Revert "Fix type annotation of Linear.bias (#142326)"

This reverts commit 81e370fc6b90f9cb98c88f3173e738aba0dc650a.

Reverted https://github.com/pytorch/pytorch/pull/142326 on behalf of https://github.com/malfet due to This introduced a graph break and regressed inductor tests, see 73622fc5fa/1 ([comment](https://github.com/pytorch/pytorch/pull/142326#issuecomment-2614196349))

torch/ao/pruning/_experimental/pruner/base_structured_sparsifier.py

@@ -265,7 +265,7 @@ class BaseStructuredSparsifier(BaseSparsifier):
                     module.prune_bias = prune_bias
 
                 module.register_forward_hook(
-                    BiasHook(module.parametrizations.weight[0], prune_bias)  # type: ignore[union-attr, index]
+                    BiasHook(module.parametrizations.weight[0], prune_bias)
                 )
 
     def prune(self) -> None:

torch/nn/modules/activation.py

@@ -7,7 +7,6 @@ import torch.nn.functional as F
 from torch import Tensor
 from torch.nn.init import constant_, xavier_normal_, xavier_uniform_
 from torch.nn.parameter import Parameter
-from torch.utils._typing_utils import not_none
 
 from .linear import NonDynamicallyQuantizableLinear
 from .module import Module
@@ -1123,7 +1122,6 @@ class MultiheadAttention(Module):
             xavier_uniform_(self.v_proj_weight)
 
         if self.in_proj_bias is not None:
-            assert self.out_proj.bias is not None
             constant_(self.in_proj_bias, 0.0)
             constant_(self.out_proj.bias, 0.0)
         if self.bias_k is not None:
@@ -1321,7 +1319,7 @@ class MultiheadAttention(Module):
                         self.in_proj_weight,
                         self.in_proj_bias,
                         self.out_proj.weight,
-                        not_none(self.out_proj.bias),
+                        self.out_proj.bias,
                         merged_mask,
                         need_weights,
                         average_attn_weights,

torch/nn/modules/linear.py

@@ -1,6 +1,6 @@
 # mypy: allow-untyped-defs
 import math
-from typing import Any, Optional, TYPE_CHECKING
+from typing import Any
 
 import torch
 from torch import Tensor
@@ -89,8 +89,6 @@ class Linear(Module):
     in_features: int
     out_features: int
     weight: Tensor
-    if TYPE_CHECKING:
-        bias: Optional[Tensor]
 
     def __init__(
         self,
@@ -192,8 +190,6 @@ class Bilinear(Module):
     in2_features: int
     out_features: int
     weight: Tensor
-    if TYPE_CHECKING:
-        bias: Optional[Tensor]
 
     def __init__(
         self,

torch/nn/modules/transformer.py

@@ -7,7 +7,6 @@ import torch
 import torch.nn.functional as F
 from torch import Tensor
 from torch.nn.init import xavier_uniform_
-from torch.utils._typing_utils import not_none
 
 from .activation import MultiheadAttention
 from .container import ModuleList
@@ -842,15 +841,15 @@ class TransformerEncoderLayer(Module):
                 self.self_attn.in_proj_weight,
                 self.self_attn.in_proj_bias,
                 self.self_attn.out_proj.weight,
-                not_none(self.self_attn.out_proj.bias),
+                self.self_attn.out_proj.bias,
                 self.norm1.weight,
                 self.norm1.bias,
                 self.norm2.weight,
                 self.norm2.bias,
                 self.linear1.weight,
-                not_none(self.linear1.bias),
+                self.linear1.bias,
                 self.linear2.weight,
-                not_none(self.linear2.bias),
+                self.linear2.bias,
             )
 
             # We have to use list comprehensions below because TorchScript does not support
@@ -886,7 +885,7 @@ class TransformerEncoderLayer(Module):
                     self.self_attn.in_proj_weight,
                     self.self_attn.in_proj_bias,
                     self.self_attn.out_proj.weight,
-                    not_none(self.self_attn.out_proj.bias),
+                    self.self_attn.out_proj.bias,
                     self.activation_relu_or_gelu == 2,
                     self.norm_first,
                     self.norm1.eps,
@@ -895,9 +894,9 @@ class TransformerEncoderLayer(Module):
                     self.norm2.weight,
                     self.norm2.bias,
                     self.linear1.weight,
-                    not_none(self.linear1.bias),
+                    self.linear1.bias,
                     self.linear2.weight,
-                    not_none(self.linear2.bias),
+                    self.linear2.bias,
                     merged_mask,
                     mask_type,
                 )

torch/utils/_typing_utils.py

@@ -1,6 +1,6 @@
 """Miscellaneous utilities to aid with typing."""
 
-from typing import Optional, TYPE_CHECKING, TypeVar
+from typing import Optional, TypeVar
 
 
 # Helper to turn Optional[T] into T when we know None either isn't
@@ -8,19 +8,7 @@ from typing import Optional, TYPE_CHECKING, TypeVar
 T = TypeVar("T")
 
 
-# TorchScript cannot handle the type signature of `not_none` at runtime, because it trips
-# over the `Optional[T]`. To allow using `not_none` from inside a TorchScript method/module,
-# we split the implementation, and hide the runtime type information from TorchScript.
-if TYPE_CHECKING:
-
 def not_none(obj: Optional[T]) -> T:
-        ...
-
-else:
-
-    def not_none(obj):
     if obj is None:
-            raise TypeError(
-                "Invariant encountered: value was None when it should not be"
-            )
+        raise TypeError("Invariant encountered: value was None when it should not be")
     return obj