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

Long overdue refactoring for quality of life and performance improvements. #140

Merged
merged 213 commits into from
Aug 24, 2023
Merged

Long overdue refactoring for quality of life and performance improvements. #140

merged 213 commits into from
Aug 24, 2023

Conversation

n8mellis
Copy link
Contributor

This PR represents a near-total refactoring of the existing Chassis code to fix bugs and add some long-awaited features. Even though this is a significant refactor, there should be no* breaking changes for users of the Chassis SDK.

Notable improvements:

  1. Chassis now supports local Docker builds. Using a remote build server is still supported but is no longer required.
  2. The dependency on MLFlow has been removed. Any function that can be cloudpickle'd is now supported (MLFlow also uses cloudpickle under the hood.)
  3. Models now support multiple inputs and outputs per inference.
  4. The number of PIP dependencies to install the Chassis SDK has been dramatically reduced. For some features (like kserve support), there are optional dependencies that can be used to bring in those extra dependencies.
  5. The use of conda to satisfy dependencies is no longer required. This, combined with the removal of MLFlow results is an up to 10x size reduction in the generated model container.
  6. Due to all of the above, container build times are now 2-5x faster.

In addition, the remote build server has been completely rewritten (in Rust). It no longer has any dependencies on object storage and has much better support for parallel builds. In addition, kaniko has been replaced by BuildKit which enables multi-platform images to be built. Finally, the Helm chart has been updated to allow changing all Kubernetes values like the amount of resources to allocate to the BuildKit pods, various timeouts, etc.

@n8mellis
Checkpoint
fed23fe
@n8mellis
Checkpoint
462eaf4
@n8mellis
Started new build service in Rust.
19b73a4
@n8mellis
Checkpoint
3221e72
@n8mellis
Moved examples folder to root directory.
a647905
@n8mellis
Checkpoint
d999770
@n8mellis
Checkpoint
dc5d70d
@n8mellis
Refactored package
5390537
@n8mellis
Changed protobuf v3.x requirement.
45eb3be 
We only really care that it's less than v4.
@n8mellis
Cleanup, refactoring, and started adding tests.
2b38d50
@n8mellis
Checkpoint
fc46b92
@n8mellis
Update type definitions.
bb96a2f 
 Use more generic `Mapping` instead of `Dict` or `dict`.
@n8mellis
Fixed error in pyproject.toml.
242334e
@n8mellis
Refactoring and added local Docker build.
84208ed
@n8mellis
Added remote builder on client side.
a6cec52
@n8mellis
Implemented deprecated publish method.
845987c
@n8mellis
Set remote build server version.
160b581
@n8mellis
Don't suffix the model version on the image name.
a2c3a7a
@n8mellis
Starting work on new remote build server.
1fc6510
@n8mellis
Checkpoint
462dd1d
@n8mellis
Added init method to ModelMetadata.
f0158ce
@n8mellis
Fixed recursion and cyclic import issues.
4d0f8ad
@n8mellis
Fixed pyproject.toml.
0ce2424
@n8mellis
Checkpoint
0ec547d
@n8mellis
Checkpoint
8504681
@n8mellis
Checkpoint
c0e4826
@n8mellis
Add prepare_context args to DockerBuilder.
7a34daf 
This means that users don't need to call prepare_context explicitly, it will happen automatically when initializing a DockerBuilder.
@n8mellis
Improve response from build_image.
5f00e3d
@n8mellis
Improve build_image logging.
e2ce3da
@n8mellis
Fix legacy predict function signatures.
c1bb6bc
bmunday3 and others added 26 commits August 17, 2023 09:35
@bmunday3
Small stylistic changes and torch installation tip
8cac25b
@n8mellis
Forgot to update one reference to the old "typing" package.
2afbc67
@caradoxical
Updates to the reference section
08dc940
@caradoxical
Merge branch 'nathan/overhaul' of https://github.com/modzy/chassis in…
365f084 
…to nathan/overhaul
@n8mellis
Optimize torch imports
99ab959 
Automatically use CPU-only version of `torch*` packages if no CUDA version is specified.
@n8mellis
Removed unused GitHub Action.
4ed1d54
@bmunday3
Updated examples to be cleaned up
d60c0e0
@bmunday3
Merge branch 'nathan/overhaul' of https://github.com/modzy/chassis in…
53d9d3a 
…to nathan/overhaul
@bmunday3
Updated chassis url for legacy examples
bb8417e
@bmunday3
deleted legacy examples folder
e466f43
@bmunday3
Reorganized examples
b85b47a
@n8mellis
Test new build automation
5446e00
@n8mellis
Try using single quotes
78a67d6
@n8mellis
Added missing -e from tox commands.
c29ae75
@n8mellis
Use explicit strings for python version.
b1844d9
@n8mellis
Comment out types test for now.
4ab00bf
@n8mellis
Remove requirements.txt from root.
360b5b8
@n8mellis
Fix python version as a string.
e8bef8d
@n8mellis
Enable latest tag for build server image.
04982d1
@n8mellis
Merge remote-tracking branch 'origin/main' into nathan/overhaul
deae87d 
# Conflicts:
#	chassisml_sdk/examples/fastai/fastai_tabular_training.ipynb
#	chassisml_sdk/examples/hugging-face/huggingface_distilbert_text_classification.ipynb
#	chassisml_sdk/examples/lightgbm/lightgbm_classification.ipynb
#	chassisml_sdk/examples/mxnet/mxnet_mobilenet_image_classification.ipynb
#	chassisml_sdk/examples/onnx/onnx_mobilenet_image_classification.ipynb
#	chassisml_sdk/examples/onnx/onnx_transformer_text_generation.ipynb
#	chassisml_sdk/examples/pmml/pmml_iris_forest_classification.ipynb
#	chassisml_sdk/examples/pmml/pmml_linear_regression.ipynb
#	chassisml_sdk/examples/pytorch/pytorch_deeplab_resnet50_semantic_segmentation.ipynb
#	chassisml_sdk/examples/pytorch/pytorch_fastrcnn_object_detection.ipynb
#	chassisml_sdk/examples/pytorch/pytorch_resnet50_image_classification.ipynb
#	chassisml_sdk/examples/pytorch/pytorch_resnet50_image_classification_batch_gpu.ipynb
#	chassisml_sdk/examples/pytorch/pytorch_tabular_shelter_animal_outcome.ipynb
#	chassisml_sdk/examples/sklearn/data/random_forest.joblib
#	chassisml_sdk/examples/sklearn/sklearn_logreg_image_classification.ipynb
#	chassisml_sdk/examples/sklearn/sklearn_svm_image_classification.ipynb
#	chassisml_sdk/examples/sklearn/sklearn_tree_wine_classification.ipynb
@n8mellis
Add tox as a dependency to build the docs.
67cfa75
@n8mellis
Attempt to use a Docker image cache to speed up build.
72955bb
@n8mellis
Remove nathan/overhaul as a default build branch.
169af04
@n8mellis
Update PR build automation.
330b9ff
@n8mellis
Use Docker buildx.
d402117
@n8mellis
Update build step name.
3a73f30 
[skip ci]
@n8mellis n8mellis requested a review from bmunday3 August 24, 2023 18:45
@n8mellis n8mellis merged commit 2567337 into main Aug 24, 2023
@n8mellis n8mellis deleted the nathan/overhaul branch August 24, 2023 18:46
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Reviewers

@bmunday3 bmunday3 bmunday3 approved these changes

Assignees
No one assigned
Labels
None yet
Projects
None yet
Milestone
No milestone
Development

Successfully merging this pull request may close these issues.
3 participants
@n8mellis @bmunday3 @caradoxical