-
Notifications
You must be signed in to change notification settings - Fork 3.7k
Logistics for ONNX Release 1.13.0
Przemyslaw Wysocki edited this page Dec 6, 2022
·
7 revisions
Release Manager: Przemyslaw Wysocki
Target Release date: 12/12/2022
- 10/13 - Create v1.13.0 release wiki with release schedule. (done)
- 10/13 - Document key v1.13.0 changes for the release in this wiki. (done)
- 11/18 - Code freeze. All PRs must be validated and merged by this date. (done)
- 11/21 - Cut the release branch. (done)
- 11/30 - Create test packages for v1.13.0rc1 (done)
| Op Name | Description | Validation status | ONNX PR |
|---|---|---|---|
| Col2Im | Rearranges input tensor into blocks | #3948 | |
| BitwiseNot | Bitwise operations set | #4497 | |
| BitwiseAnd, BitwiseOr and BitwiseXor | Bitwise operations set | #4496 |
| Op Name | Description | Validation status | ONNX PR |
|---|---|---|---|
| Resize | New attributes: antialias, axes and keep_aspect_ratio_policy
|
#4126 | |
| Resize | Allow for both scales and sizes to be provided when one of them is an empty constant |
#4388 | |
| Pad | New attribute axes
|
#4190 | |
| OptionalHasElement | New input types handling | #4326 | |
| OptionalHasElement and OptionalGetElement | Accept tensor and sequence types | #4421 | |
| ScatterElement and ScatterND | Add max and min as supported reduction attributes |
#4411 | |
| Split | Add support for uneven tensor splitting and new num_outputs attribute |
#4481 | |
| LpPool | New attributes: ceil_mode and dilations
|
#4534 |
| Function Name | Description | Validation status | ONNX PR |
|---|---|---|---|
| CenterCropPad | New function, center crops or pads an input to given dimensions | #4190 | |
| mish | New activation function | #4350 | |
| GroupNormalization | New normalization function | #4621 |
Reference Python runtime dependent on only Python and numpy has been added. #4483
ONNX now supports Python 3.11. #4490
Support for M1/M2 ARM processors has been added. #4642
| Description | PR | Status | Notes |
|---|---|---|---|
| Fix concurrent remove of .proto3.* files during build | #3317 | Merged | |
| Avoid installing empty directories | #3590 | Merged | |
| Fix bf16 Support | #4193 | Merged | |
| Make domain check in shape inference consider "ai.onnx" as "" | #3590 | Merged | |
| Upgrade python syntax with pyupgrade | #4212 | Merged | |
| Use ONNX Runtime PyPI package instead of ort-nightly | #4219 | Merged | |
Update requirements.txt
|
#4223 | Merged | |
| Fix mapping TensorProto to NumPy for bfloat16 | #4234 | Merged | |
Update ir.h to support initializer better |
#4235 | Merged | |
| Add Truncation Mode to F32->BF16 Conversion Helper (and fix big-endian support) | #4238 | Merged | |
| Print utility extension | #4246 | Merged | |
| Fix incorrect expected output in windowing and stft tests, and window function function def | #4249 | Merged | |
Fix: Window functions generate incorrect shape for with symmetric attribute |
#4256 | Merged | |
Handle raw types correctly in helper.make_tensor
|
#4262 | Merged | |
| Fix layer normalization's reference outputs | #4263 | Merged | |
| Fix sub-graph generation for LN | #4268 | Merged | |
Improve mapping and add more tests for make_tensor
|
#4270 | Merged | |
Remove reference to numpy.typing
|
#4277 | Merged | |
| Fix crash: only traverse graph to add symbols if symbol table presents | #4282 | Merged | |
| Revert #3979: Checker should validate the node's inputs/outputs have names when Variadic | #4283 | Merged | |
Update tensor_util.cc to get uint64_t data easier |
#4285 | Merged | |
Refactor shape_inference_test.py
|
#4302 | Merged | |
extract_model run time optimization |
#4324 | Merged | |
Add is_constant_initializer in ir.h
|
#4336 | Merged | |
| Specify pairwise conversion behavior for the Cast op | #4351 | Merged | |
Use RepeatedPtrField::Get instead of RepeatedPtrField::operator[], to be compatible with protobuf 3.0 API |
#4354 | Merged | |
| Clarify minimum version of Protobuf required by ONNX | #4360 | Merged | |
| Add detailed error message for saving 2GB proto | #4366 | Merged | |
| Extend printer to handle non-empty domain | #4372 | Merged | |
Make C++ and Python check_model consistent |
#4386 | Merged | |
Fix bugs and add .clang-tidy
|
#4391 | Merged | |
