The dipatcher didn't check attribute dtype, as AttributeProto is a totally different system from InputProto in ONNX. This PR introduces the mapping table for AttributeProto type to python type. And further utilize it in opschema matching.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/105104
Approved by: https://github.com/thiagocrepaldi
## ONNXRegistry
### Motivation
In #100660, we used the torchscript registry to allow dispatcher. However, it doesn't meet the needs of FX exporter. The idea of torchscript exporter is built on top of three points:
(1) Use `_SymbolicFunctionGroup` to dispatch opset version as we need ops to fall back when we don't have it in the current exporter opset version
(2) One aten maps to multiple supported opset versions, and each version maps to one symbolic function
(3) Custom symbolic function is considered prior to default symbolic function
Now that TorchLib will support all aten op across all opset versions, we don't need the opset version dispatch layer. And with onnx overloads created by torchlib, we need a way to support custom operators and prioritize them among all overloads.
### Feature
Introduce a public OnnxRegistry API initiated with fixed opset version which supports user registered operators. **The dispatching opset version is no longer needed as TorchLib is expected to provide full aten support across all opset version. And Dispatcher is expected to prioritize custome operators than the defaults.**
### API:
(1) `register_custom_op(self, function: OnnxFunction, domain: str, op_name: str, overload: Optional[str] = None)`: Register a custom operator into the current OnnxRegistry. This is expected to be used when the default operators don't mee the need of users. **For example, need a different opset version from the registry, or different calculation**.
(2) `is_registered_op(self, domain: str, op_name: str, overload: Optional[str] = None)`: Whether the aten op is registered.
(3) `get_functions(domain: str, op_name: str, overload: Optional[str] = None)`: Return a set of registered SymbolicFunctions under the aten
### TODO:
(1)`remove_op(op_name: str)`: removing the whole support for certain op allows decompose the graph to prims.
(2)Expose OnnxRegistry to users, and disable the opset_version option in export API. Export API should use the ops in registry only.
---
## OnnxDispatcher
The Changes in the function `dispatch` and `_find_the_perfect_or_nearest_match_onnxfunction` are meant to allow complex type and custom operator supports.
### Respect Custom Ops
(1) Override: Check if we can find the perfect match in custom operator overloads prior to defaults
(2) Tie breaker: If we have the same nearest match of default and custom overload, we choose the custom.
### Supplementary
[Design discussion doc](https://microsoft-my.sharepoint.com/:w:/p/thiagofc/EW-5Q3jWhFNMtQHHtPpJiAQB-P2qAcVRkYjfbmeSddnjWA?e=QUX9zG&wdOrigin=TEAMS-ELECTRON.p2p.bim&wdExp=TEAMS-TREATMENT&wdhostclicktime=1687554493295&web=1)
Please check the Registry and Dispatcher sections.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/103943
Approved by: https://github.com/BowenBao, https://github.com/justinchuby
Introduce `Analysis` to analyze fx graphmodule and emit diagnostics. This class
can be extended to interact with `Transform` (passes) to decide if a pass should
trigger based on graph analysis result. E.g., if decomp needs to run by checking
operator namespace in nodes. For now leaving it as out of scope but can revisit
if maintaining multi fx extractor becomes reality.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/100451
Approved by: https://github.com/titaiwangms
* CI Test environment to install onnx and onnx-script.
* Add symbolic function for `bitwise_or`, `convert_element_type` and `masked_fill_`.
* Update symbolic function for `slice` and `arange`.
* Update .pyi signature for `_jit_pass_onnx_graph_shape_type_inference`.
Co-authored-by: Wei-Sheng Chin <wschin@outlook.com>
Co-authored-by: Ti-Tai Wang <titaiwang@microsoft.com>
Pull Request resolved: https://github.com/pytorch/pytorch/pull/94564
Approved by: https://github.com/abock
All this time, PyTorch and ONNX has different strategy for None in output. And in internal test, we flatten the torch outputs to see if the rest of them matched. However, this doesn't work anymore in scripting after Optional node is introduced, since some of None would be kept.
