[GitHub] [incubator-mxnet] apeforest commented on issue #17331: [mxnet 2.0] [item 2.4] Turning on large tensor support by default

apeforest commented on issue #17331: URL: https://github.com/apache/incubator-mxnet/issues/17331#issuecomment-624456645 @access2rohit This result is a little surprising. In the earlier benchmark results provided by @JonTanS, there is a ~17% degradation in BERT training when large tensor (

[GitHub] [incubator-mxnet] apeforest commented on issue #17331: [mxnet 2.0] [item 2.4] Turning on large tensor support by default

apeforest commented on issue #17331: [mxnet 2.0] [item 2.4] Turning on large tensor support by default URL: https://github.com/apache/incubator-mxnet/issues/17331#issuecomment-582749284 > @eric-haibin-lin Yes I am calculating this by: 1 - (LT, MKL value / MKL value). > For the samples/

[GitHub] [incubator-mxnet] apeforest commented on issue #17331: [mxnet 2.0] [item 2.4] Turning on large tensor support by default

apeforest commented on issue #17331: [mxnet 2.0] [item 2.4] Turning on large tensor support by default URL: https://github.com/apache/incubator-mxnet/issues/17331#issuecomment-589829033 Thanks to @JonTanS for running the profiler, we have ping pointed the performance degradation in operat