Wenbing Li

@julianjungwirth , would you please propose your fixing as a PR to help other as well?

is this LSTM including a custom cell? Can you share us some details about the model?

@xadupre , can you help with that?

did it start to occur on tf 2.4.x or 2.5.x? I will try to find some time to setup the env. and fix the issue.

Are you suggestion ONNX should also something similar to keras's use_bias flag so that they could be converted into each other in your case?

If you are familiar with Protobuf, take a look at https://github.com/onnx/onnx/blob/e5e9a539f550f07ec156812484e8d4f33fb91f88/onnx/onnx.proto3#L335

_" Drilling down into the error leads to a specific line in the Attention Class: weighted_input = x * K.expand_dims(a) "_ I don't see why it is a converter issue...

@xadupre @prabhat00155

model.load_weights is to help you load the pre-trained weights file.

should be something like model = ResNet50(include_top=True, weights="") model.load_weights("your-pretrained-model-files")