Added MXNet Deployment Capability#390
Conversation
|
Can one of the admins verify this patch? |
|
jenkins ok to test |
|
Test FAILed. |
dcrankshaw
left a comment
dcrankshaw
left a comment
There was a problem hiding this comment.
This looks great. Just a couple minor comments.
Also, it looks like the tests failed because of the format checker. Run ./bin/format_code.sh to fix.
|
|
||
| except Exception as e: | ||
| print(e) | ||
| logger.warn("Error saving MXNet model: %s" % e) |
There was a problem hiding this comment.
This should be a logger.error().
| # Deploy model | ||
| clipper_conn.build_and_deploy_model(name, version, input_type, | ||
| serialization_dir, base_image, labels, | ||
| registry, num_replicas) |
There was a problem hiding this comment.
Move lines 185-190 into the try block
dcrankshaw
added
type: enhancement
status: needs revision
and removed
status: needs review
labels
|
Test FAILed. |
|
Test FAILed. |
|
Test FAILed. |
|
@chester-leung for some reason |
|
Test FAILed. |
|
Test FAILed. |
|
Hey sorry didn't mean to close this. Opened #404 to replace. |
* created virtual env and added supporting files for mxnet * added venv to .gitignore * Remove venv * implemented serialization: save/load * finished writing imxnet integration test * testing integration test * bind and fit model before saving * Remove deps directory * load model changed * reformatted code * removed duplicate code in clipper_admin/deployers/mxnet.py that was causing two models to register * fixed model loading * reformatted code * deleted commented out old code * Hopefully fixed formatting error * bind/set params after loading serialized model instead of fitting. model deployment now requires user to specify input data shape * removed cp train.data line in MXNetDockerfile * uncommented build images in build_docker_images.sh * updated variable name and added documentation for mxnet_input_shape in clipper_admin/clipper_admin/deployers/mxnet.py * changed mxnet_input_shape parameter in deploy_mxnet_model() to mxnet_data_shapes. added documentation for mxnet_data_shapes parameter * deleted unnecessary empty line from dockerfiles/ClipperTestDockerfile * reformatted code * changed description of mxnet_data_shapes in clipper_admin/clipper_admin/deployers/mxnet.py * documentation changes in clipper_admin/clipper_admin/deployers/mxnet.py
3 participants
Added capability to save/load and deploy MXNet models.
Added mxnet.py in clipper_admin/clipper_admin/deployers/ , mxnet_container.py in containers/python/ , and an integration test for MXNet. Also added necessary files to build a docker image from MXNet model.
Fixes: #364