| Clarify and add helper for making attribute references | #4393 | Merged | |
Use CMAKE_INSTALL_LIBDIR to avoid hardcoded path |
#4395 | Merged | |
| Fix errors reported by flake8 5.0.2 | #4397 | Merged | |
Do not allow to read tensor's external_data outside the model directory |
#4400 | Merged | |
Add py.typed to let mypy get stub for type information |
#4412 | Merged | |
| Modify the annotation: fix the label serial number mistake | #4426 | Merged | |
| Format all python code with black and isort - take 2 | #4427 | Merged | |
Forbid possible ir_version=0 in IR |
#4429 | Merged | |
| Set up VS Code to format python code at save | #4430 | Merged | |
| Deprecate ONNXIFI: ONNX Interface for Framework Integration | #4431 | Merged | |
| Expose a Python interface for inference functions | #4409 | Merged | |
| Ignore the format PR in git blame | #4433 | Merged | |
Remove redundant static_cast in ir_pb_converter.cc |
#4437 | Merged | |
| Make ONNX installable even with spaces in path | #4473 | Merged | |
| Fix resize shape inference | #4448 | Merged | |
Add hasInput method to InferenceContext
|
#4451 | Merged | |
Make check_model with full_check always not modify the model in place |
#4456 | Merged | |
Use find_namespace_packages to get needed subdirectories to silent lots of warnings |
#4457 | Merged | |
| Primary ops to function milestone 1 | #4458 | Merged | |
| Primary ops to function milestone 2 | #4512 | Merged | |
| Use filesystem to load filename to prevent encoding issues on Windows | #4470 | Merged | |
| Remove unnecessary import | #4484 | Merged | |
| Reduces the number of ignored mypy errors | #4495 | Merged | |
| Avoid modifying node when creating test cases | #4517 | Merged | |
| Skip version conversion if non-ONNX domain and throw explicit error | #4521 | Merged | |
| Relax protobuf version requirement for version below 4 | #4535 | Merged | |
Move mypy config from setup.cfg to pyproject.toml
|
#4542 | Merged | |
| Fix several issues regarding recent mapping update | #4551 | Merged | |
Fix shebang for protoc-gen-mypy.sh script |
#4568 | Merged | |
| Catch >2GB models in checker | #2744 | Merged | |
| Relax protobuf version requirement for version below 4 | #4535 | Merged | |
| Bump protobuf from 3.16.0 to 3.18.3 | #4544 | Merged | |
| Move check_is_experimental_op from check_node to check_graph | #4556 | Merged | |
| Find python libs only if BUILD_ONNX_PYTHON is set to ON | #4579 | Merged | |
| Enable partial data propagation for Resize's sizes | #4582 | Merged | |
| Add type inference for CategoryMapper & TreeEnsembleRegressor/Classifier | #4600 | Merged | |
| Check the rest of actual input/output type if Variadic | #4622 | Merged | |
| Add 32 and 64 bit unsigned integers as tensor types in IR | #4634 | Merged | |
| Upgrade pybind11 to 2.9.2 to improve Python 3.11 support | #4635 | Merged | |
| Output missing last new line from coverage reporter | #4636 | Merged | |
| provide macos universal2 wheel | #4642 | Merged | |
| Fix backend test for ReferenceEvaluator with numpy 1.16.6 | #4658 | Merged | |
| Consider GraphInferenceContext in inference functions: InferenceContext | #4632 | Merged |
| Description | PR | Status | Notes |
|---|---|---|---|
Enhance test_backend_test.py and simplify backend test CI |
#3393 | Merged | |
Weekly test current version_converter with all models from ONNX Model Zoo |
#4040 | Merged | |
| Verify minimum supported numpy in Windows Release CI | #4279 | Merged | |
| Create inline lints with github actions | #4296 | Merged | |
| Remove stale bot config | #4315 | Merged | |
| Enable stale GitHub action | #4316 | Merged | |
| Add new Lint Python CI in the CI document | #4332 | Merged | |
| Ignore noisy pylint warnings | #4359 | Merged | |
| Set up black and isort | #4361 | Merged | |
| Test lower protoc version in Linux CI | #4365 | Merged | |
| Remove pyflakes from CI | #4371 | Merged | |
| Print model name in the exception message | #4378 | Merged | |
Upgrade to macos-11 due to macos-10.15 deprecation |
#4379 | Merged | |
| Install ONNX before linting | #4428 | Merged | |
| Fix Windows Release Python 3.10 CI failure when building NumPy from source | #4440 | Merged | |
| Fix global LGTM Python/C++ warnings and fully enable it | #4467 | Merged | |
