RuntimeError: aten.addmm.default: got mixed torch.Tensor and DTensor, need to convert all torch.Tensor to DTensor before calling distributed operators!

[wangguangyuan]

The Python script wan_14b_text_to_video_tensor_parallel.py is reporting an error. Have you successfully run it locally? Does it have any advantages compared to USP?

[Artiprocher]

@wangguangyuan We have tested the script on Linux (8*A100). Tensor parallel is slower than USP.

[liyaxuanliyaxuan]

same issue