Note
Click here to download the full example code
torch.export Tutorial
Created On: Oct 02, 2023 | Last Updated: Jan 27, 2025 | Last Verified: Nov 05, 2024
Author: William Wen, Zhengxu Chen, Angela Yi, Pian Pawakapan
Warning
torch.export and its related features are in prototype status and are subject to backwards compatibility
breaking changes. This tutorial provides a snapshot of torch.export usage as of PyTorch 2.5.
torch.export() is the PyTorch 2.X way to export PyTorch models into
standardized model representations, intended
to be run on different (i.e. Python-less) environments. The official
documentation can be found here.
In this tutorial, you will learn how to use torch.export() to extract
ExportedProgram’s (i.e. single-graph representations) from PyTorch programs.
We also detail some considerations/modifications that you may need
to make in order to make your model compatible with torch.export.
Contents
Basic Usage
torch.export extracts single-graph representations from PyTorch programs
by tracing the target function, given example inputs.
torch.export.export() is the main entry point for torch.export.
In this tutorial, torch.export and torch.export.export() are practically synonymous,
though torch.export generally refers to the PyTorch 2.X export process, and torch.export.export()
generally refers to the actual function call.
The signature of torch.export.export() is:
export(
mod: torch.nn.Module,
args: Tuple[Any, ...],
kwargs: Optional[Dict[str, Any]] = None,
*,
dynamic_shapes: Optional[Dict[str, Dict[int, Dim]]] = None
) -> ExportedProgram
torch.export.export() traces the tensor computation graph from calling mod(*args, **kwargs)
and wraps it in an ExportedProgram, which can be serialized or executed later with
different inputs. To execute the ExportedProgram we can call .module()
on it to return a torch.nn.Module which is callable, just like the
original program.
We will detail the dynamic_shapes argument later in the tutorial.
import torch
from torch.export import export
class MyModule(torch.nn.Module):
def __init__(self):
super().__init__()
self.lin = torch.nn.Linear(100, 10)
def forward(self, x, y):
return torch.nn.functional.relu(self.lin(x + y), inplace=True)
mod = MyModule()
exported_mod = export(mod, (torch.randn(8, 100), torch.randn(8, 100)))
print(type(exported_mod))
print(exported_mod.module()(torch.randn(8, 100), torch.randn(8, 100)))
/usr/local/lib/python3.10/dist-packages/torch/_dynamo/pgo.py:465: UserWarning:
dynamo_pgo force disabled by torch._inductor.config.force_disable_caches
<class 'torch.export.exported_program.ExportedProgram'>
tensor([[0.8632, 0.8407, 0.0407, 0.0000, 0.4132, 0.0000, 0.0000, 0.1538, 0.6111,
0.0000],
[0.0000, 0.0000, 0.0273, 0.8057, 0.0000, 1.0162, 0.8042, 0.0000, 0.2660,
0.0000],
[0.9481, 0.1396, 1.0225, 0.9563, 0.5832, 0.2546, 0.4095, 0.4591, 0.0000,
2.0053],
[1.1300, 0.4873, 0.0000, 0.9663, 1.2275, 1.4015, 0.0000, 0.9444, 0.0000,
0.0000],
[0.0000, 0.8724, 1.1648, 0.6867, 0.0000, 0.2833, 0.3202, 0.5848, 0.0000,
0.0833],
[1.1311, 0.1324, 0.0000, 1.7842, 0.0000, 0.3474, 0.9916, 0.3571, 0.0000,
0.0000],
[1.4348, 1.0570, 0.1771, 0.0000, 0.9510, 0.0000, 0.0000, 0.0000, 0.2618,
0.0000],
[0.8853, 0.0000, 0.0000, 0.4486, 0.0000, 0.0000, 0.5841, 0.7604, 0.0000,
0.0000]], grad_fn=<ReluBackward0>)
Let’s review some attributes of ExportedProgram that are of interest.
The graph attribute is an FX graph
traced from the function we exported, that is, the computation graph of all PyTorch operations.
The FX graph is in “ATen IR” meaning that it contains only “ATen-level” operations.
The graph_signature attribute gives a more detailed description of the
input and output nodes in the exported graph, describing which ones are
parameters, buffers, user inputs, or user outputs.
The range_constraints attributes will be covered later.
print(exported_mod)
ExportedProgram:
class GraphModule(torch.nn.Module):
def forward(self, p_lin_weight: "f32[10, 100]", p_lin_bias: "f32[10]", x: "f32[8, 100]", y: "f32[8, 100]"):
# File: /var/lib/workspace/intermediate_source/torch_export_tutorial.py:71 in forward, code: return torch.nn.functional.relu(self.lin(x + y), inplace=True)
add: "f32[8, 100]" = torch.ops.aten.add.Tensor(x, y); x = y = None
linear: "f32[8, 10]" = torch.ops.aten.linear.default(add, p_lin_weight, p_lin_bias); add = p_lin_weight = p_lin_bias = None
relu_: "f32[8, 10]" = torch.ops.aten.relu_.default(linear); linear = None
return (relu_,)
Graph signature: ExportGraphSignature(input_specs=[InputSpec(kind=<InputKind.PARAMETER: 2>, arg=TensorArgument(name='p_lin_weight'), target='lin.weight', persistent=None), InputSpec(kind=<InputKind.PARAMETER: 2>, arg=TensorArgument(name='p_lin_bias'), target='lin.bias', persistent=None), InputSpec(kind=<InputKind.USER_INPUT: 1>, arg=TensorArgument(name='x'), target=None, persistent=None), InputSpec(kind=<InputKind.USER_INPUT: 1>, arg=TensorArgument(name='y'), target=None, persistent=None)], output_specs=[OutputSpec(kind=<OutputKind.USER_OUTPUT: 1>, arg=TensorArgument(name='relu_'), target=None)])
Range constraints: {}
See the torch.export documentation
for more details.
Graph Breaks
Although torch.export shares components with torch.compile,
the key limitation of torch.export, especially when compared to
torch.compile, is that it does not support graph breaks. This is because
handling graph breaks involves interpreting the unsupported operation with
default Python evaluation, which is incompatible with the export use case.
Therefore, in order to make your model code compatible with torch.export,
you will need to modify your code to remove graph breaks.
A graph break is necessary in cases such as:
data-dependent control flow
class Bad1(torch.nn.Module):
def forward(self, x):
if x.sum() > 0:
return torch.sin(x)
return torch.cos(x)
import traceback as tb
try:
export(Bad1(), (torch.randn(3, 3),))
except Exception:
tb.print_exc()
class GraphModule(torch.nn.Module):
def forward(self, L_x_: "f32[3, 3][3, 1]cpu"):
l_x_ = L_x_
# File: /var/lib/workspace/intermediate_source/torch_export_tutorial.py:116 in forward, code: if x.sum() > 0:
sum_1: "f32[][]cpu" = l_x_.sum(); l_x_ = None
gt: "b8[][]cpu" = sum_1 > 0; sum_1 = gt = None
Traceback (most recent call last):
File "/var/lib/workspace/intermediate_source/torch_export_tutorial.py", line 122, in <module>
export(Bad1(), (torch.randn(3, 3),))
File "/usr/local/lib/python3.10/dist-packages/torch/export/__init__.py", line 360, in export
return _export(
File "/usr/local/lib/python3.10/dist-packages/torch/export/_trace.py", line 1092, in wrapper
raise e
File "/usr/local/lib/python3.10/dist-packages/torch/export/_trace.py", line 1065, in wrapper
ep = fn(*args, **kwargs)
File "/usr/local/lib/python3.10/dist-packages/torch/export/exported_program.py", line 121, in wrapper
return fn(*args, **kwargs)
File "/usr/local/lib/python3.10/dist-packages/torch/export/_trace.py", line 2112, in _export
ep = _export_for_training(
File "/usr/local/lib/python3.10/dist-packages/torch/export/_trace.py", line 1092, in wrapper
raise e
File "/usr/local/lib/python3.10/dist-packages/torch/export/_trace.py", line 1065, in wrapper
ep = fn(*args, **kwargs)
File "/usr/local/lib/python3.10/dist-packages/torch/export/exported_program.py", line 121, in wrapper
return fn(*args, **kwargs)
File "/usr/local/lib/python3.10/dist-packages/torch/export/_trace.py", line 1975, in _export_for_training
export_artifact = export_func( # type: ignore[operator]
File "/usr/local/lib/python3.10/dist-packages/torch/export/_trace.py", line 1344, in _strict_export_lower_to_aten_ir
gm_torch_level = _export_to_torch_ir(
File "/usr/local/lib/python3.10/dist-packages/torch/export/_trace.py", line 739, in _export_to_torch_ir
gm_torch_level, _ = torch._dynamo.export(
File "/usr/local/lib/python3.10/dist-packages/torch/_dynamo/eval_frame.py", line 1677, in inner
result_traced = opt_f(*args, **kwargs)
File "/usr/local/lib/python3.10/dist-packages/torch/nn/modules/module.py", line 1751, in _wrapped_call_impl
return self._call_impl(*args, **kwargs)
File "/usr/local/lib/python3.10/dist-packages/torch/nn/modules/module.py", line 1762, in _call_impl
return forward_call(*args, **kwargs)
File "/usr/local/lib/python3.10/dist-packages/torch/_dynamo/eval_frame.py", line 659, in _fn
raise e.with_traceback(None) from None
torch._dynamo.exc.Unsupported: Data-dependent branching
Explanation: Detected data-dependent branching (e.g. `if my_tensor.sum() > 0:`). Dynamo does not support tracing dynamic control flow.
Hint: This graph break is fundamental - it is unlikely that Dynamo will ever be able to trace through your code. Consider finding a workaround.
Hint: Use `torch.cond` to express dynamic control flow.
Developer debug context: attempted to jump with TensorVariable()
from user code:
File "/var/lib/workspace/intermediate_source/torch_export_tutorial.py", line 116, in forward
if x.sum() > 0:
Set TORCHDYNAMO_VERBOSE=1 for the internal stack trace (please do this especially if you're reporting a bug to PyTorch). For even more developer context, set TORCH_LOGS="+dynamo"
accessing tensor data with
.data
class Bad2(torch.nn.Module):
def forward(self, x):
x.data[0, 0] = 3
return x
try:
export(Bad2(), (torch.randn(3, 3),))
except Exception:
tb.print_exc()
calling unsupported functions (such as many built-in functions)
class Bad3(torch.nn.Module):
def forward(self, x):
x = x + 1
return x + id(x)
try:
export(Bad3(), (torch.randn(3, 3),))
except Exception:
tb.print_exc()
Traceback (most recent call last):
File "/var/lib/workspace/intermediate_source/torch_export_tutorial.py", line 148, in <module>
export(Bad3(), (torch.randn(3, 3),))
File "/usr/local/lib/python3.10/dist-packages/torch/export/__init__.py", line 360, in export
return _export(
File "/usr/local/lib/python3.10/dist-packages/torch/export/_trace.py", line 1092, in wrapper
raise e
File "/usr/local/lib/python3.10/dist-packages/torch/export/_trace.py", line 1065, in wrapper
ep = fn(*args, **kwargs)
File "/usr/local/lib/python3.10/dist-packages/torch/export/exported_program.py", line 121, in wrapper
return fn(*args, **kwargs)
File "/usr/local/lib/python3.10/dist-packages/torch/export/_trace.py", line 2112, in _export
ep = _export_for_training(
File "/usr/local/lib/python3.10/dist-packages/torch/export/_trace.py", line 1092, in wrapper
raise e
File "/usr/local/lib/python3.10/dist-packages/torch/export/_trace.py", line 1065, in wrapper
ep = fn(*args, **kwargs)
File "/usr/local/lib/python3.10/dist-packages/torch/export/exported_program.py", line 121, in wrapper
return fn(*args, **kwargs)
File "/usr/local/lib/python3.10/dist-packages/torch/export/_trace.py", line 1975, in _export_for_training
export_artifact = export_func( # type: ignore[operator]
File "/usr/local/lib/python3.10/dist-packages/torch/export/_trace.py", line 1344, in _strict_export_lower_to_aten_ir
gm_torch_level = _export_to_torch_ir(
File "/usr/local/lib/python3.10/dist-packages/torch/export/_trace.py", line 739, in _export_to_torch_ir
gm_torch_level, _ = torch._dynamo.export(
File "/usr/local/lib/python3.10/dist-packages/torch/_dynamo/eval_frame.py", line 1677, in inner
result_traced = opt_f(*args, **kwargs)
File "/usr/local/lib/python3.10/dist-packages/torch/nn/modules/module.py", line 1751, in _wrapped_call_impl
return self._call_impl(*args, **kwargs)
File "/usr/local/lib/python3.10/dist-packages/torch/nn/modules/module.py", line 1762, in _call_impl
return forward_call(*args, **kwargs)
File "/usr/local/lib/python3.10/dist-packages/torch/_dynamo/eval_frame.py", line 659, in _fn
raise e.with_traceback(None) from None
torch._dynamo.exc.Unsupported: call_id not supported for sourceless TensorVariable
from user code:
File "/var/lib/workspace/intermediate_source/torch_export_tutorial.py", line 145, in forward
return x + id(x)
Set TORCHDYNAMO_VERBOSE=1 for the internal stack trace (please do this especially if you're reporting a bug to PyTorch). For even more developer context, set TORCH_LOGS="+dynamo"
Non-Strict Export
To trace the program, torch.export uses TorchDynamo by default, a byte
code analysis engine, to symbolically analyze the Python code and build a
graph based on the results. This analysis allows torch.export to provide
stronger guarantees about safety, but not all Python code is supported,
causing these graph breaks.
To address this issue, in PyTorch 2.3, we introduced a new mode of
exporting called non-strict mode, where we trace through the program using the
Python interpreter executing it exactly as it would in eager mode, allowing us
to skip over unsupported Python features. This is done through adding a
strict=False flag.
Looking at some of the previous examples which resulted in graph breaks:
Calling unsupported functions (such as many built-in functions) traces
through, but in this case, id(x) gets specialized as a constant integer in
the graph. This is because id(x) is not a tensor operation, so the
operation is not recorded in the graph.
class Bad3(torch.nn.Module):
def forward(self, x):
x = x + 1
return x + id(x)
bad3_nonstrict = export(Bad3(), (torch.randn(3, 3),), strict=False)
print(bad3_nonstrict)
print(bad3_nonstrict.module()(torch.ones(3, 3)))
ExportedProgram:
class GraphModule(torch.nn.Module):
def forward(self, x: "f32[3, 3]"):
# File: /var/lib/workspace/intermediate_source/torch_export_tutorial.py:179 in forward, code: x = x + 1
add: "f32[3, 3]" = torch.ops.aten.add.Tensor(x, 1); x = None
# File: /var/lib/workspace/intermediate_source/torch_export_tutorial.py:180 in forward, code: return x + id(x)
add_1: "f32[3, 3]" = torch.ops.aten.add.Tensor(add, 140323388154288); add = None
return (add_1,)
Graph signature: ExportGraphSignature(input_specs=[InputSpec(kind=<InputKind.USER_INPUT: 1>, arg=TensorArgument(name='x'), target=None, persistent=None)], output_specs=[OutputSpec(kind=<OutputKind.USER_OUTPUT: 1>, arg=TensorArgument(name='add_1'), target=None)])
Range constraints: {}
tensor([[1.4032e+14, 1.4032e+14, 1.4032e+14],
[1.4032e+14, 1.4032e+14, 1.4032e+14],
[1.4032e+14, 1.4032e+14, 1.4032e+14]])
However, there are still some features that require rewrites to the original module:
Control Flow Ops
torch.export actually does support data-dependent control flow.
But these need to be expressed using control flow ops. For example,
we can fix the control flow example above using the cond op, like so:
class Bad1Fixed(torch.nn.Module):
def forward(self, x):
def true_fn(x):
return torch.sin(x)
def false_fn(x):
return torch.cos(x)
return torch.cond(x.sum() > 0, true_fn, false_fn, [x])
exported_bad1_fixed = export(Bad1Fixed(), (torch.randn(3, 3),))
print(exported_bad1_fixed)
print(exported_bad1_fixed.module()(torch.ones(3, 3)))
print(exported_bad1_fixed.module()(-torch.ones(3, 3)))
ExportedProgram:
class GraphModule(torch.nn.Module):
def forward(self, x: "f32[3, 3]"):
# File: /var/lib/workspace/intermediate_source/torch_export_tutorial.py:205 in forward, code: return torch.cond(x.sum() > 0, true_fn, false_fn, [x])
sum_1: "f32[]" = torch.ops.aten.sum.default(x)
gt: "b8[]" = torch.ops.aten.gt.Scalar(sum_1, 0); sum_1 = None
# File: /usr/local/lib/python3.10/dist-packages/torch/_higher_order_ops/cond.py:137 in cond, code: return cond_op(pred, true_fn, false_fn, operands)
true_graph_0 = self.true_graph_0
false_graph_0 = self.false_graph_0
cond = torch.ops.higher_order.cond(gt, true_graph_0, false_graph_0, [x]); gt = true_graph_0 = false_graph_0 = x = None
getitem: "f32[3, 3]" = cond[0]; cond = None
return (getitem,)
class true_graph_0(torch.nn.Module):
def forward(self, x: "f32[3, 3]"):
# File: /var/lib/workspace/intermediate_source/torch_export_tutorial.py:202 in true_fn, code: return torch.sin(x)
sin: "f32[3, 3]" = torch.ops.aten.sin.default(x); x = None
return (sin,)
class false_graph_0(torch.nn.Module):
def forward(self, x: "f32[3, 3]"):
# File: /var/lib/workspace/intermediate_source/torch_export_tutorial.py:204 in false_fn, code: return torch.cos(x)
cos: "f32[3, 3]" = torch.ops.aten.cos.default(x); x = None
return (cos,)
Graph signature: ExportGraphSignature(input_specs=[InputSpec(kind=<InputKind.USER_INPUT: 1>, arg=TensorArgument(name='x'), target=None, persistent=None)], output_specs=[OutputSpec(kind=<OutputKind.USER_OUTPUT: 1>, arg=TensorArgument(name='getitem'), target=None)])
Range constraints: {}
tensor([[0.8415, 0.8415, 0.8415],
[0.8415, 0.8415, 0.8415],
[0.8415, 0.8415, 0.8415]])
tensor([[0.5403, 0.5403, 0.5403],
[0.5403, 0.5403, 0.5403],
[0.5403, 0.5403, 0.5403]])
There are limitations to cond that one should be aware of:
The predicate (i.e.
x.sum() > 0) must result in a boolean or a single-element tensor.The operands (i.e.
[x]) must be tensors.The branch function (i.e.
true_fnandfalse_fn) signature must match with the operands and they must both return a single tensor with the same metadata (for example,dtype,shape, etc.).Branch functions cannot mutate input or global variables.
Branch functions cannot access closure variables, except for
selfif the function is defined in the scope of a method.
For more details about cond, check out the cond documentation.
We can also use map, which applies a function across the first dimension
of the first tensor argument.
from torch._higher_order_ops.map import map as torch_map
class MapModule(torch.nn.Module):
def forward(self, xs, y, z):
def body(x, y, z):
return x + y + z
return torch_map(body, xs, y, z)
inps = (torch.ones(6, 4), torch.tensor(5), torch.tensor(4))
exported_map_example = export(MapModule(), inps)
print(exported_map_example)
print(exported_map_example.module()(*inps))
ExportedProgram:
class GraphModule(torch.nn.Module):
def forward(self, xs: "f32[6, 4]", y: "i64[]", z: "i64[]"):
# File: /var/lib/workspace/intermediate_source/torch_export_tutorial.py:236 in forward, code: return torch_map(body, xs, y, z)
body_graph_0 = self.body_graph_0
map_impl = torch.ops.higher_order.map_impl(body_graph_0, [xs], [y, z]); body_graph_0 = xs = y = z = None
getitem: "f32[6, 4]" = map_impl[0]; map_impl = None
return (getitem,)
class body_graph_0(torch.nn.Module):
def forward(self, xs: "f32[4]", y: "i64[]", z: "i64[]"):
# File: /var/lib/workspace/intermediate_source/torch_export_tutorial.py:234 in body, code: return x + y + z
add: "f32[4]" = torch.ops.aten.add.Tensor(xs, y); xs = y = None
add_1: "f32[4]" = torch.ops.aten.add.Tensor(add, z); add = z = None
return (add_1,)
Graph signature: ExportGraphSignature(input_specs=[InputSpec(kind=<InputKind.USER_INPUT: 1>, arg=TensorArgument(name='xs'), target=None, persistent=None), InputSpec(kind=<InputKind.USER_INPUT: 1>, arg=TensorArgument(name='y'), target=None, persistent=None), InputSpec(kind=<InputKind.USER_INPUT: 1>, arg=TensorArgument(name='z'), target=None, persistent=None)], output_specs=[OutputSpec(kind=<OutputKind.USER_OUTPUT: 1>, arg=TensorArgument(name='getitem'), target=None)])
Range constraints: {}
tensor([[10., 10., 10., 10.],
[10., 10., 10., 10.],
[10., 10., 10., 10.],
[10., 10., 10., 10.],
[10., 10., 10., 10.],
[10., 10., 10., 10.]])
Other control flow ops include while_loop, associative_scan, and
scan. For more documentation on each operator, please refer to
this page.
Constraints/Dynamic Shapes
This section covers dynamic behavior and representation of exported programs. Dynamic behavior is subjective to the particular model being exported, so for the most part of this tutorial, we’ll focus on this particular toy model (with the resulting tensor shapes annotated):
class DynamicModel(torch.nn.Module):
def __init__(self):
super().__init__()
self.l = torch.nn.Linear(5, 3)
def forward(
self,
w: torch.Tensor, # [6, 5]
x: torch.Tensor, # [4]
y: torch.Tensor, # [8, 4]
z: torch.Tensor, # [32]
):
x0 = x + y # [8, 4]
x1 = self.l(w) # [6, 3]
x2 = x0.flatten() # [32]
x3 = x2 + z # [32]
return x1, x3
By default, torch.export produces a static program. One consequence of this is that at runtime,
the program won’t work on inputs with different shapes, even if they’re valid in eager mode.
w = torch.randn(6, 5)
x = torch.randn(4)
y = torch.randn(8, 4)
z = torch.randn(32)
model = DynamicModel()
ep = export(model, (w, x, y, z))
model(w, x, torch.randn(3, 4), torch.randn(12))
try:
ep.module()(w, x, torch.randn(3, 4), torch.randn(12))
except Exception:
tb.print_exc()
Traceback (most recent call last):
File "/var/lib/workspace/intermediate_source/torch_export_tutorial.py", line 286, in <module>
ep.module()(w, x, torch.randn(3, 4), torch.randn(12))
File "/usr/local/lib/python3.10/dist-packages/torch/fx/graph_module.py", line 830, in call_wrapped
return self._wrapped_call(self, *args, **kwargs)
File "/usr/local/lib/python3.10/dist-packages/torch/fx/graph_module.py", line 406, in __call__
raise e
File "/usr/local/lib/python3.10/dist-packages/torch/fx/graph_module.py", line 393, in __call__
return super(self.cls, obj).__call__(*args, **kwargs) # type: ignore[misc]
File "/usr/local/lib/python3.10/dist-packages/torch/nn/modules/module.py", line 1751, in _wrapped_call_impl
return self._call_impl(*args, **kwargs)
File "/usr/local/lib/python3.10/dist-packages/torch/nn/modules/module.py", line 1857, in _call_impl
return inner()
File "/usr/local/lib/python3.10/dist-packages/torch/nn/modules/module.py", line 1784, in inner
args_kwargs_result = hook(self, args, kwargs) # type: ignore[misc]
File "/usr/local/lib/python3.10/dist-packages/torch/_dynamo/eval_frame.py", line 838, in _fn
return fn(*args, **kwargs)
File "/usr/local/lib/python3.10/dist-packages/torch/export/_unlift.py", line 55, in _check_input_constraints_pre_hook
_check_input_constraints_for_graph(
File "/usr/local/lib/python3.10/dist-packages/torch/_export/utils.py", line 398, in _check_input_constraints_for_graph
raise RuntimeError(
RuntimeError: Expected input at *args[2].shape[0] to be equal to 8, but got 3
Basic concepts: symbols and guards
To enable dynamism, export() provides a dynamic_shapes argument. The easiest way to work with
dynamic shapes is using Dim.AUTO and looking at the program that’s returned. Dynamic behavior is specified
at a input dimension-level; for each input we can specify a tuple of values:
Before we look at the program that’s produced, let’s understand what specifying dynamic_shapes entails,
and how that interacts with export. For every input dimension where a Dim object is specified, a symbol is
allocated,
taking on a range of [2, inf] (why not [0, inf] or [1, inf]? we’ll explain later in the
0/1 specialization section).
Export then runs model tracing, looking at each operation that’s performed by the model. Each individual operation can emit
what’s called “guards”; basically boolean condition that are required to be true for the program to be valid.
When guards involve symbols allocated for input dimensions, the program contains restrictions on what input shapes are valid;
i.e. the program’s dynamic behavior. The symbolic shapes subsystem is the part responsible for taking in all the emitted guards
and producing a final program representation that adheres to all of these guards. Before we see this “final representation” in
an ExportedProgram, let’s look at the guards emitted by the toy model we’re tracing.
Here, each forward input tensor is annotated with the symbol allocated at the start of tracing:
class DynamicModel(torch.nn.Module):
def __init__(self):
super().__init__()
self.l = torch.nn.Linear(5, 3)
def forward(
self,
w: torch.Tensor, # [s0, s1]
x: torch.Tensor, # [s2]
y: torch.Tensor, # [s3, s4]
z: torch.Tensor, # [s5]
):
x0 = x + y # guard: s2 == s4
x1 = self.l(w) # guard: s1 == 5
x2 = x0.flatten() # no guard added here
x3 = x2 + z # guard: s3 * s4 == s5
return x1, x3
Let’s understand each of the operations and the emitted guards:
x0 = x + y: This is an element-wise add with broadcasting, sincexis a 1-d tensor andya 2-d tensor.xis broadcasted along the last dimension ofy, emitting the guards2 == s4.x1 = self.l(w): Callingnn.Linear()performs a matrix multiplication with model parameters. In export, parameters, buffers, and constants are considered program state, which is considered static, and so this is a matmul between a dynamic input (w: [s0, s1]), and a statically-shaped tensor. This emits the guards1 == 5.x2 = x0.flatten(): This call actually doesn’t emit any guards! (at least none relevant to input shapes)x3 = x2 + z:x2has shape[s3*s4]after flattening, and this element-wise add emitss3 * s4 == s5.
Writing all of these guards down and summarizing is almost like a mathematical proof, which is what the symbolic shapes subsystem tries to do! In summary, we can conclude that the program must have the following input shapes to be valid:
w: [s0, 5]x: [s2]y: [s3, s2]z: [s2*s3]
And when we do finally print out the exported program to see our result, those shapes are what we see annotated on the corresponding inputs:
print(ep)
ExportedProgram:
class GraphModule(torch.nn.Module):
def forward(self, p_l_weight: "f32[3, 5]", p_l_bias: "f32[3]", w: "f32[s0, 5]", x: "f32[s2]", y: "f32[s3, s2]", z: "f32[s2*s3]"):
# File: /var/lib/workspace/intermediate_source/torch_export_tutorial.py:268 in forward, code: x0 = x + y # [8, 4]
add: "f32[s3, s2]" = torch.ops.aten.add.Tensor(x, y); x = y = None
# File: /var/lib/workspace/intermediate_source/torch_export_tutorial.py:269 in forward, code: x1 = self.l(w) # [6, 3]
linear: "f32[s0, 3]" = torch.ops.aten.linear.default(w, p_l_weight, p_l_bias); w = p_l_weight = p_l_bias = None
# File: /var/lib/workspace/intermediate_source/torch_export_tutorial.py:270 in forward, code: x2 = x0.flatten() # [32]
flatten: "f32[s2*s3]" = torch.ops.aten.flatten.using_ints(add); add = None
# File: /var/lib/workspace/intermediate_source/torch_export_tutorial.py:271 in forward, code: x3 = x2 + z # [32]
add_1: "f32[s2*s3]" = torch.ops.aten.add.Tensor(flatten, z); flatten = z = None
return (linear, add_1)
Graph signature: ExportGraphSignature(input_specs=[InputSpec(kind=<InputKind.PARAMETER: 2>, arg=TensorArgument(name='p_l_weight'), target='l.weight', persistent=None), InputSpec(kind=<InputKind.PARAMETER: 2>, arg=TensorArgument(name='p_l_bias'), target='l.bias', persistent=None), InputSpec(kind=<InputKind.USER_INPUT: 1>, arg=TensorArgument(name='w'), target=None, persistent=None), InputSpec(kind=<InputKind.USER_INPUT: 1>, arg=TensorArgument(name='x'), target=None, persistent=None), InputSpec(kind=<InputKind.USER_INPUT: 1>, arg=TensorArgument(name='y'), target=None, persistent=None), InputSpec(kind=<InputKind.USER_INPUT: 1>, arg=TensorArgument(name='z'), target=None, persistent=None)], output_specs=[OutputSpec(kind=<OutputKind.USER_OUTPUT: 1>, arg=TensorArgument(name='linear'), target=None), OutputSpec(kind=<OutputKind.USER_OUTPUT: 1>, arg=TensorArgument(name='add_1'), target=None)])
Range constraints: {s0: VR[2, int_oo], s2: VR[2, int_oo], s3: VR[2, int_oo], s2*s3: VR[4, int_oo]}
Another feature to notice is the range_constraints field above, which contains a valid range for each symbol. This isn’t so interesting currently, since this export call doesn’t emit any guards related to symbol bounds and each base symbol has a generic bound, but this will come up later.
