.. automodule:: torch.onnx
Open Neural Network eXchange (ONNX) is an open standard format for representing machine learning models. The torch.onnx module can export PyTorch models to ONNX. The model can then be consumed by any of the many runtimes that support ONNX.
Here is a simple script which exports a pretrained AlexNet to an ONNX file named alexnet.onnx.
The call to torch.onnx.export runs the model once to trace its execution and then exports the
traced model to the specified file:
import torch import torchvision dummy_input = torch.randn(10, 3, 224, 224, device="cuda") model = torchvision.models.alexnet(pretrained=True).cuda() # Providing input and output names sets the display names for values # within the model's graph. Setting these does not change the semantics # of the graph; it is only for readability. # # The inputs to the network consist of the flat list of inputs (i.e. # the values you would pass to the forward() method) followed by the # flat list of parameters. You can partially specify names, i.e. provide # a list here shorter than the number of inputs to the model, and we will # only set that subset of names, starting from the beginning. input_names = [ "actual_input_1" ] + [ "learned_%d" % i for i in range(16) ] output_names = [ "output1" ] torch.onnx.export(model, dummy_input, "alexnet.onnx", verbose=True, input_names=input_names, output_names=output_names)
The resulting alexnet.onnx file contains a binary protocol buffer
which contains both the network structure and parameters of the model you exported
(in this case, AlexNet). The argument verbose=True causes the
exporter to print out a human-readable representation of the model:
# These are the inputs and parameters to the network, which have taken on
# the names we specified earlier.
graph(%actual_input_1 : Float(10, 3, 224, 224)
%learned_0 : Float(64, 3, 11, 11)
%learned_1 : Float(64)
%learned_2 : Float(192, 64, 5, 5)
%learned_3 : Float(192)
# ---- omitted for brevity ----
%learned_14 : Float(1000, 4096)
%learned_15 : Float(1000)) {
# Every statement consists of some output tensors (and their types),
# the operator to be run (with its attributes, e.g., kernels, strides,
# etc.), its input tensors (%actual_input_1, %learned_0, %learned_1)
%17 : Float(10, 64, 55, 55) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[11, 11], pads=[2, 2, 2, 2], strides=[4, 4]](%actual_input_1, %learned_0, %learned_1), scope: AlexNet/Sequential[features]/Conv2d[0]
%18 : Float(10, 64, 55, 55) = onnx::Relu(%17), scope: AlexNet/Sequential[features]/ReLU[1]
%19 : Float(10, 64, 27, 27) = onnx::MaxPool[kernel_shape=[3, 3], pads=[0, 0, 0, 0], strides=[2, 2]](%18), scope: AlexNet/Sequential[features]/MaxPool2d[2]
# ---- omitted for brevity ----
%29 : Float(10, 256, 6, 6) = onnx::MaxPool[kernel_shape=[3, 3], pads=[0, 0, 0, 0], strides=[2, 2]](%28), scope: AlexNet/Sequential[features]/MaxPool2d[12]
# Dynamic means that the shape is not known. This may be because of a
# limitation of our implementation (which we would like to fix in a
# future release) or shapes which are truly dynamic.
%30 : Dynamic = onnx::Shape(%29), scope: AlexNet
%31 : Dynamic = onnx::Slice[axes=[0], ends=[1], starts=[0]](%30), scope: AlexNet
%32 : Long() = onnx::Squeeze[axes=[0]](%31), scope: AlexNet
%33 : Long() = onnx::Constant[value={9216}](), scope: AlexNet
# ---- omitted for brevity ----
%output1 : Float(10, 1000) = onnx::Gemm[alpha=1, beta=1, broadcast=1, transB=1](%45, %learned_14, %learned_15), scope: AlexNet/Sequential[classifier]/Linear[6]
return (%output1);
}
You can also verify the output using the ONNX library,
which you can install using pip:
pip install onnx
Then, you can run:
import onnx
# Load the ONNX model
model = onnx.load("alexnet.onnx")
# Check that the model is well formed
onnx.checker.check_model(model)
# Print a human readable representation of the graph
print(onnx.helper.printable_graph(model.graph))
You can also run the exported model with one of the many runtimes that support ONNX. For example after installing ONNX Runtime, you can load and run the model:
import onnxruntime as ort
ort_session = ort.InferenceSession("alexnet.onnx")
outputs = ort_session.run(
None,
{"actual_input_1": np.random.randn(10, 3, 224, 224).astype(np.float32)},
)
print(outputs[0])
Here is a more involved tutorial on exporting a model and running it with ONNX Runtime.
