases.

Tests the correctness and performance of the HCCL function.

Compresses a model, including quantization and tensor decomposition.

Locates memory problems during model training and inference.

Describes how to quickly locate and rectify faults.

Describes the RDMA-based network detection technology RPing, which is used to send detection packets, record network latency, and collect statistics on packet sending and receiving.

Describes the log content format, and how to view logs and set log levels.

Describes the environment variables that can be used to build AI applications and services based on CANN.

Provides some built-in graph fusion and UB fusion patterns of Ascend AI processors. Graph fusion and UB fusion are key methods for improving the performance of the entire network.

Describes the open ports, transport layer protocols used by the ports, authentication modes, and functions.