Skip to content
This repository has been archived by the owner on Jan 15, 2024. It is now read-only.

[v0.x] Add onnx based transformer example #1556

Open
wants to merge 5 commits into
base: v0.x
Choose a base branch
from

Conversation

Zha0q1
Copy link

@Zha0q1 Zha0q1 commented Apr 16, 2021

Adds a new notebook that runs transformer based on onnxruntime.
The onnx export feature will be added in MXNet 1.9

@szha @sxjscience @sandeep-krishnamurthy

@Zha0q1 Zha0q1 requested a review from a team as a code owner April 16, 2021 20:36
@Zha0q1
Copy link
Author

Zha0q1 commented Apr 16, 2021

Hi @barry-jin would you help identify the issue with docs ci?

for the gpu ci I think libcudnn 8 is missing

@barry-jin
Copy link
Contributor

Hi @barry-jin would you help identify the issue with docs ci?

for the gpu ci I think libcudnn 8 is missing

Hi @Zha0q1, gluon-nlp v0.x CI is running on master branch's docker, which use nvidia/cuda:10.2-cudnn7-devel-ubuntu18.04 as base. Looks like mxnet-cu102==1.8.0.post0 is forced to use libcudnn.so.8. I have created a workaround try to fix this in #1553

@Zha0q1
Copy link
Author

Zha0q1 commented Apr 16, 2021

Hi @barry-jin would you help identify the issue with docs ci?
for the gpu ci I think libcudnn 8 is missing

Hi @Zha0q1, gluon-nlp v0.x CI is running on master branch's docker, which use nvidia/cuda:10.2-cudnn7-devel-ubuntu18.04 as base. Looks like mxnet-cu102==1.8.0.post0 is forced to use libcudnn.so.8. I have created a workaround try to fix this in #1553

Thanks! I think the other two pipelines also have some issues

@barry-jin
Copy link
Contributor

Hi @barry-jin would you help identify the issue with docs ci?
for the gpu ci I think libcudnn 8 is missing

Hi @Zha0q1, gluon-nlp v0.x CI is running on master branch's docker, which use nvidia/cuda:10.2-cudnn7-devel-ubuntu18.04 as base. Looks like mxnet-cu102==1.8.0.post0 is forced to use libcudnn.so.8. I have created a workaround try to fix this in #1553

Thanks! I think the other two pipelines also have some issues

Yes, for website-build pipeline, it will submit job to AWS batch to run the script compile_notebooks.sh, make sure updating your ci folder to have this script in ci/batch. For unittest, it looks like there may be some memory related issues, more investigation is needed.

@Zha0q1
Copy link
Author

Zha0q1 commented Apr 16, 2021

Hi @barry-jin would you help identify the issue with docs ci?
for the gpu ci I think libcudnn 8 is missing

Hi @Zha0q1, gluon-nlp v0.x CI is running on master branch's docker, which use nvidia/cuda:10.2-cudnn7-devel-ubuntu18.04 as base. Looks like mxnet-cu102==1.8.0.post0 is forced to use libcudnn.so.8. I have created a workaround try to fix this in #1553

Thanks! I think the other two pipelines also have some issues

Yes, for website-build pipeline, it will submit job to AWS batch to run the script compile_notebooks.sh, make sure updating your ci folder to have this script in ci/batch. For unittest, it looks like there may be some memory related issues, more investigation is needed.

Thanks! I merged with v0.x

@github-actions
Copy link

@Zha0q1 Zha0q1 closed this Apr 19, 2021
@Zha0q1 Zha0q1 reopened this Apr 19, 2021
@github-actions
Copy link

@barry-jin
Copy link
Contributor

Hi @Zha0q1 . v0.x CI has been fixed. Feel free to merge with current v0.x and re-run the tests. Thanks!

@Zha0q1
Copy link
Author

Zha0q1 commented Apr 21, 2021

@barry-jin thanks!

Sign up for free to subscribe to this conversation on GitHub. Already have an account? Sign in.
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

2 participants