| Upgrade macOS in azp due to upcoming deprecation | #4471 | Merged | |
| Install specified mypy version first in Python Lint CI to make mypy version consistent | #4498 | Merged | |
Validate the generated number of input.pb and output.pb and new node files in CIs |
#4514 | Merged | |
| Use GitHub code scanning before retiring deprecated LGTM.com | #4524 | Merged | |
| Add flake8-bugbear in checks | #4557 | Merged | |
| Editable_mode=False for now to enable pip install -e . with latest setuptools | #4558 | Merged | |
| Build onnx in page creating CI | #4570 | Merged | |
| Implement type and shape inference for ArrayFeatureExtractor | #4572 | Merged | |
| Manually catch build succeeded for page creating CI | #4576 | Merged | |
| Only manually trigger page deployment and remove duplicate style check in Linux-CI | #4597 | Merged | |
| Fix LayerNormalization equation | #4607 | Merged | |
| Upgrade to newer Protobuf 3.20.2 | #4629 | Merged | |
| revert AttributeHasValue | #4630 | Merged | |
| Use Python 3.10 in CI job CodeQL | #4633 | Merged | |
| Fix enforce-style CI error: solve empty-body issue raised by mypy 0.990 | #4643 | Merged | |
| Fix _parse_repo_info | #4648 | Merged | |
| Fix weekly mac release CI: only submit source distribution in one CI | #4656 | Merged | |
| Correct types in GridSample def.cc | #4655 | Merged |
| Description | PR | Status | Notes |
|---|---|---|---|
| Fix some typos | #4228 | Merged | |
| Remove prerequisites from installation instruction | #4247 | Merged | |
| Update README with build modifications | #4259 | Merged | |
Update repo_guidelines.md
|
#4306 | Merged | |
| Merge is needed only there are changes directly added to the release | #4308 | Merged | |
| Clarify in the doc that ONNX should use rc as release candidate | #4333 | Merged | |
Update readme.md
|
#4349 | Merged | |
Fix typo in AddNewOp.md
|
#4353 | Merged | |
| Add docs for cpp tests requiring built libraries | #4401 | Merged | |
| Update PR template | #4434 | Merged | |
| Update issues template and ONNX Hub doc | #4441 | Merged | |
Remove sentences about deprecated onnx-docker in docs |
#4455 | Merged | |
| Create feature request form | #4479 | Merged | |
Add vcpkg instruct step |
#4531 | Merged | |
| Fix documentation generation | #4564 | Merged | |
| Format Python code in documents with black | #4530 | Merged | |
| Updating based on steering committee discussion on 2022-09-14 | #4573 | Merged | |
| Update ScatterND documentation | #4575 | Merged | |
| fix spelling new python doc | #4598 | Merged | |
| Add a link to the HTML documentation from the main page | #4594 | Merged | |
| Clarify reshape behavior | #4601 | Merged | |
| Update maxpool doc | #4612 | Merged | |
| Rename onnx_docs_folder to operator in docs | #4623 | Merged | |
| fix some spelling and grammar issues | #4637 | Merged | |
| Fix link to package documentation on README.md | #4649 | Merged | |
| update docs for onnx.save_model | #4662 | Merged | |
| Fixed Typo in concepts.rst | #4661 | Merged | |
| Fix typo in python.rst | #4667 | Merged | |
| Fixed some typos in python.rst | #4668 | Merged |
| Description | PR | Status | Notes |
|---|---|---|---|
| Use fully qualified pathname when loading DLL to prevent security vulnerability | #4377 | Merged | |
Add comment to make sure vsnprintf usage is secure |
#4377 | Merged | |
| Solve vulnerability issue while loading external tensors | #4508 | Merged |
-
TENSOR_TYPE_TO_STORAGE_TENSOR_TYPEhas been deprecated #4270 - ONNXIFI: ONNX Interface for Framework Integration has been deprecated #4431
- onnxruntime
- pytorch
- onnx-tensorflow
- tensorflow-onnx
- sklearn-onnx
- onnxmltools
- onnx-tensort
- onnx-coreml
Thanks to these individuals for their contributions in this release since last 1.12.0 release: Anand Krishna, Christian Bourjau, Chun-Wei Chen, G. Ramalingam, Gary Miguel, Gaétan Lepage, Ilya Lavrenov, Ivan Novikov, Jack·Boos·Yu, Jakub Bachurski, Jicheng Tang, Joaquin Anton, Justin Chu, Nat Kershaw (MSFT), Philip Lassen, Prasanth Pulavarthi, Przemyslaw Wysocki, SpaceIm, Stephen Neuendorffer, Takeshi Watanabe, Teodora Sechkova, Thiago Crepaldi, Xavier Dupré, Yi Zhang, Yuan Yao, andife, daquexian, kylesayrs, liqun Fu, longlee0622, sqhao, williamberman, yanbc
The list has been acquired with a script written by Aaron Bockover.