So far, because we’ve been exporting this toy model, this experience has not been representative of how hard it typically is to debug dynamic shapes guards & issues. In most cases it isn’t obvious what guards are being emitted, and which operations and parts of user code are responsible. For this toy model we pinpoint the exact lines, and the guards are rather intuitive.
In more complicated cases, a helpful first step is always to enable verbose logging. This can be done either with the environment
variable TORCH_LOGS="+dynamic", or interactively with torch._logging.set_logs(dynamic=10):
I0502 18:42:26.254000 635 torch/fx/experimental/symbolic_shapes.py:3334] [8/0] create_env
I0502 18:42:26.257000 635 torch/fx/experimental/symbolic_shapes.py:4606] [8/0] create_symbol s0 = 6 for L['w'].size()[0] [2, int_oo] (_dynamo/variables/builder.py:3033 in <lambda>), for more info run with TORCHDYNAMO_EXTENDED_DEBUG_CREATE_SYMBOL="s0" or to suppress this message run with TORCHDYNAMO_EXTENDED_ADVICE="0"
I0502 18:42:26.257000 635 torch/fx/experimental/symbolic_shapes.py:4606] [8/0] create_symbol s1 = 5 for L['w'].size()[1] [2, int_oo] (_dynamo/variables/builder.py:3033 in <lambda>), for more info run with TORCHDYNAMO_EXTENDED_DEBUG_CREATE_SYMBOL="s1" or to suppress this message run with TORCHDYNAMO_EXTENDED_ADVICE="0"
V0502 18:42:26.258000 635 torch/fx/experimental/symbolic_shapes.py:7018] [8/0] runtime_assert True == True [statically known]
I0502 18:42:26.260000 635 torch/fx/experimental/symbolic_shapes.py:4606] [8/0] create_symbol s2 = 4 for L['x'].size()[0] [2, int_oo] (_dynamo/variables/builder.py:3033 in <lambda>), for more info run with TORCHDYNAMO_EXTENDED_DEBUG_CREATE_SYMBOL="s2" or to suppress this message run with TORCHDYNAMO_EXTENDED_ADVICE="0"
I0502 18:42:26.262000 635 torch/fx/experimental/symbolic_shapes.py:4606] [8/0] create_symbol s3 = 8 for L['y'].size()[0] [2, int_oo] (_dynamo/variables/builder.py:3033 in <lambda>), for more info run with TORCHDYNAMO_EXTENDED_DEBUG_CREATE_SYMBOL="s3" or to suppress this message run with TORCHDYNAMO_EXTENDED_ADVICE="0"
I0502 18:42:26.262000 635 torch/fx/experimental/symbolic_shapes.py:4606] [8/0] create_symbol s4 = 4 for L['y'].size()[1] [2, int_oo] (_dynamo/variables/builder.py:3033 in <lambda>), for more info run with TORCHDYNAMO_EXTENDED_DEBUG_CREATE_SYMBOL="s4" or to suppress this message run with TORCHDYNAMO_EXTENDED_ADVICE="0"
I0502 18:42:26.265000 635 torch/fx/experimental/symbolic_shapes.py:4606] [8/0] create_symbol s5 = 32 for L['z'].size()[0] [2, int_oo] (_dynamo/variables/builder.py:3033 in <lambda>), for more info run with TORCHDYNAMO_EXTENDED_DEBUG_CREATE_SYMBOL="s5" or to suppress this message run with TORCHDYNAMO_EXTENDED_ADVICE="0"
V0502 18:42:26.267000 635 torch/fx/experimental/symbolic_shapes.py:6787] [8/0] eval size_oblivious(Eq(s2, 1)) == False [statically known]
V0502 18:42:26.267000 635 torch/fx/experimental/symbolic_shapes.py:7018] [8/0] runtime_assert True == True [statically known]
V0502 18:42:26.268000 635 torch/fx/experimental/symbolic_shapes.py:6787] [8/0] eval size_oblivious(Eq(s4, 1)) == False [statically known]
I0502 18:42:26.268000 635 torch/fx/experimental/symbolic_shapes.py:6630] [8/0] runtime_assert Eq(s2, s4) [guard added] x0 = x + y # [8, 4] # ar/lib/workspace/intermediate_source/torch_export_tutorial.py:268 in forward (_subclasses/fake_impls.py:881 in infer_size), for more info run with TORCHDYNAMO_EXTENDED_DEBUG_GUARD_ADDED="Eq(s2, s4)"
I0502 18:42:26.269000 635 torch/fx/experimental/symbolic_shapes.py:6234] [8/0] set_replacement s4 = s2 (solve) VR[2, int_oo]
V0502 18:42:26.271000 635 torch/fx/experimental/symbolic_shapes.py:6787] [8/0] eval size_oblivious(Ne(s2, 1)) == True [statically known]
V0502 18:42:26.272000 635 torch/fx/experimental/symbolic_shapes.py:6787] [8/0] eval size_oblivious(Ne(s3, 1)) == True [statically known]
I0502 18:42:26.277000 635 torch/fx/experimental/symbolic_shapes.py:6630] [8/0] runtime_assert Eq(s1, 5) [guard added] x1 = self.l(w) # [6, 3] # ar/lib/workspace/intermediate_source/torch_export_tutorial.py:269 in forward (_meta_registrations.py:2236 in meta_mm), for more info run with TORCHDYNAMO_EXTENDED_DEBUG_GUARD_ADDED="Eq(s1, 5)"
V0502 18:42:26.278000 635 torch/fx/experimental/symbolic_shapes.py:6071] [8/0] _update_var_to_range s1 = VR[5, 5] (update)
I0502 18:42:26.279000 635 torch/fx/experimental/symbolic_shapes.py:6234] [8/0] set_replacement s1 = 5 (range_refined_to_singleton) VR[5, 5]
V0502 18:42:26.280000 635 torch/fx/experimental/symbolic_shapes.py:6787] [8/0] eval size_oblivious(Eq(s0, 1)) == False [statically known]
V0502 18:42:26.285000 635 torch/fx/experimental/symbolic_shapes.py:6787] [8/0] eval size_oblivious(Eq(s2*s3, 1)) == False [statically known]
V0502 18:42:26.286000 635 torch/fx/experimental/symbolic_shapes.py:6787] [8/0] eval size_oblivious(Eq(s5, 1)) == False [statically known]
I0502 18:42:26.287000 635 torch/fx/experimental/symbolic_shapes.py:6630] [8/0] runtime_assert Eq(s2*s3, s5) [guard added] x3 = x2 + z # [32] # ar/lib/workspace/intermediate_source/torch_export_tutorial.py:271 in forward (_subclasses/fake_impls.py:881 in infer_size), for more info run with TORCHDYNAMO_EXTENDED_DEBUG_GUARD_ADDED="Eq(s2*s3, s5)"
V0502 18:42:26.288000 635 torch/fx/experimental/symbolic_shapes.py:6071] [8/0] _update_var_to_range s5 = VR[4, int_oo] (update)
I0502 18:42:26.289000 635 torch/fx/experimental/symbolic_shapes.py:6234] [8/0] set_replacement s5 = s2*s3 (solve) VR[4, int_oo]
V0502 18:42:26.290000 635 torch/fx/experimental/symbolic_shapes.py:6787] [8/0] eval size_oblivious(Ne(s2*s3, 1)) == True [statically known]
I0502 18:42:26.296000 635 torch/fx/experimental/symbolic_shapes.py:4734] [8/0] produce_guards
V0502 18:42:26.296000 635 torch/fx/experimental/symbolic_shapes.py:4954] [8/0] track_symint L['w'].size()[0] s0 None
V0502 18:42:26.296000 635 torch/fx/experimental/symbolic_shapes.py:4954] [8/0] track_symint L['w'].size()[1] 5 None
V0502 18:42:26.297000 635 torch/fx/experimental/symbolic_shapes.py:4954] [8/0] track_symint L['w'].stride()[0] 5 None
V0502 18:42:26.297000 635 torch/fx/experimental/symbolic_shapes.py:4954] [8/0] track_symint L['w'].stride()[1] 1 None
V0502 18:42:26.297000 635 torch/fx/experimental/symbolic_shapes.py:4954] [8/0] track_symint L['w'].storage_offset() 0 None
V0502 18:42:26.297000 635 torch/fx/experimental/symbolic_shapes.py:4954] [8/0] track_symint L['x'].size()[0] s2 None
V0502 18:42:26.298000 635 torch/fx/experimental/symbolic_shapes.py:4954] [8/0] track_symint L['x'].stride()[0] 1 None
V0502 18:42:26.298000 635 torch/fx/experimental/symbolic_shapes.py:4954] [8/0] track_symint L['x'].storage_offset() 0 None
V0502 18:42:26.298000 635 torch/fx/experimental/symbolic_shapes.py:4954] [8/0] track_symint L['y'].size()[0] s3 None
V0502 18:42:26.298000 635 torch/fx/experimental/symbolic_shapes.py:4954] [8/0] track_symint L['y'].size()[1] s2 None
V0502 18:42:26.299000 635 torch/fx/experimental/symbolic_shapes.py:4954] [8/0] track_symint L['y'].stride()[0] s2 None
V0502 18:42:26.299000 635 torch/fx/experimental/symbolic_shapes.py:4954] [8/0] track_symint L['y'].stride()[1] 1 None
V0502 18:42:26.299000 635 torch/fx/experimental/symbolic_shapes.py:4954] [8/0] track_symint L['y'].storage_offset() 0 None
V0502 18:42:26.299000 635 torch/fx/experimental/symbolic_shapes.py:4954] [8/0] track_symint L['z'].size()[0] s2*s3 None
V0502 18:42:26.300000 635 torch/fx/experimental/symbolic_shapes.py:4954] [8/0] track_symint L['z'].stride()[0] 1 None
V0502 18:42:26.300000 635 torch/fx/experimental/symbolic_shapes.py:4954] [8/0] track_symint L['z'].storage_offset() 0 None
V0502 18:42:26.332000 635 torch/fx/experimental/symbolic_shapes.py:6787] eval size_oblivious(Ne(s0, 1)) == True [statically known]
This spits out quite a handful, even with this simple toy model. The log lines here have been cut short at front and end to ignore unnecessary info, but looking through the logs we can see the lines relevant to what we described above; e.g. the allocation of symbols:
"""
create_symbol s0 = 6 for L['w'].size()[0] [2, int_oo] (_dynamo/variables/builder.py:2841 in <lambda>)
create_symbol s1 = 5 for L['w'].size()[1] [2, int_oo] (_dynamo/variables/builder.py:2841 in <lambda>)
runtime_assert True == True [statically known]
create_symbol s2 = 4 for L['x'].size()[0] [2, int_oo] (_dynamo/variables/builder.py:2841 in <lambda>)
create_symbol s3 = 8 for L['y'].size()[0] [2, int_oo] (_dynamo/variables/builder.py:2841 in <lambda>)
create_symbol s4 = 4 for L['y'].size()[1] [2, int_oo] (_dynamo/variables/builder.py:2841 in <lambda>)
create_symbol s5 = 32 for L['z'].size()[0] [2, int_oo] (_dynamo/variables/builder.py:2841 in <lambda>)
"""
"\ncreate_symbol s0 = 6 for L['w'].size()[0] [2, int_oo] (_dynamo/variables/builder.py:2841 in <lambda>)\ncreate_symbol s1 = 5 for L['w'].size()[1] [2, int_oo] (_dynamo/variables/builder.py:2841 in <lambda>)\nruntime_assert True == True [statically known]\ncreate_symbol s2 = 4 for L['x'].size()[0] [2, int_oo] (_dynamo/variables/builder.py:2841 in <lambda>)\ncreate_symbol s3 = 8 for L['y'].size()[0] [2, int_oo] (_dynamo/variables/builder.py:2841 in <lambda>)\ncreate_symbol s4 = 4 for L['y'].size()[1] [2, int_oo] (_dynamo/variables/builder.py:2841 in <lambda>)\ncreate_symbol s5 = 32 for L['z'].size()[0] [2, int_oo] (_dynamo/variables/builder.py:2841 in <lambda>)\n"
The lines with create_symbol show when a new symbol has been allocated, and the logs also identify the tensor variable names and dimensions they’ve been allocated for. In other lines we can also see the guards emitted:
"""
runtime_assert Eq(s2, s4) [guard added] x0 = x + y # output shape: [8, 4] # dynamic_shapes_tutorial.py:16 in forward (_subclasses/fake_impls.py:845 in infer_size), for more info run with TORCHDYNAMO_EXTENDED_DEBUG_GUARD_ADDED="Eq(s2, s4)"
runtime_assert Eq(s1, 5) [guard added] x1 = self.l(w) # [6, 3] # dynamic_shapes_tutorial.py:17 in forward (_meta_registrations.py:2127 in meta_mm), for more info run with TORCHDYNAMO_EXTENDED_DEBUG_GUARD_ADDED="Eq(s1, 5)"
runtime_assert Eq(s2*s3, s5) [guard added] x3 = x2 + z # [32] # dynamic_shapes_tutorial.py:19 in forward (_subclasses/fake_impls.py:845 in infer_size), for more info run with TORCHDYNAMO_EXTENDED_DEBUG_GUARD_ADDED="Eq(s2*s3, s5)"
"""
'\nruntime_assert Eq(s2, s4) [guard added] x0 = x + y # output shape: [8, 4] # dynamic_shapes_tutorial.py:16 in forward (_subclasses/fake_impls.py:845 in infer_size), for more info run with TORCHDYNAMO_EXTENDED_DEBUG_GUARD_ADDED="Eq(s2, s4)"\nruntime_assert Eq(s1, 5) [guard added] x1 = self.l(w) # [6, 3] # dynamic_shapes_tutorial.py:17 in forward (_meta_registrations.py:2127 in meta_mm), for more info run with TORCHDYNAMO_EXTENDED_DEBUG_GUARD_ADDED="Eq(s1, 5)"\nruntime_assert Eq(s2*s3, s5) [guard added] x3 = x2 + z # [32] # dynamic_shapes_tutorial.py:19 in forward (_subclasses/fake_impls.py:845 in infer_size), for more info run with TORCHDYNAMO_EXTENDED_DEBUG_GUARD_ADDED="Eq(s2*s3, s5)"\n'
Next to the [guard added] messages, we also see the responsible user lines of code - luckily here the model is simple enough.
In many real-world cases it’s not so straightforward: high-level torch operations can have complicated fake-kernel implementations
or operator decompositions that complicate where and what guards are emitted. In such cases the best way to dig deeper and investigate
is to follow the logs’ suggestion, and re-run with environment variable TORCHDYNAMO_EXTENDED_DEBUG_GUARD_ADDED="...", to further
attribute the guard of interest.
Dim.AUTO is just one of the available options for interacting with dynamic_shapes; as of writing this 2 other options are available:
Dim.DYNAMIC, and Dim.STATIC. Dim.STATIC simply marks a dimension static, while Dim.DYNAMIC is similar to Dim.AUTO in all
ways except one: it raises an error when specializing to a constant; this is designed to maintain dynamism. See for example what happens when a
static guard is emitted on a dynamically-marked dimension:
I0502 18:42:26.353000 635 torch/fx/experimental/symbolic_shapes.py:3334] [9/0] create_env
I0502 18:42:26.355000 635 torch/fx/experimental/symbolic_shapes.py:4606] [9/0] create_symbol s0 = 6 for L['w'].size()[0] [2, int_oo] (_dynamo/variables/builder.py:3033 in <lambda>), for more info run with TORCHDYNAMO_EXTENDED_DEBUG_CREATE_SYMBOL="s0" or to suppress this message run with TORCHDYNAMO_EXTENDED_ADVICE="0"
I0502 18:42:26.355000 635 torch/fx/experimental/symbolic_shapes.py:4606] [9/0] create_symbol s1 = 5 for L['w'].size()[1] [2, int_oo] (_dynamo/variables/builder.py:3033 in <lambda>), for more info run with TORCHDYNAMO_EXTENDED_DEBUG_CREATE_SYMBOL="s1" or to suppress this message run with TORCHDYNAMO_EXTENDED_ADVICE="0"
V0502 18:42:26.356000 635 torch/fx/experimental/symbolic_shapes.py:7018] [9/0] runtime_assert True == True [statically known]
I0502 18:42:26.359000 635 torch/fx/experimental/symbolic_shapes.py:4606] [9/0] create_symbol s2 = 4 for L['x'].size()[0] [2, int_oo] (_dynamo/variables/builder.py:3033 in <lambda>), for more info run with TORCHDYNAMO_EXTENDED_DEBUG_CREATE_SYMBOL="s2" or to suppress this message run with TORCHDYNAMO_EXTENDED_ADVICE="0"
I0502 18:42:26.361000 635 torch/fx/experimental/symbolic_shapes.py:4606] [9/0] create_symbol s3 = 8 for L['y'].size()[0] [2, int_oo] (_dynamo/variables/builder.py:3033 in <lambda>), for more info run with TORCHDYNAMO_EXTENDED_DEBUG_CREATE_SYMBOL="s3" or to suppress this message run with TORCHDYNAMO_EXTENDED_ADVICE="0"
I0502 18:42:26.361000 635 torch/fx/experimental/symbolic_shapes.py:4606] [9/0] create_symbol s4 = 4 for L['y'].size()[1] [2, int_oo] (_dynamo/variables/builder.py:3033 in <lambda>), for more info run with TORCHDYNAMO_EXTENDED_DEBUG_CREATE_SYMBOL="s4" or to suppress this message run with TORCHDYNAMO_EXTENDED_ADVICE="0"
I0502 18:42:26.364000 635 torch/fx/experimental/symbolic_shapes.py:4606] [9/0] create_symbol s5 = 32 for L['z'].size()[0] [2, int_oo] (_dynamo/variables/builder.py:3033 in <lambda>), for more info run with TORCHDYNAMO_EXTENDED_DEBUG_CREATE_SYMBOL="s5" or to suppress this message run with TORCHDYNAMO_EXTENDED_ADVICE="0"
V0502 18:42:26.366000 635 torch/fx/experimental/symbolic_shapes.py:6787] [9/0] eval size_oblivious(Eq(s2, 1)) == False [statically known]
V0502 18:42:26.366000 635 torch/fx/experimental/symbolic_shapes.py:7018] [9/0] runtime_assert True == True [statically known]
V0502 18:42:26.367000 635 torch/fx/experimental/symbolic_shapes.py:6787] [9/0] eval size_oblivious(Eq(s4, 1)) == False [statically known]
I0502 18:42:26.368000 635 torch/fx/experimental/symbolic_shapes.py:6630] [9/0] runtime_assert Eq(s2, s4) [guard added] x0 = x + y # [8, 4] # ar/lib/workspace/intermediate_source/torch_export_tutorial.py:268 in forward (_subclasses/fake_impls.py:881 in infer_size), for more info run with TORCHDYNAMO_EXTENDED_DEBUG_GUARD_ADDED="Eq(s2, s4)"
I0502 18:42:26.369000 635 torch/fx/experimental/symbolic_shapes.py:6234] [9/0] set_replacement s4 = s2 (solve) VR[2, int_oo]
V0502 18:42:26.370000 635 torch/fx/experimental/symbolic_shapes.py:6787] [9/0] eval size_oblivious(Ne(s2, 1)) == True [statically known]
V0502 18:42:26.371000 635 torch/fx/experimental/symbolic_shapes.py:6787] [9/0] eval size_oblivious(Ne(s3, 1)) == True [statically known]
I0502 18:42:26.377000 635 torch/fx/experimental/symbolic_shapes.py:6630] [9/0] runtime_assert Eq(s1, 5) [guard added] x1 = self.l(w) # [6, 3] # ar/lib/workspace/intermediate_source/torch_export_tutorial.py:269 in forward (_meta_registrations.py:2236 in meta_mm), for more info run with TORCHDYNAMO_EXTENDED_DEBUG_GUARD_ADDED="Eq(s1, 5)"
V0502 18:42:26.378000 635 torch/fx/experimental/symbolic_shapes.py:6071] [9/0] _update_var_to_range s1 = VR[5, 5] (update)
I0502 18:42:26.379000 635 torch/fx/experimental/symbolic_shapes.py:6234] [9/0] set_replacement s1 = 5 (range_refined_to_singleton) VR[5, 5]
V0502 18:42:26.380000 635 torch/fx/experimental/symbolic_shapes.py:6787] [9/0] eval size_oblivious(Eq(s0, 1)) == False [statically known]
V0502 18:42:26.386000 635 torch/fx/experimental/symbolic_shapes.py:6787] [9/0] eval size_oblivious(Eq(s2*s3, 1)) == False [statically known]
V0502 18:42:26.387000 635 torch/fx/experimental/symbolic_shapes.py:6787] [9/0] eval size_oblivious(Eq(s5, 1)) == False [statically known]
I0502 18:42:26.387000 635 torch/fx/experimental/symbolic_shapes.py:6630] [9/0] runtime_assert Eq(s2*s3, s5) [guard added] x3 = x2 + z # [32] # ar/lib/workspace/intermediate_source/torch_export_tutorial.py:271 in forward (_subclasses/fake_impls.py:881 in infer_size), for more info run with TORCHDYNAMO_EXTENDED_DEBUG_GUARD_ADDED="Eq(s2*s3, s5)"
V0502 18:42:26.388000 635 torch/fx/experimental/symbolic_shapes.py:6071] [9/0] _update_var_to_range s5 = VR[4, int_oo] (update)
I0502 18:42:26.390000 635 torch/fx/experimental/symbolic_shapes.py:6234] [9/0] set_replacement s5 = s2*s3 (solve) VR[4, int_oo]
V0502 18:42:26.391000 635 torch/fx/experimental/symbolic_shapes.py:6787] [9/0] eval size_oblivious(Ne(s2*s3, 1)) == True [statically known]
I0502 18:42:26.397000 635 torch/fx/experimental/symbolic_shapes.py:4734] [9/0] produce_guards
V0502 18:42:26.397000 635 torch/fx/experimental/symbolic_shapes.py:4954] [9/0] track_symint L['w'].size()[0] s0 None
V0502 18:42:26.397000 635 torch/fx/experimental/symbolic_shapes.py:4954] [9/0] track_symint L['w'].size()[1] 5 RelaxedUnspecConstraint(warn_only=False)
V0502 18:42:26.398000 635 torch/fx/experimental/symbolic_shapes.py:4954] [9/0] track_symint L['w'].stride()[0] 5 None
V0502 18:42:26.398000 635 torch/fx/experimental/symbolic_shapes.py:4954] [9/0] track_symint L['w'].stride()[1] 1 None
V0502 18:42:26.398000 635 torch/fx/experimental/symbolic_shapes.py:4954] [9/0] track_symint L['w'].storage_offset() 0 None
V0502 18:42:26.399000 635 torch/fx/experimental/symbolic_shapes.py:4954] [9/0] track_symint L['x'].size()[0] s2 None
V0502 18:42:26.399000 635 torch/fx/experimental/symbolic_shapes.py:4954] [9/0] track_symint L['x'].stride()[0] 1 None
V0502 18:42:26.399000 635 torch/fx/experimental/symbolic_shapes.py:4954] [9/0] track_symint L['x'].storage_offset() 0 None
V0502 18:42:26.400000 635 torch/fx/experimental/symbolic_shapes.py:4954] [9/0] track_symint L['y'].size()[0] s3 None
V0502 18:42:26.400000 635 torch/fx/experimental/symbolic_shapes.py:4954] [9/0] track_symint L['y'].size()[1] s2 None
V0502 18:42:26.400000 635 torch/fx/experimental/symbolic_shapes.py:4954] [9/0] track_symint L['y'].stride()[0] s2 None
V0502 18:42:26.401000 635 torch/fx/experimental/symbolic_shapes.py:4954] [9/0] track_symint L['y'].stride()[1] 1 None
V0502 18:42:26.401000 635 torch/fx/experimental/symbolic_shapes.py:4954] [9/0] track_symint L['y'].storage_offset() 0 None
V0502 18:42:26.401000 635 torch/fx/experimental/symbolic_shapes.py:4954] [9/0] track_symint L['z'].size()[0] s2*s3 None
V0502 18:42:26.402000 635 torch/fx/experimental/symbolic_shapes.py:4954] [9/0] track_symint L['z'].stride()[0] 1 None
V0502 18:42:26.402000 635 torch/fx/experimental/symbolic_shapes.py:4954] [9/0] track_symint L['z'].storage_offset() 0 None
E0502 18:42:26.404000 635 torch/_guards.py:359] [9/0] Error while creating guard:
E0502 18:42:26.404000 635 torch/_guards.py:359] [9/0] Name: ''
E0502 18:42:26.404000 635 torch/_guards.py:359] [9/0] Source: shape_env
E0502 18:42:26.404000 635 torch/_guards.py:359] [9/0] Create Function: SHAPE_ENV
E0502 18:42:26.404000 635 torch/_guards.py:359] [9/0] Guard Types: None
E0502 18:42:26.404000 635 torch/_guards.py:359] [9/0] Code List: None
E0502 18:42:26.404000 635 torch/_guards.py:359] [9/0] Object Weakref: None
E0502 18:42:26.404000 635 torch/_guards.py:359] [9/0] Guarded Class Weakref: None
E0502 18:42:26.404000 635 torch/_guards.py:359] [9/0] Traceback (most recent call last):
E0502 18:42:26.404000 635 torch/_guards.py:359] [9/0] File "/usr/local/lib/python3.10/dist-packages/torch/_guards.py", line 357, in create
E0502 18:42:26.404000 635 torch/_guards.py:359] [9/0] return self.create_fn(builder, self)
E0502 18:42:26.404000 635 torch/_guards.py:359] [9/0] File "/usr/local/lib/python3.10/dist-packages/torch/_dynamo/guards.py", line 1959, in SHAPE_ENV
E0502 18:42:26.404000 635 torch/_guards.py:359] [9/0] python_code_parts, verbose_code_parts = _get_code_parts(
E0502 18:42:26.404000 635 torch/_guards.py:359] [9/0] File "/usr/local/lib/python3.10/dist-packages/torch/_dynamo/guards.py", line 1942, in _get_code_parts
E0502 18:42:26.404000 635 torch/_guards.py:359] [9/0] return output_graph.shape_env.produce_guards_verbose(
E0502 18:42:26.404000 635 torch/_guards.py:359] [9/0] File "/usr/local/lib/python3.10/dist-packages/torch/fx/experimental/symbolic_shapes.py", line 5409, in produce_guards_verbose
E0502 18:42:26.404000 635 torch/_guards.py:359] [9/0] raise ConstraintViolationError(
E0502 18:42:26.404000 635 torch/_guards.py:359] [9/0] torch.fx.experimental.symbolic_shapes.ConstraintViolationError: Constraints violated (L['w'].size()[1])! For more information, run with TORCH_LOGS="+dynamic".
E0502 18:42:26.404000 635 torch/_guards.py:359] [9/0] - Not all values of RelaxedUnspecConstraint(L['w'].size()[1]) are valid because L['w'].size()[1] was inferred to be a constant (5).
E0502 18:42:26.407000 635 torch/_guards.py:361] [9/0] Created at:
E0502 18:42:26.407000 635 torch/_guards.py:361] [9/0] File "/usr/local/lib/python3.10/dist-packages/torch/_dynamo/convert_frame.py", line 694, in transform
E0502 18:42:26.407000 635 torch/_guards.py:361] [9/0] tracer = InstructionTranslator(
E0502 18:42:26.407000 635 torch/_guards.py:361] [9/0] File "/usr/local/lib/python3.10/dist-packages/torch/_dynamo/symbolic_convert.py", line 3329, in __init__
E0502 18:42:26.407000 635 torch/_guards.py:361] [9/0] output=OutputGraph(
E0502 18:42:26.407000 635 torch/_guards.py:361] [9/0] File "/usr/local/lib/python3.10/dist-packages/torch/_dynamo/output_graph.py", line 358, in __init__
E0502 18:42:26.407000 635 torch/_guards.py:361] [9/0] self.init_ambient_guards()
E0502 18:42:26.407000 635 torch/_guards.py:361] [9/0] File "/usr/local/lib/python3.10/dist-packages/torch/_dynamo/output_graph.py", line 512, in init_ambient_guards
E0502 18:42:26.407000 635 torch/_guards.py:361] [9/0] self.guards.add(ShapeEnvSource().make_guard(GuardBuilder.SHAPE_ENV))
Traceback (most recent call last):
File "/usr/local/lib/python3.10/dist-packages/torch/export/_trace.py", line 739, in _export_to_torch_ir
gm_torch_level, _ = torch._dynamo.export(
File "/usr/local/lib/python3.10/dist-packages/torch/_dynamo/eval_frame.py", line 1722, in inner
raise constraint_violation_error
File "/usr/local/lib/python3.10/dist-packages/torch/_dynamo/eval_frame.py", line 1677, in inner
result_traced = opt_f(*args, **kwargs)
File "/usr/local/lib/python3.10/dist-packages/torch/nn/modules/module.py", line 1751, in _wrapped_call_impl
return self._call_impl(*args, **kwargs)
File "/usr/local/lib/python3.10/dist-packages/torch/nn/modules/module.py", line 1762, in _call_impl
return forward_call(*args, **kwargs)
File "/usr/local/lib/python3.10/dist-packages/torch/_dynamo/eval_frame.py", line 655, in _fn
return fn(*args, **kwargs)
File "/usr/local/lib/python3.10/dist-packages/torch/nn/modules/module.py", line 1751, in _wrapped_call_impl
return self._call_impl(*args, **kwargs)
File "/usr/local/lib/python3.10/dist-packages/torch/nn/modules/module.py", line 1762, in _call_impl
return forward_call(*args, **kwargs)
File "/usr/local/lib/python3.10/dist-packages/torch/_dynamo/convert_frame.py", line 1432, in __call__
return self._torchdynamo_orig_callable(
File "/usr/local/lib/python3.10/dist-packages/torch/_dynamo/convert_frame.py", line 598, in __call__
return _compile(
File "/usr/local/lib/python3.10/dist-packages/torch/_dynamo/convert_frame.py", line 1059, in _compile
guarded_code = compile_inner(code, one_graph, hooks, transform)
File "/usr/local/lib/python3.10/dist-packages/torch/_utils_internal.py", line 97, in wrapper_function
return function(*args, **kwargs)
File "/usr/local/lib/python3.10/dist-packages/torch/_dynamo/convert_frame.py", line 761, in compile_inner
return _compile_inner(code, one_graph, hooks, transform)
File "/usr/local/lib/python3.10/dist-packages/torch/_dynamo/convert_frame.py", line 906, in _compile_inner
check_fn = CheckFunctionManager(
File "/usr/local/lib/python3.10/dist-packages/torch/_dynamo/guards.py", line 2481, in __init__
guard.create(builder)
File "/usr/local/lib/python3.10/dist-packages/torch/_guards.py", line 357, in create
return self.create_fn(builder, self)
File "/usr/local/lib/python3.10/dist-packages/torch/_dynamo/guards.py", line 1959, in SHAPE_ENV
python_code_parts, verbose_code_parts = _get_code_parts(
File "/usr/local/lib/python3.10/dist-packages/torch/_dynamo/guards.py", line 1942, in _get_code_parts
return output_graph.shape_env.produce_guards_verbose(
File "/usr/local/lib/python3.10/dist-packages/torch/fx/experimental/symbolic_shapes.py", line 5409, in produce_guards_verbose
raise ConstraintViolationError(
torch.fx.experimental.symbolic_shapes.ConstraintViolationError: Constraints violated (L['w'].size()[1])! For more information, run with TORCH_LOGS="+dynamic".