Internally, :func:`torch.onnx.export()` requires a :class:`torch.jit.ScriptModule` rather than
a :class:`torch.nn.Module`. If the passed-in model is not already a ScriptModule,
export() will use tracing to convert it to one:
- Tracing: If
torch.onnx.export()is called with a Module that is not already aScriptModule, it first does the equivalent of :func:`torch.jit.trace`, which executes the model once with the givenargsand records all operations that happen during that execution. This means that if your model is dynamic, e.g., changes behavior depending on input data, the exported model will not capture this dynamic behavior. We recommend examining the exported model and making sure the operators look reasonable. Tracing will unroll loops and if statements, exporting a static graph that is exactly the same as the traced run. If you want to export your model with dynamic control flow, you will need to use scripting. - Scripting: Compiling a model via scripting preserves dynamic control flow and is valid for inputs
of different sizes. To use scripting:
- Use :func:`torch.jit.script` to produce a
ScriptModule. - Call
torch.onnx.export()with theScriptModuleas the model. Theargsare still required, but they will be used internally only to produce example outputs, so that the types and shapes of the outputs can be captured. No tracing will be performed.
- Use :func:`torch.jit.script` to produce a
See Introduction to TorchScript and TorchScript for more details, including how to compose tracing and scripting to suit the particular requirements of different models.
PyTorch models can be written using NumPy or Python types and functions, but during :ref:`tracing<tracing-vs-scripting>`, any variables of NumPy or Python types (rather than torch.Tensor) are converted to constants, which will produce the wrong result if those values should change depending on the inputs.
For example, rather than using numpy functions on numpy.ndarrays:
# Bad! Will be replaced with constants during tracing. x, y = np.random.rand(1, 2), np.random.rand(1, 2) np.concatenate((x, y), axis=1)
Use torch operators on torch.Tensors:
# Good! Tensor operations will be captured during tracing. x, y = torch.randn(1, 2), torch.randn(1, 2) torch.cat((x, y), dim=1)
And rather than use :func:`torch.Tensor.item` (which converts a Tensor to a Python built-in number):
# Bad! y.item() will be replaced with a constant during tracing.
def forward(self, x, y):
return x.reshape(y.item(), -1)
Use torch's support for implicit casting of single-element tensors:
# Good! y will be preserved as a variable during tracing.
def forward(self, x, y):
return x.reshape(y, -1)
Using the Tensor.data field can produce an incorrect trace and therefore an incorrect ONNX graph. Use :func:`torch.Tensor.detach` instead. (Work is ongoing to remove Tensor.data entirely).
In tracing mode, shapes obtained from tensor.shape are traced as tensors,
and share the same memory. This might cause a mismatch the final output values.
As a workaround, avoid the use of inplace operations in these scenarios.
For example, in the model:
class Model(torch.nn.Module):
def forward(self, states):
batch_size, seq_length = states.shape[:2]
real_seq_length = seq_length
real_seq_length += 2
return real_seq_length + seq_length
real_seq_length and seq_length share the same memory in tracing mode.
This could be avoided by rewriting the inplace operation:
real_seq_length = real_seq_length + 2
- Only :class:`torch.Tensors`, numeric types that can be trivially converted to torch.Tensors (e.g. float, int),
and tuples and lists of those types are supported as model inputs or outputs. Dict and str inputs and
outputs are accepted in :ref:`tracing<tracing-vs-scripting>` mode, but:
- Any computation that depends on the value of a dict or a str input will be replaced with the constant value seen during the one traced execution.
- Any output that is a dict will be silently replaced with a flattened sequence of its values
(keys will be removed). E.g.
{"foo": 1, "bar": 2}becomes(1, 2). - Any output that is a str will be silently removed.
- Certain operations involving tuples and lists are not supported in :ref:`scripting<tracing-vs-scripting>` mode due to limited support in ONNX for nested sequences. In particular appending a tuple to a list is not supported. In tracing mode, the nested sequences will be flattened automatically during the tracing.
Due to differences in implementations of operators, running the exported model on different runtimes may produce different results from each other or from PyTorch. Normally these differences are numerically small, so this should only be a concern if your application is sensitive to these small differences.
Tensor indexing patterns that cannot be exported are listed below.
If you are experiencing issues exporting a model that does not include any of
the unsupported patterns below, please double check that you are exporting with
the latest opset_version.
When indexing into a tensor for reading, the following patterns are not supported:
# Tensor indices that includes negative values. data[torch.tensor([[1, 2], [2, -3]]), torch.tensor([-2, 3])] # Workarounds: use positive index values.
