MindSpeed-MM Generative Model Training

This Skill guides users through the end-to-end training pipeline for multimodal generative models (text-to-video, text-to-image) on Huawei Ascend NPU.

Prerequisites

Critical: Do NOT use bash scripts/install.sh for generative models. Official MindSpeed-MM docs state install.sh only fully supports Qwen3/Qwen3.5. For Wan2.1 and other generative models, follow the manual install flow below (matches examples/wan2.1/README.md).

Step P1: Clone repositories

git clone https://gitcode.com/Ascend/MindSpeed-MM.git /root/workspace/MindSpeed-MM
git clone https://github.com/NVIDIA/Megatron-LM.git /root/workspace/Megatron-LM
cd /root/workspace/Megatron-LM && git checkout core_v0.12.1
cp -r megatron /root/workspace/MindSpeed-MM/
cd /root/workspace/MindSpeed-MM

Step P2: Install PyTorch + torch_npu