#83184 forces script module to keep all Nones from Pytorch, but in ONNX, the model only keeps the ones generated with Optional node, and deletes those meaningless None.
This PR uses Optional node to keep those meaningless None in output as well, so when it comes to script module result comparison, Pytorch and ONNX should have the same amount of Nones.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/84789
Approved by: https://github.com/BowenBao
Fix#82589
Why:
1. **full_check** works in `onnx::checker::check_model` function as it turns on **strict_mode** in `onnx::shape_inference::InferShapes()` which I think that was the intention of this part of code.
2. **strict_mode** catches failed shape type inference (invalid ONNX model from onnx perspective) and ONNXRUNTIME can't run these invalid models, as ONNXRUNTIME actually rely on ONNX shape type inference to optimize ONNX graph. Why we don't set it True for default? >>> some of existing users use other platform, such as caffe2 to run ONNX model which doesn't need valid ONNX model to run.
3. This PR doesn't change the original behavior of `check_onnx_proto`, but add a warning message for those models which can't pass strict shape type inference, saying the models would fail on onnxruntime.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/83186
Approved by: https://github.com/justinchuby, https://github.com/thiagocrepaldi, https://github.com/jcwchen, https://github.com/BowenBao
* Reflect required arguments in method signature for each diagnostic rule. Previous design accepts arbitrary sized tuple which is hard to use and prone to error.
* Removed `DiagnosticTool` to keep things compact.
* Removed specifying supported rule set for tool(context) and checking if rule of reported diagnostic falls inside the set, to keep things compact.
* Initial overview markdown file.
* Change `full_description` definition. Now `text` field should not be empty. And its markdown should be stored in `markdown` field.
* Change `message_default_template` to allow only named fields (excluding numeric fields). `field_name` provides clarity on what argument is expected.
* Added `diagnose` api to `torch.onnx._internal.diagnostics`.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/87830
Approved by: https://github.com/abock
Introduce `_jit_pass_onnx_assign_node_and_value_names` to parse and assign
scoped name for nodes and values in exported onnx graph.
Module layer information is obtained from `ONNXScopeName` captured in `scope`
attribute in nodes. For nodes, the processed onnx node name are stored in
attribute `onnx_name`. For values, the processed onnx output name are stored
as `debugName`.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/82040
Approved by: https://github.com/AllenTiTaiWang, https://github.com/justinchuby, https://github.com/abock
Introduce `_jit_pass_onnx_assign_node_and_value_names` to parse and assign
scoped name for nodes and values in exported onnx graph.
Module layer information is obtained from `ONNXScopeName` captured in `scope`
attribute in nodes. For nodes, the processed onnx node name are stored in
attribute `onnx_name`. For values, the processed onnx output name are stored
as `debugName`.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/82040
Approved by: https://github.com/AllenTiTaiWang, https://github.com/justinchuby, https://github.com/abock
Add flag (inline_autograd) to enable inline export of model consisting of autograd functions. Currently, this flag should only be used in TrainingMode.EVAL and not for training.
An example:
If a model containing ``autograd.Function`` is as follows
```
class AutogradFunc(torch.autograd.Function):
@staticmethod
def forward(ctx, i):
result = i.exp()
result = result.log()
ctx.save_for_backward(result)
return result
```
Then the model is exported as
```
graph(%0 : Float):
%1 : Float = ^AutogradFunc(%0)
return (%1)
```
If inline_autograd is set to True, this will be exported as
```
graph(%0 : Float):
%1 : Float = onnx::Exp(%0)
%2 : Float = onnx::Log(%1)
return (%2)
```
If one of the ops within the autograd module is not supported, that particular node is exported as is mirroring ONNX_FALLTHROUGH mode
Fixes: #61813
Pull Request resolved: https://github.com/pytorch/pytorch/pull/74765
Approved by: https://github.com/BowenBao, https://github.com/malfet
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/73284
Some important ops won't support optional type until opset 16,
so we can't fully test things end-to-end, but I believe this should
be all that's needed. Once ONNX Runtime supports opset 16,
we can do more testing and fix any remaining bugs.