When indexing into a Tensor for writing, the following patterns are not supported:
# Multiple tensor indices if any has rank >= 2 data[torch.tensor([[1, 2], [2, 3]]), torch.tensor([2, 3])] = new_data # Workarounds: use single tensor index with rank >= 2, # or multiple consecutive tensor indices with rank == 1. # Multiple tensor indices that are not consecutive data[torch.tensor([2, 3]), :, torch.tensor([1, 2])] = new_data # Workarounds: transpose `data` such that tensor indices are consecutive. # Tensor indices that includes negative values. data[torch.tensor([1, -2]), torch.tensor([-2, 3])] = new_data # Workarounds: use positive index values. # Implicit broadcasting required for new_data. data[torch.tensor([[0, 2], [1, 1]]), 1:3] = new_data # Workarounds: expand new_data explicitly. # Example: # data shape: [3, 4, 5] # new_data shape: [5] # expected new_data shape after broadcasting: [2, 2, 2, 5]
When exporting a model that includes unsupported operators, you'll see an error message like:
RuntimeError: ONNX export failed: Couldn't export operator foo
When that happens, there are a few things you can do:
- Change the model to not use that operator.
- Create a symbolic function to convert the operator and register it as a custom symbolic function.
- Contribute to PyTorch to add the same symbolic function to :mod:`torch.onnx` itself.
If you decided to implement a symbolic function (we hope you will contribute it back to PyTorch!), here is how you can get started:
A "symbolic function" is a function that decomposes a PyTorch operator into a composition of a series of ONNX operators.
During export, each node (which contains a PyTorch operator) in the TorchScript
graph is visited by the exporter in topological order.
Upon visiting a node, the exporter looks for a registered symbolic functions for
that operator. Symbolic functions are implemented in Python. A symbolic function for
an op named foo would look something like:
def foo(
g,
input_0: torch._C.Value,
input_1: torch._C.Value) -> Union[None, torch._C.Value, List[torch._C.Value]]:
"""
Adds the ONNX operations representing this PyTorch function by updating the
graph g with `g.op()` calls.
Args:
g (Graph): graph to write the ONNX representation into.
input_0 (Value): value representing the variables which contain
the first input for this operator.
input_1 (Value): value representing the variables which contain
the second input for this operator.
Returns:
A Value or List of Values specifying the ONNX nodes that compute something
equivalent to the original PyTorch operator with the given inputs.
None if it cannot be converted to ONNX.
"""
...
The torch._C types are Python wrappers around the types defined in C++ in
ir.h.
The process for adding a symbolic function depends on the type of operator.
ATen is PyTorch's built-in tensor library.
If the operator is an ATen operator (shows up in the TorchScript graph with the prefix
aten::), make sure it is not supported already.
Visit the auto generated :doc:`list of supported TorchScript operators <../onnx_supported_aten_ops>`
for details on which operator are supported in each opset_version.
If the operator is not in the list above:
- Define the symbolic function in
torch/onnx/symbolic_opset<version>.py, for example torch/onnx/symbolic_opset9.py. Make sure the function has the same name as the ATen function, which may be declared intorch/_C/_VariableFunctions.pyiortorch/nn/functional.pyi(these files are generated at build time, so will not appear in your checkout until you build PyTorch). - By default, the first arg is the ONNX graph.
Other arg names must EXACTLY match the names in the
.pyifile, because dispatch is done with keyword arguments. - In the symbolic function, if the operator is in the ONNX standard operator set, we only need to create a node to represent the ONNX operator in the graph. If not, we can compose several standard operators that have the equivalent semantics to the ATen operator.
Here is an example of handling missing symbolic function for the ELU operator.
If we run the following code:
print(
torch.jit.trace(
torch.nn.ELU(), # module
torch.ones(1) # example input
).graph
)
We see something like:
graph(%self : __torch__.torch.nn.modules.activation.___torch_mangle_0.ELU,
%input : Float(1, strides=[1], requires_grad=0, device=cpu)):
%4 : float = prim::Constant[value=1.]()
%5 : int = prim::Constant[value=1]()
%6 : int = prim::Constant[value=1]()
%7 : Float(1, strides=[1], requires_grad=0, device=cpu) = aten::elu(%input, %4, %5, %6)
return (%7)
Since we see aten::elu in the graph, we know this is an ATen operator.
We check the ONNX operator list,
and confirm that Elu is standardized in ONNX.
We find a signature for elu in torch/nn/functional.pyi:
def elu(input: Tensor, alpha: float = ..., inplace: bool = ...) -> Tensor: ...
We add the following lines to symbolic_opset9.py:
def elu(g, input: torch.Value, alpha: torch.Value, inplace: bool = False):
return g.op("Elu", input, alpha_f=alpha)
Now PyTorch is able to export models containing the aten::elu operator!
See the torch/onnx/symbolic_opset*.py files for more examples.
If the operator is a sub-class of :class:`torch.autograd.Function`, there are three ways to export it.