- Not all values of RelaxedUnspecConstraint(L['w'].size()[1]) are valid because L['w'].size()[1] was inferred to be a constant (5).
During handling of the above exception, another exception occurred:
Traceback (most recent call last):
File "/var/lib/workspace/intermediate_source/torch_export_tutorial.py", line 418, in <module>
export(model, (w, x, y, z), dynamic_shapes=dynamic_shapes)
File "/usr/local/lib/python3.10/dist-packages/torch/export/__init__.py", line 360, in export
return _export(
File "/usr/local/lib/python3.10/dist-packages/torch/export/_trace.py", line 1092, in wrapper
raise e
File "/usr/local/lib/python3.10/dist-packages/torch/export/_trace.py", line 1065, in wrapper
ep = fn(*args, **kwargs)
File "/usr/local/lib/python3.10/dist-packages/torch/export/exported_program.py", line 121, in wrapper
return fn(*args, **kwargs)
File "/usr/local/lib/python3.10/dist-packages/torch/export/_trace.py", line 2112, in _export
ep = _export_for_training(
File "/usr/local/lib/python3.10/dist-packages/torch/export/_trace.py", line 1092, in wrapper
raise e
File "/usr/local/lib/python3.10/dist-packages/torch/export/_trace.py", line 1065, in wrapper
ep = fn(*args, **kwargs)
File "/usr/local/lib/python3.10/dist-packages/torch/export/exported_program.py", line 121, in wrapper
return fn(*args, **kwargs)
File "/usr/local/lib/python3.10/dist-packages/torch/export/_trace.py", line 1975, in _export_for_training
export_artifact = export_func( # type: ignore[operator]
File "/usr/local/lib/python3.10/dist-packages/torch/export/_trace.py", line 1344, in _strict_export_lower_to_aten_ir
gm_torch_level = _export_to_torch_ir(
File "/usr/local/lib/python3.10/dist-packages/torch/export/_trace.py", line 756, in _export_to_torch_ir
raise UserError(UserErrorType.CONSTRAINT_VIOLATION, str(e)) # noqa: B904
torch._dynamo.exc.UserError: Constraints violated (L['w'].size()[1])! For more information, run with TORCH_LOGS="+dynamic".
- Not all values of RelaxedUnspecConstraint(L['w'].size()[1]) are valid because L['w'].size()[1] was inferred to be a constant (5).
Static guards also aren’t always inherent to the model; they can also come from user specifications. In fact, a common pitfall leading to shape
specializations is when the user specifies conflicting markers for equivalent dimensions; one dynamic and another static. The same error type is
raised when this is the case for x.shape[0] and y.shape[1]:
I0502 18:42:26.425000 635 torch/fx/experimental/symbolic_shapes.py:3334] [10/0] create_env
I0502 18:42:26.427000 635 torch/fx/experimental/symbolic_shapes.py:4606] [10/0] create_symbol s0 = 6 for L['w'].size()[0] [2, int_oo] (_dynamo/variables/builder.py:3033 in <lambda>), for more info run with TORCHDYNAMO_EXTENDED_DEBUG_CREATE_SYMBOL="s0" or to suppress this message run with TORCHDYNAMO_EXTENDED_ADVICE="0"
I0502 18:42:26.428000 635 torch/fx/experimental/symbolic_shapes.py:4606] [10/0] create_symbol s1 = 5 for L['w'].size()[1] [2, int_oo] (_dynamo/variables/builder.py:3033 in <lambda>), for more info run with TORCHDYNAMO_EXTENDED_DEBUG_CREATE_SYMBOL="s1" or to suppress this message run with TORCHDYNAMO_EXTENDED_ADVICE="0"
V0502 18:42:26.429000 635 torch/fx/experimental/symbolic_shapes.py:7018] [10/0] runtime_assert True == True [statically known]
I0502 18:42:26.432000 635 torch/fx/experimental/symbolic_shapes.py:4606] [10/0] create_symbol s2 = 8 for L['y'].size()[0] [2, int_oo] (_dynamo/variables/builder.py:3033 in <lambda>), for more info run with TORCHDYNAMO_EXTENDED_DEBUG_CREATE_SYMBOL="s2" or to suppress this message run with TORCHDYNAMO_EXTENDED_ADVICE="0"
I0502 18:42:26.432000 635 torch/fx/experimental/symbolic_shapes.py:4606] [10/0] create_symbol s3 = 4 for L['y'].size()[1] [2, int_oo] (_dynamo/variables/builder.py:3033 in <lambda>), for more info run with TORCHDYNAMO_EXTENDED_DEBUG_CREATE_SYMBOL="s3" or to suppress this message run with TORCHDYNAMO_EXTENDED_ADVICE="0"
I0502 18:42:26.435000 635 torch/fx/experimental/symbolic_shapes.py:4606] [10/0] create_symbol s4 = 32 for L['z'].size()[0] [2, int_oo] (_dynamo/variables/builder.py:3033 in <lambda>), for more info run with TORCHDYNAMO_EXTENDED_DEBUG_CREATE_SYMBOL="s4" or to suppress this message run with TORCHDYNAMO_EXTENDED_ADVICE="0"
V0502 18:42:26.438000 635 torch/fx/experimental/symbolic_shapes.py:6787] [10/0] eval size_oblivious(Eq(s3, 1)) == False [statically known]
I0502 18:42:26.442000 635 torch/fx/experimental/symbolic_shapes.py:6630] [10/0] runtime_assert Eq(s3, 4) [guard added] x0 = x + y # [8, 4] # ar/lib/workspace/intermediate_source/torch_export_tutorial.py:268 in forward (_subclasses/fake_impls.py:881 in infer_size), for more info run with TORCHDYNAMO_EXTENDED_DEBUG_GUARD_ADDED="Eq(s3, 4)"
V0502 18:42:26.443000 635 torch/fx/experimental/symbolic_shapes.py:6071] [10/0] _update_var_to_range s3 = VR[4, 4] (update)
I0502 18:42:26.443000 635 torch/fx/experimental/symbolic_shapes.py:6234] [10/0] set_replacement s3 = 4 (range_refined_to_singleton) VR[4, 4]
V0502 18:42:26.445000 635 torch/fx/experimental/symbolic_shapes.py:6787] [10/0] eval size_oblivious(Ne(s2, 1)) == True [statically known]
I0502 18:42:26.451000 635 torch/fx/experimental/symbolic_shapes.py:6630] [10/0] runtime_assert Eq(s1, 5) [guard added] x1 = self.l(w) # [6, 3] # ar/lib/workspace/intermediate_source/torch_export_tutorial.py:269 in forward (_meta_registrations.py:2236 in meta_mm), for more info run with TORCHDYNAMO_EXTENDED_DEBUG_GUARD_ADDED="Eq(s1, 5)"
V0502 18:42:26.452000 635 torch/fx/experimental/symbolic_shapes.py:6071] [10/0] _update_var_to_range s1 = VR[5, 5] (update)
I0502 18:42:26.453000 635 torch/fx/experimental/symbolic_shapes.py:6234] [10/0] set_replacement s1 = 5 (range_refined_to_singleton) VR[5, 5]
V0502 18:42:26.454000 635 torch/fx/experimental/symbolic_shapes.py:6787] [10/0] eval size_oblivious(Eq(s0, 1)) == False [statically known]
V0502 18:42:26.455000 635 torch/fx/experimental/symbolic_shapes.py:7018] [10/0] runtime_assert True == True [statically known]
V0502 18:42:26.461000 635 torch/fx/experimental/symbolic_shapes.py:6787] [10/0] eval size_oblivious(Eq(s4, 1)) == False [statically known]
I0502 18:42:26.465000 635 torch/fx/experimental/symbolic_shapes.py:6630] [10/0] runtime_assert Eq(4*s2, s4) [guard added] x3 = x2 + z # [32] # ar/lib/workspace/intermediate_source/torch_export_tutorial.py:271 in forward (_subclasses/fake_impls.py:881 in infer_size), for more info run with TORCHDYNAMO_EXTENDED_DEBUG_GUARD_ADDED="Eq(4*s2, s4)"
V0502 18:42:26.467000 635 torch/fx/experimental/symbolic_shapes.py:6071] [10/0] _update_var_to_range s4 = VR[8, int_oo] (update)
I0502 18:42:26.469000 635 torch/fx/experimental/symbolic_shapes.py:6234] [10/0] set_replacement s4 = 4*s2 (solve) VR[8, int_oo]
I0502 18:42:26.476000 635 torch/fx/experimental/symbolic_shapes.py:4734] [10/0] produce_guards
V0502 18:42:26.476000 635 torch/fx/experimental/symbolic_shapes.py:4954] [10/0] track_symint L['w'].size()[0] s0 None
V0502 18:42:26.476000 635 torch/fx/experimental/symbolic_shapes.py:4954] [10/0] track_symint L['w'].size()[1] 5 None
V0502 18:42:26.477000 635 torch/fx/experimental/symbolic_shapes.py:4954] [10/0] track_symint L['w'].stride()[0] 5 None
V0502 18:42:26.477000 635 torch/fx/experimental/symbolic_shapes.py:4954] [10/0] track_symint L['w'].stride()[1] 1 None
V0502 18:42:26.477000 635 torch/fx/experimental/symbolic_shapes.py:4954] [10/0] track_symint L['w'].storage_offset() 0 None
V0502 18:42:26.478000 635 torch/fx/experimental/symbolic_shapes.py:4954] [10/0] track_symint L['x'].size()[0] 4 None
V0502 18:42:26.478000 635 torch/fx/experimental/symbolic_shapes.py:4954] [10/0] track_symint L['x'].stride()[0] 1 None
V0502 18:42:26.478000 635 torch/fx/experimental/symbolic_shapes.py:4954] [10/0] track_symint L['x'].storage_offset() 0 None
V0502 18:42:26.479000 635 torch/fx/experimental/symbolic_shapes.py:4954] [10/0] track_symint L['y'].size()[0] s2 None
V0502 18:42:26.479000 635 torch/fx/experimental/symbolic_shapes.py:4954] [10/0] track_symint L['y'].size()[1] 4 RelaxedUnspecConstraint(warn_only=False)
V0502 18:42:26.480000 635 torch/fx/experimental/symbolic_shapes.py:4954] [10/0] track_symint L['y'].stride()[0] 4 None
V0502 18:42:26.480000 635 torch/fx/experimental/symbolic_shapes.py:4954] [10/0] track_symint L['y'].stride()[1] 1 None
V0502 18:42:26.480000 635 torch/fx/experimental/symbolic_shapes.py:4954] [10/0] track_symint L['y'].storage_offset() 0 None
V0502 18:42:26.480000 635 torch/fx/experimental/symbolic_shapes.py:4954] [10/0] track_symint L['z'].size()[0] 4*s2 None
V0502 18:42:26.481000 635 torch/fx/experimental/symbolic_shapes.py:4954] [10/0] track_symint L['z'].stride()[0] 1 None
V0502 18:42:26.481000 635 torch/fx/experimental/symbolic_shapes.py:4954] [10/0] track_symint L['z'].storage_offset() 0 None
E0502 18:42:26.483000 635 torch/_guards.py:359] [10/0] Error while creating guard:
E0502 18:42:26.483000 635 torch/_guards.py:359] [10/0] Name: ''
E0502 18:42:26.483000 635 torch/_guards.py:359] [10/0] Source: shape_env
E0502 18:42:26.483000 635 torch/_guards.py:359] [10/0] Create Function: SHAPE_ENV
E0502 18:42:26.483000 635 torch/_guards.py:359] [10/0] Guard Types: None
E0502 18:42:26.483000 635 torch/_guards.py:359] [10/0] Code List: None
E0502 18:42:26.483000 635 torch/_guards.py:359] [10/0] Object Weakref: None
E0502 18:42:26.483000 635 torch/_guards.py:359] [10/0] Guarded Class Weakref: None
E0502 18:42:26.483000 635 torch/_guards.py:359] [10/0] Traceback (most recent call last):
E0502 18:42:26.483000 635 torch/_guards.py:359] [10/0] File "/usr/local/lib/python3.10/dist-packages/torch/_guards.py", line 357, in create
E0502 18:42:26.483000 635 torch/_guards.py:359] [10/0] return self.create_fn(builder, self)
E0502 18:42:26.483000 635 torch/_guards.py:359] [10/0] File "/usr/local/lib/python3.10/dist-packages/torch/_dynamo/guards.py", line 1959, in SHAPE_ENV
E0502 18:42:26.483000 635 torch/_guards.py:359] [10/0] python_code_parts, verbose_code_parts = _get_code_parts(
E0502 18:42:26.483000 635 torch/_guards.py:359] [10/0] File "/usr/local/lib/python3.10/dist-packages/torch/_dynamo/guards.py", line 1942, in _get_code_parts
E0502 18:42:26.483000 635 torch/_guards.py:359] [10/0] return output_graph.shape_env.produce_guards_verbose(
E0502 18:42:26.483000 635 torch/_guards.py:359] [10/0] File "/usr/local/lib/python3.10/dist-packages/torch/fx/experimental/symbolic_shapes.py", line 5409, in produce_guards_verbose
E0502 18:42:26.483000 635 torch/_guards.py:359] [10/0] raise ConstraintViolationError(
E0502 18:42:26.483000 635 torch/_guards.py:359] [10/0] torch.fx.experimental.symbolic_shapes.ConstraintViolationError: Constraints violated (L['y'].size()[1])! For more information, run with TORCH_LOGS="+dynamic".
E0502 18:42:26.483000 635 torch/_guards.py:359] [10/0] - Not all values of RelaxedUnspecConstraint(L['y'].size()[1]) are valid because L['y'].size()[1] was inferred to be a constant (4).
E0502 18:42:26.484000 635 torch/_guards.py:361] [10/0] Created at:
E0502 18:42:26.484000 635 torch/_guards.py:361] [10/0] File "/usr/local/lib/python3.10/dist-packages/torch/_dynamo/convert_frame.py", line 694, in transform
E0502 18:42:26.484000 635 torch/_guards.py:361] [10/0] tracer = InstructionTranslator(
E0502 18:42:26.484000 635 torch/_guards.py:361] [10/0] File "/usr/local/lib/python3.10/dist-packages/torch/_dynamo/symbolic_convert.py", line 3329, in __init__
E0502 18:42:26.484000 635 torch/_guards.py:361] [10/0] output=OutputGraph(
E0502 18:42:26.484000 635 torch/_guards.py:361] [10/0] File "/usr/local/lib/python3.10/dist-packages/torch/_dynamo/output_graph.py", line 358, in __init__
E0502 18:42:26.484000 635 torch/_guards.py:361] [10/0] self.init_ambient_guards()
E0502 18:42:26.484000 635 torch/_guards.py:361] [10/0] File "/usr/local/lib/python3.10/dist-packages/torch/_dynamo/output_graph.py", line 512, in init_ambient_guards
E0502 18:42:26.484000 635 torch/_guards.py:361] [10/0] self.guards.add(ShapeEnvSource().make_guard(GuardBuilder.SHAPE_ENV))
Traceback (most recent call last):
File "/usr/local/lib/python3.10/dist-packages/torch/export/_trace.py", line 739, in _export_to_torch_ir
gm_torch_level, _ = torch._dynamo.export(
File "/usr/local/lib/python3.10/dist-packages/torch/_dynamo/eval_frame.py", line 1722, in inner
raise constraint_violation_error
File "/usr/local/lib/python3.10/dist-packages/torch/_dynamo/eval_frame.py", line 1677, in inner
result_traced = opt_f(*args, **kwargs)
File "/usr/local/lib/python3.10/dist-packages/torch/nn/modules/module.py", line 1751, in _wrapped_call_impl
return self._call_impl(*args, **kwargs)
File "/usr/local/lib/python3.10/dist-packages/torch/nn/modules/module.py", line 1762, in _call_impl
return forward_call(*args, **kwargs)
File "/usr/local/lib/python3.10/dist-packages/torch/_dynamo/eval_frame.py", line 655, in _fn
return fn(*args, **kwargs)
File "/usr/local/lib/python3.10/dist-packages/torch/nn/modules/module.py", line 1751, in _wrapped_call_impl
return self._call_impl(*args, **kwargs)
File "/usr/local/lib/python3.10/dist-packages/torch/nn/modules/module.py", line 1762, in _call_impl
return forward_call(*args, **kwargs)
File "/usr/local/lib/python3.10/dist-packages/torch/_dynamo/convert_frame.py", line 1432, in __call__
return self._torchdynamo_orig_callable(
File "/usr/local/lib/python3.10/dist-packages/torch/_dynamo/convert_frame.py", line 598, in __call__
return _compile(
File "/usr/local/lib/python3.10/dist-packages/torch/_dynamo/convert_frame.py", line 1059, in _compile
guarded_code = compile_inner(code, one_graph, hooks, transform)
File "/usr/local/lib/python3.10/dist-packages/torch/_utils_internal.py", line 97, in wrapper_function
return function(*args, **kwargs)
File "/usr/local/lib/python3.10/dist-packages/torch/_dynamo/convert_frame.py", line 761, in compile_inner
return _compile_inner(code, one_graph, hooks, transform)
File "/usr/local/lib/python3.10/dist-packages/torch/_dynamo/convert_frame.py", line 906, in _compile_inner
check_fn = CheckFunctionManager(
File "/usr/local/lib/python3.10/dist-packages/torch/_dynamo/guards.py", line 2481, in __init__
guard.create(builder)
File "/usr/local/lib/python3.10/dist-packages/torch/_guards.py", line 357, in create
return self.create_fn(builder, self)
File "/usr/local/lib/python3.10/dist-packages/torch/_dynamo/guards.py", line 1959, in SHAPE_ENV
python_code_parts, verbose_code_parts = _get_code_parts(
File "/usr/local/lib/python3.10/dist-packages/torch/_dynamo/guards.py", line 1942, in _get_code_parts
return output_graph.shape_env.produce_guards_verbose(
File "/usr/local/lib/python3.10/dist-packages/torch/fx/experimental/symbolic_shapes.py", line 5409, in produce_guards_verbose
raise ConstraintViolationError(
torch.fx.experimental.symbolic_shapes.ConstraintViolationError: Constraints violated (L['y'].size()[1])! For more information, run with TORCH_LOGS="+dynamic".
- Not all values of RelaxedUnspecConstraint(L['y'].size()[1]) are valid because L['y'].size()[1] was inferred to be a constant (4).
During handling of the above exception, another exception occurred:
Traceback (most recent call last):
File "/var/lib/workspace/intermediate_source/torch_export_tutorial.py", line 431, in <module>
export(model, (w, x, y, z), dynamic_shapes=dynamic_shapes)
File "/usr/local/lib/python3.10/dist-packages/torch/export/__init__.py", line 360, in export
return _export(
File "/usr/local/lib/python3.10/dist-packages/torch/export/_trace.py", line 1092, in wrapper
raise e
File "/usr/local/lib/python3.10/dist-packages/torch/export/_trace.py", line 1065, in wrapper
ep = fn(*args, **kwargs)
File "/usr/local/lib/python3.10/dist-packages/torch/export/exported_program.py", line 121, in wrapper
return fn(*args, **kwargs)
File "/usr/local/lib/python3.10/dist-packages/torch/export/_trace.py", line 2112, in _export
ep = _export_for_training(
File "/usr/local/lib/python3.10/dist-packages/torch/export/_trace.py", line 1092, in wrapper
raise e
File "/usr/local/lib/python3.10/dist-packages/torch/export/_trace.py", line 1065, in wrapper
ep = fn(*args, **kwargs)
File "/usr/local/lib/python3.10/dist-packages/torch/export/exported_program.py", line 121, in wrapper
return fn(*args, **kwargs)
File "/usr/local/lib/python3.10/dist-packages/torch/export/_trace.py", line 1975, in _export_for_training
export_artifact = export_func( # type: ignore[operator]
File "/usr/local/lib/python3.10/dist-packages/torch/export/_trace.py", line 1344, in _strict_export_lower_to_aten_ir
gm_torch_level = _export_to_torch_ir(
File "/usr/local/lib/python3.10/dist-packages/torch/export/_trace.py", line 756, in _export_to_torch_ir
raise UserError(UserErrorType.CONSTRAINT_VIOLATION, str(e)) # noqa: B904
torch._dynamo.exc.UserError: Constraints violated (L['y'].size()[1])! For more information, run with TORCH_LOGS="+dynamic".
- Not all values of RelaxedUnspecConstraint(L['y'].size()[1]) are valid because L['y'].size()[1] was inferred to be a constant (4).
Here you might ask why export “specializes”, i.e. why we resolve this static/dynamic conflict by going with the static route. The answer is because
of the symbolic shapes system described above, of symbols and guards. When x.shape[0] is marked static, we don’t allocate a symbol, and compile
treating this shape as a concrete integer 4. A symbol is allocated for y.shape[1], and so we finally emit the guard s3 == 4, leading to
specialization.
One feature of export is that during tracing, statements like asserts, torch._check(), and if/else conditions will also emit guards.
See what happens when we augment the existing model with such statements:
class DynamicModel(torch.nn.Module):
def __init__(self):
super().__init__()
self.l = torch.nn.Linear(5, 3)
def forward(self, w, x, y, z):
assert w.shape[0] <= 512
torch._check(x.shape[0] >= 4)
if w.shape[0] == x.shape[0] + 2:
x0 = x + y
x1 = self.l(w)
x2 = x0.flatten()
x3 = x2 + z
return x1, x3
else:
return w
dynamic_shapes = {
"w": (Dim.AUTO, Dim.AUTO),
"x": (Dim.AUTO,),
"y": (Dim.AUTO, Dim.AUTO),
"z": (Dim.AUTO,),
}
try:
ep = export(DynamicModel(), (w, x, y, z), dynamic_shapes=dynamic_shapes)
except Exception:
tb.print_exc()
I0502 18:42:26.501000 635 torch/fx/experimental/symbolic_shapes.py:3334] [11/0] create_env
I0502 18:42:26.503000 635 torch/fx/experimental/symbolic_shapes.py:4606] [11/0] create_symbol s0 = 6 for L['w'].size()[0] [2, int_oo] (_dynamo/variables/builder.py:3033 in <lambda>), for more info run with TORCHDYNAMO_EXTENDED_DEBUG_CREATE_SYMBOL="s0" or to suppress this message run with TORCHDYNAMO_EXTENDED_ADVICE="0"
I0502 18:42:26.503000 635 torch/fx/experimental/symbolic_shapes.py:4606] [11/0] create_symbol s1 = 5 for L['w'].size()[1] [2, int_oo] (_dynamo/variables/builder.py:3033 in <lambda>), for more info run with TORCHDYNAMO_EXTENDED_DEBUG_CREATE_SYMBOL="s1" or to suppress this message run with TORCHDYNAMO_EXTENDED_ADVICE="0"
V0502 18:42:26.504000 635 torch/fx/experimental/symbolic_shapes.py:7018] [11/0] runtime_assert True == True [statically known]
I0502 18:42:26.507000 635 torch/fx/experimental/symbolic_shapes.py:4606] [11/0] create_symbol s2 = 4 for L['x'].size()[0] [2, int_oo] (_dynamo/variables/builder.py:3033 in <lambda>), for more info run with TORCHDYNAMO_EXTENDED_DEBUG_CREATE_SYMBOL="s2" or to suppress this message run with TORCHDYNAMO_EXTENDED_ADVICE="0"
I0502 18:42:26.509000 635 torch/fx/experimental/symbolic_shapes.py:4606] [11/0] create_symbol s3 = 8 for L['y'].size()[0] [2, int_oo] (_dynamo/variables/builder.py:3033 in <lambda>), for more info run with TORCHDYNAMO_EXTENDED_DEBUG_CREATE_SYMBOL="s3" or to suppress this message run with TORCHDYNAMO_EXTENDED_ADVICE="0"
I0502 18:42:26.509000 635 torch/fx/experimental/symbolic_shapes.py:4606] [11/0] create_symbol s4 = 4 for L['y'].size()[1] [2, int_oo] (_dynamo/variables/builder.py:3033 in <lambda>), for more info run with TORCHDYNAMO_EXTENDED_DEBUG_CREATE_SYMBOL="s4" or to suppress this message run with TORCHDYNAMO_EXTENDED_ADVICE="0"
I0502 18:42:26.512000 635 torch/fx/experimental/symbolic_shapes.py:4606] [11/0] create_symbol s5 = 32 for L['z'].size()[0] [2, int_oo] (_dynamo/variables/builder.py:3033 in <lambda>), for more info run with TORCHDYNAMO_EXTENDED_DEBUG_CREATE_SYMBOL="s5" or to suppress this message run with TORCHDYNAMO_EXTENDED_ADVICE="0"
I0502 18:42:26.518000 635 torch/fx/experimental/symbolic_shapes.py:6630] [11/0] runtime_assert s0 <= 512 [guard added] assert w.shape[0] <= 512 # ar/lib/workspace/intermediate_source/torch_export_tutorial.py:450 in forward (_dynamo/symbolic_convert.py:669 in inner), for more info run with TORCHDYNAMO_EXTENDED_DEBUG_GUARD_ADDED="s0 <= 512"
V0502 18:42:26.519000 635 torch/fx/experimental/symbolic_shapes.py:6071] [11/0] _update_var_to_range s0 = VR[2, 512] (update)
I0502 18:42:26.524000 635 torch/fx/experimental/symbolic_shapes.py:6630] [11/0] runtime_assert s2 >= 4 [guard added] torch._check(x.shape[0] >= 4) # ar/lib/workspace/intermediate_source/torch_export_tutorial.py:451 in forward (_dynamo/utils.py:3284 in run_node), for more info run with TORCHDYNAMO_EXTENDED_DEBUG_GUARD_ADDED="s2 >= 4"
V0502 18:42:26.524000 635 torch/fx/experimental/symbolic_shapes.py:6071] [11/0] _update_var_to_range s2 = VR[4, int_oo] (update)
I0502 18:42:26.530000 635 torch/fx/experimental/symbolic_shapes.py:6630] [11/0] eval Eq(s0, s2 + 2) [guard added] if w.shape[0] == x.shape[0] + 2: # ar/lib/workspace/intermediate_source/torch_export_tutorial.py:452 in forward (_dynamo/variables/tensor.py:1245 in evaluate_expr), for more info run with TORCHDYNAMO_EXTENDED_DEBUG_GUARD_ADDED="Eq(s0, s2 + 2)"
V0502 18:42:26.531000 635 torch/fx/experimental/symbolic_shapes.py:6071] [11/0] _update_var_to_range s0 = VR[6, 512] (update)
V0502 18:42:26.534000 635 torch/fx/experimental/symbolic_shapes.py:6071] [11/0] _update_var_to_range s2 = VR[4, 510] (update)
I0502 18:42:26.534000 635 torch/fx/experimental/symbolic_shapes.py:6234] [11/0] set_replacement s0 = s2 + 2 (solve) VR[6, 512]
V0502 18:42:26.536000 635 torch/fx/experimental/symbolic_shapes.py:6787] [11/0] eval size_oblivious(Eq(s2, 1)) == False [statically known]
V0502 18:42:26.537000 635 torch/fx/experimental/symbolic_shapes.py:7018] [11/0] runtime_assert True == True [statically known]
V0502 18:42:26.538000 635 torch/fx/experimental/symbolic_shapes.py:6787] [11/0] eval size_oblivious(Eq(s4, 1)) == False [statically known]
I0502 18:42:26.540000 635 torch/fx/experimental/symbolic_shapes.py:6630] [11/0] runtime_assert Eq(s2, s4) [guard added] x0 = x + y # ar/lib/workspace/intermediate_source/torch_export_tutorial.py:453 in forward (_subclasses/fake_impls.py:881 in infer_size), for more info run with TORCHDYNAMO_EXTENDED_DEBUG_GUARD_ADDED="Eq(s2, s4)"
V0502 18:42:26.541000 635 torch/fx/experimental/symbolic_shapes.py:6071] [11/0] _update_var_to_range s4 = VR[4, 510] (update)
I0502 18:42:26.542000 635 torch/fx/experimental/symbolic_shapes.py:6234] [11/0] set_replacement s4 = s2 (solve) VR[4, 510]
V0502 18:42:26.543000 635 torch/fx/experimental/symbolic_shapes.py:6787] [11/0] eval size_oblivious(Ne(s2, 1)) == True [statically known]
V0502 18:42:26.544000 635 torch/fx/experimental/symbolic_shapes.py:6787] [11/0] eval size_oblivious(Ne(s3, 1)) == True [statically known]
I0502 18:42:26.551000 635 torch/fx/experimental/symbolic_shapes.py:6630] [11/0] runtime_assert Eq(s1, 5) [guard added] x1 = self.l(w) # ar/lib/workspace/intermediate_source/torch_export_tutorial.py:454 in forward (_meta_registrations.py:2236 in meta_mm), for more info run with TORCHDYNAMO_EXTENDED_DEBUG_GUARD_ADDED="Eq(s1, 5)"
V0502 18:42:26.552000 635 torch/fx/experimental/symbolic_shapes.py:6071] [11/0] _update_var_to_range s1 = VR[5, 5] (update)
I0502 18:42:26.552000 635 torch/fx/experimental/symbolic_shapes.py:6234] [11/0] set_replacement s1 = 5 (range_refined_to_singleton) VR[5, 5]
V0502 18:42:26.562000 635 torch/fx/experimental/symbolic_shapes.py:6787] [11/0] eval size_oblivious(Eq(s2*s3, 1)) == False [statically known]
V0502 18:42:26.563000 635 torch/fx/experimental/symbolic_shapes.py:6787] [11/0] eval size_oblivious(Eq(s5, 1)) == False [statically known]
I0502 18:42:26.569000 635 torch/fx/experimental/symbolic_shapes.py:6630] [11/0] runtime_assert Eq(s2*s3, s5) [guard added] x3 = x2 + z # ar/lib/workspace/intermediate_source/torch_export_tutorial.py:456 in forward (_subclasses/fake_impls.py:881 in infer_size), for more info run with TORCHDYNAMO_EXTENDED_DEBUG_GUARD_ADDED="Eq(s2*s3, s5)"
V0502 18:42:26.570000 635 torch/fx/experimental/symbolic_shapes.py:6071] [11/0] _update_var_to_range s5 = VR[8, int_oo] (update)
I0502 18:42:26.572000 635 torch/fx/experimental/symbolic_shapes.py:6234] [11/0] set_replacement s5 = s2*s3 (solve) VR[8, int_oo]
V0502 18:42:26.573000 635 torch/fx/experimental/symbolic_shapes.py:6787] [11/0] eval size_oblivious(Ne(s2*s3, 1)) == True [statically known]
V0502 18:42:26.577000 635 torch/fx/experimental/symbolic_shapes.py:7018] [11/0] runtime_assert s2 >= 4 == True [statically known]
I0502 18:42:26.582000 635 torch/fx/experimental/symbolic_shapes.py:4734] [11/0] produce_guards
V0502 18:42:26.583000 635 torch/fx/experimental/symbolic_shapes.py:4954] [11/0] track_symint L['w'].size()[0] s2 + 2 None
V0502 18:42:26.583000 635 torch/fx/experimental/symbolic_shapes.py:4954] [11/0] track_symint L['w'].size()[1] 5 None
V0502 18:42:26.584000 635 torch/fx/experimental/symbolic_shapes.py:4954] [11/0] track_symint L['w'].stride()[0] 5 None
V0502 18:42:26.584000 635 torch/fx/experimental/symbolic_shapes.py:4954] [11/0] track_symint L['w'].stride()[1] 1 None
V0502 18:42:26.584000 635 torch/fx/experimental/symbolic_shapes.py:4954] [11/0] track_symint L['w'].storage_offset() 0 None
V0502 18:42:26.585000 635 torch/fx/experimental/symbolic_shapes.py:4954] [11/0] track_symint L['x'].size()[0] s2 None
V0502 18:42:26.585000 635 torch/fx/experimental/symbolic_shapes.py:4954] [11/0] track_symint L['x'].stride()[0] 1 None
V0502 18:42:26.585000 635 torch/fx/experimental/symbolic_shapes.py:4954] [11/0] track_symint L['x'].storage_offset() 0 None
V0502 18:42:26.586000 635 torch/fx/experimental/symbolic_shapes.py:4954] [11/0] track_symint L['y'].size()[0] s3 None
V0502 18:42:26.586000 635 torch/fx/experimental/symbolic_shapes.py:4954] [11/0] track_symint L['y'].size()[1] s2 None
V0502 18:42:26.586000 635 torch/fx/experimental/symbolic_shapes.py:4954] [11/0] track_symint L['y'].stride()[0] s2 None
V0502 18:42:26.587000 635 torch/fx/experimental/symbolic_shapes.py:4954] [11/0] track_symint L['y'].stride()[1] 1 None
V0502 18:42:26.587000 635 torch/fx/experimental/symbolic_shapes.py:4954] [11/0] track_symint L['y'].storage_offset() 0 None
V0502 18:42:26.587000 635 torch/fx/experimental/symbolic_shapes.py:4954] [11/0] track_symint L['z'].size()[0] s2*s3 None
V0502 18:42:26.588000 635 torch/fx/experimental/symbolic_shapes.py:4954] [11/0] track_symint L['z'].stride()[0] 1 None
V0502 18:42:26.588000 635 torch/fx/experimental/symbolic_shapes.py:4954] [11/0] track_symint L['z'].storage_offset() 0 None
Each of these statements emits an additional guard, and the exported program shows the changes; s0 is eliminated in favor of s2 + 2,
and s2 now contains lower and upper bounds, reflected in range_constraints.
