PyTorch Functions and Modules¶

TorchScript supports a subset of the tensor and neural network functions that PyTorch provides. Most methods on Tensor as well as functions in the torch namespace, all functions in torch.nn.functional and most modules from torch.nn are supported in TorchScript.

See TorchScript Unsupported Pytorch Constructs for a list of unsupported PyTorch functions and modules.

Python Functions and Modules¶

Many of Python’s built-in functions are supported in TorchScript. The math module is also supported (see math Module for details), but no other Python modules (built-in or third party) are supported.

Python Language Reference Comparison¶

For a full listing of supported Python features, see Python Language Reference Coverage.

Disable JIT for Debugging¶

PYTORCH_JIT¶

Setting the environment variable PYTORCH_JIT=0 will disable all script and tracing annotations. If there is hard-to-debug error in one of your TorchScript models, you can use this flag to force everything to run using native Python. Since TorchScript (scripting and tracing) is disabled with this flag, you can use tools like pdb to debug the model code. For example:

@torch.jit.script
def scripted_fn(x : torch.Tensor):
    for i in range(12):
        x = x + x
    return x

def fn(x):
    x = torch.neg(x)
    import pdb; pdb.set_trace()
    return scripted_fn(x)

traced_fn = torch.jit.trace(fn, (torch.rand(4, 5),))
traced_fn(torch.rand(3, 4))

Debugging this script with pdb works except for when we invoke the @torch.jit.script function. We can globally disable JIT, so that we can call the @torch.jit.script function as a normal Python function and not compile it. If the above script is called disable_jit_example.py, we can invoke it like so:

$ PYTORCH_JIT=0 python disable_jit_example.py

and we will be able to step into the @torch.jit.script function as a normal Python function. To disable the TorchScript compiler for a specific function, see @torch.jit.ignore.

Inspecting Code¶

TorchScript provides a code pretty-printer for all ScriptModule instances. This pretty-printer gives an interpretation of the script method’s code as valid Python syntax. For example:

@torch.jit.script
def foo(len):
    # type: (int) -> torch.Tensor
    rv = torch.zeros(3, 4)
    for i in range(len):
        if i < 10:
            rv = rv - 1.0
        else:
            rv = rv + 1.0
    return rv

print(foo.code)

A ScriptModule with a single forward method will have an attribute code, which you can use to inspect the ScriptModule’s code. If the ScriptModule has more than one method, you will need to access .code on the method itself and not the module. We can inspect the code of a method named foo on a ScriptModule by accessing .foo.code. The example above produces this output:

def foo(len: int) -> Tensor:
    rv = torch.zeros([3, 4], dtype=None, layout=None, device=None, pin_memory=None)
    rv0 = rv
    for i in range(len):
        if torch.lt(i, 10):
            rv1 = torch.sub(rv0, 1., 1)
        else:
            rv1 = torch.add(rv0, 1., 1)
        rv0 = rv1
    return rv0

This is TorchScript’s compilation of the code for the forward method. You can use this to ensure TorchScript (tracing or scripting) has captured your model code correctly.

Interpreting Graphs¶

TorchScript also has a representation at a lower level than the code pretty- printer, in the form of IR graphs.

TorchScript uses a static single assignment (SSA) intermediate representation (IR) to represent computation. The instructions in this format consist of ATen (the C++ backend of PyTorch) operators and other primitive operators, including control flow operators for loops and conditionals. As an example:

@torch.jit.script
def foo(len):
    # type: (int) -> torch.Tensor
    rv = torch.zeros(3, 4)
    for i in range(len):
        if i < 10:
            rv = rv - 1.0
        else:
            rv = rv + 1.0
    return rv

print(foo.graph)

graph follows the same rules described in the Inspecting Code section with regard to forward method lookup.

The example script above produces the graph:

graph(%len.1 : int):
  %24 : int = prim::Constant[value=1]()
  %17 : bool = prim::Constant[value=1]() # test.py:10:5
  %12 : bool? = prim::Constant()
  %10 : Device? = prim::Constant()
  %6 : int? = prim::Constant()
  %1 : int = prim::Constant[value=3]() # test.py:9:22
  %2 : int = prim::Constant[value=4]() # test.py:9:25
  %20 : int = prim::Constant[value=10]() # test.py:11:16
  %23 : float = prim::Constant[value=1]() # test.py:12:23
  %4 : int[] = prim::ListConstruct(%1, %2)
  %rv.1 : Tensor = aten::zeros(%4, %6, %6, %10, %12) # test.py:9:10
  %rv : Tensor = prim::Loop(%len.1, %17, %rv.1) # test.py:10:5
    block0(%i.1 : int, %rv.14 : Tensor):
      %21 : bool = aten::lt(%i.1, %20) # test.py:11:12
      %rv.13 : Tensor = prim::If(%21) # test.py:11:9
        block0():
          %rv.3 : Tensor = aten::sub(%rv.14, %23, %24) # test.py:12:18
          -> (%rv.3)
        block1():
          %rv.6 : Tensor = aten::add(%rv.14, %23, %24) # test.py:14:18
          -> (%rv.6)
      -> (%17, %rv.13)
  return (%rv)

Take the instruction %rv.1 : Tensor = aten::zeros(%4, %6, %6, %10, %12) # test.py:9:10 for example.

