masahi opened a new pull request #8810: URL: https://github.com/apache/tvm/pull/8810
Testing fp16 conversion on the EfficientDet D2 model from [TF2 detection zoo](https://github.com/tensorflow/models/blob/master/research/object_detection/g3doc/tf2_detection_zoo.md) reveals that ops that involve large summation, like `global_avgpool` or `adaptive_avgpool` end up producing `nan` when done in fp16. To be on the safe side, we should disable these ops for fp16 conversion. I also disabled `sum` and `mean` for the same reason, but `avg_pool2d` etc are not modified for now. -- 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]
