During the process of converting a Paddle model with NHWC data format, TVM crashed. Through debugging, I confirmed that the crash occurred during the model compilation stage rather than during model loading. Upon visualizing the NHWC data format Paddle model using Netron, I observed that the weight shape of the convolution operator in the model is in the OIWH format instead of OWHI format. Could the issue be related to this discrepancy?
I would like to submit a PR to fix this issue. Can anyone guide me on how to troubleshoot this problem?