[ONNX] Tried to trace <submodule> but it is not part of the active trace

[zetyquickly (Emil Bogomolov)]

Have a look at this thread maybe you’ll find something relevant

[ONNX] Quantized fused Conv2d won't trace

Please re-try with pytorch nightly build, we recently fixed this so you shouldn’t be seeing this error anymore. Seems like the conv layer is not quantized so it produces onnx::Conv as opposed to the _caffe2::Int8Conv operator. Currently the onnx export path to caffe2 does not support partially quantized model, so it expects the entire pytorch model to be able to get quantized.