You can add a static method named symbolic to your function class. It should return
ONNX operators that represent the function's behavior in ONNX. For example:
class MyRelu(torch.autograd.Function):
@staticmethod
def forward(ctx, input: torch.Tensor) -> torch.Tensor:
ctx.save_for_backward(input)
return input.clamp(min=0)
@staticmethod
def symbolic(g: torch.Graph, input: torch.Value) -> torch.Value:
return g.op("Clip", input, g.op("Constant", value_t=torch.tensor(0, dtype=torch.float)))
In cases where a static symbolic method is not provided for its subsequent :class:`torch.autograd.Function` or
where a function to register prim::PythonOp as custom symbolic functions is not provided,
:func:`torch.onnx.export` tries to inline the graph that corresponds to that :class:`torch.autograd.Function` such that
this function is broken down into individual operators that were used within the function.
The export should be successful as long as these individual operators are supported. For example:
class MyLogExp(torch.autograd.Function):
@staticmethod
def forward(ctx, input: torch.Tensor) -> torch.Tensor:
ctx.save_for_backward(input)
h = input.exp()
return h.log().log()
There is no static symbolic method present for this model, yet it is exported as follows:
graph(%input : Float(1, strides=[1], requires_grad=0, device=cpu)):
%1 : float = onnx::Exp[](%input)
%2 : float = onnx::Log[](%1)
%3 : float = onnx::Log[](%2)
return (%3)
If you need to avoid inlining of :class:`torch.autograd.Function`, you should export models with
operator_export_type set to ONNX_FALLTHROUGH or ONNX_ATEN_FALLBACK.
If a model uses a custom operator implemented in C++ as described in Extending TorchScript with Custom C++ Operators, you can export it by following this example:
from torch.onnx import symbolic_helper
# Define custom symbolic function
@symbolic_helper.parse_args("v", "v", "f", "i")
def symbolic_foo_forward(g, input1, input2, attr1, attr2):
return g.op("custom_domain::Foo", input1, input2, attr1_f=attr1, attr2_i=attr2)
# Register custom symbolic function
torch.onnx.register_custom_op_symbolic("custom_ops::foo_forward", symbolic_foo_forward, 9)
class FooModel(torch.nn.Module):
def __init__(self, attr1, attr2):
super().__init__()
self.attr1 = attr1
self.attr2 = attr2
def forward(self, input1, input2):
# Calling custom op
return torch.ops.custom_ops.foo_forward(input1, input2, self.attr1, self.attr2)
model = FooModel(attr1, attr2)
torch.onnx.export(
model,
(example_input1, example_input1),
"model.onnx",
# only needed if you want to specify an opset version > 1.
custom_opsets={"custom_domain": 2}
)
You can export your model as one or a combination of many standard ONNX ops, or as a custom ONNX operator.
The example above exports it as a custom operator in the "custom_domain" opset.
When exporting a custom operator, you can specify the custom domain version using the
custom_opsets dictionary at export. If not specified, the custom opset version defaults to 1.
The runtime that consumes the model needs to support the custom op. See Caffe2 custom ops, ONNX Runtime custom ops, or your runtime of choice's documentation.
When export fails due to an unconvertible ATen op, there may in fact be more than one such op but the error message only mentions the first. To discover all of the unconvertible ops in one go you can:
# prepare model, args, opset_version
...
torch_script_graph, unconvertible_ops = torch.onnx.utils.unconvertible_ops(
model, args, opset_version=opset_version
)
print(set(unconvertible_ops))
The set is approximated because some ops may be removed during the conversion process and don't need to be converted. Some other ops may have partial support that will fail conversion with particular inputs, but this should give you a general idea of what ops are not supported. Please feel free to open Github Issues for op support requests.
Q: I have exported my LSTM model, but its input size seems to be fixed?
The tracer records the shapes of the example inputs. If the model should accept
inputs of dynamic shapes, set dynamic_axes when calling :func:`torch.onnx.export`.
Q: How to export models containing loops?
See Tracing vs Scripting.
Q: How to export models with primitive type inputs (e.g. int, float)?
Support for primitive numeric type inputs was added in PyTorch 1.9. However, the exporter does not support models with str inputs.
Q: Does ONNX support implicit scalar datatype casting?
The ONNX standard does not, but the exporter will try to handle that part. Scalars are exported as constant tensors. The exporter will figure out the right data type for scalars. In rare cases when it is unable to do so, you will need to manually specify the datatype with e.g. dtype=torch.float32. If you see any errors, please [create a GitHub issue](https://github.com/pytorch/pytorch/issues).
Q: Are lists of Tensors exportable to ONNX?
Yes, for opset_version >= 11, since ONNX introduced the Sequence type in opset 11.
.. autofunction:: export
.. autofunction:: export_to_pretty_string
.. autofunction:: register_custom_op_symbolic
.. autofunction:: unregister_custom_op_symbolic
.. autofunction:: select_model_mode_for_export
.. autofunction:: is_in_onnx_export
.. autofunction:: enable_log
.. autofunction:: disable_log
.. autosummary::
:toctree: generated
:nosignatures:
:template: classtemplate.rst
JitScalarType