You are viewing a plain text version of this content. The canonical link for it is here.

Posted to commits@tvm.apache.org by GitBox <gi...@apache.org> on 2021/04/19 08:29:23 UTC

[GitHub] [tvm] masahi edited a comment on issue #7878: PyTorch quantized group conv2d raise an error when converting to tvm

masahi edited a comment on issue #7878:
URL: https://github.com/apache/tvm/issues/7878#issuecomment-822278347


   @anijain2305 Is group conv supported by QNN? The following line that does multiplication seems incorrect for group conv case. Here, the weight shape is (24, 3, 3, 3), and multiplying 24 * 3 results in the error message above because weight scale shape is (24,).
   https://github.com/apache/tvm/blob/813136401a11a49d6c15e6013c34dd822a5c4ff6/src/relay/qnn/op/convolution.cc#L81
   
   @eleflea For now you can do per tensor quantization to workaround this problem (the error happens if you use per channel weight quantization by `get_default_qconfig('fbgemm')`). You can force per tensor Q by `get_default_qconfig('qnnpack')` for example.


-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

For queries about this service, please contact Infrastructure at:
users@infra.apache.org