Xilinx/Vitis-AI

In the UG1414, the operators supported by PyTorch are not available in the process of training and quantification

Ruiii1 opened this issue · 0 comments

Ruiii1 commented

Software Environment:
System: Ubuntu 20.04
Docker Image: Vitis AI 3.0
Manual: ug1414(v3.0)

Hello everyone, I have encountered the following questions while modifying network operators according to the manual:

  1. I have seen the supported operators in the manual, but I am unable to use them. For example, I can import Add, AddScalar, Cat, Mul, Matmul, Max, Sum, etc., from pytorch_nndct/nn/modules/functional.py, but I am unable to use operators like sub and neg mentioned in the manual. How should I use them?

  2. If an operator is not supported, how can I customize it? I have checked the official manual but still do not understand how to proceed. For example, how should I customize the torch.div operation?

I hope the official team can provide some ideas or relevant guidelines. Thank you very much.