beartype has served us well in identifying type errors and ensuring we call internal functions with the correct arguments (thanks!). However, the value of having beartype is diminished because of the following:
1. When beartype improves support for better Dict[] type checking, it discovered typing mistakes in some functions that were previously uncaught. This caused the exporter to fail with newer versions beartype when it used to succeed. Since we cannot fix PyTorch and release a new version just because of this, it creates confusion for users that have beartype in their environment from using torch.onnx
2. beartype adds an additional call line in the traceback, which makes the already thick dynamo stack even larger, affecting readability when users diagnose errors with the traceback.
3. Since the typing annotations need to be evaluated, we cannot use new syntaxes like `|` because we need to maintain compatibility with Python 3.8. We don't want to wait for PyTorch take py310 as the lowest supported Python before using the new typing syntaxes.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/130484
Approved by: https://github.com/titaiwangms
beartype has served us well in identifying type errors and ensuring we call internal functions with the correct arguments (thanks!). However, the value of having beartype is diminished because of the following:
1. When beartype improves support for better Dict[] type checking, it discovered typing mistakes in some functions that were previously uncaught. This caused the exporter to fail with newer versions beartype when it used to succeed. Since we cannot fix PyTorch and release a new version just because of this, it creates confusion for users that have beartype in their environment from using torch.onnx
2. beartype adds an additional call line in the traceback, which makes the already thick dynamo stack even larger, affecting readability when users diagnose errors with the traceback.
3. Since the typing annotations need to be evaluated, we cannot use new syntaxes like `|` because we need to maintain compatibility with Python 3.8. We don't want to wait for PyTorch take py310 as the lowest supported Python before using the new typing syntaxes.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/130484
Approved by: https://github.com/titaiwangms
beartype has served us well in identifying type errors and ensuring we call internal functions with the correct arguments (thanks!). However, the value of having beartype is diminished because of the following:
1. When beartype improves support for better Dict[] type checking, it discovered typing mistakes in some functions that were previously uncaught. This caused the exporter to fail with newer versions beartype when it used to succeed. Since we cannot fix PyTorch and release a new version just because of this, it creates confusion for users that have beartype in their environment from using torch.onnx
2. beartype adds an additional call line in the traceback, which makes the already thick dynamo stack even larger, affecting readability when users diagnose errors with the traceback.
3. Since the typing annotations need to be evaluated, we cannot use new syntaxes like `|` because we need to maintain compatibility with Python 3.8. We don't want to wait for PyTorch take py310 as the lowest supported Python before using the new typing syntaxes.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/130484
Approved by: https://github.com/titaiwangms
The `usort` config in `pyproject.toml` has no effect due to a typo. Fixing the typo make `usort` do more and generate the changes in the PR. Except `pyproject.toml`, all changes are generated by `lintrunner -a --take UFMT --all-files`.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/127122
Approved by: https://github.com/kit1980
This is a partial revert of https://github.com/pytorch/pytorch/pull/124059
Like in #124297, profiling has revealed that testing equality on *every* output is kind of expensive. So we only test equality when we know there is an unbacked binding. This is the same playbook as the previous PR, just on FakeTensorProp instead of PropagateUnbackedSymInts. Note that we also need to populate `unbacked_bindings` in proxy_tensor.py, since we're generating an entirely new graph in that case.
We now have enough propagation that we're able to trigger a bug related to divisibility replacement. In https://github.com/pytorch/pytorch/pull/113165 we allowed to replace `u0` with `u1 * c` for some constant c, when we have determined that u0 is divisible by c. However, where does the binding for u1 come from? What we will have in practice is that there is some node that is supposed to have bound u1, but which actually is getting a `u1 * c` in its output. So, to get u1, we must divide out c. Fortunately, under the divisibility condition, this is always possible (but remember, we must test divisibility at runtime!)
Because we have tightened up asserts, it is now an error to allocate unbacked SymInts and then fail to track them under unbacked_bindings. In torch/_dynamo/eval_frame.py and torch/_functorch/_aot_autograd/collect_metadata_analysis.py there are examples of benign cases where we repropagated fake tensors but then immediately threw away the results. In these cases, it's not appropriate to rebind, since we're still using the old FX graph that has all of the old symbols. So we just manually clear it. It is possible that other cases will need to be updated, so this PR is "risky" from the perspective of hitting fbcode.
Signed-off-by: Edward Z. Yang <ezyang@meta.com>
Pull Request resolved: https://github.com/pytorch/pytorch/pull/124310
Approved by: https://github.com/lezcano
A global transorm list is created. All backend instances call the transform functions in that list sequentially to modify the exported ONNX model before sending model to ORT session. For example, `record_onnx_model_transform` below is a no-op transform and only records the ONNX graphs sent to ONNXRuntime.