For the if/else condition, you might ask why the True branch was taken, and why it wasn’t the w.shape[0] != x.shape[0] + 2 guard that
got emitted from tracing. The answer is that export is guided by the sample inputs provided by tracing, and specializes on the branches taken.
If different sample input shapes were provided that fail the if condition, export would trace and emit guards corresponding to the else branch.
Additionally, you might ask why we traced only the if branch, and if it’s possible to maintain control-flow in your program and keep both branches
alive. For that, refer to rewriting your model code following the Control Flow Ops section above.
0/1 specialization
Since we’re talking about guards and specializations, it’s a good time to talk about the 0/1 specialization issue we brought up earlier. The bottom line is that export will specialize on sample input dimensions with value 0 or 1, because these shapes have trace-time properties that don’t generalize to other shapes. For example, size 1 tensors can broadcast while other sizes fail; and size 0 … . This just means that you should specify 0/1 sample inputs when you’d like your program to hardcode them, and non-0/1 sample inputs when dynamic behavior is desirable. See what happens at runtime when we export this linear layer:
ep = export(
torch.nn.Linear(4, 3),
(torch.randn(1, 4),),
dynamic_shapes={
"input": (Dim.AUTO, Dim.STATIC),
},
)
try:
ep.module()(torch.randn(2, 4))
except Exception:
tb.print_exc()
I0502 18:42:26.650000 635 torch/fx/experimental/symbolic_shapes.py:3334] [12/0] create_env
I0502 18:42:26.664000 635 torch/fx/experimental/symbolic_shapes.py:4734] [12/0] produce_guards
V0502 18:42:26.664000 635 torch/fx/experimental/symbolic_shapes.py:4954] [12/0] track_symint L['args'][0].size()[0] 1 None
V0502 18:42:26.664000 635 torch/fx/experimental/symbolic_shapes.py:4954] [12/0] track_symint L['args'][0].size()[1] 4 None
V0502 18:42:26.665000 635 torch/fx/experimental/symbolic_shapes.py:4954] [12/0] track_symint L['args'][0].stride()[0] 4 None
V0502 18:42:26.665000 635 torch/fx/experimental/symbolic_shapes.py:4954] [12/0] track_symint L['args'][0].stride()[1] 1 None
V0502 18:42:26.665000 635 torch/fx/experimental/symbolic_shapes.py:4954] [12/0] track_symint L['args'][0].storage_offset() 0 None
Traceback (most recent call last):
File "/var/lib/workspace/intermediate_source/torch_export_tutorial.py", line 500, in <module>
ep.module()(torch.randn(2, 4))
File "/usr/local/lib/python3.10/dist-packages/torch/fx/graph_module.py", line 830, in call_wrapped
return self._wrapped_call(self, *args, **kwargs)
File "/usr/local/lib/python3.10/dist-packages/torch/fx/graph_module.py", line 406, in __call__
raise e
File "/usr/local/lib/python3.10/dist-packages/torch/fx/graph_module.py", line 393, in __call__
return super(self.cls, obj).__call__(*args, **kwargs) # type: ignore[misc]
File "/usr/local/lib/python3.10/dist-packages/torch/nn/modules/module.py", line 1751, in _wrapped_call_impl
return self._call_impl(*args, **kwargs)
File "/usr/local/lib/python3.10/dist-packages/torch/nn/modules/module.py", line 1857, in _call_impl
return inner()
File "/usr/local/lib/python3.10/dist-packages/torch/nn/modules/module.py", line 1784, in inner
args_kwargs_result = hook(self, args, kwargs) # type: ignore[misc]
File "/usr/local/lib/python3.10/dist-packages/torch/_dynamo/eval_frame.py", line 838, in _fn
return fn(*args, **kwargs)
File "/usr/local/lib/python3.10/dist-packages/torch/export/_unlift.py", line 55, in _check_input_constraints_pre_hook
_check_input_constraints_for_graph(
File "/usr/local/lib/python3.10/dist-packages/torch/_export/utils.py", line 398, in _check_input_constraints_for_graph
raise RuntimeError(
RuntimeError: Expected input at *args[0].shape[0] to be equal to 1, but got 2
Named Dims
So far we’ve only been talking about 3 ways to specify dynamic shapes: Dim.AUTO, Dim.DYNAMIC, and Dim.STATIC. The attraction of these is the
low-friction user experience; all the guards emitted during model tracing are adhered to, and dynamic behavior like min/max ranges, relations, and static/dynamic
dimensions are automatically figured out underneath export. The dynamic shapes subsystem essentially acts as a “discovery” process, summarizing these guards
and presenting what export believes is the overall dynamic behavior of the program. The drawback of this design appears once the user has stronger expectations or
beliefs about the dynamic behavior of these models - maybe there is a strong desire on dynamism and specializations on particular dimensions are to be avoided at
all costs, or maybe we just want to catch changes in dynamic behavior with changes to the original model code, or possibly underlying decompositions or meta-kernels.
These changes won’t be detected and the export() call will most likely succeed, unless tests are in place that check the resulting ExportedProgram representation.
For such cases, our stance is to recommend the “traditional” way of specifying dynamic shapes, which longer-term users of export might be familiar with: named Dims:
This style of dynamic shapes allows the user to specify what symbols are allocated for input dimensions, min/max bounds on those symbols, and places restrictions on the
dynamic behavior of the ExportedProgram produced; ConstraintViolation errors will be raised if model tracing emits guards that conflict with the relations or static/dynamic
specifications given. For example, in the above specification, the following is asserted:
x.shape[0]is to have range[4, 256], and related toy.shape[0]byy.shape[0] == 2 * x.shape[0].x.shape[1]is static.y.shape[1]has range[2, 512], and is unrelated to any other dimension.
In this design, we allow relations between dimensions to be specified with univariate linear expressions: A * dim + B can be specified for any dimension. This allows users
to specify more complex constraints like integer divisibility for dynamic dimensions:
dx = Dim("dx", min=4, max=512)
dynamic_shapes = {
"x": (4 * dx, None) # x.shape[0] has range [16, 2048], and is divisible by 4.
}
Constraint violations, suggested fixes
One common issue with this specification style (before Dim.AUTO was introduced), is that the specification would often be mismatched with what was produced by model tracing.
That would lead to ConstraintViolation errors and export suggested fixes - see for example with this model & specification, where the model inherently requires equality between
dimensions 0 of x and y, and requires dimension 1 to be static.
class Foo(torch.nn.Module):
def forward(self, x, y):
w = x + y
return w + torch.ones(4)
dx, dy, d1 = torch.export.dims("dx", "dy", "d1")
try:
ep = export(
Foo(),
(torch.randn(6, 4), torch.randn(6, 4)),
dynamic_shapes={
"x": (dx, d1),
"y": (dy, d1),
},
)
except Exception:
tb.print_exc()
I0502 18:42:26.769000 635 torch/fx/experimental/symbolic_shapes.py:3334] [13/0] create_env
I0502 18:42:26.771000 635 torch/fx/experimental/symbolic_shapes.py:4606] [13/0] create_symbol s0 = 6 for L['x'].size()[0] [2, int_oo] (_dynamo/variables/builder.py:3033 in <lambda>), for more info run with TORCHDYNAMO_EXTENDED_DEBUG_CREATE_SYMBOL="s0" or to suppress this message run with TORCHDYNAMO_EXTENDED_ADVICE="0"
I0502 18:42:26.772000 635 torch/fx/experimental/symbolic_shapes.py:4606] [13/0] create_symbol s1 = 4 for L['x'].size()[1] [2, int_oo] (_dynamo/variables/builder.py:3033 in <lambda>), for more info run with TORCHDYNAMO_EXTENDED_DEBUG_CREATE_SYMBOL="s1" or to suppress this message run with TORCHDYNAMO_EXTENDED_ADVICE="0"
V0502 18:42:26.773000 635 torch/fx/experimental/symbolic_shapes.py:7018] [13/0] runtime_assert True == True [statically known]
I0502 18:42:26.776000 635 torch/fx/experimental/symbolic_shapes.py:4606] [13/0] create_symbol s2 = 6 for L['y'].size()[0] [2, int_oo] (_dynamo/variables/builder.py:3033 in <lambda>), for more info run with TORCHDYNAMO_EXTENDED_DEBUG_CREATE_SYMBOL="s2" or to suppress this message run with TORCHDYNAMO_EXTENDED_ADVICE="0"
I0502 18:42:26.776000 635 torch/fx/experimental/symbolic_shapes.py:4606] [13/0] create_symbol s3 = 4 for L['y'].size()[1] [2, int_oo] (_dynamo/variables/builder.py:3033 in <lambda>), for more info run with TORCHDYNAMO_EXTENDED_DEBUG_CREATE_SYMBOL="s3" or to suppress this message run with TORCHDYNAMO_EXTENDED_ADVICE="0"
V0502 18:42:26.780000 635 torch/fx/experimental/symbolic_shapes.py:6787] [13/0] eval size_oblivious(Eq(s1, 1)) == False [statically known]
V0502 18:42:26.781000 635 torch/fx/experimental/symbolic_shapes.py:7018] [13/0] runtime_assert True == True [statically known]
V0502 18:42:26.782000 635 torch/fx/experimental/symbolic_shapes.py:6787] [13/0] eval size_oblivious(Eq(s0, 1)) == False [statically known]
V0502 18:42:26.783000 635 torch/fx/experimental/symbolic_shapes.py:6787] [13/0] eval size_oblivious(Eq(s3, 1)) == False [statically known]
I0502 18:42:26.784000 635 torch/fx/experimental/symbolic_shapes.py:6630] [13/0] runtime_assert Eq(s1, s3) [guard added] w = x + y # ar/lib/workspace/intermediate_source/torch_export_tutorial.py:552 in forward (_subclasses/fake_impls.py:881 in infer_size), for more info run with TORCHDYNAMO_EXTENDED_DEBUG_GUARD_ADDED="Eq(s1, s3)"
I0502 18:42:26.786000 635 torch/fx/experimental/symbolic_shapes.py:6234] [13/0] set_replacement s3 = s1 (solve) VR[2, int_oo]
V0502 18:42:26.787000 635 torch/fx/experimental/symbolic_shapes.py:6787] [13/0] eval size_oblivious(Eq(s2, 1)) == False [statically known]
I0502 18:42:26.788000 635 torch/fx/experimental/symbolic_shapes.py:6630] [13/0] runtime_assert Eq(s0, s2) [guard added] w = x + y # ar/lib/workspace/intermediate_source/torch_export_tutorial.py:552 in forward (_subclasses/fake_impls.py:881 in infer_size), for more info run with TORCHDYNAMO_EXTENDED_DEBUG_GUARD_ADDED="Eq(s0, s2)"
I0502 18:42:26.789000 635 torch/fx/experimental/symbolic_shapes.py:6234] [13/0] set_replacement s2 = s0 (solve) VR[2, int_oo]
V0502 18:42:26.791000 635 torch/fx/experimental/symbolic_shapes.py:6787] [13/0] eval size_oblivious(Ne(s1, 1)) == True [statically known]
V0502 18:42:26.792000 635 torch/fx/experimental/symbolic_shapes.py:6787] [13/0] eval size_oblivious(Ne(s0, 1)) == True [statically known]
I0502 18:42:26.799000 635 torch/fx/experimental/symbolic_shapes.py:6630] [13/0] runtime_assert Eq(s1, 4) [guard added] return w + torch.ones(4) # ar/lib/workspace/intermediate_source/torch_export_tutorial.py:553 in forward (_subclasses/fake_impls.py:881 in infer_size), for more info run with TORCHDYNAMO_EXTENDED_DEBUG_GUARD_ADDED="Eq(s1, 4)"
V0502 18:42:26.799000 635 torch/fx/experimental/symbolic_shapes.py:6071] [13/0] _update_var_to_range s1 = VR[4, 4] (update)
I0502 18:42:26.800000 635 torch/fx/experimental/symbolic_shapes.py:6234] [13/0] set_replacement s1 = 4 (range_refined_to_singleton) VR[4, 4]
V0502 18:42:26.804000 635 torch/fx/experimental/symbolic_shapes.py:6071] [13/0] _update_var_to_range s3 = VR[4, 4] (update)
I0502 18:42:26.804000 635 torch/fx/experimental/symbolic_shapes.py:6234] [13/0] set_replacement s3 = 4 (find) VR[4, 4]
I0502 18:42:26.807000 635 torch/fx/experimental/symbolic_shapes.py:4734] [13/0] produce_guards
V0502 18:42:26.808000 635 torch/fx/experimental/symbolic_shapes.py:4954] [13/0] track_symint L['x'].size()[0] s0 StrictMinMaxConstraint(warn_only=False, vr=VR[0, int_oo])
V0502 18:42:26.808000 635 torch/fx/experimental/symbolic_shapes.py:4954] [13/0] track_symint L['x'].size()[1] 4 StrictMinMaxConstraint(warn_only=False, vr=VR[0, int_oo])
V0502 18:42:26.808000 635 torch/fx/experimental/symbolic_shapes.py:4954] [13/0] track_symint L['x'].stride()[0] 4 None
V0502 18:42:26.809000 635 torch/fx/experimental/symbolic_shapes.py:4954] [13/0] track_symint L['x'].stride()[1] 1 None
V0502 18:42:26.809000 635 torch/fx/experimental/symbolic_shapes.py:4954] [13/0] track_symint L['x'].storage_offset() 0 None
V0502 18:42:26.809000 635 torch/fx/experimental/symbolic_shapes.py:4954] [13/0] track_symint L['y'].size()[0] s0 StrictMinMaxConstraint(warn_only=False, vr=VR[0, int_oo])
V0502 18:42:26.810000 635 torch/fx/experimental/symbolic_shapes.py:4954] [13/0] track_symint L['y'].size()[1] 4 StrictMinMaxConstraint(warn_only=False, vr=VR[0, int_oo])
V0502 18:42:26.810000 635 torch/fx/experimental/symbolic_shapes.py:4954] [13/0] track_symint L['y'].stride()[0] 4 None
V0502 18:42:26.811000 635 torch/fx/experimental/symbolic_shapes.py:4954] [13/0] track_symint L['y'].stride()[1] 1 None
V0502 18:42:26.811000 635 torch/fx/experimental/symbolic_shapes.py:4954] [13/0] track_symint L['y'].storage_offset() 0 None
E0502 18:42:26.812000 635 torch/_guards.py:359] [13/0] Error while creating guard:
E0502 18:42:26.812000 635 torch/_guards.py:359] [13/0] Name: ''
E0502 18:42:26.812000 635 torch/_guards.py:359] [13/0] Source: shape_env
E0502 18:42:26.812000 635 torch/_guards.py:359] [13/0] Create Function: SHAPE_ENV
E0502 18:42:26.812000 635 torch/_guards.py:359] [13/0] Guard Types: None
E0502 18:42:26.812000 635 torch/_guards.py:359] [13/0] Code List: None
E0502 18:42:26.812000 635 torch/_guards.py:359] [13/0] Object Weakref: None
E0502 18:42:26.812000 635 torch/_guards.py:359] [13/0] Guarded Class Weakref: None
E0502 18:42:26.812000 635 torch/_guards.py:359] [13/0] Traceback (most recent call last):
E0502 18:42:26.812000 635 torch/_guards.py:359] [13/0] File "/usr/local/lib/python3.10/dist-packages/torch/_guards.py", line 357, in create
E0502 18:42:26.812000 635 torch/_guards.py:359] [13/0] return self.create_fn(builder, self)
E0502 18:42:26.812000 635 torch/_guards.py:359] [13/0] File "/usr/local/lib/python3.10/dist-packages/torch/_dynamo/guards.py", line 1959, in SHAPE_ENV
E0502 18:42:26.812000 635 torch/_guards.py:359] [13/0] python_code_parts, verbose_code_parts = _get_code_parts(
E0502 18:42:26.812000 635 torch/_guards.py:359] [13/0] File "/usr/local/lib/python3.10/dist-packages/torch/_dynamo/guards.py", line 1942, in _get_code_parts
E0502 18:42:26.812000 635 torch/_guards.py:359] [13/0] return output_graph.shape_env.produce_guards_verbose(
E0502 18:42:26.812000 635 torch/_guards.py:359] [13/0] File "/usr/local/lib/python3.10/dist-packages/torch/fx/experimental/symbolic_shapes.py", line 5409, in produce_guards_verbose
E0502 18:42:26.812000 635 torch/_guards.py:359] [13/0] raise ConstraintViolationError(
E0502 18:42:26.812000 635 torch/_guards.py:359] [13/0] torch.fx.experimental.symbolic_shapes.ConstraintViolationError: Constraints violated (d1, dy)! For more information, run with TORCH_LOGS="+dynamic".
E0502 18:42:26.812000 635 torch/_guards.py:359] [13/0] - Not all values of d1 = L['x'].size()[1] in the specified range are valid because d1 was inferred to be a constant (4).
E0502 18:42:26.812000 635 torch/_guards.py:359] [13/0] - Not all values of d1 = L['y'].size()[1] in the specified range are valid because d1 was inferred to be a constant (4).
E0502 18:42:26.812000 635 torch/_guards.py:359] [13/0] - The values of dy = L['y'].size()[0] and dx = L['x'].size()[0] must always be equal.
E0502 18:42:26.814000 635 torch/_guards.py:361] [13/0] Created at:
E0502 18:42:26.814000 635 torch/_guards.py:361] [13/0] File "/usr/local/lib/python3.10/dist-packages/torch/_dynamo/convert_frame.py", line 694, in transform
E0502 18:42:26.814000 635 torch/_guards.py:361] [13/0] tracer = InstructionTranslator(
E0502 18:42:26.814000 635 torch/_guards.py:361] [13/0] File "/usr/local/lib/python3.10/dist-packages/torch/_dynamo/symbolic_convert.py", line 3329, in __init__
E0502 18:42:26.814000 635 torch/_guards.py:361] [13/0] output=OutputGraph(
E0502 18:42:26.814000 635 torch/_guards.py:361] [13/0] File "/usr/local/lib/python3.10/dist-packages/torch/_dynamo/output_graph.py", line 358, in __init__
E0502 18:42:26.814000 635 torch/_guards.py:361] [13/0] self.init_ambient_guards()
E0502 18:42:26.814000 635 torch/_guards.py:361] [13/0] File "/usr/local/lib/python3.10/dist-packages/torch/_dynamo/output_graph.py", line 512, in init_ambient_guards
E0502 18:42:26.814000 635 torch/_guards.py:361] [13/0] self.guards.add(ShapeEnvSource().make_guard(GuardBuilder.SHAPE_ENV))
Traceback (most recent call last):
File "/usr/local/lib/python3.10/dist-packages/torch/export/_trace.py", line 739, in _export_to_torch_ir
gm_torch_level, _ = torch._dynamo.export(
File "/usr/local/lib/python3.10/dist-packages/torch/_dynamo/eval_frame.py", line 1722, in inner
raise constraint_violation_error
File "/usr/local/lib/python3.10/dist-packages/torch/_dynamo/eval_frame.py", line 1677, in inner
result_traced = opt_f(*args, **kwargs)
File "/usr/local/lib/python3.10/dist-packages/torch/nn/modules/module.py", line 1751, in _wrapped_call_impl
return self._call_impl(*args, **kwargs)
File "/usr/local/lib/python3.10/dist-packages/torch/nn/modules/module.py", line 1762, in _call_impl
return forward_call(*args, **kwargs)
File "/usr/local/lib/python3.10/dist-packages/torch/_dynamo/eval_frame.py", line 655, in _fn
return fn(*args, **kwargs)
File "/usr/local/lib/python3.10/dist-packages/torch/nn/modules/module.py", line 1751, in _wrapped_call_impl
return self._call_impl(*args, **kwargs)
File "/usr/local/lib/python3.10/dist-packages/torch/nn/modules/module.py", line 1762, in _call_impl
return forward_call(*args, **kwargs)
File "/usr/local/lib/python3.10/dist-packages/torch/_dynamo/convert_frame.py", line 1432, in __call__
return self._torchdynamo_orig_callable(
File "/usr/local/lib/python3.10/dist-packages/torch/_dynamo/convert_frame.py", line 598, in __call__
return _compile(
File "/usr/local/lib/python3.10/dist-packages/torch/_dynamo/convert_frame.py", line 1059, in _compile
guarded_code = compile_inner(code, one_graph, hooks, transform)
File "/usr/local/lib/python3.10/dist-packages/torch/_utils_internal.py", line 97, in wrapper_function
return function(*args, **kwargs)
File "/usr/local/lib/python3.10/dist-packages/torch/_dynamo/convert_frame.py", line 761, in compile_inner
return _compile_inner(code, one_graph, hooks, transform)
File "/usr/local/lib/python3.10/dist-packages/torch/_dynamo/convert_frame.py", line 906, in _compile_inner
check_fn = CheckFunctionManager(
File "/usr/local/lib/python3.10/dist-packages/torch/_dynamo/guards.py", line 2481, in __init__
guard.create(builder)
File "/usr/local/lib/python3.10/dist-packages/torch/_guards.py", line 357, in create
return self.create_fn(builder, self)
File "/usr/local/lib/python3.10/dist-packages/torch/_dynamo/guards.py", line 1959, in SHAPE_ENV
python_code_parts, verbose_code_parts = _get_code_parts(
File "/usr/local/lib/python3.10/dist-packages/torch/_dynamo/guards.py", line 1942, in _get_code_parts
return output_graph.shape_env.produce_guards_verbose(
File "/usr/local/lib/python3.10/dist-packages/torch/fx/experimental/symbolic_shapes.py", line 5409, in produce_guards_verbose
raise ConstraintViolationError(
torch.fx.experimental.symbolic_shapes.ConstraintViolationError: Constraints violated (d1, dy)! For more information, run with TORCH_LOGS="+dynamic".
- Not all values of d1 = L['x'].size()[1] in the specified range are valid because d1 was inferred to be a constant (4).
- Not all values of d1 = L['y'].size()[1] in the specified range are valid because d1 was inferred to be a constant (4).
- The values of dy = L['y'].size()[0] and dx = L['x'].size()[0] must always be equal.
Suggested fixes:
d1 = 4
dy = dx
During handling of the above exception, another exception occurred:
Traceback (most recent call last):
File "/var/lib/workspace/intermediate_source/torch_export_tutorial.py", line 557, in <module>
ep = export(
File "/usr/local/lib/python3.10/dist-packages/torch/export/__init__.py", line 360, in export
return _export(
File "/usr/local/lib/python3.10/dist-packages/torch/export/_trace.py", line 1092, in wrapper
raise e
File "/usr/local/lib/python3.10/dist-packages/torch/export/_trace.py", line 1065, in wrapper
ep = fn(*args, **kwargs)
File "/usr/local/lib/python3.10/dist-packages/torch/export/exported_program.py", line 121, in wrapper
return fn(*args, **kwargs)
File "/usr/local/lib/python3.10/dist-packages/torch/export/_trace.py", line 2112, in _export
ep = _export_for_training(
File "/usr/local/lib/python3.10/dist-packages/torch/export/_trace.py", line 1092, in wrapper
raise e
File "/usr/local/lib/python3.10/dist-packages/torch/export/_trace.py", line 1065, in wrapper
ep = fn(*args, **kwargs)
File "/usr/local/lib/python3.10/dist-packages/torch/export/exported_program.py", line 121, in wrapper
return fn(*args, **kwargs)
File "/usr/local/lib/python3.10/dist-packages/torch/export/_trace.py", line 1975, in _export_for_training
export_artifact = export_func( # type: ignore[operator]
File "/usr/local/lib/python3.10/dist-packages/torch/export/_trace.py", line 1344, in _strict_export_lower_to_aten_ir
gm_torch_level = _export_to_torch_ir(
File "/usr/local/lib/python3.10/dist-packages/torch/export/_trace.py", line 756, in _export_to_torch_ir
raise UserError(UserErrorType.CONSTRAINT_VIOLATION, str(e)) # noqa: B904
torch._dynamo.exc.UserError: Constraints violated (d1, dy)! For more information, run with TORCH_LOGS="+dynamic".
- Not all values of d1 = L['x'].size()[1] in the specified range are valid because d1 was inferred to be a constant (4).
- Not all values of d1 = L['y'].size()[1] in the specified range are valid because d1 was inferred to be a constant (4).
- The values of dy = L['y'].size()[0] and dx = L['x'].size()[0] must always be equal.
Suggested fixes:
d1 = 4
dy = dx
The expectation with suggested fixes is that the user can interactively copy-paste the changes into their dynamic shapes specification, and successfully export afterwards.
