[GitHub] zheng-da commented on issue #10921: Test cases improvement for MKLDNN on Gluon
zheng-da commented on issue #10921: Test cases improvement for MKLDNN on Gluon URL: https://github.com/apache/incubator-mxnet/pull/10921#issuecomment-392415271 +1 for C++ unit test This is an automated message from the Apache Git Service. To respond to the message, please log on 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
[GitHub] zheng-da commented on issue #10921: Test cases improvement for MKLDNN on Gluon
zheng-da commented on issue #10921: Test cases improvement for MKLDNN on Gluon URL: https://github.com/apache/incubator-mxnet/pull/10921#issuecomment-392403691 @marcoabreu I think we should remove the tests for generating huge weight matrices, as @juliusshufan is doing right now for Dense. It's clear that the code will run into out-of-memory errors, regardless of the backend (native MXNet, MKLDNN or CuDNN). This is an automated message from the Apache Git Service. To respond to the message, please log on 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
[GitHub] zheng-da commented on issue #10921: Test cases improvement for MKLDNN on Gluon
zheng-da commented on issue #10921: Test cases improvement for MKLDNN on Gluon URL: https://github.com/apache/incubator-mxnet/pull/10921#issuecomment-392218718 @marcoabreu i'm not sure what you mean. do you mean we should support NDArrays of 85GB? i don't think we can use malloc to allocate a single piece of memory of this size? This is an automated message from the Apache Git Service. To respond to the message, please log on 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
[GitHub] zheng-da commented on issue #10921: Test cases improvement for MKLDNN on Gluon
zheng-da commented on issue #10921: Test cases improvement for MKLDNN on Gluon URL: https://github.com/apache/incubator-mxnet/pull/10921#issuecomment-392087536 your input size is too large for Dense. Very few systems can actually allocate a single piece of memory of 85GB. There is no reason to test the error in https://github.com/apache/incubator-mxnet/issues/10807. No system can handle it. This is an automated message from the Apache Git Service. To respond to the message, please log on 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
[GitHub] zheng-da commented on issue #10921: Test cases improvement for MKLDNN on Gluon
zheng-da commented on issue #10921: Test cases improvement for MKLDNN on Gluon URL: https://github.com/apache/incubator-mxnet/pull/10921#issuecomment-391919796 @juliusshufan https://github.com/apache/incubator-mxnet/issues/10807 does this case work on CPU without mkldnn? This is an automated message from the Apache Git Service. To respond to the message, please log on 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
[GitHub] zheng-da commented on issue #10921: Test cases improvement for MKLDNN on Gluon
zheng-da commented on issue #10921: Test cases improvement for MKLDNN on Gluon URL: https://github.com/apache/incubator-mxnet/pull/10921#issuecomment-391802884 why does it fail in so many tests? This is an automated message from the Apache Git Service. To respond to the message, please log on 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
[GitHub] zheng-da commented on issue #10921: Test cases improvement for MKLDNN on Gluon
zheng-da commented on issue #10921: Test cases improvement for MKLDNN on Gluon URL: https://github.com/apache/incubator-mxnet/pull/10921#issuecomment-391400600 could you rebase your code? it seems something is messed up This is an automated message from the Apache Git Service. To respond to the message, please log on 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
[GitHub] zheng-da commented on issue #10921: Test cases improvement for MKLDNN on Gluon
zheng-da commented on issue #10921: Test cases improvement for MKLDNN on Gluon URL: https://github.com/apache/incubator-mxnet/pull/10921#issuecomment-388640918 @marcoabreu i agree with you that we need cpp tests for MKLDNN to explicitly cover different input and output NDArrays. I'm writing such unit tests. These tests will cover more different cases and many of them are not covered by the python unit tests. However, writing such cpp unit tests is much more difficult and probably can only be used for testing a single operator. We'll need more people to write C++ unit tests. The tests on in this PR. https://github.com/apache/incubator-mxnet/pull/10651 I'll probably split the PR and submit another PR for adding cpp unit tests. That's why the python tests can be useful. It can easily cover many different operator combinations, which may cause problems in MKLDNN and was never tested in our unit tests. This is an automated message from the Apache Git Service. To respond to the message, please log on 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
[GitHub] zheng-da commented on issue #10921: Test cases improvement for MKLDNN on Gluon
zheng-da commented on issue #10921: Test cases improvement for MKLDNN on Gluon URL: https://github.com/apache/incubator-mxnet/pull/10921#issuecomment-388640918 @marcoabreu i agree with you that we need cpp tests for MKLDNN to explicitly cover different input and output NDArrays. I'm writing such unit tests. These tests will cover more different cases and many of them are not covered by the python unit tests. However, writing such cpp unit tests is much more difficult and probably can only be used for testing a single operator. We'll need more people to write unit tests. The tests on in this PR. https://github.com/apache/incubator-mxnet/pull/10651 I'll probably split the PR and submit another PR for adding cpp unit tests. That's why the python tests can be useful. It can easily cover many different operator combinations, which may cause problems in MKLDNN and was never tested in our unit tests. This is an automated message from the Apache Git Service. To respond to the message, please log on 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