ptrendx commented on issue #17278: [Performance] CachedOp/Gluon performance thread URL: https://github.com/apache/incubator-mxnet/issues/17278#issuecomment-574321205 There are a few performance issues that we saw in `CachedOp` compared to graph executor. - https://github.com/apache/incubator-mxnet/blob/master/python/mxnet/gluon/block.py#L1013-L1014 is constructing a Python list (inclusing memory allocation) every time, which added ~100us for BERT inference - Bulking of ops is broken for ops using inputs to the graph / producing outputs of the graph (I believe it is referred in this comment: https://github.com/apache/incubator-mxnet/blob/master/src/imperative/imperative_utils.h#L1030)
---------------------------------------------------------------- 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: us...@infra.apache.org With regards, Apache Git Services