• %rv.1 : Tensor means we assign the output to a (unique) value named rv.1, that value is of Tensor type and that we do not know its concrete shape.

• aten::zeros is the operator (equivalent to torch.zeros) and the input list (%4, %6, %6, %10, %12) specifies which values in scope should be passed as inputs. The schema for built-in functions like aten::zeros can be found at Builtin Functions.

• # test.py:9:10 is the location in the original source file that generated this instruction. In this case, it is a file named test.py, on line 9, and at character 10.

Notice that operators can also have associated blocks, namely the prim::Loop and prim::If operators. In the graph print-out, these operators are formatted to reflect their equivalent source code forms to facilitate easy debugging.

Graphs can be inspected as shown to confirm that the computation described by a ScriptModule is correct, in both automated and manual fashion, as described below.

Tracer¶

def loop_in_traced_fn(x):
    result = x[0]
    for i in range(x.size(0)):
        result = result * x[i]
    return result

inputs = (torch.rand(3, 4, 5),)
check_inputs = [(torch.rand(4, 5, 6),), (torch.rand(2, 3, 4),)]

traced = torch.jit.trace(loop_in_traced_fn, inputs, check_inputs=check_inputs)

ERROR: Graphs differed across invocations!
Graph diff:

            graph(%x : Tensor) {
            %1 : int = prim::Constant[value=0]()
            %2 : int = prim::Constant[value=0]()
            %result.1 : Tensor = aten::select(%x, %1, %2)
            %4 : int = prim::Constant[value=0]()
            %5 : int = prim::Constant[value=0]()
            %6 : Tensor = aten::select(%x, %4, %5)
            %result.2 : Tensor = aten::mul(%result.1, %6)
            %8 : int = prim::Constant[value=0]()
            %9 : int = prim::Constant[value=1]()
            %10 : Tensor = aten::select(%x, %8, %9)
        -   %result : Tensor = aten::mul(%result.2, %10)
        +   %result.3 : Tensor = aten::mul(%result.2, %10)
        ?          ++
            %12 : int = prim::Constant[value=0]()
            %13 : int = prim::Constant[value=2]()
            %14 : Tensor = aten::select(%x, %12, %13)
        +   %result : Tensor = aten::mul(%result.3, %14)
        +   %16 : int = prim::Constant[value=0]()
        +   %17 : int = prim::Constant[value=3]()
        +   %18 : Tensor = aten::select(%x, %16, %17)
        -   %15 : Tensor = aten::mul(%result, %14)
        ?     ^                                 ^
        +   %19 : Tensor = aten::mul(%result, %18)
        ?     ^                                 ^
        -   return (%15);
        ?             ^
        +   return (%19);
        ?             ^
            }

def fn(x):
    result = x[0]
    for i in range(x.size(0)):
        result = result * x[i]
    return result

inputs = (torch.rand(3, 4, 5),)
check_inputs = [(torch.rand(4, 5, 6),), (torch.rand(2, 3, 4),)]

scripted_fn = torch.jit.script(fn)
print(scripted_fn.graph)
#print(str(scripted_fn.graph).strip())

for input_tuple in [inputs] + check_inputs:
    torch.testing.assert_allclose(fn(*input_tuple), scripted_fn(*input_tuple))

graph(%x : Tensor) {
    %5 : bool = prim::Constant[value=1]()
    %1 : int = prim::Constant[value=0]()
    %result.1 : Tensor = aten::select(%x, %1, %1)
    %4 : int = aten::size(%x, %1)
    %result : Tensor = prim::Loop(%4, %5, %result.1)
    block0(%i : int, %7 : Tensor) {
        %10 : Tensor = aten::select(%x, %1, %i)
        %result.2 : Tensor = aten::mul(%7, %10)
        -> (%5, %result.2)
    }
    return (%result);
}

def fill_row_zero(x):
    x[0] = torch.rand(*x.shape[1:2])
    return x

traced = torch.jit.trace(fill_row_zero, (torch.rand(3, 4),))
print(traced.graph)

fill_row_zero.py:4: TracerWarning: There are 2 live references to the data region being modified when tracing in-place operator copy_ (possibly due to an assignment). This might cause the trace to be incorrect, because all other views that also reference this data will not reflect this change in the trace! On the other hand, if all other views use the same memory chunk, but are disjoint (e.g. are outputs of torch.split), this might still be safe.
    x[0] = torch.rand(*x.shape[1:2])
fill_row_zero.py:6: TracerWarning: Output nr 1. of the traced function does not match the corresponding output of the Python function. Detailed error:
Not within tolerance rtol=1e-05 atol=1e-05 at input[0, 1] (0.09115803241729736 vs. 0.6782537698745728) and 3 other locations (33.00%)
    traced = torch.jit.trace(fill_row_zero, (torch.rand(3, 4),))
graph(%0 : Float(3, 4)) {
    return (%0);
}

def fill_row_zero(x):
    x = torch.cat((torch.rand(1, *x.shape[1:2]), x[1:2]), dim=0)
    return x

traced = torch.jit.trace(fill_row_zero, (torch.rand(3, 4),))
print(traced.graph)

