### Description
The PR implements FloatE4M3FN, FloatE5M2, FloatE4MEFNUZ, FloatE5M2FNUZ
as described in PR https://github.com/onnx/onnx/pull/4805. It uses CUDA
API to cast float/half to float8 if CUDA>=11.8, a custom implementation
if CUDA<11.8.
* It implements, Cast, QuantizeLinear, DequantizeLinear for all types on
CPU, only for types FloatE4M3FN, FloatE5M2 on CUDA.
* It extends the supported types for control flow operator, Shape,
Reshape, Identity, If, Loop, Scan, Reshape
* It implements Equal(19).
* Cast, QuantizeLinear, DequantizeLinear operators now support a
parameter `saturate` only valid for float 8 types. It is true by
default. In that case, any value out of range is converted into the
maximum float 8 value. If false, it is infinite.
* QuantizeLinear, DequantizeLinear now supports multiple scales on CUDA
(and ROCm by extension), scale = 1D tensor with one scale per channel
### Motivation and Context
Supports latest onnx version.
Fixes
[AB#15395](https://aiinfra.visualstudio.com/6a833879-cd9b-44a4-a9de-adc2d818f13c/_workitems/edit/15395)
---------
Co-authored-by: Xavier Dupre <xadupre@microsoft.com@orttrainingdev8.d32nl1ml4oruzj4qz3bqlggovf.px.internal.cloudapp.net>
Co-authored-by: Randy Shuai <rashuai@microsoft.com>
Co-authored-by: Edward Chen <18449977+edgchen1@users.noreply.github.com>
Co-authored-by: Scott McKay <Scott.McKay@microsoft.com>
### Description
Bump ruff version in CI and fixed new lint errors.
- This change enables the flake8-implicit-str-concat rules which helps
detect unintended string concatenations:
https://beta.ruff.rs/docs/rules/#flake8-implicit-str-concat-isc
- Update gitignore to include common python files that we want to
exclude.
### Motivation and Context
Code quality
### Description
<!-- Describe your changes. -->
Add required graph transformer to duplicate DQ nodes to ensure that QDQ
node units have unique DQ nodes. This condition is necessary for QDQ
node unit processing.
### Motivation and Context
<!-- - Why is this change required? What problem does it solve?
- If it fixes an open issue, please link to the issue here. -->
There is an existing Python utility that does this:
c7ced7a5e9/tools/python/util/qdq_helpers/qdq_model_utils.py (L77)
This PR implements it as a graph transformer so it is integrated into
ORT and does not require a separate step to update the model. There are
also tests to ensure that its effects are not undone by basic level
graph optimizations.
### Description
`lintrunner` is a linter runner successfully used by pytorch, onnx and
onnx-script. It provides a uniform experience running linters locally
and in CI. It supports all major dev systems: Windows, Linux and MacOs.
The checks are enforced by the `Python format` workflow.
This PR adopts `lintrunner` to onnxruntime and fixed ~2000 flake8 errors
in Python code. `lintrunner` now runs all required python lints
including `ruff`(replacing `flake8`), `black` and `isort`. Future lints
like `clang-format` can be added.
Most errors are auto-fixed by `ruff` and the fixes should be considered
robust.
Lints that are more complicated to fix are applied `# noqa` for now and
should be fixed in follow up PRs.
### Notable changes
1. This PR **removed some suboptimal patterns**:
- `not xxx in` -> `xxx not in` membership checks
- bare excepts (`except:` -> `except Exception`)
- unused imports
The follow up PR will remove:
- `import *`
- mutable values as default in function definitions (`def func(a=[])`)
- more unused imports
- unused local variables
2. Use `ruff` to replace `flake8`. `ruff` is much (40x) faster than
flake8 and is more robust. We are using it successfully in onnx and
onnx-script. It also supports auto-fixing many flake8 errors.
3. Removed the legacy flake8 ci flow and updated docs.
4. The added workflow supports SARIF code scanning reports on github,
example snapshot:
5. Removed `onnxruntime-python-checks-ci-pipeline` as redundant
### Motivation and Context
<!-- - Why is this change required? What problem does it solve?
- If it fixes an open issue, please link to the issue here. -->
Unified linting experience in CI and local.
Replacing https://github.com/microsoft/onnxruntime/pull/14306
---------
Signed-off-by: Justin Chu <justinchu@microsoft.com>
### Description
<!-- Describe your changes. -->
I fixed some broken links in the C API documentation, but then did a
quick pass over all of the links I could find and then fixed those.
### Motivation and Context
<!-- - Why is this change required? What problem does it solve?
- If it fixes an open issue, please link to the issue here. -->
I got some 404's when exploring the documentation and wanted to fix it.
Description: Set black's target version to be py37 - py310
Motivation and Context
Black by default targets its format for py3.10. Since our project supports python 3.7, we need to target version to all the python versions supported.
Re-ran black. 13 files reformatted.
Description: Format all python files under onnxruntime with black and isort.
After checking in, we can use .git-blame-ignore-revs to ignore the formatting PR in git blame.