Lastly, there’s couple nice-to-knows about the options for specification:
Noneis a good option for static behavior: -dynamic_shapes=None(default) exports with the entire model being static. - specifyingNoneat an input-level exports with all tensor dimensions static, and is also required for non-tensor inputs. - specifyingNoneat a dimension-level specializes that dimension, though this is deprecated in favor ofDim.STATIC.specifying per-dimension integer values also produces static behavior, and will additionally check that the provided sample input matches the specification.
These options are combined in the inputs & dynamic shapes spec below:
inputs = (
torch.randn(4, 4),
torch.randn(3, 3),
16,
False,
)
dynamic_shapes = {
"tensor_0": (Dim.AUTO, None),
"tensor_1": None,
"int_val": None,
"bool_val": None,
}
Data-dependent errors
While trying to export models, you have may have encountered errors like “Could not guard on data-dependent expression”, or Could not extract specialized integer from data-dependent expression”.
These errors exist because torch.export() compiles programs using FakeTensors, which symbolically represent their real tensor counterparts. While these have equivalent symbolic properties
(e.g. sizes, strides, dtypes), they diverge in that FakeTensors do not contain any data values. While this avoids unnecessary memory usage and expensive computation, it does mean that export may be
unable to out-of-the-box compile parts of user code where compilation relies on data values. In short, if the compiler requires a concrete, data-dependent value in order to proceed, it will error out,
complaining that the value is not available.
Data-dependent values appear in many places, and common sources are calls like item(), tolist(), or torch.unbind() that extract scalar values from tensors.
How are these values represented in the exported program? In the Constraints/Dynamic Shapes
section, we talked about allocating symbols to represent dynamic input dimensions.
The same happens here: we allocate symbols for every data-dependent value that appears in the program. The important distinction is that these are “unbacked” symbols,
in contrast to the “backed” symbols allocated for input dimensions. The “backed/unbacked”
nomenclature refers to the presence/absence of a “hint” for the symbol: a concrete value backing the symbol, that can inform the compiler on how to proceed.
In the input shape symbol case (backed symbols), these hints are simply the sample input shapes provided, which explains why control-flow branching is determined by the sample input properties. For data-dependent values, the symbols are taken from FakeTensor “data” during tracing, and so the compiler doesn’t know the actual values (hints) that these symbols would take on.
Let’s see how these show up in exported programs:
class Foo(torch.nn.Module):
def forward(self, x, y):
a = x.item()
b = y.tolist()
return b + [a]
inps = (
torch.tensor(1),
torch.tensor([2, 3]),
)
ep = export(Foo(), inps)
print(ep)
I0502 18:42:26.829000 635 torch/fx/experimental/symbolic_shapes.py:3334] [14/0] create_env
I0502 18:42:26.833000 635 torch/fx/experimental/symbolic_shapes.py:4276] [14/0] create_unbacked_symint u0 [-int_oo, int_oo] a = x.item() # ar/lib/workspace/intermediate_source/torch_export_tutorial.py:618 in forward (_subclasses/fake_impls.py:422 in local_scalar_dense)
I0502 18:42:26.833000 635 torch/fx/experimental/symbolic_shapes.py:1130] [14/0] compute_unbacked_bindings [u0]
I0502 18:42:26.836000 635 torch/fx/experimental/symbolic_shapes.py:4276] [14/0] create_unbacked_symint u1 [-int_oo, int_oo] b = y.tolist() # ar/lib/workspace/intermediate_source/torch_export_tutorial.py:619 in forward (_subclasses/fake_impls.py:422 in local_scalar_dense)
I0502 18:42:26.836000 635 torch/fx/experimental/symbolic_shapes.py:1130] [14/0] compute_unbacked_bindings [u1]
I0502 18:42:26.838000 635 torch/fx/experimental/symbolic_shapes.py:4276] [14/0] create_unbacked_symint u2 [-int_oo, int_oo] b = y.tolist() # ar/lib/workspace/intermediate_source/torch_export_tutorial.py:619 in forward (_subclasses/fake_impls.py:422 in local_scalar_dense)
I0502 18:42:26.838000 635 torch/fx/experimental/symbolic_shapes.py:1130] [14/0] compute_unbacked_bindings [u2]
I0502 18:42:26.842000 635 torch/fx/experimental/symbolic_shapes.py:4734] [14/0] produce_guards
V0502 18:42:26.842000 635 torch/fx/experimental/symbolic_shapes.py:4954] [14/0] track_symint L['x'].storage_offset() 0 None
V0502 18:42:26.842000 635 torch/fx/experimental/symbolic_shapes.py:4954] [14/0] track_symint L['y'].size()[0] 2 None
V0502 18:42:26.843000 635 torch/fx/experimental/symbolic_shapes.py:4954] [14/0] track_symint L['y'].stride()[0] 1 None
V0502 18:42:26.843000 635 torch/fx/experimental/symbolic_shapes.py:4954] [14/0] track_symint L['y'].storage_offset() 0 None
I0502 18:42:26.849000 635 torch/fx/experimental/symbolic_shapes.py:4276] create_unbacked_symint u3 [-int_oo, int_oo] (_subclasses/fake_impls.py:422 in local_scalar_dense)
I0502 18:42:26.850000 635 torch/fx/experimental/symbolic_shapes.py:4276] create_unbacked_symint u4 [-int_oo, int_oo] (_subclasses/fake_impls.py:422 in local_scalar_dense)
I0502 18:42:26.857000 635 torch/fx/experimental/symbolic_shapes.py:4276] create_unbacked_symint u5 [-int_oo, int_oo] (_subclasses/fake_impls.py:422 in local_scalar_dense)
I0502 18:42:26.857000 635 torch/fx/experimental/symbolic_shapes.py:1130] compute_unbacked_bindings [u5]
I0502 18:42:26.857000 635 torch/fx/experimental/symbolic_shapes.py:6234] set_replacement u5 = u0 (rename_unbacked_to) VR[-int_oo, int_oo]
I0502 18:42:26.859000 635 torch/fx/experimental/symbolic_shapes.py:4276] create_unbacked_symint u6 [-int_oo, int_oo] (_subclasses/fake_impls.py:422 in local_scalar_dense)
I0502 18:42:26.859000 635 torch/fx/experimental/symbolic_shapes.py:1130] compute_unbacked_bindings [u6]
I0502 18:42:26.860000 635 torch/fx/experimental/symbolic_shapes.py:6234] set_replacement u6 = u1 (rename_unbacked_to) VR[-int_oo, int_oo]
I0502 18:42:26.862000 635 torch/fx/experimental/symbolic_shapes.py:4276] create_unbacked_symint u7 [-int_oo, int_oo] (_subclasses/fake_impls.py:422 in local_scalar_dense)
I0502 18:42:26.862000 635 torch/fx/experimental/symbolic_shapes.py:1130] compute_unbacked_bindings [u7]
I0502 18:42:26.862000 635 torch/fx/experimental/symbolic_shapes.py:6234] set_replacement u7 = u2 (rename_unbacked_to) VR[-int_oo, int_oo]
ExportedProgram:
class GraphModule(torch.nn.Module):
def forward(self, x: "i64[]", y: "i64[2]"):
# File: /var/lib/workspace/intermediate_source/torch_export_tutorial.py:618 in forward, code: a = x.item()
item: "Sym(u0)" = torch.ops.aten.item.default(x); x = None
# File: /var/lib/workspace/intermediate_source/torch_export_tutorial.py:619 in forward, code: b = y.tolist()
select: "i64[]" = torch.ops.aten.select.int(y, 0, 0)
item_1: "Sym(u1)" = torch.ops.aten.item.default(select); select = None
select_1: "i64[]" = torch.ops.aten.select.int(y, 0, 1); y = None
item_2: "Sym(u2)" = torch.ops.aten.item.default(select_1); select_1 = None
return (item_1, item_2, item)
Graph signature: ExportGraphSignature(input_specs=[InputSpec(kind=<InputKind.USER_INPUT: 1>, arg=TensorArgument(name='x'), target=None, persistent=None), InputSpec(kind=<InputKind.USER_INPUT: 1>, arg=TensorArgument(name='y'), target=None, persistent=None)], output_specs=[OutputSpec(kind=<OutputKind.USER_OUTPUT: 1>, arg=SymIntArgument(name='item_1'), target=None), OutputSpec(kind=<OutputKind.USER_OUTPUT: 1>, arg=SymIntArgument(name='item_2'), target=None), OutputSpec(kind=<OutputKind.USER_OUTPUT: 1>, arg=SymIntArgument(name='item'), target=None)])
Range constraints: {u0: VR[-int_oo, int_oo], u1: VR[-int_oo, int_oo], u2: VR[-int_oo, int_oo], u3: VR[-int_oo, int_oo], u4: VR[-int_oo, int_oo], u5: VR[-int_oo, int_oo], u6: VR[-int_oo, int_oo], u7: VR[-int_oo, int_oo]}
The result is that 3 unbacked symbols (notice they’re prefixed with “u”, instead of the usual “s” for input shape/backed symbols) are allocated and returned:
1 for the item() call, and 1 for each of the elements of y with the tolist() call.
Note from the range constraints field that these take on ranges of [-int_oo, int_oo], not the default [0, int_oo] range allocated to input shape symbols,
since we have no information on what these values are - they don’t represent sizes, so don’t necessarily have positive values.
Guards, torch._check()
But the case above is easy to export, because the concrete values of these symbols aren’t used in any compiler decision-making; all that’s relevant is that the return values are unbacked symbols. The data-dependent errors highlighted in this section are cases like the following, where data-dependent guards are encountered:
Here we actually need the “hint”, or the concrete value of a for the compiler to decide whether to trace return y + 2 or return y * 5 as the output.
Because we trace with FakeTensors, we don’t know what a // 2 >= 5 actually evaluates to, and export errors out with “Could not guard on data-dependent expression u0 // 2 >= 5 (unhinted)”.
So how do we export this toy model? Unlike torch.compile(), export requires full graph compilation, and we can’t just graph break on this. Here are some basic options:
Manual specialization: we could intervene by selecting the branch to trace, either by removing the control-flow code to contain only the specialized branch, or using
torch.compiler.is_compiling()to guard what’s traced at compile-time.torch.cond(): we could rewrite the control-flow code to usetorch.cond()so we don’t specialize on a branch.
While these options are valid, they have their pitfalls. Option 1 sometimes requires drastic, invasive rewrites of the model code to specialize, and torch.cond() is not a comprehensive system for handling data-dependent errors.
As we will see, there are data-dependent errors that do not involve control-flow.
The generally recommended approach is to start with torch._check() calls. While these give the impression of purely being assert statements, they are in fact a system of informing the compiler on properties of symbols.
While a torch._check() call does act as an assertion at runtime, when traced at compile-time, the checked expression is sent to the symbolic shapes subsystem for reasoning, and any symbol properties that follow from the expression being true,
are stored as symbol properties (provided it’s smart enough to infer those properties). So even if unbacked symbols don’t have hints, if we’re able to communicate properties that are generally true for these symbols via
torch._check() calls, we can potentially bypass data-dependent guards without rewriting the offending model code.
For example in the model above, inserting torch._check(a >= 10) would tell the compiler that y + 2 can always be returned, and torch._check(a == 4) tells it to return y * 5.
See what happens when we re-export this model.
class Foo(torch.nn.Module):
def forward(self, x, y):
a = x.item()
torch._check(a >= 10)
torch._check(a <= 60)
if a // 2 >= 5:
return y + 2
else:
return y * 5
inps = (
torch.tensor(32),
torch.randn(4),
)
ep = export(Foo(), inps)
print(ep)
I0502 18:42:26.872000 635 torch/fx/experimental/symbolic_shapes.py:3334] [15/0] create_env
I0502 18:42:26.876000 635 torch/fx/experimental/symbolic_shapes.py:4276] [15/0] create_unbacked_symint u0 [-int_oo, int_oo] a = x.item() # ar/lib/workspace/intermediate_source/torch_export_tutorial.py:672 in forward (_subclasses/fake_impls.py:422 in local_scalar_dense)
I0502 18:42:26.877000 635 torch/fx/experimental/symbolic_shapes.py:1130] [15/0] compute_unbacked_bindings [u0]
I0502 18:42:26.879000 635 torch/fx/experimental/symbolic_shapes.py:6630] [15/0] runtime_assert u0 >= 10 [guard added] torch._check(a >= 10) # ar/lib/workspace/intermediate_source/torch_export_tutorial.py:673 in forward (_dynamo/utils.py:3284 in run_node), for more info run with TORCHDYNAMO_EXTENDED_DEBUG_GUARD_ADDED="u0 >= 10"
V0502 18:42:26.880000 635 torch/fx/experimental/symbolic_shapes.py:6071] [15/0] _update_var_to_range u0 = VR[10, int_oo] (update)
I0502 18:42:26.885000 635 torch/fx/experimental/symbolic_shapes.py:6630] [15/0] runtime_assert u0 <= 60 [guard added] torch._check(a <= 60) # ar/lib/workspace/intermediate_source/torch_export_tutorial.py:674 in forward (_dynamo/utils.py:3284 in run_node), for more info run with TORCHDYNAMO_EXTENDED_DEBUG_GUARD_ADDED="u0 <= 60"
V0502 18:42:26.886000 635 torch/fx/experimental/symbolic_shapes.py:6071] [15/0] _update_var_to_range u0 = VR[10, 60] (update)
V0502 18:42:26.891000 635 torch/fx/experimental/symbolic_shapes.py:6787] [15/0] eval False == True [statically known]
V0502 18:42:26.894000 635 torch/fx/experimental/symbolic_shapes.py:7018] [15/0] runtime_assert u0 >= 10 == True [statically known]
V0502 18:42:26.895000 635 torch/fx/experimental/symbolic_shapes.py:7018] [15/0] runtime_assert u0 <= 60 == True [statically known]
I0502 18:42:26.897000 635 torch/fx/experimental/symbolic_shapes.py:4734] [15/0] produce_guards
V0502 18:42:26.898000 635 torch/fx/experimental/symbolic_shapes.py:4954] [15/0] track_symint L['x'].storage_offset() 0 None
V0502 18:42:26.898000 635 torch/fx/experimental/symbolic_shapes.py:4954] [15/0] track_symint L['y'].size()[0] 4 None
V0502 18:42:26.898000 635 torch/fx/experimental/symbolic_shapes.py:4954] [15/0] track_symint L['y'].stride()[0] 1 None
V0502 18:42:26.899000 635 torch/fx/experimental/symbolic_shapes.py:4954] [15/0] track_symint L['y'].storage_offset() 0 None
I0502 18:42:26.912000 635 torch/fx/experimental/symbolic_shapes.py:4276] create_unbacked_symint u1 [-int_oo, int_oo] (_subclasses/fake_impls.py:422 in local_scalar_dense)
I0502 18:42:26.912000 635 torch/fx/experimental/symbolic_shapes.py:1130] compute_unbacked_bindings [u1]
V0502 18:42:26.913000 635 torch/fx/experimental/symbolic_shapes.py:6071] _update_var_to_range u1 = VR[10, 60] (update)
I0502 18:42:26.913000 635 torch/fx/experimental/symbolic_shapes.py:6234] set_replacement u1 = u0 (rename_unbacked_to) VR[10, 60]
ExportedProgram:
class GraphModule(torch.nn.Module):
def forward(self, x: "i64[]", y: "f32[4]"):
# File: /var/lib/workspace/intermediate_source/torch_export_tutorial.py:672 in forward, code: a = x.item()
item: "Sym(u0)" = torch.ops.aten.item.default(x); x = None
ge_1: "Sym(u0 >= 10)" = item >= 10
_assert_scalar_default = torch.ops.aten._assert_scalar.default(ge_1, "Runtime assertion failed for expression u0 >= 10 on node 'ge_1'"); ge_1 = _assert_scalar_default = None
le_1: "Sym(u0 <= 60)" = item <= 60; item = None
_assert_scalar_default_1 = torch.ops.aten._assert_scalar.default(le_1, "Runtime assertion failed for expression u0 <= 60 on node 'le_1'"); le_1 = _assert_scalar_default_1 = None
# File: /var/lib/workspace/intermediate_source/torch_export_tutorial.py:676 in forward, code: return y + 2
add: "f32[4]" = torch.ops.aten.add.Tensor(y, 2); y = None
return (add,)
Graph signature: ExportGraphSignature(input_specs=[InputSpec(kind=<InputKind.USER_INPUT: 1>, arg=TensorArgument(name='x'), target=None, persistent=None), InputSpec(kind=<InputKind.USER_INPUT: 1>, arg=TensorArgument(name='y'), target=None, persistent=None)], output_specs=[OutputSpec(kind=<OutputKind.USER_OUTPUT: 1>, arg=TensorArgument(name='add'), target=None)])
Range constraints: {u0: VR[10, 60], u1: VR[10, 60]}
Export succeeds, and note from the range constraints field that u0 takes on a range of [10, 60].
So what information do torch._check() calls actually communicate? This varies as the symbolic shapes subsystem gets smarter, but at a fundamental level, these are generally true:
Equality with non-data-dependent expressions:
torch._check()calls that communicate equalities likeu0 == s0 + 4oru0 == 5.Range refinement: calls that provide lower or upper bounds for symbols, like the above.
Some basic reasoning around more complicated expressions: inserting
torch._check(a < 4)will typically tell the compiler thata >= 4is false. Checks on complex expressions liketorch._check(a ** 2 - 3 * a <= 10)will typically get you past identical guards.
As mentioned previously, torch._check() calls have applicability outside of data-dependent control flow. For example, here’s a model where torch._check() insertion
prevails while manual specialization & torch.cond() do not:
class Foo(torch.nn.Module):
def forward(self, x, y):
a = x.item()
return y[a]
inps = (
torch.tensor(32),
torch.randn(60),
)
try:
export(Foo(), inps)
except Exception:
tb.print_exc()
I0502 18:42:26.927000 635 torch/fx/experimental/symbolic_shapes.py:3334] [16/0] create_env
I0502 18:42:26.931000 635 torch/fx/experimental/symbolic_shapes.py:4276] [16/0] create_unbacked_symint u0 [-int_oo, int_oo] a = x.item() # ar/lib/workspace/intermediate_source/torch_export_tutorial.py:701 in forward (_subclasses/fake_impls.py:422 in local_scalar_dense)
I0502 18:42:26.932000 635 torch/fx/experimental/symbolic_shapes.py:1130] [16/0] compute_unbacked_bindings [u0]
V0502 18:42:26.935000 635 torch/fx/experimental/symbolic_shapes.py:5984] [16/0] Data dependent variable 'u0' allocated at:
V0502 18:42:26.935000 635 torch/fx/experimental/symbolic_shapes.py:5984] [16/0] File "/usr/local/bin/sphinx-build", line 8, in <module>
V0502 18:42:26.935000 635 torch/fx/experimental/symbolic_shapes.py:5984] [16/0] sys.exit(main())
V0502 18:42:26.935000 635 torch/fx/experimental/symbolic_shapes.py:5984] [16/0] File "/usr/local/lib/python3.10/dist-packages/sphinx/cmd/build.py", line 288, in main
V0502 18:42:26.935000 635 torch/fx/experimental/symbolic_shapes.py:5984] [16/0] return make_main(argv)
V0502 18:42:26.935000 635 torch/fx/experimental/symbolic_shapes.py:5984] [16/0] File "/usr/local/lib/python3.10/dist-packages/sphinx/cmd/build.py", line 193, in make_main
V0502 18:42:26.935000 635 torch/fx/experimental/symbolic_shapes.py:5984] [16/0] return make_mode.run_make_mode(argv[1:])
V0502 18:42:26.935000 635 torch/fx/experimental/symbolic_shapes.py:5984] [16/0] File "/usr/local/lib/python3.10/dist-packages/sphinx/cmd/make_mode.py", line 160, in run_make_mode
V0502 18:42:26.935000 635 torch/fx/experimental/symbolic_shapes.py:5984] [16/0] return make.run_generic_build(args[0])
V0502 18:42:26.935000 635 torch/fx/experimental/symbolic_shapes.py:5984] [16/0] File "/usr/local/lib/python3.10/dist-packages/sphinx/cmd/make_mode.py", line 148, in run_generic_build
V0502 18:42:26.935000 635 torch/fx/experimental/symbolic_shapes.py:5984] [16/0] return build_main(args + opts)
V0502 18:42:26.935000 635 torch/fx/experimental/symbolic_shapes.py:5984] [16/0] File "/usr/local/lib/python3.10/dist-packages/sphinx/cmd/build.py", line 272, in build_main
V0502 18:42:26.935000 635 torch/fx/experimental/symbolic_shapes.py:5984] [16/0] app = Sphinx(args.sourcedir, args.confdir, args.outputdir,
V0502 18:42:26.935000 635 torch/fx/experimental/symbolic_shapes.py:5984] [16/0] File "/usr/local/lib/python3.10/dist-packages/sphinx/application.py", line 256, in __init__
V0502 18:42:26.935000 635 torch/fx/experimental/symbolic_shapes.py:5984] [16/0] self._init_builder()
V0502 18:42:26.935000 635 torch/fx/experimental/symbolic_shapes.py:5984] [16/0] File "/usr/local/lib/python3.10/dist-packages/sphinx/application.py", line 314, in _init_builder
V0502 18:42:26.935000 635 torch/fx/experimental/symbolic_shapes.py:5984] [16/0] self.events.emit('builder-inited')
V0502 18:42:26.935000 635 torch/fx/experimental/symbolic_shapes.py:5984] [16/0] File "/usr/local/lib/python3.10/dist-packages/sphinx/events.py", line 94, in emit
V0502 18:42:26.935000 635 torch/fx/experimental/symbolic_shapes.py:5984] [16/0] results.append(listener.handler(self.app, *args))
V0502 18:42:26.935000 635 torch/fx/experimental/symbolic_shapes.py:5984] [16/0] File "/usr/local/lib/python3.10/dist-packages/sphinx_gallery/gen_gallery.py", line 491, in generate_gallery_rst
V0502 18:42:26.935000 635 torch/fx/experimental/symbolic_shapes.py:5984] [16/0] ) = generate_dir_rst(
V0502 18:42:26.935000 635 torch/fx/experimental/symbolic_shapes.py:5984] [16/0] File "/usr/local/lib/python3.10/dist-packages/sphinx_gallery/gen_rst.py", line 431, in generate_dir_rst
V0502 18:42:26.935000 635 torch/fx/experimental/symbolic_shapes.py:5984] [16/0] intro, title, cost = generate_file_rst(
V0502 18:42:26.935000 635 torch/fx/experimental/symbolic_shapes.py:5984] [16/0] File "/usr/local/lib/python3.10/dist-packages/sphinx_gallery/gen_rst.py", line 1027, in generate_file_rst
V0502 18:42:26.935000 635 torch/fx/experimental/symbolic_shapes.py:5984] [16/0] output_blocks, time_elapsed = execute_script(script_blocks,
V0502 18:42:26.935000 635 torch/fx/experimental/symbolic_shapes.py:5984] [16/0] File "/usr/local/lib/python3.10/dist-packages/sphinx_gallery/gen_rst.py", line 945, in execute_script
V0502 18:42:26.935000 635 torch/fx/experimental/symbolic_shapes.py:5984] [16/0] output_blocks.append(execute_code_block(
V0502 18:42:26.935000 635 torch/fx/experimental/symbolic_shapes.py:5984] [16/0] File "/usr/local/lib/python3.10/dist-packages/sphinx_gallery/gen_rst.py", line 810, in execute_code_block
V0502 18:42:26.935000 635 torch/fx/experimental/symbolic_shapes.py:5984] [16/0] is_last_expr, mem_max = _exec_and_get_memory(
V0502 18:42:26.935000 635 torch/fx/experimental/symbolic_shapes.py:5984] [16/0] File "/usr/local/lib/python3.10/dist-packages/sphinx_gallery/gen_rst.py", line 676, in _exec_and_get_memory
V0502 18:42:26.935000 635 torch/fx/experimental/symbolic_shapes.py:5984] [16/0] mem_max, _ = gallery_conf['call_memory'](
V0502 18:42:26.935000 635 torch/fx/experimental/symbolic_shapes.py:5984] [16/0] File "/usr/local/lib/python3.10/dist-packages/sphinx_gallery/gen_gallery.py", line 223, in call_memory
V0502 18:42:26.935000 635 torch/fx/experimental/symbolic_shapes.py:5984] [16/0] return 0., func()
V0502 18:42:26.935000 635 torch/fx/experimental/symbolic_shapes.py:5984] [16/0] File "/usr/local/lib/python3.10/dist-packages/sphinx_gallery/gen_rst.py", line 600, in __call__
V0502 18:42:26.935000 635 torch/fx/experimental/symbolic_shapes.py:5984] [16/0] exec(self.code, self.fake_main.__dict__)
V0502 18:42:26.935000 635 torch/fx/experimental/symbolic_shapes.py:5984] [16/0] File "/var/lib/workspace/intermediate_source/torch_export_tutorial.py", line 709, in <module>
V0502 18:42:26.935000 635 torch/fx/experimental/symbolic_shapes.py:5984] [16/0] export(Foo(), inps)
V0502 18:42:26.935000 635 torch/fx/experimental/symbolic_shapes.py:5984] [16/0] File "/usr/local/lib/python3.10/dist-packages/torch/export/__init__.py", line 360, in export
V0502 18:42:26.935000 635 torch/fx/experimental/symbolic_shapes.py:5984] [16/0] return _export(
V0502 18:42:26.935000 635 torch/fx/experimental/symbolic_shapes.py:5984] [16/0] File "/usr/local/lib/python3.10/dist-packages/torch/export/_trace.py", line 1065, in wrapper
V0502 18:42:26.935000 635 torch/fx/experimental/symbolic_shapes.py:5984] [16/0] ep = fn(*args, **kwargs)
V0502 18:42:26.935000 635 torch/fx/experimental/symbolic_shapes.py:5984] [16/0] File "/usr/local/lib/python3.10/dist-packages/torch/export/exported_program.py", line 121, in wrapper
V0502 18:42:26.935000 635 torch/fx/experimental/symbolic_shapes.py:5984] [16/0] return fn(*args, **kwargs)
V0502 18:42:26.935000 635 torch/fx/experimental/symbolic_shapes.py:5984] [16/0] File "/usr/local/lib/python3.10/dist-packages/torch/export/_trace.py", line 2112, in _export
V0502 18:42:26.935000 635 torch/fx/experimental/symbolic_shapes.py:5984] [16/0] ep = _export_for_training(
V0502 18:42:26.935000 635 torch/fx/experimental/symbolic_shapes.py:5984] [16/0] File "/usr/local/lib/python3.10/dist-packages/torch/export/_trace.py", line 1065, in wrapper
V0502 18:42:26.935000 635 torch/fx/experimental/symbolic_shapes.py:5984] [16/0] ep = fn(*args, **kwargs)
V0502 18:42:26.935000 635 torch/fx/experimental/symbolic_shapes.py:5984] [16/0] File "/usr/local/lib/python3.10/dist-packages/torch/export/exported_program.py", line 121, in wrapper
V0502 18:42:26.935000 635 torch/fx/experimental/symbolic_shapes.py:5984] [16/0] return fn(*args, **kwargs)
V0502 18:42:26.935000 635 torch/fx/experimental/symbolic_shapes.py:5984] [16/0] File "/usr/local/lib/python3.10/dist-packages/torch/export/_trace.py", line 1975, in _export_for_training
V0502 18:42:26.935000 635 torch/fx/experimental/symbolic_shapes.py:5984] [16/0] export_artifact = export_func( # type: ignore[operator]
V0502 18:42:26.935000 635 torch/fx/experimental/symbolic_shapes.py:5984] [16/0] File "/usr/local/lib/python3.10/dist-packages/torch/export/_trace.py", line 1344, in _strict_export_lower_to_aten_ir
