utils

Classes

ActivityContext

Functions

dynamo_skip	Skip frames associated with the function code, but still process recursively invoked frames
dynamo_disable	Decorator to disable TorchDynamo
is_submodule_of
is_submodule_or_same
fake_tensor
fake_tensor_like
fake_tensors
real_tensors
has_fake_tensor
distributed_isend_obj
distributed_send_obj
distributed_recv_obj

class ActivityContext

Bases: Generic[T]

__init__(max_depth=None, no_duplicates=False, reversed=False)

Parameters:

max_depth (int | None)

get_active()

Return type:

T | None

is_active()

Return type:

bool

exception ActivityContextDuplicateException

Bases: Exception

distributed_isend_obj(obj, dst=0, group=None)

Parameters:

• obj (Any)

• dst (int)

• group (ProcessGroup | None)

Return type:

list[Work | None]

distributed_recv_obj(src=None, group=None)

Parameters:

• src (int | None)

• group (ProcessGroup | None)

Return type:

Any

distributed_send_obj(obj, dst=0, group=None)

Parameters:

• obj (Any)

• dst (int)

• group (ProcessGroup | None)

dynamo_disable(fn=None, recursive=True, *, reason=None, wrapping=True)

Decorator to disable TorchDynamo

If recursive=True, Dynamo is completely skipped on the decorated function frame as well as the recursively invoked functions.

If recursive=False, Dynamo skips frames associated with the function code, but still process recursively invoked frames.

If reason is provided, it will be printed when Dynamo attempts to trace the disabled function.

dynamo_skip(fn=None)

Skip frames associated with the function code, but still process recursively invoked frames

Parameters:

fn (Callable[[~_P], _R] | None)

Return type:

Callable[[…], Any]

fake_tensor(t: Tensor, *, dtype: dtype | None = None, use_meta=False) → Tensor

fake_tensor(size: Sequence[int] | torch.Size, *, dtype: dtype | None = None, use_meta=False) → Tensor

fake_tensor(*args: int, dtype: dtype | None = None, use_meta=False) → Tensor

fake_tensor_like(t, use_meta=False)

Parameters:

t (Tensor)

Return type:

Tensor

fake_tensors(value, use_meta=False)

Parameters:

value (T)

Return type:

T

has_fake_tensor(v)

Parameters:

v (Any)

Return type:

bool

is_submodule_of(module_name, other_module_name)

Parameters:

• module_name (str)

• other_module_name (str)

Return type:

bool

is_submodule_or_same(module_name, other_module_name)

Parameters:

• module_name (str)

• other_module_name (str)

Return type:

bool

real_tensors(value)

Parameters:

value (Any)

Return type:

Any