MindSpore

Huawei's open-source deep-learning framework, built mainly to pair with Ascend chips.

MindSpore (Chinese name 昇思) is a deep-learning framework Huawei open-sourced in 2020, occupying the same role as PyTorch or TensorFlow — defining networks, and running training and inference. It supports CPU and GPU, but is optimized primarily for Huawei's Ascend NPUs, calling into the hardware underneath through Ascend CANN (Huawei's chip operator and compiler software stack). It's something you're likely to encounter when training or deploying models in a domestic Chinese compute environment; running a VLA or similar model on an Ascend server or Ascend edge chip for embodied AI generally means either porting PyTorch code over to MindSpore or using PyTorch's Ascend adaptation plugin.

Also called

Shengsi

Related

Compute Architecture for Neural Networks (Huawei Ascend, CANN) · PyTorch · Huawei · Inference Deployment · Huawei Cloud CloudRobo Embodied AI Platform

Sources

昇思 MindSpore 官网 (Chinese)
MindSpore GitHub