cancel
Showing results for 
Show  only  | Search instead for 
Did you mean: 
Nirvana
Visitor
Visitor
200 Views
Registered: ‎04-03-2021

Quantization issue (Pytorch) : " aten_op '_convolution' parse failed(list index out of range)"

Jump to solution

Hello, 

I'm facing this issue while i'm trying to quantize my CNN model with VITIS-AI using this command : 

python quant_model.py --quant_mode calib --subset_len 200

I am getting this output : 

[NNDCT_NOTE]: Loading NNDCT kernels...
-------- Start model test

[NNDCT_NOTE]: Quantization calibration process start up...

[NNDCT_NOTE]: =>Quant Module is in 'cpu'.

[NNDCT_NOTE]: =>Parsing Model...
aten_op '_convolution' parse failed(list index out of range)
(vitis-ai-pytorch) Vitis-AI /workspace/workspace/model >

 While debuging, i found that the bug is cause by the torch_quantizer() function call.

The first layer of my CNN model is torch.nn.Conv1d type, is it because vitis-ai doesn't support this function ?

Does any one have any idea of how to fix this issue ?

 

Thank you.

0 Kudos
1 Solution

Accepted Solutions
niuxj
Xilinx Employee
Xilinx Employee
90 Views
Registered: ‎08-10-2020

Hi, @Nirvana :

So far vai_q_pytorch does not support conv1D. And it will be supported in Vitis-AI 1.4.

You can try another model if just learn about the how to use vai_q_pytorch.

View solution in original post

2 Replies
niuxj
Xilinx Employee
Xilinx Employee
91 Views
Registered: ‎08-10-2020

Hi, @Nirvana :

So far vai_q_pytorch does not support conv1D. And it will be supported in Vitis-AI 1.4.

You can try another model if just learn about the how to use vai_q_pytorch.

View solution in original post

Nirvana
Visitor
Visitor
62 Views
Registered: ‎04-03-2021

Thank you @niuxj 

0 Kudos