V0502 18:42:26.935000 635 torch/fx/experimental/symbolic_shapes.py:5984] [16/0] gm_torch_level = _export_to_torch_ir(
V0502 18:42:26.935000 635 torch/fx/experimental/symbolic_shapes.py:5984] [16/0] File "/usr/local/lib/python3.10/dist-packages/torch/export/_trace.py", line 739, in _export_to_torch_ir
V0502 18:42:26.935000 635 torch/fx/experimental/symbolic_shapes.py:5984] [16/0] gm_torch_level, _ = torch._dynamo.export(
V0502 18:42:26.935000 635 torch/fx/experimental/symbolic_shapes.py:5984] [16/0] File "/usr/local/lib/python3.10/dist-packages/torch/_dynamo/eval_frame.py", line 1677, in inner
V0502 18:42:26.935000 635 torch/fx/experimental/symbolic_shapes.py:5984] [16/0] result_traced = opt_f(*args, **kwargs)
V0502 18:42:26.935000 635 torch/fx/experimental/symbolic_shapes.py:5984] [16/0] File "/usr/local/lib/python3.10/dist-packages/torch/nn/modules/module.py", line 1751, in _wrapped_call_impl
V0502 18:42:26.935000 635 torch/fx/experimental/symbolic_shapes.py:5984] [16/0] return self._call_impl(*args, **kwargs)
V0502 18:42:26.935000 635 torch/fx/experimental/symbolic_shapes.py:5984] [16/0] File "/usr/local/lib/python3.10/dist-packages/torch/nn/modules/module.py", line 1762, in _call_impl
V0502 18:42:26.935000 635 torch/fx/experimental/symbolic_shapes.py:5984] [16/0] return forward_call(*args, **kwargs)
V0502 18:42:26.935000 635 torch/fx/experimental/symbolic_shapes.py:5984] [16/0] File "/usr/local/lib/python3.10/dist-packages/torch/_dynamo/eval_frame.py", line 655, in _fn
V0502 18:42:26.935000 635 torch/fx/experimental/symbolic_shapes.py:5984] [16/0] return fn(*args, **kwargs)
V0502 18:42:26.935000 635 torch/fx/experimental/symbolic_shapes.py:5984] [16/0] File "/usr/local/lib/python3.10/dist-packages/torch/nn/modules/module.py", line 1751, in _wrapped_call_impl
V0502 18:42:26.935000 635 torch/fx/experimental/symbolic_shapes.py:5984] [16/0] return self._call_impl(*args, **kwargs)
V0502 18:42:26.935000 635 torch/fx/experimental/symbolic_shapes.py:5984] [16/0] File "/usr/local/lib/python3.10/dist-packages/torch/nn/modules/module.py", line 1762, in _call_impl
V0502 18:42:26.935000 635 torch/fx/experimental/symbolic_shapes.py:5984] [16/0] return forward_call(*args, **kwargs)
V0502 18:42:26.935000 635 torch/fx/experimental/symbolic_shapes.py:5984] [16/0] File "/usr/local/lib/python3.10/dist-packages/torch/_dynamo/convert_frame.py", line 1432, in __call__
V0502 18:42:26.935000 635 torch/fx/experimental/symbolic_shapes.py:5984] [16/0] return self._torchdynamo_orig_callable(
V0502 18:42:26.935000 635 torch/fx/experimental/symbolic_shapes.py:5984] [16/0] File "/usr/local/lib/python3.10/dist-packages/torch/_dynamo/convert_frame.py", line 598, in __call__
V0502 18:42:26.935000 635 torch/fx/experimental/symbolic_shapes.py:5984] [16/0] return _compile(
V0502 18:42:26.935000 635 torch/fx/experimental/symbolic_shapes.py:5984] [16/0] File "/usr/local/lib/python3.10/dist-packages/torch/_dynamo/convert_frame.py", line 1059, in _compile
V0502 18:42:26.935000 635 torch/fx/experimental/symbolic_shapes.py:5984] [16/0] guarded_code = compile_inner(code, one_graph, hooks, transform)
V0502 18:42:26.935000 635 torch/fx/experimental/symbolic_shapes.py:5984] [16/0] File "/usr/local/lib/python3.10/dist-packages/torch/_utils_internal.py", line 97, in wrapper_function
V0502 18:42:26.935000 635 torch/fx/experimental/symbolic_shapes.py:5984] [16/0] return function(*args, **kwargs)
V0502 18:42:26.935000 635 torch/fx/experimental/symbolic_shapes.py:5984] [16/0] File "/usr/local/lib/python3.10/dist-packages/torch/_dynamo/convert_frame.py", line 761, in compile_inner
V0502 18:42:26.935000 635 torch/fx/experimental/symbolic_shapes.py:5984] [16/0] return _compile_inner(code, one_graph, hooks, transform)
V0502 18:42:26.935000 635 torch/fx/experimental/symbolic_shapes.py:5984] [16/0] File "/usr/local/lib/python3.10/dist-packages/torch/_dynamo/convert_frame.py", line 797, in _compile_inner
V0502 18:42:26.935000 635 torch/fx/experimental/symbolic_shapes.py:5984] [16/0] out_code = transform_code_object(code, transform)
V0502 18:42:26.935000 635 torch/fx/experimental/symbolic_shapes.py:5984] [16/0] File "/usr/local/lib/python3.10/dist-packages/torch/_dynamo/bytecode_transformation.py", line 1422, in transform_code_object
V0502 18:42:26.935000 635 torch/fx/experimental/symbolic_shapes.py:5984] [16/0] transformations(instructions, code_options)
V0502 18:42:26.935000 635 torch/fx/experimental/symbolic_shapes.py:5984] [16/0] File "/usr/local/lib/python3.10/dist-packages/torch/_dynamo/convert_frame.py", line 257, in _fn
V0502 18:42:26.935000 635 torch/fx/experimental/symbolic_shapes.py:5984] [16/0] return fn(*args, **kwargs)
V0502 18:42:26.935000 635 torch/fx/experimental/symbolic_shapes.py:5984] [16/0] File "/usr/local/lib/python3.10/dist-packages/torch/_dynamo/convert_frame.py", line 715, in transform
V0502 18:42:26.935000 635 torch/fx/experimental/symbolic_shapes.py:5984] [16/0] tracer.run()
V0502 18:42:26.935000 635 torch/fx/experimental/symbolic_shapes.py:5984] [16/0] File "/usr/local/lib/python3.10/dist-packages/torch/_dynamo/symbolic_convert.py", line 3500, in run
V0502 18:42:26.935000 635 torch/fx/experimental/symbolic_shapes.py:5984] [16/0] super().run()
V0502 18:42:26.935000 635 torch/fx/experimental/symbolic_shapes.py:5984] [16/0] File "/usr/local/lib/python3.10/dist-packages/torch/_dynamo/symbolic_convert.py", line 1337, in run
V0502 18:42:26.935000 635 torch/fx/experimental/symbolic_shapes.py:5984] [16/0] while self.step():
V0502 18:42:26.935000 635 torch/fx/experimental/symbolic_shapes.py:5984] [16/0] File "/usr/local/lib/python3.10/dist-packages/torch/_dynamo/symbolic_convert.py", line 1246, in step
V0502 18:42:26.935000 635 torch/fx/experimental/symbolic_shapes.py:5984] [16/0] self.dispatch_table[inst.opcode](self, inst)
V0502 18:42:26.935000 635 torch/fx/experimental/symbolic_shapes.py:5984] [16/0] File "/usr/local/lib/python3.10/dist-packages/torch/_dynamo/symbolic_convert.py", line 819, in wrapper
V0502 18:42:26.935000 635 torch/fx/experimental/symbolic_shapes.py:5984] [16/0] return inner_fn(self, inst)
V0502 18:42:26.935000 635 torch/fx/experimental/symbolic_shapes.py:5984] [16/0] File "/usr/local/lib/python3.10/dist-packages/torch/_dynamo/symbolic_convert.py", line 2168, in CALL_FUNCTION
V0502 18:42:26.935000 635 torch/fx/experimental/symbolic_shapes.py:5984] [16/0] self.call_function(fn, args, {})
V0502 18:42:26.935000 635 torch/fx/experimental/symbolic_shapes.py:5984] [16/0] File "/usr/local/lib/python3.10/dist-packages/torch/_dynamo/symbolic_convert.py", line 1170, in call_function
V0502 18:42:26.935000 635 torch/fx/experimental/symbolic_shapes.py:5984] [16/0] self.push(fn.call_function(self, args, kwargs)) # type: ignore[arg-type]
V0502 18:42:26.935000 635 torch/fx/experimental/symbolic_shapes.py:5984] [16/0] File "/usr/local/lib/python3.10/dist-packages/torch/_dynamo/variables/misc.py", line 903, in call_function
V0502 18:42:26.935000 635 torch/fx/experimental/symbolic_shapes.py:5984] [16/0] return self.obj.call_method(tx, self.name, args, kwargs)
V0502 18:42:26.935000 635 torch/fx/experimental/symbolic_shapes.py:5984] [16/0] File "/usr/local/lib/python3.10/dist-packages/torch/_dynamo/variables/tensor.py", line 632, in call_method
V0502 18:42:26.935000 635 torch/fx/experimental/symbolic_shapes.py:5984] [16/0] return wrap_fx_proxy(
V0502 18:42:26.935000 635 torch/fx/experimental/symbolic_shapes.py:5984] [16/0] File "/usr/local/lib/python3.10/dist-packages/torch/_dynamo/variables/builder.py", line 2302, in wrap_fx_proxy
V0502 18:42:26.935000 635 torch/fx/experimental/symbolic_shapes.py:5984] [16/0] return wrap_fx_proxy_cls(target_cls=TensorVariable, **kwargs)
V0502 18:42:26.935000 635 torch/fx/experimental/symbolic_shapes.py:5984] [16/0] File "/usr/local/lib/python3.10/dist-packages/torch/_dynamo/variables/builder.py", line 2368, in wrap_fx_proxy_cls
V0502 18:42:26.935000 635 torch/fx/experimental/symbolic_shapes.py:5984] [16/0] return _wrap_fx_proxy(
V0502 18:42:26.935000 635 torch/fx/experimental/symbolic_shapes.py:5984] [16/0] File "/usr/local/lib/python3.10/dist-packages/torch/_dynamo/variables/builder.py", line 2464, in _wrap_fx_proxy
V0502 18:42:26.935000 635 torch/fx/experimental/symbolic_shapes.py:5984] [16/0] example_value = get_fake_value(proxy.node, tx, allow_non_graph_fake=True)
V0502 18:42:26.935000 635 torch/fx/experimental/symbolic_shapes.py:5984] [16/0] File "/usr/local/lib/python3.10/dist-packages/torch/_dynamo/utils.py", line 3127, in get_fake_value
V0502 18:42:26.935000 635 torch/fx/experimental/symbolic_shapes.py:5984] [16/0] ret_val = wrap_fake_exception(
V0502 18:42:26.935000 635 torch/fx/experimental/symbolic_shapes.py:5984] [16/0] File "/usr/local/lib/python3.10/dist-packages/torch/_dynamo/utils.py", line 2641, in wrap_fake_exception
V0502 18:42:26.935000 635 torch/fx/experimental/symbolic_shapes.py:5984] [16/0] return fn()
V0502 18:42:26.935000 635 torch/fx/experimental/symbolic_shapes.py:5984] [16/0] File "/usr/local/lib/python3.10/dist-packages/torch/_dynamo/utils.py", line 3128, in <lambda>
V0502 18:42:26.935000 635 torch/fx/experimental/symbolic_shapes.py:5984] [16/0] lambda: run_node(tx.output, node, args, kwargs, nnmodule)
V0502 18:42:26.935000 635 torch/fx/experimental/symbolic_shapes.py:5984] [16/0] File "/usr/local/lib/python3.10/dist-packages/torch/_dynamo/utils.py", line 3295, in run_node
V0502 18:42:26.935000 635 torch/fx/experimental/symbolic_shapes.py:5984] [16/0] return getattr(args[0], node.target)(*args[1:], **kwargs)
V0502 18:42:26.935000 635 torch/fx/experimental/symbolic_shapes.py:5984] [16/0] File "/usr/local/lib/python3.10/dist-packages/torch/utils/_stats.py", line 27, in wrapper
V0502 18:42:26.935000 635 torch/fx/experimental/symbolic_shapes.py:5984] [16/0] return fn(*args, **kwargs)
V0502 18:42:26.935000 635 torch/fx/experimental/symbolic_shapes.py:5984] [16/0] File "/usr/local/lib/python3.10/dist-packages/torch/_subclasses/fake_tensor.py", line 1282, in __torch_dispatch__
V0502 18:42:26.935000 635 torch/fx/experimental/symbolic_shapes.py:5984] [16/0] return self.dispatch(func, types, args, kwargs)
V0502 18:42:26.935000 635 torch/fx/experimental/symbolic_shapes.py:5984] [16/0] File "/usr/local/lib/python3.10/dist-packages/torch/_subclasses/fake_tensor.py", line 1823, in dispatch
V0502 18:42:26.935000 635 torch/fx/experimental/symbolic_shapes.py:5984] [16/0] return self._cached_dispatch_impl(func, types, args, kwargs)
V0502 18:42:26.935000 635 torch/fx/experimental/symbolic_shapes.py:5984] [16/0] File "/usr/local/lib/python3.10/dist-packages/torch/_subclasses/fake_tensor.py", line 1393, in _cached_dispatch_impl
V0502 18:42:26.935000 635 torch/fx/experimental/symbolic_shapes.py:5984] [16/0] output = self._dispatch_impl(func, types, args, kwargs)
V0502 18:42:26.935000 635 torch/fx/experimental/symbolic_shapes.py:5984] [16/0] File "/usr/local/lib/python3.10/dist-packages/torch/_subclasses/fake_tensor.py", line 2397, in _dispatch_impl
V0502 18:42:26.935000 635 torch/fx/experimental/symbolic_shapes.py:5984] [16/0] op_impl_out = op_impl(self, func, *args, **kwargs)
V0502 18:42:26.935000 635 torch/fx/experimental/symbolic_shapes.py:5984] [16/0] File "/usr/local/lib/python3.10/dist-packages/torch/_subclasses/fake_impls.py", line 160, in dispatch_to_op_implementations_dict
V0502 18:42:26.935000 635 torch/fx/experimental/symbolic_shapes.py:5984] [16/0] return op_implementations_dict[func](fake_mode, func, *args, **kwargs)
V0502 18:42:26.935000 635 torch/fx/experimental/symbolic_shapes.py:5984] [16/0] File "/usr/local/lib/python3.10/dist-packages/torch/_subclasses/fake_impls.py", line 422, in local_scalar_dense
V0502 18:42:26.935000 635 torch/fx/experimental/symbolic_shapes.py:5984] [16/0] r = fake_mode.shape_env.create_unbacked_symint()
V0502 18:42:26.935000 635 torch/fx/experimental/symbolic_shapes.py:5984] [16/0] File "/usr/local/lib/python3.10/dist-packages/torch/fx/experimental/recording.py", line 263, in wrapper
V0502 18:42:26.935000 635 torch/fx/experimental/symbolic_shapes.py:5984] [16/0] return retlog(fn(*args, **kwargs))
V0502 18:42:26.935000 635 torch/fx/experimental/symbolic_shapes.py:5984] [16/0]
W0502 18:42:26.943000 635 torch/fx/experimental/symbolic_shapes.py:6679] [16/0] failed during evaluate_expr(-u0 > 60, hint=None, size_oblivious=True, forcing_spec=False
E0502 18:42:26.943000 635 torch/fx/experimental/recording.py:299] [16/0] failed while running evaluate_expr(*(-u0 > 60, None, False, True), **{})
E0502 18:42:26.943000 635 torch/fx/experimental/recording.py:299] [16/0] Traceback (most recent call last):
E0502 18:42:26.943000 635 torch/fx/experimental/recording.py:299] [16/0] File "/usr/local/lib/python3.10/dist-packages/torch/fx/experimental/recording.py", line 263, in wrapper
E0502 18:42:26.943000 635 torch/fx/experimental/recording.py:299] [16/0] return retlog(fn(*args, **kwargs))
E0502 18:42:26.943000 635 torch/fx/experimental/recording.py:299] [16/0] File "/usr/local/lib/python3.10/dist-packages/torch/fx/experimental/symbolic_shapes.py", line 6671, in evaluate_expr
E0502 18:42:26.943000 635 torch/fx/experimental/recording.py:299] [16/0] return self._evaluate_expr(
E0502 18:42:26.943000 635 torch/fx/experimental/recording.py:299] [16/0] File "/usr/local/lib/python3.10/dist-packages/torch/fx/experimental/symbolic_shapes.py", line 6894, in _evaluate_expr
E0502 18:42:26.943000 635 torch/fx/experimental/recording.py:299] [16/0] raise self._make_data_dependent_error(
E0502 18:42:26.943000 635 torch/fx/experimental/recording.py:299] [16/0] torch.fx.experimental.symbolic_shapes.GuardOnDataDependentSymNode: Could not guard on data-dependent expression -u0 > 60 (unhinted: -u0 > 60). (Size-like symbols: none)
E0502 18:42:26.943000 635 torch/fx/experimental/recording.py:299] [16/0]
E0502 18:42:26.943000 635 torch/fx/experimental/recording.py:299] [16/0] Caused by: return y[a] # ar/lib/workspace/intermediate_source/torch_export_tutorial.py:702 in forward (_meta_registrations.py:5278 in meta_select)
E0502 18:42:26.943000 635 torch/fx/experimental/recording.py:299] [16/0] For more information, run with TORCH_LOGS="dynamic"
E0502 18:42:26.943000 635 torch/fx/experimental/recording.py:299] [16/0] For extended logs when we create symbols, also add TORCHDYNAMO_EXTENDED_DEBUG_CREATE_SYMBOL="u0"
E0502 18:42:26.943000 635 torch/fx/experimental/recording.py:299] [16/0] If you suspect the guard was triggered from C++, add TORCHDYNAMO_EXTENDED_DEBUG_CPP=1
E0502 18:42:26.943000 635 torch/fx/experimental/recording.py:299] [16/0] For more debugging help, see https://docs.google.com/document/d/1HSuTTVvYH1pTew89Rtpeu84Ht3nQEFTYhAX3Ypa_xJs/edit?usp=sharing
E0502 18:42:26.943000 635 torch/fx/experimental/recording.py:299] [16/0]
E0502 18:42:26.943000 635 torch/fx/experimental/recording.py:299] [16/0] User Stack (most recent call last):
E0502 18:42:26.943000 635 torch/fx/experimental/recording.py:299] [16/0] (snipped, see stack below for prefix)
E0502 18:42:26.943000 635 torch/fx/experimental/recording.py:299] [16/0] File "/var/lib/workspace/intermediate_source/torch_export_tutorial.py", line 702, in forward
E0502 18:42:26.943000 635 torch/fx/experimental/recording.py:299] [16/0] return y[a]
E0502 18:42:26.943000 635 torch/fx/experimental/recording.py:299] [16/0]
E0502 18:42:26.943000 635 torch/fx/experimental/recording.py:299] [16/0] For C++ stack trace, run with TORCHDYNAMO_EXTENDED_DEBUG_CPP=1
E0502 18:42:26.945000 635 torch/_subclasses/fake_tensor.py:2431] [16/0] failed while attempting to run meta for aten.select.int
E0502 18:42:26.945000 635 torch/_subclasses/fake_tensor.py:2431] [16/0] Traceback (most recent call last):
E0502 18:42:26.945000 635 torch/_subclasses/fake_tensor.py:2431] [16/0] File "/usr/local/lib/python3.10/dist-packages/torch/_subclasses/fake_tensor.py", line 2427, in _dispatch_impl
E0502 18:42:26.945000 635 torch/_subclasses/fake_tensor.py:2431] [16/0] r = func(*args, **kwargs)
E0502 18:42:26.945000 635 torch/_subclasses/fake_tensor.py:2431] [16/0] File "/usr/local/lib/python3.10/dist-packages/torch/_ops.py", line 756, in __call__
E0502 18:42:26.945000 635 torch/_subclasses/fake_tensor.py:2431] [16/0] return self._op(*args, **kwargs)
E0502 18:42:26.945000 635 torch/_subclasses/fake_tensor.py:2431] [16/0] File "/usr/local/lib/python3.10/dist-packages/torch/_meta_registrations.py", line 5278, in meta_select
E0502 18:42:26.945000 635 torch/_subclasses/fake_tensor.py:2431] [16/0] guard_size_oblivious(-index > size) or guard_size_oblivious(index >= size)
E0502 18:42:26.945000 635 torch/_subclasses/fake_tensor.py:2431] [16/0] File "/usr/local/lib/python3.10/dist-packages/torch/fx/experimental/symbolic_shapes.py", line 408, in guard_size_oblivious
E0502 18:42:26.945000 635 torch/_subclasses/fake_tensor.py:2431] [16/0] return expr.node.guard_size_oblivious("", 0)
E0502 18:42:26.945000 635 torch/_subclasses/fake_tensor.py:2431] [16/0] File "/usr/local/lib/python3.10/dist-packages/torch/fx/experimental/sym_node.py", line 588, in guard_size_oblivious
E0502 18:42:26.945000 635 torch/_subclasses/fake_tensor.py:2431] [16/0] r = self.evaluate(size_oblivious=True)
E0502 18:42:26.945000 635 torch/_subclasses/fake_tensor.py:2431] [16/0] File "/usr/local/lib/python3.10/dist-packages/torch/fx/experimental/sym_node.py", line 510, in evaluate
E0502 18:42:26.945000 635 torch/_subclasses/fake_tensor.py:2431] [16/0] return self.shape_env.evaluate_sym_node(self, size_oblivious)
E0502 18:42:26.945000 635 torch/_subclasses/fake_tensor.py:2431] [16/0] File "/usr/local/lib/python3.10/dist-packages/torch/fx/experimental/symbolic_shapes.py", line 6655, in evaluate_sym_node
E0502 18:42:26.945000 635 torch/_subclasses/fake_tensor.py:2431] [16/0] return self.evaluate_expr(
E0502 18:42:26.945000 635 torch/_subclasses/fake_tensor.py:2431] [16/0] File "/usr/local/lib/python3.10/dist-packages/torch/fx/experimental/recording.py", line 263, in wrapper
E0502 18:42:26.945000 635 torch/_subclasses/fake_tensor.py:2431] [16/0] return retlog(fn(*args, **kwargs))
E0502 18:42:26.945000 635 torch/_subclasses/fake_tensor.py:2431] [16/0] File "/usr/local/lib/python3.10/dist-packages/torch/fx/experimental/symbolic_shapes.py", line 6671, in evaluate_expr
E0502 18:42:26.945000 635 torch/_subclasses/fake_tensor.py:2431] [16/0] return self._evaluate_expr(
E0502 18:42:26.945000 635 torch/_subclasses/fake_tensor.py:2431] [16/0] File "/usr/local/lib/python3.10/dist-packages/torch/fx/experimental/symbolic_shapes.py", line 6894, in _evaluate_expr
E0502 18:42:26.945000 635 torch/_subclasses/fake_tensor.py:2431] [16/0] raise self._make_data_dependent_error(
E0502 18:42:26.945000 635 torch/_subclasses/fake_tensor.py:2431] [16/0] torch.fx.experimental.symbolic_shapes.GuardOnDataDependentSymNode: Could not guard on data-dependent expression -u0 > 60 (unhinted: -u0 > 60). (Size-like symbols: none)
E0502 18:42:26.945000 635 torch/_subclasses/fake_tensor.py:2431] [16/0]
E0502 18:42:26.945000 635 torch/_subclasses/fake_tensor.py:2431] [16/0] Caused by: return y[a] # ar/lib/workspace/intermediate_source/torch_export_tutorial.py:702 in forward (_meta_registrations.py:5278 in meta_select)
E0502 18:42:26.945000 635 torch/_subclasses/fake_tensor.py:2431] [16/0] For more information, run with TORCH_LOGS="dynamic"
E0502 18:42:26.945000 635 torch/_subclasses/fake_tensor.py:2431] [16/0] For extended logs when we create symbols, also add TORCHDYNAMO_EXTENDED_DEBUG_CREATE_SYMBOL="u0"
E0502 18:42:26.945000 635 torch/_subclasses/fake_tensor.py:2431] [16/0] If you suspect the guard was triggered from C++, add TORCHDYNAMO_EXTENDED_DEBUG_CPP=1
E0502 18:42:26.945000 635 torch/_subclasses/fake_tensor.py:2431] [16/0] For more debugging help, see https://docs.google.com/document/d/1HSuTTVvYH1pTew89Rtpeu84Ht3nQEFTYhAX3Ypa_xJs/edit?usp=sharing
E0502 18:42:26.945000 635 torch/_subclasses/fake_tensor.py:2431] [16/0]
E0502 18:42:26.945000 635 torch/_subclasses/fake_tensor.py:2431] [16/0] User Stack (most recent call last):
E0502 18:42:26.945000 635 torch/_subclasses/fake_tensor.py:2431] [16/0] (snipped, see stack below for prefix)
E0502 18:42:26.945000 635 torch/_subclasses/fake_tensor.py:2431] [16/0] File "/var/lib/workspace/intermediate_source/torch_export_tutorial.py", line 702, in forward
E0502 18:42:26.945000 635 torch/_subclasses/fake_tensor.py:2431] [16/0] return y[a]
E0502 18:42:26.945000 635 torch/_subclasses/fake_tensor.py:2431] [16/0]
E0502 18:42:26.945000 635 torch/_subclasses/fake_tensor.py:2431] [16/0] For C++ stack trace, run with TORCHDYNAMO_EXTENDED_DEBUG_CPP=1
Traceback (most recent call last):
File "/var/lib/workspace/intermediate_source/torch_export_tutorial.py", line 709, in <module>
export(Foo(), inps)
File "/usr/local/lib/python3.10/dist-packages/torch/export/__init__.py", line 360, in export
return _export(
File "/usr/local/lib/python3.10/dist-packages/torch/export/_trace.py", line 1092, in wrapper
raise e
File "/usr/local/lib/python3.10/dist-packages/torch/export/_trace.py", line 1065, in wrapper
ep = fn(*args, **kwargs)
File "/usr/local/lib/python3.10/dist-packages/torch/export/exported_program.py", line 121, in wrapper
return fn(*args, **kwargs)
File "/usr/local/lib/python3.10/dist-packages/torch/export/_trace.py", line 2112, in _export
ep = _export_for_training(
File "/usr/local/lib/python3.10/dist-packages/torch/export/_trace.py", line 1092, in wrapper
raise e
File "/usr/local/lib/python3.10/dist-packages/torch/export/_trace.py", line 1065, in wrapper
ep = fn(*args, **kwargs)
File "/usr/local/lib/python3.10/dist-packages/torch/export/exported_program.py", line 121, in wrapper
return fn(*args, **kwargs)
File "/usr/local/lib/python3.10/dist-packages/torch/export/_trace.py", line 1975, in _export_for_training
export_artifact = export_func( # type: ignore[operator]
File "/usr/local/lib/python3.10/dist-packages/torch/export/_trace.py", line 1344, in _strict_export_lower_to_aten_ir
gm_torch_level = _export_to_torch_ir(
File "/usr/local/lib/python3.10/dist-packages/torch/export/_trace.py", line 739, in _export_to_torch_ir
gm_torch_level, _ = torch._dynamo.export(
File "/usr/local/lib/python3.10/dist-packages/torch/_dynamo/eval_frame.py", line 1677, in inner
result_traced = opt_f(*args, **kwargs)
File "/usr/local/lib/python3.10/dist-packages/torch/nn/modules/module.py", line 1751, in _wrapped_call_impl
return self._call_impl(*args, **kwargs)
File "/usr/local/lib/python3.10/dist-packages/torch/nn/modules/module.py", line 1762, in _call_impl
return forward_call(*args, **kwargs)
File "/usr/local/lib/python3.10/dist-packages/torch/_dynamo/eval_frame.py", line 659, in _fn
raise e.with_traceback(None) from None
torch._dynamo.exc.UserError: Could not guard on data-dependent expression -u0 > 60 (unhinted: -u0 > 60). (Size-like symbols: none)
Caused by: return y[a] # ar/lib/workspace/intermediate_source/torch_export_tutorial.py:702 in forward (_meta_registrations.py:5278 in meta_select)
For more information, run with TORCH_LOGS="dynamic"
For extended logs when we create symbols, also add TORCHDYNAMO_EXTENDED_DEBUG_CREATE_SYMBOL="u0"
If you suspect the guard was triggered from C++, add TORCHDYNAMO_EXTENDED_DEBUG_CPP=1
For more debugging help, see https://docs.google.com/document/d/1HSuTTVvYH1pTew89Rtpeu84Ht3nQEFTYhAX3Ypa_xJs/edit?usp=sharing
User Stack (most recent call last):
(snipped, see stack below for prefix)
File "/var/lib/workspace/intermediate_source/torch_export_tutorial.py", line 702, in forward
return y[a]
For C++ stack trace, run with TORCHDYNAMO_EXTENDED_DEBUG_CPP=1
For more information about this error, see: https://pytorch.org/docs/main/generated/exportdb/index.html#constrain-as-size-example
from user code:
File "/var/lib/workspace/intermediate_source/torch_export_tutorial.py", line 702, in forward
return y[a]
Set TORCHDYNAMO_VERBOSE=1 for the internal stack trace (please do this especially if you're reporting a bug to PyTorch). For even more developer context, set TORCH_LOGS="+dynamo"
Here is a scenario where torch._check() insertion is required simply to prevent an operation from failing. The export call will fail with
“Could not guard on data-dependent expression -u0 > 60”, implying that the compiler doesn’t know if this is a valid indexing operation -
if the value of x is out-of-bounds for y or not. Here, manual specialization is too prohibitive, and torch.cond() has no place.
