Installation Guide

Python Package Installation

MindIE SD is a Python package built on PyTorch that can be easily integrated into Python applications.

Dependencies

• OS: Linux
• Python: >=3.10
• PyTorch: 2.6, 2.7, 2.8, 2.9, 2.10
• torch-npu: 2.6, 2.7, 2.8, 2.9, 2.10
• CANN: 9.0.0
• triton: 3.5.0
• triton-ascend: 3.2.1

CANN Installation

MindIE SD depends on the CANN Toolkit development package and CANN ops operator package. See the CANN Software Installation Guide for installation instructions. Choose the installation scenario based on your installation method and operating system, then click "Start Reading" and follow the "Install CANN" section.

After installation, run the following command to set environment variables (using the default installation path as an example):

source /usr/local/Ascend/ascend-toolkit/set_env.sh

Notes

1. MindIE SD primarily depends on the torch-npu version and will try to meet the CANN and Python version requirements mandated by torch-npu.
2. After CANN installation, the installation path provides a process-level environment variable setup script set_env.sh to automatically configure environment variables. This script includes LD_LIBRARY_PATH and ASCEND_CUSTOM_OPP_PATH as shown in Table 1 Environment Variables. These settings automatically expire when the user process ends.

Table 1 Environment Variables

Environment Variable	Description
LD_LIBRARY_PATH	Dynamic library search path.
ASCEND_CUSTOM_OPP_PATH	Custom operator package installation path for inference engine.
ASCEND_RT_VISIBLE_DEVICES	Specifies the logical IDs of Ascend AI processors used by the current process. Configure as needed. Example: "0,1,2" or "0-2"; Ascend AI processor logical IDs are separated by "," and consecutive IDs use "-".

Quick Install

Currently the simplest way is to install via pip. Our package is named mindiesd, which differs from the repository name.

pip install mindiesd

Source Build

In some cases, you may need to install MindIE SD from source to try the latest features or customize the library for your specific needs.

Follow these steps to install MindIE SD from source:

1. Clone the repository and enter the project:

   git clone https://github.com/MindIE-SD/MindIE-SD.git && cd MindIE-SD
   

2. [Optional] Install dependencies:

   pip install -r requirements.txt
   

3. Build and install:

   python setup.py bdist_wheel
   cd dist
   pip install mindiesd-*.whl
   

Nightly Build Installation

Nightly builds are available for testing the latest features:

Coming soon...