sxjscience commented on issue #17654: [LayerNorm] Missing the mismatch cues of in_channels URL: https://github.com/apache/incubator-mxnet/issues/17654#issuecomment-589741619 @ZheyuYe The C++ side implementation of the shape inferring logic is here: https://github.com/apache/incubator-mxnet/blob/9dcf71d8fe33f77ed316a95fcffaf1f7f883ff70/src/operator/nn/layer_norm.cc#L39-L66 The python side is here: https://github.com/apache/incubator-mxnet/blob/9dcf71d8fe33f77ed316a95fcffaf1f7f883ff70/python/mxnet/gluon/nn/basic_layers.py#L609-L614 The problem should be to check the shape of gamma and beta: https://github.com/apache/incubator-mxnet/blob/9dcf71d8fe33f77ed316a95fcffaf1f7f883ff70/src/operator/nn/layer_norm.cc#L56-L57 Would you try to investigate the issue? You can append `std::cout << in_shape->at(layernorm::kGamma)`, which should not be empty when `in_channel` is given. I think one way to solve the prioblem is to use the same `SHAPE_ASSIGN_CHECK` as here: https://github.com/apache/incubator-mxnet/blob/9dcf71d8fe33f77ed316a95fcffaf1f7f883ff70/src/operator/numpy/np_where_op.cc#L42
---------------------------------------------------------------- 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: us...@infra.apache.org With regards, Apache Git Services