Instead, informing the compiler of u0’s range is sufficient:
class Foo(torch.nn.Module):
def forward(self, x, y):
a = x.item()
torch._check(a >= 0)
torch._check(a < y.shape[0])
return y[a]
inps = (
torch.tensor(32),
torch.randn(60),
)
ep = export(Foo(), inps)
print(ep)
I0502 18:42:26.957000 635 torch/fx/experimental/symbolic_shapes.py:3334] [17/0] create_env
I0502 18:42:26.961000 635 torch/fx/experimental/symbolic_shapes.py:4276] [17/0] create_unbacked_symint u0 [-int_oo, int_oo] a = x.item() # ar/lib/workspace/intermediate_source/torch_export_tutorial.py:721 in forward (_subclasses/fake_impls.py:422 in local_scalar_dense)
I0502 18:42:26.961000 635 torch/fx/experimental/symbolic_shapes.py:1130] [17/0] compute_unbacked_bindings [u0]
I0502 18:42:26.963000 635 torch/fx/experimental/symbolic_shapes.py:6630] [17/0] runtime_assert u0 >= 0 [guard added] torch._check(a >= 0) # ar/lib/workspace/intermediate_source/torch_export_tutorial.py:722 in forward (_dynamo/utils.py:3284 in run_node), for more info run with TORCHDYNAMO_EXTENDED_DEBUG_GUARD_ADDED="u0 >= 0"
V0502 18:42:26.964000 635 torch/fx/experimental/symbolic_shapes.py:6071] [17/0] _update_var_to_range u0 = VR[0, int_oo] (update)
I0502 18:42:26.967000 635 torch/fx/experimental/symbolic_shapes.py:6630] [17/0] runtime_assert u0 < 60 [guard added] torch._check(a < y.shape[0]) # ar/lib/workspace/intermediate_source/torch_export_tutorial.py:723 in forward (_dynamo/utils.py:3284 in run_node), for more info run with TORCHDYNAMO_EXTENDED_DEBUG_GUARD_ADDED="u0 < 60"
V0502 18:42:26.968000 635 torch/fx/experimental/symbolic_shapes.py:6071] [17/0] _update_var_to_range u0 = VR[0, 59] (update)
V0502 18:42:26.970000 635 torch/fx/experimental/symbolic_shapes.py:6787] [17/0] eval size_oblivious(-u0 > 60) == False [statically known]
V0502 18:42:26.970000 635 torch/fx/experimental/symbolic_shapes.py:6787] [17/0] eval size_oblivious(u0 >= 60) == False [statically known]
V0502 18:42:26.971000 635 torch/fx/experimental/symbolic_shapes.py:6787] [17/0] eval False == True [statically known]
V0502 18:42:26.974000 635 torch/fx/experimental/symbolic_shapes.py:7018] [17/0] runtime_assert u0 >= 0 == True [statically known]
V0502 18:42:26.975000 635 torch/fx/experimental/symbolic_shapes.py:7018] [17/0] runtime_assert u0 <= 59 == True [statically known]
V0502 18:42:26.976000 635 torch/fx/experimental/symbolic_shapes.py:7018] [17/0] runtime_assert u0 < 60 == True [statically known]
I0502 18:42:26.978000 635 torch/fx/experimental/symbolic_shapes.py:4734] [17/0] produce_guards
V0502 18:42:26.979000 635 torch/fx/experimental/symbolic_shapes.py:4954] [17/0] track_symint L['x'].storage_offset() 0 None
V0502 18:42:26.979000 635 torch/fx/experimental/symbolic_shapes.py:4954] [17/0] track_symint L['y'].size()[0] 60 None
V0502 18:42:26.979000 635 torch/fx/experimental/symbolic_shapes.py:4954] [17/0] track_symint L['y'].stride()[0] 1 None
V0502 18:42:26.980000 635 torch/fx/experimental/symbolic_shapes.py:4954] [17/0] track_symint L['y'].storage_offset() 0 None
I0502 18:42:26.993000 635 torch/fx/experimental/symbolic_shapes.py:4276] create_unbacked_symint u1 [-int_oo, int_oo] (_subclasses/fake_impls.py:422 in local_scalar_dense)
I0502 18:42:26.994000 635 torch/fx/experimental/symbolic_shapes.py:1130] compute_unbacked_bindings [u1]
V0502 18:42:26.994000 635 torch/fx/experimental/symbolic_shapes.py:6071] _update_var_to_range u1 = VR[0, 59] (update)
I0502 18:42:26.994000 635 torch/fx/experimental/symbolic_shapes.py:6234] set_replacement u1 = u0 (rename_unbacked_to) VR[0, 59]
ExportedProgram:
class GraphModule(torch.nn.Module):
def forward(self, x: "i64[]", y: "f32[60]"):
# File: /var/lib/workspace/intermediate_source/torch_export_tutorial.py:721 in forward, code: a = x.item()
item: "Sym(u0)" = torch.ops.aten.item.default(x); x = None
ge_1: "Sym(u0 >= 0)" = item >= 0
_assert_scalar_default = torch.ops.aten._assert_scalar.default(ge_1, "Runtime assertion failed for expression u0 >= 0 on node 'ge_1'"); ge_1 = _assert_scalar_default = None
le_1: "Sym(u0 <= 59)" = item <= 59
_assert_scalar_default_1 = torch.ops.aten._assert_scalar.default(le_1, "Runtime assertion failed for expression u0 <= 59 on node 'le_1'"); le_1 = _assert_scalar_default_1 = None
#
lt_1: "Sym(u0 < 60)" = item < 60
_assert_scalar_default_2 = torch.ops.aten._assert_scalar.default(lt_1, "Runtime assertion failed for expression u0 < 60 on node 'lt_1'"); lt_1 = _assert_scalar_default_2 = None
# File: /var/lib/workspace/intermediate_source/torch_export_tutorial.py:724 in forward, code: return y[a]
select: "f32[]" = torch.ops.aten.select.int(y, 0, item); y = item = None
return (select,)
Graph signature: ExportGraphSignature(input_specs=[InputSpec(kind=<InputKind.USER_INPUT: 1>, arg=TensorArgument(name='x'), target=None, persistent=None), InputSpec(kind=<InputKind.USER_INPUT: 1>, arg=TensorArgument(name='y'), target=None, persistent=None)], output_specs=[OutputSpec(kind=<OutputKind.USER_OUTPUT: 1>, arg=TensorArgument(name='select'), target=None)])
Range constraints: {u0: VR[0, 59], u1: VR[0, 59]}
Specialized values
Another category of data-dependent error happens when the program attempts to extract a concrete data-dependent integer/float value while tracing. This looks something like “Could not extract specialized integer from data-dependent expression”, and is analogous to the previous class of errors - if these occur when attempting to evaluate concrete integer/float values, data-dependent guard errors arise with evaluating concrete boolean values.
This error typically occurs when there is an explicit or implicit int() cast on a data-dependent expression. For example, this list comprehension
has a range() call that implicitly does an int() cast on the size of the list:
class Foo(torch.nn.Module):
def forward(self, x, y):
a = x.item()
b = torch.cat([y for y in range(a)], dim=0)
return b + int(a)
inps = (
torch.tensor(32),
torch.randn(60),
)
try:
export(Foo(), inps, strict=False)
except Exception:
tb.print_exc()
I0502 18:42:27.010000 635 torch/fx/experimental/symbolic_shapes.py:3334] create_env
I0502 18:42:27.016000 635 torch/fx/experimental/symbolic_shapes.py:4276] create_unbacked_symint u0 [-int_oo, int_oo] (_subclasses/fake_impls.py:422 in local_scalar_dense)
I0502 18:42:27.016000 635 torch/fx/experimental/symbolic_shapes.py:1130] compute_unbacked_bindings [u0]
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] Data dependent variable 'u0' allocated at:
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] File "/usr/local/bin/sphinx-build", line 8, in <module>
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] sys.exit(main())
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] File "/usr/local/lib/python3.10/dist-packages/sphinx/cmd/build.py", line 288, in main
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] return make_main(argv)
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] File "/usr/local/lib/python3.10/dist-packages/sphinx/cmd/build.py", line 193, in make_main
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] return make_mode.run_make_mode(argv[1:])
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] File "/usr/local/lib/python3.10/dist-packages/sphinx/cmd/make_mode.py", line 160, in run_make_mode
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] return make.run_generic_build(args[0])
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] File "/usr/local/lib/python3.10/dist-packages/sphinx/cmd/make_mode.py", line 148, in run_generic_build
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] return build_main(args + opts)
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] File "/usr/local/lib/python3.10/dist-packages/sphinx/cmd/build.py", line 272, in build_main
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] app = Sphinx(args.sourcedir, args.confdir, args.outputdir,
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] File "/usr/local/lib/python3.10/dist-packages/sphinx/application.py", line 256, in __init__
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] self._init_builder()
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] File "/usr/local/lib/python3.10/dist-packages/sphinx/application.py", line 314, in _init_builder
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] self.events.emit('builder-inited')
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] File "/usr/local/lib/python3.10/dist-packages/sphinx/events.py", line 94, in emit
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] results.append(listener.handler(self.app, *args))
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] File "/usr/local/lib/python3.10/dist-packages/sphinx_gallery/gen_gallery.py", line 491, in generate_gallery_rst
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] ) = generate_dir_rst(
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] File "/usr/local/lib/python3.10/dist-packages/sphinx_gallery/gen_rst.py", line 431, in generate_dir_rst
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] intro, title, cost = generate_file_rst(
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] File "/usr/local/lib/python3.10/dist-packages/sphinx_gallery/gen_rst.py", line 1027, in generate_file_rst
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] output_blocks, time_elapsed = execute_script(script_blocks,
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] File "/usr/local/lib/python3.10/dist-packages/sphinx_gallery/gen_rst.py", line 945, in execute_script
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] output_blocks.append(execute_code_block(
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] File "/usr/local/lib/python3.10/dist-packages/sphinx_gallery/gen_rst.py", line 810, in execute_code_block
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] is_last_expr, mem_max = _exec_and_get_memory(
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] File "/usr/local/lib/python3.10/dist-packages/sphinx_gallery/gen_rst.py", line 676, in _exec_and_get_memory
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] mem_max, _ = gallery_conf['call_memory'](
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] File "/usr/local/lib/python3.10/dist-packages/sphinx_gallery/gen_gallery.py", line 223, in call_memory
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] return 0., func()
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] File "/usr/local/lib/python3.10/dist-packages/sphinx_gallery/gen_rst.py", line 600, in __call__
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] exec(self.code, self.fake_main.__dict__)
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] File "/var/lib/workspace/intermediate_source/torch_export_tutorial.py", line 756, in <module>
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] export(Foo(), inps, strict=False)
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] File "/usr/local/lib/python3.10/dist-packages/torch/export/__init__.py", line 360, in export
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] return _export(
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] File "/usr/local/lib/python3.10/dist-packages/torch/export/_trace.py", line 1065, in wrapper
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] ep = fn(*args, **kwargs)
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] File "/usr/local/lib/python3.10/dist-packages/torch/export/exported_program.py", line 121, in wrapper
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] return fn(*args, **kwargs)
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] File "/usr/local/lib/python3.10/dist-packages/torch/export/_trace.py", line 2112, in _export
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] ep = _export_for_training(
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] File "/usr/local/lib/python3.10/dist-packages/torch/export/_trace.py", line 1065, in wrapper
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] ep = fn(*args, **kwargs)
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] File "/usr/local/lib/python3.10/dist-packages/torch/export/exported_program.py", line 121, in wrapper
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] return fn(*args, **kwargs)
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] File "/usr/local/lib/python3.10/dist-packages/torch/export/_trace.py", line 1975, in _export_for_training
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] export_artifact = export_func( # type: ignore[operator]
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] File "/usr/local/lib/python3.10/dist-packages/torch/export/_trace.py", line 1910, in _non_strict_export
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] aten_export_artifact = _to_aten_func( # type: ignore[operator]
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] File "/usr/local/lib/python3.10/dist-packages/torch/export/_trace.py", line 1696, in _export_to_aten_ir_make_fx
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] gm, graph_signature = transform(_make_fx_helper)(
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] File "/usr/local/lib/python3.10/dist-packages/torch/export/_trace.py", line 1840, in _aot_export_non_strict
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] gm, sig = aot_export(wrapped_mod, args, kwargs=kwargs, **flags)
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] File "/usr/local/lib/python3.10/dist-packages/torch/export/_trace.py", line 1616, in _make_fx_helper
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] gm = make_fx(
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] File "/usr/local/lib/python3.10/dist-packages/torch/fx/experimental/proxy_tensor.py", line 2240, in wrapped
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] return make_fx_tracer.trace(f, *args)
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] File "/usr/local/lib/python3.10/dist-packages/torch/fx/experimental/proxy_tensor.py", line 2178, in trace
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] return self._trace_inner(f, *args)
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] File "/usr/local/lib/python3.10/dist-packages/torch/fx/experimental/proxy_tensor.py", line 2149, in _trace_inner
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] t = dispatch_trace(
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] File "/usr/local/lib/python3.10/dist-packages/torch/_compile.py", line 51, in inner
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] return disable_fn(*args, **kwargs)
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] File "/usr/local/lib/python3.10/dist-packages/torch/_dynamo/eval_frame.py", line 838, in _fn
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] return fn(*args, **kwargs)
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] File "/usr/local/lib/python3.10/dist-packages/torch/fx/experimental/proxy_tensor.py", line 1174, in dispatch_trace
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] graph = tracer.trace(root, concrete_args) # type: ignore[arg-type]
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] File "/usr/local/lib/python3.10/dist-packages/torch/fx/experimental/proxy_tensor.py", line 1738, in trace
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] res = super().trace(root, concrete_args)
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] File "/usr/local/lib/python3.10/dist-packages/torch/_dynamo/eval_frame.py", line 838, in _fn
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] return fn(*args, **kwargs)
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] File "/usr/local/lib/python3.10/dist-packages/torch/fx/_symbolic_trace.py", line 838, in trace
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] (self.create_arg(fn(*args)),),
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] File "/usr/local/lib/python3.10/dist-packages/torch/fx/experimental/proxy_tensor.py", line 1229, in wrapped
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] out = f(*tensors) # type:ignore[call-arg]
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] File "<string>", line 1, in <lambda>
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] File "/usr/local/lib/python3.10/dist-packages/torch/export/_trace.py", line 1520, in wrapped_fn
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] return tuple(flat_fn(*args))
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] File "/usr/local/lib/python3.10/dist-packages/torch/_functorch/_aot_autograd/utils.py", line 184, in flat_fn
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] tree_out = fn(*args, **kwargs)
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] File "/usr/local/lib/python3.10/dist-packages/torch/_functorch/_aot_autograd/traced_function_transforms.py", line 903, in functional_call
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] out = mod(*args[params_len:], **kwargs)
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] File "/usr/local/lib/python3.10/dist-packages/torch/fx/_symbolic_trace.py", line 813, in module_call_wrapper
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] return self.call_module(mod, forward, args, kwargs)
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] File "/usr/local/lib/python3.10/dist-packages/torch/fx/experimental/proxy_tensor.py", line 1808, in call_module
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] return Tracer.call_module(self, m, forward, args, kwargs)
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] File "/usr/local/lib/python3.10/dist-packages/torch/fx/_symbolic_trace.py", line 531, in call_module
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] ret_val = forward(*args, **kwargs)
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] File "/usr/local/lib/python3.10/dist-packages/torch/fx/_symbolic_trace.py", line 806, in forward
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] return _orig_module_call(mod, *args, **kwargs)
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] File "/usr/local/lib/python3.10/dist-packages/torch/nn/modules/module.py", line 1751, in _wrapped_call_impl
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] return self._call_impl(*args, **kwargs)
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] File "/usr/local/lib/python3.10/dist-packages/torch/nn/modules/module.py", line 1762, in _call_impl
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] return forward_call(*args, **kwargs)
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] File "/usr/local/lib/python3.10/dist-packages/torch/export/_trace.py", line 1824, in forward
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] tree_out = mod(*args, **kwargs)
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] File "/usr/local/lib/python3.10/dist-packages/torch/fx/_symbolic_trace.py", line 813, in module_call_wrapper
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] return self.call_module(mod, forward, args, kwargs)
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] File "/usr/local/lib/python3.10/dist-packages/torch/fx/experimental/proxy_tensor.py", line 1808, in call_module
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] return Tracer.call_module(self, m, forward, args, kwargs)
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] File "/usr/local/lib/python3.10/dist-packages/torch/fx/_symbolic_trace.py", line 531, in call_module
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] ret_val = forward(*args, **kwargs)
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] File "/usr/local/lib/python3.10/dist-packages/torch/fx/_symbolic_trace.py", line 806, in forward
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] return _orig_module_call(mod, *args, **kwargs)
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] File "/usr/local/lib/python3.10/dist-packages/torch/nn/modules/module.py", line 1751, in _wrapped_call_impl
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] return self._call_impl(*args, **kwargs)
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] File "/usr/local/lib/python3.10/dist-packages/torch/nn/modules/module.py", line 1762, in _call_impl
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] return forward_call(*args, **kwargs)
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] File "/var/lib/workspace/intermediate_source/torch_export_tutorial.py", line 747, in forward
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] a = x.item()
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] File "/usr/local/lib/python3.10/dist-packages/torch/fx/experimental/proxy_tensor.py", line 1277, in __torch_function__
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] return func(*args, **kwargs)
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] File "/usr/local/lib/python3.10/dist-packages/torch/fx/experimental/proxy_tensor.py", line 1324, in __torch_function__
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] return func(*args, **kwargs)
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] File "/usr/local/lib/python3.10/dist-packages/torch/_export/non_strict_utils.py", line 683, in __torch_function__
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] return func(*args, **kwargs)
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] File "/usr/local/lib/python3.10/dist-packages/torch/_ops.py", line 875, in handler
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] return torch._library.utils.handle_dispatch_mode(
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] File "/usr/local/lib/python3.10/dist-packages/torch/_library/utils.py", line 296, in handle_dispatch_mode
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] return curr_mode.__torch_dispatch__(op_overload, overload_types, args, kwargs)
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] File "/usr/local/lib/python3.10/dist-packages/torch/utils/_stats.py", line 27, in wrapper
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] return fn(*args, **kwargs)
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] File "/usr/local/lib/python3.10/dist-packages/torch/fx/experimental/proxy_tensor.py", line 1379, in __torch_dispatch__
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] return proxy_call(self, func, self.pre_dispatch, args, kwargs)
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] File "/usr/local/lib/python3.10/dist-packages/torch/fx/experimental/proxy_tensor.py", line 914, in proxy_call
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] out = func(*args, **kwargs)
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] File "/usr/local/lib/python3.10/dist-packages/torch/_ops.py", line 756, in __call__
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] return self._op(*args, **kwargs)
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] File "/usr/local/lib/python3.10/dist-packages/torch/utils/_stats.py", line 27, in wrapper
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] return fn(*args, **kwargs)
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] File "/usr/local/lib/python3.10/dist-packages/torch/_subclasses/fake_tensor.py", line 1282, in __torch_dispatch__
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] return self.dispatch(func, types, args, kwargs)
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] File "/usr/local/lib/python3.10/dist-packages/torch/_subclasses/fake_tensor.py", line 1823, in dispatch
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] return self._cached_dispatch_impl(func, types, args, kwargs)
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] File "/usr/local/lib/python3.10/dist-packages/torch/_subclasses/fake_tensor.py", line 1393, in _cached_dispatch_impl
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] output = self._dispatch_impl(func, types, args, kwargs)
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] File "/usr/local/lib/python3.10/dist-packages/torch/_subclasses/fake_tensor.py", line 2397, in _dispatch_impl
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] op_impl_out = op_impl(self, func, *args, **kwargs)
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] File "/usr/local/lib/python3.10/dist-packages/torch/_subclasses/fake_impls.py", line 160, in dispatch_to_op_implementations_dict
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] return op_implementations_dict[func](fake_mode, func, *args, **kwargs)
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] File "/usr/local/lib/python3.10/dist-packages/torch/_subclasses/fake_impls.py", line 422, in local_scalar_dense
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] r = fake_mode.shape_env.create_unbacked_symint()
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] File "/usr/local/lib/python3.10/dist-packages/torch/fx/experimental/recording.py", line 263, in wrapper
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984] return retlog(fn(*args, **kwargs))
V0502 18:42:27.017000 635 torch/fx/experimental/symbolic_shapes.py:5984]
W0502 18:42:27.026000 635 torch/fx/experimental/symbolic_shapes.py:6679] failed during evaluate_expr(u0, hint=None, size_oblivious=False, forcing_spec=False
E0502 18:42:27.026000 635 torch/fx/experimental/recording.py:299] failed while running evaluate_expr(*(u0, None, False, False), **{})
E0502 18:42:27.026000 635 torch/fx/experimental/recording.py:299] Traceback (most recent call last):
E0502 18:42:27.026000 635 torch/fx/experimental/recording.py:299] File "/usr/local/lib/python3.10/dist-packages/torch/fx/experimental/recording.py", line 263, in wrapper
E0502 18:42:27.026000 635 torch/fx/experimental/recording.py:299] return retlog(fn(*args, **kwargs))
E0502 18:42:27.026000 635 torch/fx/experimental/recording.py:299] File "/usr/local/lib/python3.10/dist-packages/torch/fx/experimental/symbolic_shapes.py", line 6671, in evaluate_expr
E0502 18:42:27.026000 635 torch/fx/experimental/recording.py:299] return self._evaluate_expr(
E0502 18:42:27.026000 635 torch/fx/experimental/recording.py:299] File "/usr/local/lib/python3.10/dist-packages/torch/fx/experimental/symbolic_shapes.py", line 6894, in _evaluate_expr
E0502 18:42:27.026000 635 torch/fx/experimental/recording.py:299] raise self._make_data_dependent_error(
E0502 18:42:27.026000 635 torch/fx/experimental/recording.py:299] torch.fx.experimental.symbolic_shapes.GuardOnDataDependentSymNode: Could not extract specialized integer from data-dependent expression u0 (unhinted: u0). (Size-like symbols: none)
E0502 18:42:27.026000 635 torch/fx/experimental/recording.py:299]
E0502 18:42:27.026000 635 torch/fx/experimental/recording.py:299] Caused by: (ar/lib/workspace/intermediate_source/torch_export_tutorial.py:748 in forward)
E0502 18:42:27.026000 635 torch/fx/experimental/recording.py:299] For more information, run with TORCH_LOGS="dynamic"
E0502 18:42:27.026000 635 torch/fx/experimental/recording.py:299] For extended logs when we create symbols, also add TORCHDYNAMO_EXTENDED_DEBUG_CREATE_SYMBOL="u0"
E0502 18:42:27.026000 635 torch/fx/experimental/recording.py:299] If you suspect the guard was triggered from C++, add TORCHDYNAMO_EXTENDED_DEBUG_CPP=1
E0502 18:42:27.026000 635 torch/fx/experimental/recording.py:299] For more debugging help, see https://docs.google.com/document/d/1HSuTTVvYH1pTew89Rtpeu84Ht3nQEFTYhAX3Ypa_xJs/edit?usp=sharing
E0502 18:42:27.026000 635 torch/fx/experimental/recording.py:299]
E0502 18:42:27.026000 635 torch/fx/experimental/recording.py:299] For C++ stack trace, run with TORCHDYNAMO_EXTENDED_DEBUG_CPP=1
def forward(self, arg0_1: "i64[]", arg1_1: "f32[60]"):
# File: /var/lib/workspace/intermediate_source/torch_export_tutorial.py:747 in forward, code: a = x.item()
item: "Sym(u0)" = torch.ops.aten.item.default(arg0_1); arg0_1 = item = None
Traceback (most recent call last):
File "/var/lib/workspace/intermediate_source/torch_export_tutorial.py", line 756, in <module>
export(Foo(), inps, strict=False)
File "/usr/local/lib/python3.10/dist-packages/torch/export/__init__.py", line 360, in export
return _export(
File "/usr/local/lib/python3.10/dist-packages/torch/export/_trace.py", line 1092, in wrapper
raise e
File "/usr/local/lib/python3.10/dist-packages/torch/export/_trace.py", line 1065, in wrapper
ep = fn(*args, **kwargs)
File "/usr/local/lib/python3.10/dist-packages/torch/export/exported_program.py", line 121, in wrapper
return fn(*args, **kwargs)
File "/usr/local/lib/python3.10/dist-packages/torch/export/_trace.py", line 2112, in _export
ep = _export_for_training(
File "/usr/local/lib/python3.10/dist-packages/torch/export/_trace.py", line 1092, in wrapper
raise e
File "/usr/local/lib/python3.10/dist-packages/torch/export/_trace.py", line 1065, in wrapper
ep = fn(*args, **kwargs)
File "/usr/local/lib/python3.10/dist-packages/torch/export/exported_program.py", line 121, in wrapper
return fn(*args, **kwargs)
File "/usr/local/lib/python3.10/dist-packages/torch/export/_trace.py", line 1975, in _export_for_training
export_artifact = export_func( # type: ignore[operator]
File "/usr/local/lib/python3.10/dist-packages/torch/export/_trace.py", line 1910, in _non_strict_export
aten_export_artifact = _to_aten_func( # type: ignore[operator]
File "/usr/local/lib/python3.10/dist-packages/torch/export/_trace.py", line 1696, in _export_to_aten_ir_make_fx
gm, graph_signature = transform(_make_fx_helper)(
File "/usr/local/lib/python3.10/dist-packages/torch/export/_trace.py", line 1840, in _aot_export_non_strict
gm, sig = aot_export(wrapped_mod, args, kwargs=kwargs, **flags)
File "/usr/local/lib/python3.10/dist-packages/torch/export/_trace.py", line 1616, in _make_fx_helper
gm = make_fx(
File "/usr/local/lib/python3.10/dist-packages/torch/fx/experimental/proxy_tensor.py", line 2240, in wrapped
return make_fx_tracer.trace(f, *args)
File "/usr/local/lib/python3.10/dist-packages/torch/fx/experimental/proxy_tensor.py", line 2178, in trace
return self._trace_inner(f, *args)
File "/usr/local/lib/python3.10/dist-packages/torch/fx/experimental/proxy_tensor.py", line 2149, in _trace_inner
t = dispatch_trace(
File "/usr/local/lib/python3.10/dist-packages/torch/_compile.py", line 51, in inner
return disable_fn(*args, **kwargs)
File "/usr/local/lib/python3.10/dist-packages/torch/_dynamo/eval_frame.py", line 838, in _fn
return fn(*args, **kwargs)
File "/usr/local/lib/python3.10/dist-packages/torch/fx/experimental/proxy_tensor.py", line 1174, in dispatch_trace
graph = tracer.trace(root, concrete_args) # type: ignore[arg-type]
File "/usr/local/lib/python3.10/dist-packages/torch/fx/experimental/proxy_tensor.py", line 1738, in trace
res = super().trace(root, concrete_args)
File "/usr/local/lib/python3.10/dist-packages/torch/_dynamo/eval_frame.py", line 838, in _fn
return fn(*args, **kwargs)
File "/usr/local/lib/python3.10/dist-packages/torch/fx/_symbolic_trace.py", line 838, in trace
(self.create_arg(fn(*args)),),
File "/usr/local/lib/python3.10/dist-packages/torch/fx/experimental/proxy_tensor.py", line 1229, in wrapped
out = f(*tensors) # type:ignore[call-arg]
File "<string>", line 1, in <lambda>
File "/usr/local/lib/python3.10/dist-packages/torch/export/_trace.py", line 1520, in wrapped_fn
return tuple(flat_fn(*args))
File "/usr/local/lib/python3.10/dist-packages/torch/_functorch/_aot_autograd/utils.py", line 184, in flat_fn
tree_out = fn(*args, **kwargs)
File "/usr/local/lib/python3.10/dist-packages/torch/_functorch/_aot_autograd/traced_function_transforms.py", line 903, in functional_call
out = mod(*args[params_len:], **kwargs)
File "/usr/local/lib/python3.10/dist-packages/torch/fx/_symbolic_trace.py", line 813, in module_call_wrapper
return self.call_module(mod, forward, args, kwargs)
File "/usr/local/lib/python3.10/dist-packages/torch/fx/experimental/proxy_tensor.py", line 1808, in call_module
return Tracer.call_module(self, m, forward, args, kwargs)
File "/usr/local/lib/python3.10/dist-packages/torch/fx/_symbolic_trace.py", line 531, in call_module
ret_val = forward(*args, **kwargs)
File "/usr/local/lib/python3.10/dist-packages/torch/fx/_symbolic_trace.py", line 806, in forward
return _orig_module_call(mod, *args, **kwargs)
File "/usr/local/lib/python3.10/dist-packages/torch/nn/modules/module.py", line 1751, in _wrapped_call_impl
return self._call_impl(*args, **kwargs)
File "/usr/local/lib/python3.10/dist-packages/torch/nn/modules/module.py", line 1762, in _call_impl
return forward_call(*args, **kwargs)
File "/usr/local/lib/python3.10/dist-packages/torch/export/_trace.py", line 1824, in forward
tree_out = mod(*args, **kwargs)
File "/usr/local/lib/python3.10/dist-packages/torch/fx/_symbolic_trace.py", line 813, in module_call_wrapper
return self.call_module(mod, forward, args, kwargs)
File "/usr/local/lib/python3.10/dist-packages/torch/fx/experimental/proxy_tensor.py", line 1808, in call_module
return Tracer.call_module(self, m, forward, args, kwargs)
File "/usr/local/lib/python3.10/dist-packages/torch/fx/_symbolic_trace.py", line 531, in call_module
ret_val = forward(*args, **kwargs)
File "/usr/local/lib/python3.10/dist-packages/torch/fx/_symbolic_trace.py", line 806, in forward
return _orig_module_call(mod, *args, **kwargs)
File "/usr/local/lib/python3.10/dist-packages/torch/nn/modules/module.py", line 1751, in _wrapped_call_impl
return self._call_impl(*args, **kwargs)
File "/usr/local/lib/python3.10/dist-packages/torch/nn/modules/module.py", line 1762, in _call_impl
return forward_call(*args, **kwargs)
File "/var/lib/workspace/intermediate_source/torch_export_tutorial.py", line 748, in forward
b = torch.cat([y for y in range(a)], dim=0)
File "/usr/local/lib/python3.10/dist-packages/torch/__init__.py", line 431, in __index__
return self.node.int_()
File "/usr/local/lib/python3.10/dist-packages/torch/fx/experimental/sym_node.py", line 466, in int_
return self.guard_int("", 0) # NB: uses Python backtrace
File "/usr/local/lib/python3.10/dist-packages/torch/fx/experimental/sym_node.py", line 516, in guard_int
r = self.evaluate()
File "/usr/local/lib/python3.10/dist-packages/torch/fx/experimental/sym_node.py", line 510, in evaluate
return self.shape_env.evaluate_sym_node(self, size_oblivious)
File "/usr/local/lib/python3.10/dist-packages/torch/fx/experimental/symbolic_shapes.py", line 6655, in evaluate_sym_node
return self.evaluate_expr(
File "/usr/local/lib/python3.10/dist-packages/torch/fx/experimental/recording.py", line 263, in wrapper
return retlog(fn(*args, **kwargs))
File "/usr/local/lib/python3.10/dist-packages/torch/fx/experimental/symbolic_shapes.py", line 6671, in evaluate_expr
return self._evaluate_expr(
File "/usr/local/lib/python3.10/dist-packages/torch/fx/experimental/symbolic_shapes.py", line 6894, in _evaluate_expr
raise self._make_data_dependent_error(
torch.fx.experimental.symbolic_shapes.GuardOnDataDependentSymNode: Could not extract specialized integer from data-dependent expression u0 (unhinted: u0). (Size-like symbols: none)
Caused by: (ar/lib/workspace/intermediate_source/torch_export_tutorial.py:748 in forward)
For more information, run with TORCH_LOGS="dynamic"
For extended logs when we create symbols, also add TORCHDYNAMO_EXTENDED_DEBUG_CREATE_SYMBOL="u0"
If you suspect the guard was triggered from C++, add TORCHDYNAMO_EXTENDED_DEBUG_CPP=1
For more debugging help, see https://docs.google.com/document/d/1HSuTTVvYH1pTew89Rtpeu84Ht3nQEFTYhAX3Ypa_xJs/edit?usp=sharing
For C++ stack trace, run with TORCHDYNAMO_EXTENDED_DEBUG_CPP=1
For these errors, some basic options you have are:
Avoid unnecessary
int()cast calls, in this case theint(a)in the return statement.Use
torch._check()calls; unfortunately all you may be able to do in this case is specialize (withtorch._check(a == 60)).Rewrite the offending code at a higher level. For example, the list comprehension is semantically a
repeat()op, which doesn’t involve anint()cast. The following rewrite avoids data-dependent errors:
class Foo(torch.nn.Module):
def forward(self, x, y):
a = x.item()
b = y.unsqueeze(0).repeat(a, 1)
return b + a
inps = (
torch.tensor(32),
torch.randn(60),
)
ep = export(Foo(), inps, strict=False)
print(ep)
I0502 18:42:27.037000 635 torch/fx/experimental/symbolic_shapes.py:3334] create_env
I0502 18:42:27.043000 635 torch/fx/experimental/symbolic_shapes.py:4276] create_unbacked_symint u0 [-int_oo, int_oo] (_subclasses/fake_impls.py:422 in local_scalar_dense)
I0502 18:42:27.043000 635 torch/fx/experimental/symbolic_shapes.py:1130] compute_unbacked_bindings [u0]
I0502 18:42:27.047000 635 torch/fx/experimental/symbolic_shapes.py:6630] runtime_assert u0 >= 0 [guard added] (_refs/__init__.py:4796 in new_empty), for more info run with TORCHDYNAMO_EXTENDED_DEBUG_GUARD_ADDED="u0 >= 0"
V0502 18:42:27.047000 635 torch/fx/experimental/symbolic_shapes.py:6071] _update_var_to_range u0 = VR[0, int_oo] (update)
V0502 18:42:27.050000 635 torch/fx/experimental/symbolic_shapes.py:6787] eval size_oblivious(Eq(u0, 0)) == False [statically known]
V0502 18:42:27.053000 635 torch/fx/experimental/symbolic_shapes.py:6787] eval size_oblivious(Eq(u0, 1)) == False [statically known]
V0502 18:42:27.053000 635 torch/fx/experimental/symbolic_shapes.py:7018] runtime_assert True == True [statically known]
I0502 18:42:27.057000 635 torch/fx/experimental/symbolic_shapes.py:4734] produce_guards
V0502 18:42:27.057000 635 torch/fx/experimental/symbolic_shapes.py:4954] track_symint L['args'][0][0].storage_offset() 0 None
V0502 18:42:27.057000 635 torch/fx/experimental/symbolic_shapes.py:4954] track_symint L['args'][0][1].size()[0] 60 None
V0502 18:42:27.058000 635 torch/fx/experimental/symbolic_shapes.py:4954] track_symint L['args'][0][1].stride()[0] 1 None
V0502 18:42:27.058000 635 torch/fx/experimental/symbolic_shapes.py:4954] track_symint L['args'][0][1].storage_offset() 0 None
V0502 18:42:27.059000 635 torch/fx/experimental/symbolic_shapes.py:7018] runtime_assert u0 >= 0 == True [statically known]
ExportedProgram:
class GraphModule(torch.nn.Module):
def forward(self, x: "i64[]", y: "f32[60]"):
# File: /var/lib/workspace/intermediate_source/torch_export_tutorial.py:769 in forward, code: a = x.item()
item: "Sym(u0)" = torch.ops.aten.item.default(x); x = None
#
sym_constrain_range_for_size_default = torch.ops.aten.sym_constrain_range_for_size.default(item); sym_constrain_range_for_size_default = None
# File: /var/lib/workspace/intermediate_source/torch_export_tutorial.py:769 in forward, code: a = x.item()
ge: "Sym(u0 >= 0)" = item >= 0
_assert_scalar_default = torch.ops.aten._assert_scalar.default(ge, "Runtime assertion failed for expression u0 >= 0 on node 'ge'"); ge = _assert_scalar_default = None
# File: /var/lib/workspace/intermediate_source/torch_export_tutorial.py:770 in forward, code: b = y.unsqueeze(0).repeat(a, 1)
unsqueeze: "f32[1, 60]" = torch.ops.aten.unsqueeze.default(y, 0); y = None
repeat: "f32[u0, 60]" = torch.ops.aten.repeat.default(unsqueeze, [item, 1]); unsqueeze = None
# File: /var/lib/workspace/intermediate_source/torch_export_tutorial.py:771 in forward, code: return b + a
add: "f32[u0, 60]" = torch.ops.aten.add.Tensor(repeat, item); repeat = item = None
return (add,)
Graph signature: ExportGraphSignature(input_specs=[InputSpec(kind=<InputKind.USER_INPUT: 1>, arg=TensorArgument(name='x'), target=None, persistent=None), InputSpec(kind=<InputKind.USER_INPUT: 1>, arg=TensorArgument(name='y'), target=None, persistent=None)], output_specs=[OutputSpec(kind=<OutputKind.USER_OUTPUT: 1>, arg=TensorArgument(name='add'), target=None)])
Range constraints: {u0: VR[0, int_oo]}
Data-dependent errors can be much more involved, and there are many more options in your toolkit to deal with them: torch._check_is_size(), guard_size_oblivious(), or real-tensor tracing, as starters.
