to_onnx and a custom operator registered with a function

This example shows how to convert a custom operator, inspired from Python Custom Operators.

A model with a custom ops

from typing import Any, Dict, List
import numpy as np
import torch
from onnx_array_api.plotting.graphviz_helper import plot_dot
from experimental_experiment.xbuilder import GraphBuilder
from experimental_experiment.helpers import pretty_onnx
from experimental_experiment.torch_interpreter import to_onnx, Dispatcher

We define a model with a custom operator.

def numpy_sin(x: torch.Tensor) -> torch.Tensor:
    assert x.device.type == "cpu"
    x_np = x.numpy()
    return torch.from_numpy(np.sin(x_np))


class ModuleWithACustomOperator(torch.nn.Module):
    def forward(self, x):
        return numpy_sin(x)


model = ModuleWithACustomOperator()

Let’s check it runs.

x = torch.randn(1, 3)
model(x)

tensor([[ 0.4876,  0.6273, -0.4967]])

As expected, it does not export.

try:
    torch.export.export(model, (x,))
    raise AssertionError("This export should failed unless pytorch now supports this model.")
except Exception as e:
    print(e)

This export should failed unless pytorch now supports this model.

The exporter fails with the same eror as it expects torch.export.export to work.

try:
    to_onnx(model, (x,))
except Exception as e:
    print(e)

Unable to interpret function <class 'torch._ops.OpOverload'>: <OpOverload(op='aten.view_as', overload='default')>, searched for ['aten::view_as', 'view_as_default'] and attributes ['__qualname__', '__name__'], args=(x, x), kwargs={}
--DEBUG--
[GraphBuilder-YNG] Message starts, there are 0 initializers, 0 nodes, 1 inputs, 1 outputs.
--PARAMETERS--
dynamic_examples=
--SHAPE--
dynamic_examples=
dynamic_objects=
dynamic_objects_rev=
dynamic_dimensions_source={}
dynamic_alias={}
dynamic_shapes=None
_known_value_shape={}
_known_types={'x': 1}
_known_shapes={'x': (1, 3)}
_known_constants=[]
_known_ranks={}
--TORCH-USERS--
view_as -> {sin}
x -> {view_as}
--TORCH-SHAPES--
x: ('run_node', ('', ('val', torch.float32, torch.Size([1, 3])))) --- 1:2:(1, 3):
view_as: ('run_node', ('', ('val', torch.float32, torch.Size([1, 3])))) --- :::
--ONNX--
-- process.graph_module --
graph():
    %x : [num_users=1] = placeholder[target=x]
    %view_as : [num_users=1] = call_function[target=torch.ops.aten.view_as.default](args = (%x, %x), kwargs = {})
    %sin : [num_users=1] = call_function[target=torch.ops.aten.sin.default](args = (%view_as,), kwargs = {})
    %to : [num_users=1] = call_function[target=torch.ops.aten.to.device](args = (%sin, cpu, torch.float32), kwargs = {})
    return (to,)
-- process.progress --
node 1/5 target=aten.view_as.default
--
[GraphBuilder-YNG.make_tensor_input] x[1:1x3]
[GraphBuilder-YNG] Message completed, there are 0 initializers, 0 nodes, 1 inputs, 1 outputs.

Registration

The exporter how to convert the new exporter into ONNX. This must be defined. The first piece is to tell the exporter that the shape of the output is the same as x. input names must be the same.

def register(fct, fct_shape, namespace, fname):
    schema_str = torch.library.infer_schema(fct, mutates_args=())
    custom_def = torch.library.CustomOpDef(namespace, fname, schema_str, fct)
    custom_def.register_kernel("cpu")(fct)
    custom_def._abstract_fn = fct_shape


register(numpy_sin, lambda x: torch.empty_like(x), "mylib", "numpy_sin")


class ModuleWithACustomOperator(torch.nn.Module):
    def forward(self, x):
        return torch.ops.mylib.numpy_sin(x)


model = ModuleWithACustomOperator()

Let’s check it runs again.

model(x)

tensor([[ 0.4876,  0.6273, -0.4967]])

Let’s see what the fx graph looks like.

print(torch.export.export(model, (x,)).graph)

graph():
    %x : [num_users=1] = placeholder[target=x]
    %numpy_sin : [num_users=1] = call_function[target=torch.ops.mylib.numpy_sin.default](args = (%x,), kwargs = {})
    return (numpy_sin,)

Next is the conversion to onnx.

T = str  # a tensor name


def numpy_sin_to_onnx(
    g: GraphBuilder,
    sts: Dict[str, Any],
    outputs: List[str],
    x: T,
    name: str = "mylib.numpy_sin",
) -> T:
    # name= ... lets the user know when the node comes from
    return g.op.Sin(x, name=name, outputs=outputs)

We create a Dispatcher.

dispatcher = Dispatcher({"mylib::numpy_sin": numpy_sin_to_onnx})

And we convert again.

onx = to_onnx(model, (x,), dispatcher=dispatcher, optimize=False)
print(pretty_onnx(onx))

opset: domain='' version=18
doc_string: large_model=False, inline=False, external_threshold=102...
input: name='x' type=dtype('float32') shape=[1, 3]
Sin(x) -> numpy_sin
  Identity(numpy_sin) -> output_0
output: name='output_0' type=dtype('float32') shape=[1, 3]

And we convert again with optimization this time.

onx = to_onnx(model, (x,), dispatcher=dispatcher, optimize=True)
print(pretty_onnx(onx))

opset: domain='' version=18
doc_string: large_model=False, inline=False, external_threshold=102...
input: name='x' type=dtype('float32') shape=[1, 3]
Sin(x) -> output_0
output: name='output_0' type=dtype('float32') shape=[1, 3]

Let’s make sure the node was produce was the user defined converter for numpy_sin. The name should be ‘mylib.numpy_sin’.

print(onx.graph.node[0])

input: "x"
output: "output_0"
name: "mylib.numpy_sin"
op_type: "Sin"
domain: ""

And visually.

plot_dot(onx)

Related examples

to_onnx and a custom operator inplace

to_onnx and a custom operator inplace

to_onnx and a model with a test

to_onnx and a model with a test

Export a model using a custom type as input

Export a model using a custom type as input