未验证 提交 2e8f9882 编写于 作者: K kuizhiqing 提交者: GitHub

test=document_fix , fix launch doc (#40848)

* test=document_fix , fix launch doc

* test=document_fix , fix typo
上级 cc8e98c7
...@@ -40,9 +40,9 @@ def launch(): ...@@ -40,9 +40,9 @@ def launch():
- ``--rank``: The rank of the node, can be auto assigned by master. Default ``--rank=-1``. - ``--rank``: The rank of the node, can be auto assigned by master. Default ``--rank=-1``.
- ``--log_level``: The log levl to set for logging.setLevel. Default ``--log_level=INFO``. - ``--log_level``: The log level to set for logging.setLevel which can be CRITICAL/ERROR/WARNING/INFO/DEBUG/NOTSET, case insensitive. The rank 0 log will not print in the terminal by default, while you can enable it by adding --log_level=debug. Default ``--log_level=INFO``.
- ``--nnodes``: The number of nodes for a distributed job, it can be a range in elastic mode, e.g., ``--nnnodes=2:3``. Default ``--nnodes=1``. - ``--nnodes``: The number of nodes for a distributed job, it can be a range in elastic mode, e.g., ``--nnodes=2:3``. Default ``--nnodes=1``.
- ``--nproc_per_node``: The number of processes to launch on a node. In gpu training, it should be less or equal to the gpus number of you system. e.g., ``--nproc_per_node=8`` - ``--nproc_per_node``: The number of processes to launch on a node. In gpu training, it should be less or equal to the gpus number of you system. e.g., ``--nproc_per_node=8``
...@@ -93,9 +93,11 @@ def launch(): ...@@ -93,9 +93,11 @@ def launch():
Returns: Returns:
``None`` - ``None``
Examples 0 (master, ip/port auto detection): Examples 0 (master, ip/port auto detection):
.. code-block:: bash
:name: code-block-example-bash0
# For training on multi node, run the following command in one of the nodes # For training on multi node, run the following command in one of the nodes
...@@ -171,7 +173,7 @@ def launch(): ...@@ -171,7 +173,7 @@ def launch():
.. code-block:: bash .. code-block:: bash
:name: code-block-example-bash5 :name: code-block-example-bash5
# To simulate distributed environment using single node, e.g., 2 servers and 4 workers, each worker use single gpu. # To simulate distributed environment using single node, e.g., 2 servers and 4 workers, each worker use single gpu.
export CUDA_VISIBLE_DEVICES=0,1,2,3 export CUDA_VISIBLE_DEVICES=0,1,2,3
python -m paddle.distributed.launch --server_num=2 --worker_num=4 train.py --lr=0.01 python -m paddle.distributed.launch --server_num=2 --worker_num=4 train.py --lr=0.01
...@@ -226,7 +228,7 @@ def launch(): ...@@ -226,7 +228,7 @@ def launch():
python -m paddle.distributed.launch --master etcd://10.0.0.1:2379 --nnodes 2:4 train.py python -m paddle.distributed.launch --master etcd://10.0.0.1:2379 --nnodes 2:4 train.py
# once the number of nodes changes between 2:4 during training, the strategy holds # once the number of nodes changes between 2:4 during training, the strategy holds
""" """
# initialize the context to run # initialize the context to run
......
Markdown is supported
0% .
You are about to add 0 people to the discussion. Proceed with caution.
先完成此消息的编辑!
想要评论请 注册