For more in-depth guides, please refer to the Export Programming Model,
or Dealing with GuardOnDataDependentSymNode errors.
Custom Ops
torch.export can export PyTorch programs with custom operators. Please
refer to this page
on how to author a custom operator in either C++ or Python.
The following is an example of registering a custom operator in python to be
used by torch.export. The important thing to note is that the custom op
must have a FakeTensor kernel.
@torch.library.custom_op("my_custom_library::custom_op", mutates_args={})
def custom_op(x: torch.Tensor) -> torch.Tensor:
print("custom_op called!")
return torch.relu(x)
@custom_op.register_fake
def custom_op_meta(x):
# Returns an empty tensor with the same shape as the expected output
return torch.empty_like(x)
Here is an example of exporting a program with the custom op.
class CustomOpExample(torch.nn.Module):
def forward(self, x):
x = torch.sin(x)
x = torch.ops.my_custom_library.custom_op(x)
x = torch.cos(x)
return x
exported_custom_op_example = export(CustomOpExample(), (torch.randn(3, 3),))
print(exported_custom_op_example)
print(exported_custom_op_example.module()(torch.randn(3, 3)))
I0502 18:42:27.075000 635 torch/fx/experimental/symbolic_shapes.py:3334] [18/0] create_env
I0502 18:42:27.085000 635 torch/fx/experimental/symbolic_shapes.py:4734] [18/0] produce_guards
V0502 18:42:27.085000 635 torch/fx/experimental/symbolic_shapes.py:4954] [18/0] track_symint L['x'].size()[0] 3 None
V0502 18:42:27.085000 635 torch/fx/experimental/symbolic_shapes.py:4954] [18/0] track_symint L['x'].size()[1] 3 None
V0502 18:42:27.086000 635 torch/fx/experimental/symbolic_shapes.py:4954] [18/0] track_symint L['x'].stride()[0] 3 None
V0502 18:42:27.086000 635 torch/fx/experimental/symbolic_shapes.py:4954] [18/0] track_symint L['x'].stride()[1] 1 None
V0502 18:42:27.086000 635 torch/fx/experimental/symbolic_shapes.py:4954] [18/0] track_symint L['x'].storage_offset() 0 None
ExportedProgram:
class GraphModule(torch.nn.Module):
def forward(self, x: "f32[3, 3]"):
# File: /var/lib/workspace/intermediate_source/torch_export_tutorial.py:812 in forward, code: x = torch.sin(x)
sin: "f32[3, 3]" = torch.ops.aten.sin.default(x); x = None
# File: /var/lib/workspace/intermediate_source/torch_export_tutorial.py:813 in forward, code: x = torch.ops.my_custom_library.custom_op(x)
custom_op: "f32[3, 3]" = torch.ops.my_custom_library.custom_op.default(sin); sin = None
# File: /var/lib/workspace/intermediate_source/torch_export_tutorial.py:814 in forward, code: x = torch.cos(x)
cos: "f32[3, 3]" = torch.ops.aten.cos.default(custom_op); custom_op = None
return (cos,)
Graph signature: ExportGraphSignature(input_specs=[InputSpec(kind=<InputKind.USER_INPUT: 1>, arg=TensorArgument(name='x'), target=None, persistent=None)], output_specs=[OutputSpec(kind=<OutputKind.USER_OUTPUT: 1>, arg=TensorArgument(name='cos'), target=None)])
Range constraints: {}
custom_op called!
tensor([[0.5499, 0.6889, 0.7180],
[0.5413, 1.0000, 1.0000],
[0.8332, 0.5524, 1.0000]])
Note that in the ExportedProgram, the custom operator is included in the graph.
IR/Decompositions
The graph produced by torch.export returns a graph containing only
ATen operators, which are the
basic unit of computation in PyTorch. As there are over 3000 ATen operators,
export provides a way to narrow down the operator set used in the graph based
on certain characteristics, creating different IRs.
By default, export produces the most generic IR which contains all ATen
operators, including both functional and non-functional operators. A functional
operator is one that does not contain any mutations or aliasing of the inputs.
You can find a list of all ATen operators
here
and you can inspect if an operator is functional by checking
op._schema.is_mutable, for example:
print(torch.ops.aten.add.Tensor._schema.is_mutable)
print(torch.ops.aten.add_.Tensor._schema.is_mutable)
False
True
This generic IR can be used to train in eager PyTorch Autograd. This IR can be
more explicitly reached through the API torch.export.export_for_training,
which was introduced in PyTorch 2.5, but calling torch.export.export
should produce the same graph as of PyTorch 2.6.
class DecompExample(torch.nn.Module):
def __init__(self) -> None:
super().__init__()
self.conv = torch.nn.Conv2d(1, 3, 1, 1)
self.bn = torch.nn.BatchNorm2d(3)
def forward(self, x):
x = self.conv(x)
x = self.bn(x)
return (x,)
ep_for_training = torch.export.export_for_training(DecompExample(), (torch.randn(1, 1, 3, 3),))
print(ep_for_training.graph)
I0502 18:42:27.111000 635 torch/fx/experimental/symbolic_shapes.py:3334] [19/0] create_env
I0502 18:42:27.140000 635 torch/fx/experimental/symbolic_shapes.py:4734] [19/0] produce_guards
V0502 18:42:27.141000 635 torch/fx/experimental/symbolic_shapes.py:4954] [19/0] track_symint L['x'].size()[0] 1 None
V0502 18:42:27.141000 635 torch/fx/experimental/symbolic_shapes.py:4954] [19/0] track_symint L['x'].size()[1] 1 None
V0502 18:42:27.141000 635 torch/fx/experimental/symbolic_shapes.py:4954] [19/0] track_symint L['x'].size()[2] 3 None
V0502 18:42:27.141000 635 torch/fx/experimental/symbolic_shapes.py:4954] [19/0] track_symint L['x'].size()[3] 3 None
V0502 18:42:27.142000 635 torch/fx/experimental/symbolic_shapes.py:4954] [19/0] track_symint L['x'].stride()[0] 9 None
V0502 18:42:27.142000 635 torch/fx/experimental/symbolic_shapes.py:4954] [19/0] track_symint L['x'].stride()[1] 9 None
V0502 18:42:27.142000 635 torch/fx/experimental/symbolic_shapes.py:4954] [19/0] track_symint L['x'].stride()[2] 3 None
V0502 18:42:27.142000 635 torch/fx/experimental/symbolic_shapes.py:4954] [19/0] track_symint L['x'].stride()[3] 1 None
V0502 18:42:27.143000 635 torch/fx/experimental/symbolic_shapes.py:4954] [19/0] track_symint L['x'].storage_offset() 0 None
graph():
%p_conv_weight : [num_users=1] = placeholder[target=p_conv_weight]
%p_conv_bias : [num_users=1] = placeholder[target=p_conv_bias]
%p_bn_weight : [num_users=1] = placeholder[target=p_bn_weight]
%p_bn_bias : [num_users=1] = placeholder[target=p_bn_bias]
%b_bn_running_mean : [num_users=1] = placeholder[target=b_bn_running_mean]
%b_bn_running_var : [num_users=1] = placeholder[target=b_bn_running_var]
%b_bn_num_batches_tracked : [num_users=1] = placeholder[target=b_bn_num_batches_tracked]
%x : [num_users=1] = placeholder[target=x]
%conv2d : [num_users=1] = call_function[target=torch.ops.aten.conv2d.default](args = (%x, %p_conv_weight, %p_conv_bias), kwargs = {})
%add_ : [num_users=0] = call_function[target=torch.ops.aten.add_.Tensor](args = (%b_bn_num_batches_tracked, 1), kwargs = {})
%batch_norm : [num_users=1] = call_function[target=torch.ops.aten.batch_norm.default](args = (%conv2d, %p_bn_weight, %p_bn_bias, %b_bn_running_mean, %b_bn_running_var, True, 0.1, 1e-05, True), kwargs = {})
return (batch_norm,)
We can then lower this exported program to an operator set which only contains
functional ATen operators through the API run_decompositions, which
decomposes the ATen operators into the ones specified in the decomposition
table, and functionalizes the graph. By specifying an empty set, we’re only
performing functionalization, and does not do any additional decompositions.
This results in an IR which contains ~2000 operators (instead of the 3000
operators above), and is ideal for inference cases.
ep_for_inference = ep_for_training.run_decompositions(decomp_table={})
print(ep_for_inference.graph)
graph():
%p_conv_weight : [num_users=1] = placeholder[target=p_conv_weight]
%p_conv_bias : [num_users=1] = placeholder[target=p_conv_bias]
%p_bn_weight : [num_users=1] = placeholder[target=p_bn_weight]
%p_bn_bias : [num_users=1] = placeholder[target=p_bn_bias]
%b_bn_running_mean : [num_users=1] = placeholder[target=b_bn_running_mean]
%b_bn_running_var : [num_users=1] = placeholder[target=b_bn_running_var]
%b_bn_num_batches_tracked : [num_users=1] = placeholder[target=b_bn_num_batches_tracked]
%x : [num_users=1] = placeholder[target=x]
%conv2d : [num_users=1] = call_function[target=torch.ops.aten.conv2d.default](args = (%x, %p_conv_weight, %p_conv_bias), kwargs = {})
%add : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%b_bn_num_batches_tracked, 1), kwargs = {})
%_native_batch_norm_legit_functional : [num_users=3] = call_function[target=torch.ops.aten._native_batch_norm_legit_functional.default](args = (%conv2d, %p_bn_weight, %p_bn_bias, %b_bn_running_mean, %b_bn_running_var, True, 0.1, 1e-05), kwargs = {})
%getitem : [num_users=1] = call_function[target=operator.getitem](args = (%_native_batch_norm_legit_functional, 0), kwargs = {})
%getitem_3 : [num_users=1] = call_function[target=operator.getitem](args = (%_native_batch_norm_legit_functional, 3), kwargs = {})
%getitem_4 : [num_users=1] = call_function[target=operator.getitem](args = (%_native_batch_norm_legit_functional, 4), kwargs = {})
return (getitem_3, getitem_4, add, getitem)
As we can see, the previously mutable operator,
torch.ops.aten.add_.default has now been replaced with
torch.ops.aten.add.default, a l operator.
We can also further lower this exported program to an operator set which only contains the Core ATen Operator Set, which is a collection of only ~180 operators. This IR is optimal for backends who do not want to reimplement all ATen operators.
from torch.export import default_decompositions
core_aten_decomp_table = default_decompositions()
core_aten_ep = ep_for_training.run_decompositions(decomp_table=core_aten_decomp_table)
print(core_aten_ep.graph)
graph():
%p_conv_weight : [num_users=1] = placeholder[target=p_conv_weight]
%p_conv_bias : [num_users=1] = placeholder[target=p_conv_bias]
%p_bn_weight : [num_users=1] = placeholder[target=p_bn_weight]
%p_bn_bias : [num_users=1] = placeholder[target=p_bn_bias]
%b_bn_running_mean : [num_users=1] = placeholder[target=b_bn_running_mean]
%b_bn_running_var : [num_users=1] = placeholder[target=b_bn_running_var]
%b_bn_num_batches_tracked : [num_users=1] = placeholder[target=b_bn_num_batches_tracked]
%x : [num_users=1] = placeholder[target=x]
%convolution : [num_users=1] = call_function[target=torch.ops.aten.convolution.default](args = (%x, %p_conv_weight, %p_conv_bias, [1, 1], [0, 0], [1, 1], False, [0, 0], 1), kwargs = {})
%add : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%b_bn_num_batches_tracked, 1), kwargs = {})
%_native_batch_norm_legit_functional : [num_users=3] = call_function[target=torch.ops.aten._native_batch_norm_legit_functional.default](args = (%convolution, %p_bn_weight, %p_bn_bias, %b_bn_running_mean, %b_bn_running_var, True, 0.1, 1e-05), kwargs = {})
%getitem : [num_users=1] = call_function[target=operator.getitem](args = (%_native_batch_norm_legit_functional, 0), kwargs = {})
%getitem_3 : [num_users=1] = call_function[target=operator.getitem](args = (%_native_batch_norm_legit_functional, 3), kwargs = {})
%getitem_4 : [num_users=1] = call_function[target=operator.getitem](args = (%_native_batch_norm_legit_functional, 4), kwargs = {})
return (getitem_3, getitem_4, add, getitem)
We now see that torch.ops.aten.conv2d.default has been decomposed
into torch.ops.aten.convolution.default. This is because convolution
is a more “core” operator, as operations like conv1d and conv2d can be
implemented using the same op.
We can also specify our own decomposition behaviors:
my_decomp_table = torch.export.default_decompositions()
def my_awesome_custom_conv2d_function(x, weight, bias, stride=[1, 1], padding=[0, 0], dilation=[1, 1], groups=1):
return 2 * torch.ops.aten.convolution(x, weight, bias, stride, padding, dilation, False, [0, 0], groups)
my_decomp_table[torch.ops.aten.conv2d.default] = my_awesome_custom_conv2d_function
my_ep = ep_for_training.run_decompositions(my_decomp_table)
print(my_ep.graph)
graph():
%p_conv_weight : [num_users=1] = placeholder[target=p_conv_weight]
%p_conv_bias : [num_users=1] = placeholder[target=p_conv_bias]
%p_bn_weight : [num_users=1] = placeholder[target=p_bn_weight]
%p_bn_bias : [num_users=1] = placeholder[target=p_bn_bias]
%b_bn_running_mean : [num_users=1] = placeholder[target=b_bn_running_mean]
%b_bn_running_var : [num_users=1] = placeholder[target=b_bn_running_var]
%b_bn_num_batches_tracked : [num_users=1] = placeholder[target=b_bn_num_batches_tracked]
%x : [num_users=1] = placeholder[target=x]
%convolution : [num_users=1] = call_function[target=torch.ops.aten.convolution.default](args = (%x, %p_conv_weight, %p_conv_bias, [1, 1], [0, 0], [1, 1], False, [0, 0], 1), kwargs = {})
%mul : [num_users=1] = call_function[target=torch.ops.aten.mul.Tensor](args = (%convolution, 2), kwargs = {})
%add : [num_users=1] = call_function[target=torch.ops.aten.add.Tensor](args = (%b_bn_num_batches_tracked, 1), kwargs = {})
%_native_batch_norm_legit_functional : [num_users=3] = call_function[target=torch.ops.aten._native_batch_norm_legit_functional.default](args = (%mul, %p_bn_weight, %p_bn_bias, %b_bn_running_mean, %b_bn_running_var, True, 0.1, 1e-05), kwargs = {})
%getitem : [num_users=1] = call_function[target=operator.getitem](args = (%_native_batch_norm_legit_functional, 0), kwargs = {})
%getitem_3 : [num_users=1] = call_function[target=operator.getitem](args = (%_native_batch_norm_legit_functional, 3), kwargs = {})
%getitem_4 : [num_users=1] = call_function[target=operator.getitem](args = (%_native_batch_norm_legit_functional, 4), kwargs = {})
return (getitem_3, getitem_4, add, getitem)
Notice that instead of torch.ops.aten.conv2d.default being decomposed
into torch.ops.aten.convolution.default, it is now decomposed into
torch.ops.aten.convolution.default and torch.ops.aten.mul.Tensor,
which matches our custom decomposition rule.
ExportDB
torch.export will only ever export a single computation graph from a PyTorch program. Because of this requirement,
there will be Python or PyTorch features that are not compatible with torch.export, which will require users to
rewrite parts of their model code. We have seen examples of this earlier in the tutorial – for example, rewriting
if-statements using cond.
ExportDB is the standard reference that documents
supported and unsupported Python/PyTorch features for torch.export. It is essentially a list a program samples, each
of which represents the usage of one particular Python/PyTorch feature and its interaction with torch.export.
Examples are also tagged by category so that they can be more easily searched.
For example, let’s use ExportDB to get a better understanding of how the predicate works in the cond operator.
We can look at the example called cond_predicate, which has a torch.cond tag. The example code looks like:
def cond_predicate(x):
"""
The conditional statement (aka predicate) passed to ``cond()`` must be one of the following:
- ``torch.Tensor`` with a single element
- boolean expression
NOTE: If the `pred` is test on a dim with batch size < 2, it will be specialized.
"""
pred = x.dim() > 2 and x.shape[2] > 10
return cond(pred, lambda x: x.cos(), lambda y: y.sin(), [x])
More generally, ExportDB can be used as a reference when one of the following occurs:
Before attempting
torch.export, you know ahead of time that your model uses some tricky Python/PyTorch features and you want to know iftorch.exportcovers that feature.When attempting
torch.export, there is a failure and it’s unclear how to work around it.
ExportDB is not exhaustive, but is intended to cover all use cases found in typical PyTorch code. Feel free to reach
out if there is an important Python/PyTorch feature that should be added to ExportDB or supported by torch.export.
Running the Exported Program
As torch.export is only a graph capturing mechanism, calling the artifact
produced by torch.export eagerly will be equivalent to running the eager
module. To optimize the execution of the Exported Program, we can pass this
exported artifact to backends such as Inductor through torch.compile,
AOTInductor,
or TensorRT.
class M(torch.nn.Module):
def __init__(self):
super().__init__()
self.linear = torch.nn.Linear(3, 3)
def forward(self, x):
x = self.linear(x)
return x
inp = torch.randn(2, 3, device="cuda")
m = M().to(device="cuda")
ep = torch.export.export(m, (inp,))
# Run it eagerly
res = ep.module()(inp)
print(res)
# Run it with torch.compile
res = torch.compile(ep.module(), backend="inductor")(inp)
print(res)
I0502 18:42:27.805000 635 torch/fx/experimental/symbolic_shapes.py:3334] [20/0] create_env
I0502 18:42:27.819000 635 torch/fx/experimental/symbolic_shapes.py:4734] [20/0] produce_guards
V0502 18:42:27.819000 635 torch/fx/experimental/symbolic_shapes.py:4954] [20/0] track_symint L['x'].size()[0] 2 None
V0502 18:42:27.819000 635 torch/fx/experimental/symbolic_shapes.py:4954] [20/0] track_symint L['x'].size()[1] 3 None
V0502 18:42:27.820000 635 torch/fx/experimental/symbolic_shapes.py:4954] [20/0] track_symint L['x'].stride()[0] 3 None
V0502 18:42:27.820000 635 torch/fx/experimental/symbolic_shapes.py:4954] [20/0] track_symint L['x'].stride()[1] 1 None
V0502 18:42:27.820000 635 torch/fx/experimental/symbolic_shapes.py:4954] [20/0] track_symint L['x'].storage_offset() 0 None
tensor([[ 0.4830, -0.5149, 0.3888],
[-0.9247, 0.8408, -0.2184]], device='cuda:0',
grad_fn=<AddmmBackward0>)
I0502 18:42:27.843000 635 torch/fx/experimental/symbolic_shapes.py:3334] [21/0] create_env
I0502 18:42:28.114000 635 torch/fx/experimental/symbolic_shapes.py:4734] [21/0] produce_guards
V0502 18:42:28.114000 635 torch/fx/experimental/symbolic_shapes.py:4954] [21/0] track_symint L['x'].size()[0] 2 None
V0502 18:42:28.115000 635 torch/fx/experimental/symbolic_shapes.py:4954] [21/0] track_symint L['x'].size()[1] 3 None
V0502 18:42:28.115000 635 torch/fx/experimental/symbolic_shapes.py:4954] [21/0] track_symint L['x'].stride()[0] 3 None
V0502 18:42:28.115000 635 torch/fx/experimental/symbolic_shapes.py:4954] [21/0] track_symint L['x'].stride()[1] 1 None
V0502 18:42:28.115000 635 torch/fx/experimental/symbolic_shapes.py:4954] [21/0] track_symint L['x'].storage_offset() 0 None
V0502 18:42:28.116000 635 torch/fx/experimental/symbolic_shapes.py:4954] [21/0] track_symint L['self']._modules['linear']._parameters['weight'].size()[0] 3 None
V0502 18:42:28.116000 635 torch/fx/experimental/symbolic_shapes.py:4954] [21/0] track_symint L['self']._modules['linear']._parameters['weight'].size()[1] 3 None
V0502 18:42:28.116000 635 torch/fx/experimental/symbolic_shapes.py:4954] [21/0] track_symint L['self']._modules['linear']._parameters['weight'].stride()[0] 3 None
V0502 18:42:28.116000 635 torch/fx/experimental/symbolic_shapes.py:4954] [21/0] track_symint L['self']._modules['linear']._parameters['weight'].stride()[1] 1 None
V0502 18:42:28.117000 635 torch/fx/experimental/symbolic_shapes.py:4954] [21/0] track_symint L['self']._modules['linear']._parameters['weight'].storage_offset() 0 None
V0502 18:42:28.117000 635 torch/fx/experimental/symbolic_shapes.py:4954] [21/0] track_symint L['self']._modules['linear']._parameters['bias'].size()[0] 3 None
V0502 18:42:28.117000 635 torch/fx/experimental/symbolic_shapes.py:4954] [21/0] track_symint L['self']._modules['linear']._parameters['bias'].stride()[0] 1 None
V0502 18:42:28.118000 635 torch/fx/experimental/symbolic_shapes.py:4954] [21/0] track_symint L['self']._modules['linear']._parameters['bias'].storage_offset() 0 None
V0502 18:42:28.118000 635 torch/fx/experimental/symbolic_shapes.py:5156] [21/0] Skipping guard L['x'].size()[0] == 2
V0502 18:42:28.118000 635 torch/fx/experimental/symbolic_shapes.py:5156] [21/0] Skipping guard L['x'].size()[1] == 3
V0502 18:42:28.118000 635 torch/fx/experimental/symbolic_shapes.py:5156] [21/0] Skipping guard L['x'].stride()[0] == 3
V0502 18:42:28.119000 635 torch/fx/experimental/symbolic_shapes.py:5156] [21/0] Skipping guard L['x'].stride()[1] == 1
V0502 18:42:28.119000 635 torch/fx/experimental/symbolic_shapes.py:5156] [21/0] Skipping guard L['x'].storage_offset() == 0
V0502 18:42:28.119000 635 torch/fx/experimental/symbolic_shapes.py:5156] [21/0] Skipping guard L['self']._modules['linear']._parameters['weight'].size()[0] == 3
V0502 18:42:28.120000 635 torch/fx/experimental/symbolic_shapes.py:5156] [21/0] Skipping guard L['self']._modules['linear']._parameters['weight'].size()[1] == 3
V0502 18:42:28.120000 635 torch/fx/experimental/symbolic_shapes.py:5156] [21/0] Skipping guard L['self']._modules['linear']._parameters['weight'].stride()[0] == 3
V0502 18:42:28.120000 635 torch/fx/experimental/symbolic_shapes.py:5156] [21/0] Skipping guard L['self']._modules['linear']._parameters['weight'].stride()[1] == 1
V0502 18:42:28.120000 635 torch/fx/experimental/symbolic_shapes.py:5156] [21/0] Skipping guard L['self']._modules['linear']._parameters['weight'].storage_offset() == 0
V0502 18:42:28.121000 635 torch/fx/experimental/symbolic_shapes.py:5156] [21/0] Skipping guard L['self']._modules['linear']._parameters['bias'].size()[0] == 3
V0502 18:42:28.121000 635 torch/fx/experimental/symbolic_shapes.py:5156] [21/0] Skipping guard L['self']._modules['linear']._parameters['bias'].stride()[0] == 1
V0502 18:42:28.121000 635 torch/fx/experimental/symbolic_shapes.py:5156] [21/0] Skipping guard L['self']._modules['linear']._parameters['bias'].storage_offset() == 0
tensor([[ 0.4830, -0.5149, 0.3888],
[-0.9247, 0.8408, -0.2184]], device='cuda:0',
grad_fn=<CompiledFunctionBackward>)
import torch._inductor
# Note: these APIs are subject to change
# Compile the exported program to a PT2 archive using ``AOTInductor``
with torch.no_grad():
pt2_path = torch._inductor.aoti_compile_and_package(ep)
# Load and run the .so file in Python.
# To load and run it in a C++ environment, see:
# https://pytorch.org/docs/main/torch.compiler_aot_inductor.html
aoti_compiled = torch._inductor.aoti_load_package(pt2_path)
res = aoti_compiled(inp)
Conclusion
We introduced torch.export, the new PyTorch 2.X way to export single computation
graphs from PyTorch programs. In particular, we demonstrate several code modifications
and considerations (control flow ops, constraints, etc.) that need to be made in order to export a graph.
Total running time of the script: ( 0 minutes 2.351 seconds)
