frozenleaves/pytorch
1
0
Fork 0
mirror of https://github.com/pytorch/pytorch.git synced 2025-10-20 12:54:11 +08:00
Code Issues Packages Projects Releases Wiki Activity

[BE][2/6] fix typos in test/ (test/test_*.py) (#157636)

Browse Source 
Pull Request resolved: https://github.com/pytorch/pytorch/pull/157636
Approved by: https://github.com/yewentao256, https://github.com/mlazos
ghstack dependencies: #156311, #156609
This commit is contained in:
Xuehai Pan
2025-07-09 13:23:55 +08:00
committed by PyTorch MergeBot
parent ffe11b2bf2
commit fc0376e8b1
57 changed files with 194 additions and 195 deletions
Download Patch File Download Diff File Expand all files Collapse all files

16
test/test_autograd.py
View File

@ -4129,7 +4129,7 @@ class TestAutograd(TestCase):
self.assertIsNone(y.grad_fn)
def test_backward_copy(self):
# This tests checks backward engine for a very subtle bug that appreared
# This tests checks backward engine for a very subtle bug that appeared
# in one of the initial versions of autograd. Gradients tensors were
# simply stored in lists while the function waited for all its gradients
# to be computed. However, sometimes an output was used multiple times,
@ -4312,7 +4312,7 @@ class TestAutograd(TestCase):
ctx.output_var.sum().backward()
return ctx.x.grad * grad_output
# Reentrant starts on CPU thread, finishs on GPU thread
# Reentrant starts on CPU thread, finishes on GPU thread
x = torch.randn(2, 2, requires_grad=True)
out = Reenter.apply(x)
out.sum().backward()
@ -10728,7 +10728,7 @@ class TestAutogradForwardMode(TestCase):
dual = fwAD.make_dual(foo, tangent)
self.assertFalse(tangent_ref.expired())
# Make sure that the tangent we provided has been re-used as is
# Make sure that the tangent we provided has been reused as is
self.assertTrue(fwAD.unpack_dual(dual)[1] is tangent)
# Make sure that dual is keeping the tangent alive
@ -11087,7 +11087,7 @@ class TestAutogradForwardMode(TestCase):
self.assertEqual(
dual_tangent.storage().data_ptr(), bar.storage().data_ptr()
)
# And the tangent is actually re-used as-is so it is still the same Tensor
# And the tangent is actually reused as-is so it is still the same Tensor
self.assertIs(dual_tangent, bar)
# Ensure we properly share the version counter
@ -11969,19 +11969,19 @@ class TestAutogradDeviceType(TestCase):
(new_param**2).sum().backward()
return grad_output
# Reentrant starts on GPU thread, finishs on GPU thread
# Reentrant starts on GPU thread, finishes on GPU thread
x = torch.randn(2, 2, device=device, requires_grad=True)
out = ReentrantFunc.apply(x)
out.sum().backward()
# Reentrant starts on CPU thread, finishs on GPU thread
# Reentrant starts on CPU thread, finishes on GPU thread
x = torch.randn(2, 2, requires_grad=True)
# set ReentrantFunc node to GPU to emit tasks to GPU queue
ReentrantFunc._cpu_mode = False
out = ReentrantFunc.apply(x)
out.sum().backward()
# Reentrant starts on GPU thread, finishs on CPU thread
# Reentrant starts on GPU thread, finishes on CPU thread
x = torch.randn(2, 2, device=device, requires_grad=True)
# set ReentrantFunc node to CPU to emit tasks to CPU queue
ReentrantFunc._cpu_mode = True
@ -13665,7 +13665,7 @@ class TestMultithreadAutograd(TestCase):
y = x * x
if torch.cuda.device_count() >= 2:
# DataParallel is calling the forward in different threads
# without progating TLS, so hooks should not be called here
# without propagating TLS, so hooks should not be called here
_self.assertEqual(len(w), 0)
else:
# DataParallel only uses one thread