#11315, #11316
Follow up to #10904.
- Move node EP assignment for ORT format into SessionState::FinalizeSessionState().
- Add unit test for #10904.
- Make convert_onnx_models_to_ort.py optimization level configurable via environment variable.
* Tweaks to the model utils
* Add handling for a dim_value of -1 when replacing the entire input shape. This occurs in models exported from PaddlePaddle
* make pytorch helpers accessible in package
* make QDQ helpers accessible in package
Add runtime optimization support to ONNX -> ORT format conversion script.
Replace `--optimization_level`, `--use_nnapi`, and `--use_coreml` with a new `--optimization_style` option.
* Add initial helper for optimizing a QDQ format model for usage with ORT.
If a DQ node has multiple consumers it will end up in multiple QDQ node units. This is complicated to handle as each qdq unit could end up being handled by different execution providers. By duplicating the DQ node we simplify this logic.
Generally the duplicate nodes will disappear when the qdq node unit is converted to a single node with a quantized operator. If there are qdq node units that are not able to be converted to use a quantized operator the ORT cleanup (pending) to drop remaining Q->DQ pairs between fp32 nodes can remove any remaining DQ nodes.
* Fix pep8 warning
Co-authored-by: Guoyu Wang <wanggy@outlook.com>
* Update required operators for prebuilt package to add opsets 14 and 15.
Add helper script to check if the prebuilt package will support the model and if not why not.
* Add support for multiple opsets being specified on a single line in the required operators config. This makes it easier to update the pre-built package config.
It's also required for validation tools to work as they only have a single opset from the model and not per-operator opsets. If we only list the incremental ops we could merge in the ops from the previous opset, but that wouldn't give a way to drop an operator from being supported.
Left the info on which ops changed though so we have a better feel for the cost of supporting each opset.
* schema change
* cc channges
* remove temp debug code
* Adding fbs namespace to session_state_flatbuffers_utils.h
* Add fbs namepsace to all ort format utils
* Change the strided copy to switch on data size not data type.
Move to header so we can reduce on the enabled types.
Setup type reduction for Concat now that it's using this implementation.
* Include ORT format model conversion scripts and infrastructure in ORT python package.
- tweak existing script setup so it can be easily run directly and from the ORT python package
Add config file and readme for Android minimal build package
Update ORT Mobile doco
Disable warning if 'all' optimizations are enabled but NCHWc transformer is excluded (device specific optimizations don't apply in this scenario so the warning is moot).
* Address PR comments
* Updates to some operators to always support int32 and int64 based on testing of Android package build config with a minimal build.
If an operator can be used for shape manipulation (int64) it is frequently used for indices manipulation (int32), so we enable both types for that set of ops.
- e.g. BERT models take indices as input
- Scatter/Gather ops utilize indices
Misc. fix to python bindings to exclude call that fails in a minimal build.
Enable type reduction for Scatter/ScatterElements CPU kernels. Some refactoring to reduce binary size.
Add MLTypeCallDispatcher methods.
Minor cleanup for Pad CPU kernel.
Enable type reduction for Shrink, Sign, SplitToSequence CPU kernels.
Some other type reduction changes including refactoring to specify element types in a single place.
Update the kernel def hashing in ORT format models. The new hashing logic ignores the ordering of type constraint types.
This is a backward compatibility breaking change, but we don't guarantee backward compatibility yet.
* Add support for custom ops library to the ORT model conversion script
Simplify model conversion now that we read ops from the ORT format model.
Enable custom ops in the python bindings if custom ops are turned on in a minimal build.
* Add test of model conversion involving custom ops.
* Add type reduction support to Min, Max and Pow
Update the C++ type reduction infrastructure to allow specifying an opset for the supported types list, as those can change across opset versions.
Minor updates to the type usage tracking script
* Add 'all opsets' macros and constant
* Add ability to generate configuration that includes required types for individual operators, to allow build size reduction based on that.
- Add python bindings for ORT format models
- Add script to update bindings and help info
- Add parsing of ORT format models
- Add ability to enable type reduction to config generation
- Update build.py to only allow operator/type reduction via config
- simpler to require config to be generated first
- can't mix a type aware (ORT format model only) and non-type aware config as that may result in insufficient types being enabled
- Add script to create reduced build config
- Update CIs
Follow up to #5811 to automate cleanup of the build docker image cache.
Added a script and build definition to clean up docker images that haven't been accessed recently.
This PR adds infrastructure to automatically cache docker images used in CI builds in a container registry.
Currently, build images are pulled from a container registry for some builds and built every time for others. The container registry requires maintenance to keep the images up to date and building images every time wastes build agent resources.
With this change, a given build image can be looked up in a cache container registry and if present, pulled, and otherwise, built and pushed. The uniqueness of a build image is determined by a hash digest of the dockerfile, docker build context directory, and certain "docker build" options. This digest is part of the image tag in the cache container repository.
The cache container registry will need to be cleaned up periodically. This is not automated yet.
