barry-jin edited a comment on issue #20687: URL: https://github.com/apache/incubator-mxnet/issues/20687#issuecomment-948924317
Hi @matteosal, the implementations of executor are totally different between v1.6 and v2.0. In v1.6, we are using executor C-APIs exposed from backend ([example](https://github.com/apache/incubator-mxnet/blob/c76593162c2a5e421243dd09a8c3cde044efaa3f/python/mxnet/executor.py#L150)), but in v2.0, we are using CachedOp APIs ([example](https://github.com/apache/incubator-mxnet/blob/5f0efbbe33a1ef2af140e91a8fd367cd3bf92373/python/mxnet/executor.py#L183)). You could try to add some parameters when initializing [CachedOp](https://github.com/apache/incubator-mxnet/blob/5f0efbbe33a1ef2af140e91a8fd367cd3bf92373/python/mxnet/executor.py#L124) to do hybridization (reference: [ref1](https://github.com/apache/incubator-mxnet/blob/5f0efbbe33a1ef2af140e91a8fd367cd3bf92373/python/mxnet/gluon/block.py#L1391-L1406), [ref2](https://github.com/apache/incubator-mxnet/pull/20462/files#diff-0b3b9d55159f8526d638cd6f3d571c1eeed96f3bcdc37492425a0749d5a3f208L124)) to see the memory allocation optimizations. For the GPU memory allocation increase, it is probably related to loading the kernels to GPU. -- 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: issues-unsubscr...@mxnet.apache.org For queries about this service, please contact Infrastructure at: us...@infra.apache.org --------------------------------------------------------------------- To unsubscribe, e-mail: issues-unsubscr...@mxnet.apache.org For additional commands, e-mail: issues-h...@mxnet.apache.org