Document that no_grad is thread local. (#21755)

ezyang · facebook-github-bot · commit b858f42e164b · 2019-06-13T13:47:09.000-07:00
Summary: Pull Request resolved: #21755 ghimport-source-id: dfb5375 Differential Revision: D15811172 Pulled By: ezyang fbshipit-source-id: c8c7c1c15277d8fe8cc513e20af449257d7ff15c
diff --git a/docs/source/torch.rst b/docs/source/torch.rst
@@ -142,7 +142,9 @@ Locally disabling gradient computation
 The context managers :func:`torch.no_grad`, :func:`torch.enable_grad`, and
 :func:`torch.set_grad_enabled` are helpful for locally disabling and enabling
 gradient computation. See :ref:`locally-disable-grad` for more details on
-their usage.
+their usage.  These context managers are thread local, so they won't
+work if you send work to another thread using the :module:`threading`
+module, etc.
 
 Examples::
 
diff --git a/torch/autograd/grad_mode.py b/torch/autograd/grad_mode.py
@@ -10,6 +10,8 @@ class no_grad(object):
     consumption for computations that would otherwise have `requires_grad=True`.
     In this mode, the result of every computation will have
     `requires_grad=False`, even when the inputs have `requires_grad=True`.
+    This context manager is thread local; it will not affect computation
+    in other threads.
 
     Also functions as a decorator.
 
@@ -49,6 +51,8 @@ class enable_grad(object):
 
     Enables gradient calculation inside a :class:`~no_grad` context. This has
     no effect outside of :class:`~no_grad`.
+    This context manager is thread local; it will not affect computation
+    in other threads.
 
     Also functions as a decorator.
 
@@ -93,6 +97,8 @@ class set_grad_enabled(object):
 
     ``set_grad_enabled`` will enable or disable grads based on its argument :attr:`mode`.
     It can be used as a context-manager or as a function.
+    This context manager is thread local; it will not affect computation
+    in other threads.
 
     Arguments:
         mode (bool): Flag whether to enable grad (``True``), or disable