Migrating to PyTorch 1.2 Recursive Scripting API¶

This section details the changes to TorchScript in PyTorch 1.2. If you are new to TorchScript you can skip this section. There are two main changes to the TorchScript API with PyTorch 1.2.

1. torch.jit.script will now attempt to recursively compile functions, methods, and classes that it encounters. Once you call torch.jit.script, compilation is “opt-out”, rather than “opt-in”.

2. torch.jit.script(nn_module_instance) is now the preferred way to create ScriptModules, instead of inheriting from torch.jit.ScriptModule. These changes combine to provide a simpler, easier-to-use API for converting your nn.Modules into ScriptModules, ready to be optimized and executed in a non-Python environment.

The new usage looks like this:

import torch
import torch.nn as nn
import torch.nn.functional as F

class Model(nn.Module):
    def __init__(self):
        super(Model, self).__init__()
        self.conv1 = nn.Conv2d(1, 20, 5)
        self.conv2 = nn.Conv2d(20, 20, 5)

    def forward(self, x):
        x = F.relu(self.conv1(x))
        return F.relu(self.conv2(x))

my_model = Model()
my_scripted_model = torch.jit.script(my_model)

• The module’s forward is compiled by default. Methods called from forward are lazily compiled in the order they are used in forward.

• To compile a method other than forward that is not called from forward, add @torch.jit.export.

• To stop the compiler from compiling a method, add @torch.jit.ignore or @torch.jit.unused. @ignore leaves the

• method as a call to python, and @unused replaces it with an exception. @ignored cannot be exported; @unused can.

• Most attribute types can be inferred, so torch.jit.Attribute is not necessary. For empty container types, annotate their types using PEP 526-style class annotations.

• Constants can be marked with a Final class annotation instead of adding the name of the member to __constants__.

• Python 3 type hints can be used in place of torch.jit.annotate

As a result of these changes, the following items are considered deprecated and should not appear in new code:

• The @torch.jit.script_method decorator

• Classes that inherit from torch.jit.ScriptModule

• The torch.jit.Attribute wrapper class

• The __constants__ array

• The torch.jit.annotate function

# Same behavior as pre-PyTorch 1.2
@torch.jit.script
def some_fn():
    return 2

# Marks a function as ignored, if nothing
# ever calls it then this has no effect
@torch.jit.ignore
def some_fn2():
    return 2

# As with ignore, if nothing calls it then it has no effect.
# If it is called in script it is replaced with an exception.
@torch.jit.unused
def some_fn3():
  import pdb; pdb.set_trace()
  return 4

# Doesn't do anything, this function is already
# the main entry point
@torch.jit.export
def some_fn4():
    return 2

from typing import Dict
import torch

class MyModule(torch.jit.ScriptModule):
    def __init__(self):
        super(MyModule, self).__init__()
        self.my_dict = torch.jit.Attribute({}, Dict[str, int])
        self.my_int = torch.jit.Attribute(20, int)

m = MyModule()

from typing import Dict

class MyModule(torch.nn.Module):
    my_dict: Dict[str, int]

    def __init__(self):
        super(MyModule, self).__init__()
        # This type cannot be inferred and must be specified
        self.my_dict = {}

        # The attribute type here is inferred to be `int`
        self.my_int = 20

    def forward(self):
        pass

m = torch.jit.script(MyModule())

class MyModule(torch.jit.ScriptModule):
    __constants__ = ['my_constant']

    def __init__(self):
        super(MyModule, self).__init__()
        self.my_constant = 2

    def forward(self):
        pass
m = MyModule()

try:
    from typing_extensions import Final
except:
    # If you don't have `typing_extensions` installed, you can use a
    # polyfill from `torch.jit`.
    from torch.jit import Final

class MyModule(torch.nn.Module):

    my_constant: Final[int]

    def __init__(self):
        super(MyModule, self).__init__()
        self.my_constant = 2

    def forward(self):
        pass

m = torch.jit.script(MyModule())

import torch
from typing import Dict, Optional

@torch.jit.script
def make_dict(flag: bool):
    x: Dict[str, int] = {}
    x['hi'] = 2
    b: Optional[int] = None
    if flag:
        b = 2
    return x, b

References¶