mindiesd 2.9.0

MindIE Stable Diffusion inference engine for Ascend NPUs

MindIE SD

MindIE SD is an Ascend-focused inference acceleration toolkit for Stable Diffusion and related multimodal generation workloads.

Highlights

• Ascend-friendly custom operators and fused kernels
• Quantization, sparse quantization, cache, and parallel execution features
• torch.compile integration for graph-level acceleration

Quick start

Complete the environment preparation and MindIE SD installation first, then install model-specific dependencies and run an example:

git clone https://modelers.cn/MindIE/Wan2.1.git
cd Wan2.1
pip install -r requirements.txt

Documentation

• English overview: docs/en/README.md
• English quick start: docs/en/quick_start.md
• English installation guide: docs/en/installing_guide.md
• English user guide: docs/en/menu_user_manual.md
• English architecture: docs/en/architecture.md
• English environment variables: docs/en/environment_variable_configuration.md
• English support matrix: docs/en/features/supported_matrix.md
• English cache features: docs/en/features/cache.md
• English parallelism features: docs/en/features/parallelism.md
• English sparse quantization: docs/en/features/sparse_quantization.md
• Chinese overview: docs/zh/index.md
• Chinese quick start: docs/zh/quick_start.md
• Chinese installation guide: docs/zh/installing_guide.md
• Chinese architecture: docs/zh/architecture.md

Community and governance

• Contributing: contributing.md
• Code of conduct: CODE_OF_CONDUCT.md
• Governance: docs/en/community/governance.md