```python
recorded_models = []
def record_onnx_model_transform(onnx_model):
# Record the ONNX model seen by the transform.
recorded_models.append(onnx_model)
from torch.onnx import (
register_backend_graph_transform,
unregister_backend_graph_transform,
)
# Register so that `onnxrt` backend calls it to modify ONNX model.
register_backend_graph_transform(record_onnx_model_transform)
def example_model(x: torch.Tensor):
y = torch.sigmoid(x)
z = x + y
return z
# During the compilation, the exported ONNX model will be
# modified by calling `record_onnx_model_transform` before
# sending the model to `onnxruntime.InferenceSession`.
compiled_model = torch.compile(
example_model,
backend="onnxrt",
dynamic=True,
)
# Now, `recorded_models` should contain one `onnx.ModelProto` representing
# `example_model(x: torch.Tensor)`.
# Remove the pass when not needed. If `record_onnx_model_transform` is not
# removed, it will be applied to all models compiled by `backend="onnxrt"`.
unregister_backend_graph_transform(record_onnx_model_transform)
```
In the future, we plan to use this mechanism to register all graph transforms such ash graph fusion and general ONNX optimization for `onnxrt`.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/120854
Approved by: https://github.com/BowenBao, https://github.com/thiagocrepaldi
As title. This PR enables dynamic shapes for running llama with ORT. Both forward and backward are captured as a single graph with this PR.
Summary of changes:
- Test llama attention, llama decoder, llama model to ensure (1) no graph breaks (2) models exported with dynamic shapes with onnxrt dynamo backend
- Reshape SymInt to tensor with shape (1,) to align with the cast done for int in fx_onnx_interpreter.py
- Create an util function to map Python types (e.g., float) to ONNX tensor element type (e.g., onnx.TensorProto.FLOAT).
- Return `hint` for torch.Sym* in type promotion pass.
- Remove _replace_to_copy_with_to since exporter supports aten::_to_copy it now.
- Modify _get_onnx_devices to return CPU device for torch.Sym*.
- Introduce _adjust_scalar_from_fx_to_onnx (e.g., change 0 to tensor(0)) and _adjust_scalar_from_onnx_to_fx (e.g., change tensor(0) to 0) for adjusting scalars when passing values to and receive values from ORT.
- Now, ValueInfoProto of graph inputs (i.e., input_value_infos) are stored and used as `ORT-expected type` when calling `_adjust_scalar_from_fx_to_onnx`.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/117009
Approved by: https://github.com/titaiwangms
This reworks the DORT backend factory function to support the options kwarg of torch.compile, and defines a concrete OrtBackendOptions type that can be used to influence the backend.
Caching is also implemented in order to reuse backends with equal options.
Wrapping the backend in auto_autograd also becomes an option, which allows `OrtBackend` to always be returned as the callable for torch.compile; wrapping happens internally if opted into (True by default).
Lastly, expose options for configuring preferred execution providers (will be attempted first), whether or not to attempt to infer an ORT EP from a torch found device in the graph or inputs, and finally the default/fallback EPs.
### Demo
The following demo runs `Gelu` through `torch.compile(backend="onnxrt")` using various backend options through a dictionary form and a strongly typed form. It additionally exports the model through both the ONNX TorchScript exporter and the new TorchDynamo exporter.
```python
import math
import onnx.inliner
import onnxruntime
import torch
import torch.onnx
torch.manual_seed(0)
class Gelu(torch.nn.Module):
def forward(self, x):
return x * (0.5 * torch.erf(math.sqrt(0.5) * x) + 1.0)
@torch.compile(
backend="onnxrt",
options={
"preferred_execution_providers": [
"NotARealEP",
"CPUExecutionProvider",
],
"export_options": torch.onnx.ExportOptions(dynamic_shapes=True),
},
)
def dort_gelu(x):
return Gelu()(x)
ort_session_options = onnxruntime.SessionOptions()
ort_session_options.log_severity_level = 0
dort_gelu2 = torch.compile(
Gelu(),
backend="onnxrt",
options=torch.onnx._OrtBackendOptions(
preferred_execution_providers=[
"NotARealEP",
"CPUExecutionProvider",
],
export_options=torch.onnx.ExportOptions(dynamic_shapes=True),
ort_session_options=ort_session_options,
),
)
x = torch.randn(10)
torch.onnx.export(Gelu(), (x,), "gelu_ts.onnx")
export_output = torch.onnx.dynamo_export(Gelu(), x)
export_output.save("gelu_dynamo.onnx")
inlined_model = onnx.inliner.inline_local_functions(export_output.model_proto)
onnx.save_model(inlined_model, "gelu_dynamo_inlined.onnx")
print("Torch Eager:")
print(Gelu()(x))
print("DORT:")
print(dort_gelu(x))
print(dort_gelu2(x))
```
Pull Request resolved: https://github.com/pytorch/pytorch/pull/107973
Approved by: https://github.com/BowenBao
