python.data-structure

dictionary

Note

Tags: python.data-structure

Support Level: SUPPORTED

Original source code:

import torch



def dictionary(x, y):
    """
    Dictionary structures are inlined and flattened along tracing.
    """
    elements = {}
    elements["x2"] = x * x
    y = y * elements["x2"]
    return {"y": y}

Result:

ExportedProgram:
    class GraphModule(torch.nn.Module):
        def forward(self, arg0_1: "f32[3, 2]", arg1_1: "i64[]"):
                mul: "f32[3, 2]" = torch.ops.aten.mul.Tensor(arg0_1, arg0_1);  arg0_1 = None

                mul_1: "f32[3, 2]" = torch.ops.aten.mul.Tensor(arg1_1, mul);  arg1_1 = mul = None
            return (mul_1,)

Graph signature: ExportGraphSignature(input_specs=[InputSpec(kind=<InputKind.USER_INPUT: 1>, arg=TensorArgument(name='arg0_1'), target=None), InputSpec(kind=<InputKind.USER_INPUT: 1>, arg=TensorArgument(name='arg1_1'), target=None)], output_specs=[OutputSpec(kind=<OutputKind.USER_OUTPUT: 1>, arg=TensorArgument(name='mul_1'), target=None)])
Range constraints: {}
Equality constraints: []

fn_with_kwargs

Note

Tags: python.data-structure

Support Level: SUPPORTED

Original source code:

import torch



def fn_with_kwargs(pos0, tuple0, *myargs, mykw0, **mykwargs):
    """
    Keyword arguments are not supported at the moment.
    """
    out = pos0
    for arg in tuple0:
        out = out * arg
    for arg in myargs:
        out = out * arg
    out = out * mykw0
    out = out * mykwargs["input0"] * mykwargs["input1"]
    return out

Result:

ExportedProgram:
    class GraphModule(torch.nn.Module):
        def forward(self, arg0_1: "f32[4]", arg1_1: "f32[4]", arg2_1: "f32[4]", arg3_1: "f32[4]", arg4_1: "f32[4]", arg5_1: "f32[4]", arg6_1: "f32[4]", arg7_1: "f32[4]"):
                mul: "f32[4]" = torch.ops.aten.mul.Tensor(arg0_1, arg1_1);  arg0_1 = arg1_1 = None
            mul_1: "f32[4]" = torch.ops.aten.mul.Tensor(mul, arg2_1);  mul = arg2_1 = None

                mul_2: "f32[4]" = torch.ops.aten.mul.Tensor(mul_1, arg3_1);  mul_1 = arg3_1 = None
            mul_3: "f32[4]" = torch.ops.aten.mul.Tensor(mul_2, arg4_1);  mul_2 = arg4_1 = None

                mul_4: "f32[4]" = torch.ops.aten.mul.Tensor(mul_3, arg5_1);  mul_3 = arg5_1 = None

                mul_5: "f32[4]" = torch.ops.aten.mul.Tensor(mul_4, arg6_1);  mul_4 = arg6_1 = None
            mul_6: "f32[4]" = torch.ops.aten.mul.Tensor(mul_5, arg7_1);  mul_5 = arg7_1 = None
            return (mul_6,)

Graph signature: ExportGraphSignature(input_specs=[InputSpec(kind=<InputKind.USER_INPUT: 1>, arg=TensorArgument(name='arg0_1'), target=None), InputSpec(kind=<InputKind.USER_INPUT: 1>, arg=TensorArgument(name='arg1_1'), target=None), InputSpec(kind=<InputKind.USER_INPUT: 1>, arg=TensorArgument(name='arg2_1'), target=None), InputSpec(kind=<InputKind.USER_INPUT: 1>, arg=TensorArgument(name='arg3_1'), target=None), InputSpec(kind=<InputKind.USER_INPUT: 1>, arg=TensorArgument(name='arg4_1'), target=None), InputSpec(kind=<InputKind.USER_INPUT: 1>, arg=TensorArgument(name='arg5_1'), target=None), InputSpec(kind=<InputKind.USER_INPUT: 1>, arg=TensorArgument(name='arg6_1'), target=None), InputSpec(kind=<InputKind.USER_INPUT: 1>, arg=TensorArgument(name='arg7_1'), target=None)], output_specs=[OutputSpec(kind=<OutputKind.USER_OUTPUT: 1>, arg=TensorArgument(name='mul_6'), target=None)])
Range constraints: {}
Equality constraints: []

list_contains

Note

Tags: python.data-structure, python.assert, torch.dynamic-shape

Support Level: SUPPORTED

Original source code:

import torch



def list_contains(x):
    """
    List containment relation can be checked on a dynamic shape or constants.
    """
    assert x.size(-1) in [6, 2]
    assert x.size(0) not in [4, 5, 6]
    assert "monkey" not in ["cow", "pig"]
    return x + x

Result:

ExportedProgram:
    class GraphModule(torch.nn.Module):
        def forward(self, arg0_1: "f32[3, 2]"):
                add: "f32[3, 2]" = torch.ops.aten.add.Tensor(arg0_1, arg0_1);  arg0_1 = None
            return (add,)

Graph signature: ExportGraphSignature(input_specs=[InputSpec(kind=<InputKind.USER_INPUT: 1>, arg=TensorArgument(name='arg0_1'), target=None)], output_specs=[OutputSpec(kind=<OutputKind.USER_OUTPUT: 1>, arg=TensorArgument(name='add'), target=None)])
Range constraints: {}
Equality constraints: []

list_unpack

Note

Tags: python.data-structure, python.control-flow

Support Level: SUPPORTED

Original source code:

from typing import List

import torch



def list_unpack(args: List[torch.Tensor]):
    """
    Lists are treated as static construct, therefore unpacking should be
    erased after tracing.
    """
    x, *y = args
    return x + y[0]

Result:

ExportedProgram:
    class GraphModule(torch.nn.Module):
        def forward(self, arg0_1: "f32[3, 2]", arg1_1: "i64[]", arg2_1: "i64[]"):
                add: "f32[3, 2]" = torch.ops.aten.add.Tensor(arg0_1, arg1_1);  arg0_1 = arg1_1 = None
            return (add,)

Graph signature: ExportGraphSignature(input_specs=[InputSpec(kind=<InputKind.USER_INPUT: 1>, arg=TensorArgument(name='arg0_1'), target=None), InputSpec(kind=<InputKind.USER_INPUT: 1>, arg=TensorArgument(name='arg1_1'), target=None), InputSpec(kind=<InputKind.USER_INPUT: 1>, arg=TensorArgument(name='arg2_1'), target=None)], output_specs=[OutputSpec(kind=<OutputKind.USER_OUTPUT: 1>, arg=TensorArgument(name='add'), target=None)])
Range constraints: {}
Equality constraints: []