How to bring in new_tensor into cuda kernel

Seungsu · September 9, 2021, 4:38pm

Hi.

I study about EmbeddingBag execution and i want to convey new tensor into ATen/native/cuda kernel.

I proceeded by adding parameters in related functions while following the method of passing weights.

But i got this error.

RuntimeError: isDifferentiableType(variable.scalar_type()) INTERNAL ASSERT FAILED at “torch/csrc/autograd/functions/utils.h”

Do you give me a advice when this error occurs exactly??