icemelon9 commented on issue #4972: Performance regression of quantization on CUDA after [Relay][AutoTVM] Relay op strategy (#4644) URL: https://github.com/apache/incubator-tvm/issues/4972#issuecomment-593545631 I guess the performance regression could be due to https://github.com/apache/incubator-tvm/blob/master/python/tvm/relay/op/strategy/cuda.py#L88. Previously TVM might use `conv2d_NCHWc_int8` by default. However `conv2d_NCHWc_int8` is not enabled now when layout is "NCHW" because its output shape is 5-dim and could be lead to wrong shape if using opt_level=2 as `alter_op_layout` pass wouldn't be enabled. We need to fix it by following x86 `conv2d_NCHWc`. I don't have any idea what causes the accuracy problem. Could you share more about what happens in the `quantize` function in your script?
---------------------------------------------------------------- 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: [email protected] With regards, Apache Git Services