Test Plan: Imported from OSS
Reviewed By: albanD
Differential Revision: D34625646
Pulled By: malfet
fbshipit-source-id: 537fcbc1e9d87686cc61f5bd66a997e99cec287b
Co-authored-by: BowenBao <bowbao@microsoft.com>
Co-authored-by: neginraoof <neginmr@utexas.edu>
Co-authored-by: Nikita Shulga <nshulga@fb.com>
(cherry picked from commit 822e79f31ae54d73407f34f166b654f4ba115ea5)
Summary:
Add ONNX exporter logging facility. Supporting both C++/Python logging api. Logging can be turned on/off. Logging output stream can be either set to `stdout` or `stderr`.
A few other changes:
* When exception is raised in passes, the current IR graph being processed will be logged.
* When exception is raised from `_jit_pass_onnx` (the pass that converts nodes from namespace `ATen` to `ONNX`), both ATen IR graph and ONNX IR graph under construction will be logged.
* Exception message for ConstantFolding is truncated to avoid being too verbose.
* Update the final printed IR graph with node name in ONNX ModelProto as node attribute. Torch IR Node does not have name. Adding this to printed IR graph helps debugging.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/71342
Reviewed By: msaroufim
Differential Revision: D34433473
Pulled By: malfet
fbshipit-source-id: 4b137dfd6a33eb681a5f2612f19aadf5dfe3d84a
(cherry picked from commit 67a8ebed5192c266f604bdcca931df6fe589699f)
Enables local function export to capture annotated attributes.
For example:
```python
class M(torch.nn.Module):
num_layers: int
def __init__(self, num_layers):
super().__init__()
self.num_layers = num_layers
def forward(self, args):
...
```
`num_layers` will now be captured as attribute of local function `M`.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/72883
Summary:
Based on past PRs, here is an non-exhaustive list of files to consider for extension. The PR is not meant to be final. Based on feedback and discussion, files could be dropped from the list, or PR could be updated to move code around such that extension is no longer needed.
List of files below and description:
* These files are for converting from IR to ONNX proto. These should be used only for ONNX.
```
"torch/csrc/jit/serialization/export.*",
"torch/csrc/jit/serialization/onnx.*",
```
* This file is touched whenever pass signature is updated.
```
"torch/_C/__init__.pyi.in",
```
* These files are touched whenever pass signature is updated. Somehow it's been convention that onnx passes are also added here, but it could be possible to move them. Let me know what you think.
~~"torch/csrc/jit/python/init.cpp",~~
~~"torch/csrc/jit/python/script_init.cpp",~~
Update: Bowen will move onnx passes to files under onnx folder.
* ~~Touched when need new attr::xxx, or onnx::xxx.~~
~~"aten/src/ATen/core/interned_strings.h"~~
Update: Nikita will help separate this file.
malfet
Pull Request resolved: https://github.com/pytorch/pytorch/pull/72297
Reviewed By: H-Huang
Differential Revision: D34254666
Pulled By: malfet
fbshipit-source-id: 032cfa590cbedf4648b7335fe8f09a2380ab14cb
(cherry picked from commit 88653eadbf)
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/68490
The use of ATEN as a fallback operator during ONNX conversion is important for increasing operator coverage or even provide more efficient implementations over some ONNX ops.
Currently this feature is available through `OperatorExportTypes.ONNX_ATEN_FALLBACK`,
but it also performs changes to the graph that are runnable by Caffe2, only.
This PR introduces restricts caffe2-specific graph transformations for `ONNX_ATEN_FALLBACK`
operator export type for when pytorch is built with caffe2 support (aka BUILD_CAFFE2=1 during build)
The first version of this PR introduced a new operator export type `ONNX_ATEN__STRICT_FALLBACK`,
which essentially is the same as `ONNX_ATEN_FALLBACK` but without caffe2 transformations.
