Merge branch 'main' into add-api-checker

Author: svekars

intermediate_source/pinmem_nonblock.py

@@ -127,7 +127,7 @@
 #   1. The device must have at least one free DMA (Direct Memory Access) engine. Modern GPU architectures such as Volterra,
 #      Tesla, or H100 devices have more than one DMA engine.
 #
-#   2. The transfer must be done on a separate, non-default cuda stream. In PyTorch, cuda streams can be handles using
+#   2. The transfer must be done on a separate, non-default cuda stream. In PyTorch, cuda streams can be handled using
 #      :class:`~torch.cuda.Stream`.
 #
 #   3. The source data must be in pinned memory.

intermediate_source/torch_compile_tutorial.py

@@ -330,7 +330,7 @@ def bar(a, b):
 #
 # The second time we run ``bar``, we take the other branch of the if statement
 # and we get 1 traced graph corresponding to the code ``b = b * -1; return x * b``.
-# We do not see a graph of ``x = a / (torch.abs(a) + 1)`` outputted the second time
+# We do not see a graph of ``x = a / (torch.abs(a) + 1); b.sum()`` outputted the second time
 # since ``torch.compile`` cached this graph from the first run and re-used it.
 #
 # Let's investigate by example how TorchDynamo would step through ``bar``.