summaryrefslogtreecommitdiff
path: root/docs
diff options
context:
space:
mode:
authorSoumith Chintala <soumith@gmail.com>2017-07-12 14:47:02 -0400
committerSoumith Chintala <soumith@gmail.com>2017-07-12 14:47:36 -0400
commit81fd2bf2d0cc64ffed50d3777194675824914c03 (patch)
treec20824067256ac7bdf9f327312b85b105c283cc0 /docs
parent8915e2710c57cefd4bb5a7876216f08d3bc8c92f (diff)
downloadpytorch-81fd2bf2d0cc64ffed50d3777194675824914c03.tar.gz
pytorch-81fd2bf2d0cc64ffed50d3777194675824914c03.tar.bz2
pytorch-81fd2bf2d0cc64ffed50d3777194675824914c03.zip
fix some language / typos
Diffstat (limited to 'docs')
-rw-r--r--docs/source/distributed.rst25
1 files changed, 15 insertions, 10 deletions
diff --git a/docs/source/distributed.rst b/docs/source/distributed.rst
index aaebec55bc..02803dee53 100644
--- a/docs/source/distributed.rst
+++ b/docs/source/distributed.rst
@@ -27,10 +27,12 @@ TCP initialization
^^^^^^^^^^^^^^^^^^
Initialization will utilize a network address reachable from all processes.
-If the address belongs to one of the machines, it requires that all processes
-have manually specified ranks. Otherwise, the it has to be a valid IP multicast
-address, and ranks can be assigned automatically. Multicast init also supports
-a ``group_name`` argument, which allows to use the same address for multiple jobs,
+If the address belongs to one of the machines, initialization requires that all processes
+have manually specified ranks.
+
+Alternatively, the address has to be a valid IP multicast address, in which case,
+ranks can be assigned automatically. Multicast initialization also supports
+a ``group_name`` argument, which allows you to use the same address for multiple jobs,
as long as they use different group names.
::
@@ -39,6 +41,7 @@ as long as they use different group names.
# Use address of one of the machines
dist.init_process_group(init_method='tcp://10.1.1.20:23456', rank=args.rank, world_size=4)
+
# or a multicast address - rank will be assigned automatically if unspecified
dist.init_process_group(init_method='tcp://[ff15:1e18:5d4c:4cf0:d02d:b659:53ba:b0a7]:23456',
world_size=4)
@@ -46,10 +49,10 @@ as long as they use different group names.
Shared file-system initialization
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
-Another initialization method makes use of a file system shared and visible for
+Another initialization method makes use of a file system shared and visible from
all machines in a group. The URL should start with ``file://`` and contain a path
to a non-existent file (in an existing directory) on a shared file system.
-This initialization method also supports a ``group_name`` argument, which allows to
+This initialization method also supports a ``group_name`` argument, which allows you to
use the same shared file path for multiple jobs, as long as they use different
group names.
@@ -77,19 +80,21 @@ are:
* ``WORLD_SIZE`` - required; can be set either here, or in a call to init function
* ``RANK`` - required; can be set either here, or in a call to init function
-The machine with rank 0 will be used to set up all connections. This is the
-default method, meaning that ``init_method`` does not have to be specified (or
+The machine with rank 0 will be used to set up all connections.
+
+This is the default method, meaning that ``init_method`` does not have to be specified (or
can be ``env://``).
Groups
------
By default collectives operate on the default group (also called the world) and
-require all processes to enter the call. However, some workloads can benefit
+require all processes to enter the distributed function call. However, some workloads can benefit
from more fine-grained communication. This is where distributed groups come
into play. :func:`~torch.distributed.new_group` function can be
used to create new groups, with arbitrary subsets of all processes. It returns
-an opaque group handle that can be given as a ``group`` argument to all collectives.
+an opaque group handle that can be given as a ``group`` argument to all collectives
+(collectives are distributed functions to exchange information in certain well-known programming patterns).
.. autofunction:: new_group