search

dmlc / gluon-nlp

NLP made easy
https://nlp.gluon.ai/
Apache License 2.0
2.56k stars 538 forks source link

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

Open Zha0q1 opened 3 years ago

Zha0q1 commented 3 years ago

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 commented 3 years ago

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

for the gpu ci I think libcudnn 8 is missing

barry-jin commented 3 years ago

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 commented 3 years ago

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 commented 3 years ago

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 commented 3 years ago

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[bot] commented 3 years ago

The documentation website for preview: http://gluon-nlp-staging.s3-accelerate.dualstack.amazonaws.com/PR1556/d4caa0b65b3733fc5fe6fc516bf47894b6482b8c/index.html

github-actions[bot] commented 3 years ago

The documentation website for preview: http://gluon-nlp-staging.s3-accelerate.dualstack.amazonaws.com/PR1556/d4caa0b65b3733fc5fe6fc516bf47894b6482b8c/index.html

barry-jin commented 3 years ago

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

Zha0q1 commented 3 years ago

@barry-jin thanks!