domin1985 opened a new pull request #6271: URL: https://github.com/apache/incubator-tvm/pull/6271
Hi, A tvm error occurred when I imported a pre-trained keras model:  TVMError: Error(s) have occurred. The program has been annotated with them: In `main`: %0 = negative(%v_param_3); %1 = nn.conv2d(%input_1, %v_param_1, padding=[4, 4], channels=64, kernel_size=[9, 9], data_layout="NHWC", kernel_layout="HWIO"); %2 = nn.bias_add(%1, %v_param_2, axis=-1); %3 = negative(%2); %4 = nn.relu(%3); %5 = multiply(%0, %4) Incompatible broadcast type TensorType([64, 1, 1], float32) and TensorType([1, (int64)96, (int64)96, 64], float32); ; %6 = nn.relu(%2); There is something wrong with the prelu convert function in Keras frontend which would result in this shape inferred mis-match error. We would better use relay.nn.prelu op directly. Tests passed using tests/python/frontend/keras/test_forward.py. @siju-samuel @yongwww ---------------------------------------------------------------- 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]
