[shiftingleft]

Is it really the full pipeline running on Huawei hardware? That is training and inference?

The report only talks about validating the "fine-grained EP scheme" on Huawei hardware.