Re: [VOTE] Release Apache MXNet (incubating) version 1.2.0.RC0

Thank you Anirudh for working on the fix! Update on the current status of RC1 release: Need to merge the following PRs: #10677 ,#10661 ,#10676

Re: [VOTE] Release Apache MXNet (incubating) version 1.2.0.RC0

Yes ideally, the fix should be - 1. take the padding tuple from onnx models and pass the appropriately sized tuple to mxnet. And if the padding in the onnx model is asymmetric we should throw an exception, saying it is not yet supported. 2. And also explore supporting asymmetric padding in mxnet.

Re: [VOTE] Release Apache MXNet (incubating) version 1.2.0.RC0

Here is a temporary fix, so MKLDNN conv follows the behavior of MXNet conv (always uses the first two elements in the tuple as padding). https://github.com/apache/incubator-mxnet/pull/10666 I also created an issue regarding this problem so that we'll fix it properly in the future. https://github.c

Re: [VOTE] Release Apache MXNet (incubating) version 1.2.0.RC0

Hi, Currently mkldnn support adds the check for tuple length for padding and raises the exception. Even though this is the correct approach, this is not consistent with MXNet behavior without MKLDNN enabled. I agree with Da and Haibin that the right approach in the longer term would be to: 1. rai

Re: [VOTE] Release Apache MXNet (incubating) version 1.2.0.RC0

Hi Da, After looking at your detailed description in github issue https://github.com/apache/incubator-mxnet/issues/10663, I would argue that the fix should go to mxnet-onnx instead of mxnet-mkldnn. In onnx padding params are in the form of (left, right, top, bottom), which supports asymmetric pad

Re: [VOTE] Release Apache MXNet (incubating) version 1.2.0.RC0

I think I have found the root of the problem. The tutorial loads a model from onnx, which uses padding (left, right, top, bottom). But mxnet uses padding (height, width). Currently, when an ONNX model is loaded to MXNet, the padding is converted correctly. MXNet conv doesn't check the number of

Re: [VOTE] Release Apache MXNet (incubating) version 1.2.0.RC0

Hi all, Thank you Thomas for reporting this issue, and thanks Da for working on the fix! We(Da, Steffen, Haibin and me) had a discussion offline about this. Next steps are the following: 1. Da will be working on the fix for the tutorials(onnx/super_resolution) and marking MKL-DNN as "Experimental

Re: [VOTE] Release Apache MXNet (incubating) version 1.2.0.RC0

I can reproduce the bug now. I'm working on a fix for the bug. Currently, there are a few more bug fixes for MKLDNN. https://github.com/apache/incubator-mxnet/pull/10651 https://github.com/apache/incubator-mxnet/pull/10624 https://github.com/apache/incubator-mxnet/pull/10619/files https://github.c

Re: [VOTE] Release Apache MXNet (incubating) version 1.2.0.RC0

@ThomasDelteil Sorry for the late reply. Yes, it should indeed work when compiled with mkldnn and cuda. On Sat, Apr 21, 2018 at 5:15 PM, Thomas DELTEIL wrote: > @Anirudh, thanks for looking into it! However I do not understand what you > mean by 'set as CPU and not GPU'? MXNet compiled with mkld

Re: [VOTE] Release Apache MXNet (incubating) version 1.2.0.RC0

@Anirudh, thanks for looking into it! However I do not understand what you mean by 'set as CPU and not GPU'? MXNet compiled with mkldnn and cuda is supposed to be able to work with both context no? There are other tutorials that are running successfully on both CPU and GPU context. @Da to reproduc

Re: [VOTE] Release Apache MXNet (incubating) version 1.2.0.RC0

@ThomasDelteil could you show me how to reproduce the problem? I'll take it a look as well. Best, Da Sent from my iPhone On Apr 21, 2018, at 1:12 PM, Anirudh Acharya mailto:anirudhk...@gmail.com>> wrote: @ThomasDelteil that might be due to the fact that in the example, the context is being se

Re: [VOTE] Release Apache MXNet (incubating) version 1.2.0.RC0

@ThomasDelteil that might be due to the fact that in the example, the context is being set as CPU and not GPU. But I will still take a look as soon as possible. Regards Anirudh On Sat, Apr 21, 2018 at 11:10 AM, Thomas DELTEIL wrote: > *-0* > > compiled from source on GPU CUDA/CUDNN, tutorials

Re: [VOTE] Release Apache MXNet (incubating) version 1.2.0.RC0

*-0* compiled from source on GPU CUDA/CUDNN, tutorials run fine. However: Compiled from source and adding USE_MKLDNN=1, the onnx/super_resolution tutorial is crashing on this line: ``` from collections import namedtuple Batch = namedtuple('Batch', ['data']) # forward on the provided data batch

Re: [VOTE] Release Apache MXNet (incubating) version 1.2.0.RC0

+1 Compiled from source. Ran the model quantization example. Both quantized model generation and inference can run successfully. On Fri, Apr 20, 2018 at 5:14 PM, Indhu wrote: > +1 > > Compiled from source on P3 instance. Tested the SSD example and some Gluon > examples. > > On Wed, Apr 18, 2018

Re: [VOTE] Release Apache MXNet (incubating) version 1.2.0.RC0

+1 Compiled from source on P3 instance. Tested the SSD example and some Gluon examples. On Wed, Apr 18, 2018, 7:40 PM Anirudh wrote: > Hi everyone, > > This is a vote to release Apache MXNet (incubating) version 1.2.0. Voting > will start now (Wednesday, April 18th) and end at 7:40 PM PDT, Satu

Re: [VOTE] Release Apache MXNet (incubating) version 1.2.0.RC0

+1 Compiled from source and verified distributed training with float32 and float16 Regards, Rahul On Thu, Apr 19, 2018 at 4:30 PM, Anirudh Acharya wrote: > +1 > > Checked the following - > >- source compiles >- the tests for the onnx import API passes. > > > Regards > Anirudh > > > On

Re: [VOTE] Release Apache MXNet (incubating) version 1.2.0.RC0

+1 Checked the following - - source compiles - the tests for the onnx import API passes. Regards Anirudh On Thu, Apr 19, 2018 at 1:11 PM, Meghna Baijal wrote: > +1 (non-binding) > > > I Checked the following: > > 1. Signatures are ok > > 2. Source compiles > > 3. mnist test passes > >

Re: [VOTE] Release Apache MXNet (incubating) version 1.2.0.RC0

+1 (non-binding) I Checked the following: 1. Signatures are ok 2. Source compiles 3. mnist test passes Regards, Meghna On Thu, Apr 19, 2018 at 10:12 AM, Anirudh wrote: > Hi all, > > Given the weekend, I am extending the vote deadline to Sunday evening, > April 22nd 7:40 PM PDT, consideri

Re: [VOTE] Release Apache MXNet (incubating) version 1.2.0.RC0

Hi all, Given the weekend, I am extending the vote deadline to Sunday evening, April 22nd 7:40 PM PDT, considering Saturday and Sunday as half days(as done before). Anirudh On Wed, Apr 18, 2018 at 7:40 PM, Anirudh wrote: > Hi everyone, > > This is a vote to release Apache MXNet (incubating) ve

[VOTE] Release Apache MXNet (incubating) version 1.2.0.RC0

Hi everyone, This is a vote to release Apache MXNet (incubating) version 1.2.0. Voting will start now (Wednesday, April 18th) and end at 7:40 PM PDT, Saturday, April 21st. Link to the release notes: https://cwiki.apache.org/confluence/display/MXNET/Apache+MXNet+%28incubating%29+1.2.0+Release+Not