Software Package Preparation

The MindIE, CANN, and Ascend Extension for PyTorch versions must match. Table 1 lists the version mapping.

**Table 1** Version mapping
MindIE	CANN	Ascend Extension for PyTorch
2.3.0	8.5.0	PyTorch framework and torch_npu plugin (2.1.0): 7.2.0 PyTorch framework and torch_npu plugin (2.6.0): 7.3.0

Image Installation

For details about how to obtain and install the MindIE image, see Mode 1: Image Installation.

PM/Container Installation

Once you download this software, you agree to the terms and conditions of Huawei Enterprise End User License Agreement (EULA). For details about the PM and container installation modes, see Mode 2: PM Installation and Mode 3: Container Installation, respectively.

**Table 2** Software package list
Software	Package Name	Description	How to Obtain
MindIE	Ascend-mindie_<version>_linux-<arch>_<abi>.run The PyTorch framework version matching Ascend-mindie_<version>_linux-<arch>_abi0.run is 2.1.0. The PyTorch framework version matching Ascend-mindie_<version>_linux-<arch>_abi1.run is 2.6.0.	Inference engine software package, which is used to develop applications based on MindIE.	Link
CANN	Ascend-cann-toolkit_<version>_linux-<arch>.run	CANN development kit (Toolkit).
	Ascend-cann-<chip_type>-ops_<version>_linux-<arch>.run	CANN binary operator package (ops). NOTE: Before installing the ops, the Toolkit software package of the same version must be installed. Select the ops software package corresponding to the running device.
	Ascend-cann-nnal_<version>_linux-<arch>.run	CANN neural network acceleration library (NNAL).
ATB Models	Ascend-mindie-atb-models_<version>_linux-<arch>_pyxxx_torchx.x.x-<abi>.tar.gz	ATB Models installation package. This component needs to be installed when MindIE Motor and MindIE LLM are used.
Ascend Extension for PyTorch	torch_npu-<torch_version>.post<post_id>-cpxxx-cpxxx-manylinux_<arch>.whl	WHL package of the torch_npu plugin.	Link NOTE: To obtain torch_npu 2.1.0, select 7.2.0 from the PyTorch drop-down list in the Matching Resources area on the upper left of the page for downloading resources of the community edition. In the PyTorch area, click the Obtain Source Code button next to your target version to go to the GitCode repository of PyTorch, and then download torch_npu.
	apex-<apex_version>_ascend-cpxxx-cpxxx-<arch>.whl	WHL package of the Apex module.	Compile it based on Python3.10. For details, see "Installing the Apex Module" in Ascend Extension for PyTorch Software Installation Guide.
	torch-<torch_version>+cpu-cpxxx-cpxxx-linux_<arch>.whl	WHL package of the PyTorch framework. NOTE: Version 2.1.0 is recommended.	PyTorch framework and torch_npu plugin (2.1.0): Obtain it from "Installing the PyTorch Framework" in Ascend Extension for PyTorch Configuration and Installation. PyTorch framework and torch_npu plugin (2.6.0): Obtain it from "Installing PyTorch" in Ascend Extension for PyTorch Software Installation Guide.
Note: <version>, <torch_version>, and <apex_version> indicate the software versions. <arch> indicates the CPU architecture. <chip_type> indicates the processor type. <abi> indicates the ABI version.

To avoid using software packages that have been tampered with during transmission or storage, download their digital signature files for integrity check while downloading the software packages.

Click PGP Verify to obtain the tool package, decompress it and then verify the PGP digital signature. For details, see OpenPGP Signature Verification Guide. If the verification fails, do not use the software package. Visit the support website to get help from the community or submit a service ticket.