AndrewZhaoLuo commented on issue #8918: URL: https://github.com/apache/tvm/issues/8918#issuecomment-912841581
This is quite a weird bug! It seems to result from multiplying elementwise two tensors which do not contain NaN values! https://github.com/apache/tvm/blob/main/python/tvm/relay/frontend/onnx.py#L3541 This operation seems to be the cause. Interestingly enough `loss` and `mask_tensor` dont seem to contain NaNs but after multiplying them the result does! Specifically it occurs where the mask_tensor = 0 only (the intention is to set the indices where the mask_tensor = 0 to 0 in the loss tensor) -- 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. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