It was preferred to not introduce a new operator export type, but to refine the existing aten fallback one
## BC-breaking note
### The global constant `torch.onnx.PYTORCH_ONNX_CAFFE2_BUNDLE` is removed in favor of
a less visible `torch.onnx._CAFFE2_ATEN_FALLBACK`.
`PYTORCH_ONNX_CAFFE2_BUNDLE` is really a dead code flag always set to False.
One alternative would be fixing it, but #66658 disables Caffe2 build by default.
Making a Caffe2 feature a private one seems to make more sense for future deprecation.
### The method `torch.onnx.export` now defaults to ONNX when `operator_export_type` is not specified.
Previously `torch.onnx.export's operator_export_type` intended to default to `ONNX_ATEN_FALLBACK` when `PYTORCH_ONNX_CAFFE2_BUNDLE` was set, but it would never happen as `PYTORCH_ONNX_CAFFE2_BUNDLE` is always undefined
Co-authored-by: Nikita Shulga <nshulga@fb.com>
Test Plan: Imported from OSS
Reviewed By: jansel
Differential Revision: D32483781
Pulled By: malfet
fbshipit-source-id: e9b447db9466b369e77d747188685495aec3f124
(cherry picked from commit 5fb1eb1b19)
Summary:
- PyTorch and ONNX has supported BFloat16, add this to unblock some mixed-precision training model.
- Support PyTorch TNLG model to use BFloat16 tensors for the inputs/outputs of the layers that run on the NPU.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/66788
Reviewed By: jansel
Differential Revision: D32283510
Pulled By: malfet
fbshipit-source-id: 150d69b1465b2b917dd6554505eca58042c1262a
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/67803
* Addresses comments from #63589
[ONNX] remove torch::onnx::PRODUCER_VERSION (#67107)
Use constants from version.h instead.
This simplifies things since we no longer have to update
PRODUCER_VERSION for each release.
Also add TORCH_VERSION to version.h so that a string is available for
this purpose.
[ONNX] Set `ir_version` based on opset_version. (#67128)
This increases the odds that the exported ONNX model will be usable.
Before this change, we were setting the IR version to a value which may
be higher than what the model consumer supports.
Also some minor clean-up in the test code:
* Fix string replacement.
* Use a temporary file so as to not leave files around in the test
current working directory.
Test Plan: Imported from OSS
Reviewed By: msaroufim
Differential Revision: D32181306
Pulled By: malfet
fbshipit-source-id: 02f136d34ef8f664ade0bc1985a584f0e8c2b663
Co-authored-by: BowenBao <bowbao@microsoft.com>
Co-authored-by: Gary Miguel <garymiguel@microsoft.com>
Co-authored-by: Nikita Shulga <nshulga@fb.com>
Summary:
Delete `-Wno-unused-variable` from top level `CMakeLists.txt`
Still suppress those warnings for tests and `torch_python`
Delete number of unused variables from caffe2 code
Use `(void)var;` to suppress unused variable in range loops
Use `C10_UNUSED` for global constructors and use `constexpr` instead of `static` for global constants
Do not delete `caffe2::OperatorBase::Output` calls as they have side effects
Pull Request resolved: https://github.com/pytorch/pytorch/pull/66041
Reviewed By: ngimel
Differential Revision: D31360142
Pulled By: malfet
fbshipit-source-id: 6fdfb9f91efdc49ca984a2f2a17ee377d28210c8
Summary:
Delete `-Wno-unused-variable` from top level `CMakeLists.txt`
Still suppress those warnings for tests and `torch_python`
Delete number of unused variables from caffe2 code
Use `(void)var;` to suppress unused variable in range loops
Use `C10_UNUSED` for global constructors and use `constexpr` instead of `static` for global constants
Pull Request resolved: https://github.com/pytorch/pytorch/pull/65954
Reviewed By: ngimel
Differential Revision: D31326599
Pulled By: malfet
fbshipit-source-id: 924155f1257a2ba1896c50512f615e45ca1f61f3
