Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

Support for PyTorch model #346

Merged
merged 35 commits into from
Jan 10, 2018
Merged

Support for PyTorch model #346

merged 35 commits into from
Jan 10, 2018

Conversation

Corey-Zumar
Copy link
Contributor

@Corey-Zumar Corey-Zumar commented Dec 16, 2017 •
edited
Loading

@haofanwang Added some updates to fix the class serialization issue and accidentally closed #322 in the process. Take a look and let me know if you have any questions.

This should be good to go once @dcrankshaw takes a final pass.

Fixes #343, #314

haofanwang and others added 23 commits November 20, 2017 15:44
@haofanwang
Add files via upload
2256349
@haofanwang
Add files via upload
5a7bda9
@haofanwang
Add files via upload
261f67a
@haofanwang
Update build_docker_images.sh
512f126
@haofanwang
Update pytorch.py
3afa439
@haofanwang
Update pytorch_container.py
6f74280
@haofanwang
Add files via upload
cd259a3
@haofanwang
Add files via upload
1591ab7
@haofanwang
Create deploy_pytorch_models.py
5501c95
@haofanwang
Update deploy_pytorch_models.py
f7e2fbe
@haofanwang
Update deploy_pytorch_models.py
cf849f9
@haofanwang
Update deploy_pytorch_models.py
fa5aecd
@haofanwang
Update deploy_pytorch_models.py
acd9675
@haofanwang
Update deploy_pytorch_models.py
932df78
@haofanwang
Update pytorch.py
a00d527
@haofanwang
Update deploy_pytorch_models.py
b38c388
@haofanwang
Update deploy_pytorch_models.py
1101f16
@haofanwang
Update deploy_pytorch_models.py
0abd22b
@haofanwang
Update pytorch_container.py
b5411c2
@haofanwang
Update deploy_pytorch_models.py
6022a6b
@haofanwang
Update deploy_pytorch_models.py
b9a309e
@Corey-Zumar
Merge branch 'develop' into HEAD
3073a06
@Corey-Zumar
Capture class dependencies
f9bc385
@AmplabJenkins
Copy link

Test FAILed.
Refer to this link for build results (access rights to CI server needed):
https://amplab.cs.berkeley.edu/jenkins//job/Clipper-PRB/815/
Test FAILed.

Corey-Zumar
Corey-Zumar commented Dec 16, 2017
View reviewed changes
clipper_admin/clipper_admin/deployers/cloudpickle.py
@@ -1,15 +1,20 @@
"""
Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Pulled the latest cloudpickle. This fixes issues with recursive dependencies that caused issues when serializing pytorch models.

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

If we're replacing Cloudpickle, let's just switch to using the cloudpickle pip package (pip install cloudpickle).

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

See #343

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Done

Corey-Zumar
Corey-Zumar commented Dec 16, 2017
View reviewed changes
clipper_admin/clipper_admin/deployers/pytorch.py
PYTORCH_MODEL_RELATIVE_PATH)

try:
torch.save(pytorch_model.state_dict(), torch_weights_save_loc)
Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

We now save the model weights and class definition separately. Weights are saved via PyTorch, and the class is pickled with CloudPickle

@AmplabJenkins
Copy link

Test PASSed.
Refer to this link for build results (access rights to CI server needed):
https://amplab.cs.berkeley.edu/jenkins//job/Clipper-PRB/816/
Test PASSed.

dcrankshaw
dcrankshaw requested changes Jan 3, 2018
View reviewed changes
Copy link
Contributor

@dcrankshaw dcrankshaw left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Two small changes then this is good to go. @Corey-Zumar can you go ahead and make those?

clipper_admin/clipper_admin/deployers/cloudpickle.py
@@ -1,15 +1,20 @@
"""
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

If we're replacing Cloudpickle, let's just switch to using the cloudpickle pip package (pip install cloudpickle).

dockerfiles/PyTorchContainerDockerfile Outdated
&& apt-get update --fix-missing \
&& apt-get install -yqq -t jessie-backports openjdk-8-jdk \
&& conda install -y --file /lib/python_container_conda_deps.txt \
&& conda install pytorch torchvision -c soumith
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

They created a PyTorch conda channel recently, so this should be conda install pytorch torchvision -c pytorch

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Done

@AmplabJenkins
Copy link

Test FAILed.
Refer to this link for build results (access rights to CI server needed):
https://amplab.cs.berkeley.edu/jenkins//job/Clipper-PRB/833/
Test FAILed.

@AmplabJenkins
Copy link

Test FAILed.
Refer to this link for build results (access rights to CI server needed):
https://amplab.cs.berkeley.edu/jenkins//job/Clipper-PRB/834/
Test FAILed.

@Corey-Zumar Corey-Zumar mentioned this pull request Jan 9, 2018
Closed
@AmplabJenkins
Copy link

Test FAILed.
Refer to this link for build results (access rights to CI server needed):
https://amplab.cs.berkeley.edu/jenkins//job/Clipper-PRB/835/
Test FAILed.

@AmplabJenkins
Copy link

Test FAILed.
Refer to this link for build results (access rights to CI server needed):
https://amplab.cs.berkeley.edu/jenkins//job/Clipper-PRB/838/
Test FAILed.

@AmplabJenkins
Copy link

Test FAILed.
Refer to this link for build results (access rights to CI server needed):
https://amplab.cs.berkeley.edu/jenkins//job/Clipper-PRB/839/
Test FAILed.

@AmplabJenkins
Copy link

Test FAILed.
Refer to this link for build results (access rights to CI server needed):
https://amplab.cs.berkeley.edu/jenkins//job/Clipper-PRB/840/
Test FAILed.

@Corey-Zumar
Copy link
Contributor Author

jenkins test this please

@AmplabJenkins
Copy link

Test FAILed.
Refer to this link for build results (access rights to CI server needed):
https://amplab.cs.berkeley.edu/jenkins//job/Clipper-PRB/844/
Test FAILed.

@AmplabJenkins
Copy link

Test PASSed.
Refer to this link for build results (access rights to CI server needed):
https://amplab.cs.berkeley.edu/jenkins//job/Clipper-PRB/845/
Test PASSed.

dcrankshaw
dcrankshaw approved these changes Jan 10, 2018
View reviewed changes
Copy link
Contributor

@dcrankshaw dcrankshaw left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

lgtm

@dcrankshaw dcrankshaw merged commit 3b44a91 into ucbrise:develop Jan 10, 2018
@Corey-Zumar Corey-Zumar mentioned this pull request Jan 18, 2018
Closed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Reviewers

@dcrankshaw dcrankshaw dcrankshaw approved these changes

@haofanwang haofanwang Awaiting requested review from haofanwang

Assignees
No one assigned
Labels
status: needs review type: enhancement
Projects
None yet
Milestone
No milestone
Development

Successfully merging this pull request may close these issues.

Replace custom fork of Cloudpickle with the maintained pip package
4 participants
@Corey-Zumar @AmplabJenkins @dcrankshaw @haofanwang