From 5e9b6695bd71ce7dbb2de624cbb43ecda3baf2e6 Mon Sep 17 00:00:00 2001
From: lg <lg4869@outlook.com>
Date: Fri, 6 Nov 2020 13:02:43 +0000
Subject: [PATCH 1/2] split doc (wip)

---
 docs/api.rst   | 52 +++++++++++++++++++++++++++++++++++++++++++++++++
 docs/index.rst | 53 +++-----------------------------------------------
 2 files changed, 55 insertions(+), 50 deletions(-)
 create mode 100644 docs/api.rst

diff --git a/docs/api.rst b/docs/api.rst
new file mode 100644
index 000000000..54fe1d3a6
--- /dev/null
+++ b/docs/api.rst
@@ -0,0 +1,52 @@
+KungFu APIs
+===========
+
+KungFu has the high-level optimizer APIs that
+allows you to transparently scale out training.
+It also has a low-level API that allows an easy implementation
+of distributed training strategies.
+The following is the public API we released so far.
+
+Distributed optimizers
+----------------------
+
+KungFu provides optimizers that implement various distributed training algorithms.
+These optimizers are used for transparently scaling out the training of
+`tf.train.Optimizer <https://www.tensorflow.org/versions/r1.15/api_docs/python/tf/train/Optimizer>`_
+and `tf.keras.optimizers.Optimizer <https://www.tensorflow.org/api_docs/python/tf/keras/optimizers/Optimizer>`_
+
+.. automodule:: kungfu.tensorflow.optimizers
+   :members:
+
+Global variable initializers
+----------------------------
+
+KungFu provide various initializers to help you synchronize
+the global variables of distributed training workers at
+the beginning of training. These initializers are used
+with ``tf.session``, ``tf.estimator``, ``tf.GradientTape``
+and ``tf.keras``, respectively.
+
+.. automodule:: kungfu.tensorflow.initializer
+   :members:
+
+Cluster management
+------------------
+
+When scaling out training, you often want to adjust
+the parameters of your training program, for example,
+sharding the training dataset or scaling the learning rate
+of the optimizer. This can be achieved using the following
+cluster management APIs.
+
+.. automodule:: kungfu.python
+   :members:
+
+TensorFlow operators
+--------------------
+
+KungFu provides TensorFlow operators to help you realise
+new distributed training optimizers.
+
+.. automodule:: kungfu.tensorflow.ops
+   :members:
diff --git a/docs/index.rst b/docs/index.rst
index f19dc7107..5a8d9da75 100644
--- a/docs/index.rst
+++ b/docs/index.rst
@@ -190,58 +190,11 @@ You simply pass an extra `with_keras` flag to both KungFu optimizers and
 Keras callback to tell KungFu you are using Keras not TensorFlow.
 Here is a full Keras training example: `Keras <https://github.com/lsds/KungFu/blob/master/examples/keras_mnist.py>`_
 
-KungFu APIs
-===========
 
-KungFu has the high-level optimizer APIs that
-allows you to transparently scale out training.
-It also has a low-level API that allows an easy implementation
-of distributed training strategies.
-The following is the public API we released so far.
+.. toctree::
+  :maxdepth: 2
 
-Distributed optimizers
-----------------------
-
-KungFu provides optimizers that implement various distributed training algorithms.
-These optimizers are used for transparently scaling out the training of
-`tf.train.Optimizer <https://www.tensorflow.org/versions/r1.15/api_docs/python/tf/train/Optimizer>`_
-and `tf.keras.optimizers.Optimizer <https://www.tensorflow.org/api_docs/python/tf/keras/optimizers/Optimizer>`_
-
-.. automodule:: kungfu.tensorflow.optimizers
-   :members:
-
-Global variable initializers
-----------------------------
-
-KungFu provide various initializers to help you synchronize
-the global variables of distributed training workers at
-the beginning of training. These initializers are used
-with ``tf.session``, ``tf.estimator``, ``tf.GradientTape``
-and ``tf.keras``, respectively.
-
-.. automodule:: kungfu.tensorflow.initializer
-   :members:
-
-Cluster management
-------------------
-
-When scaling out training, you often want to adjust
-the parameters of your training program, for example,
-sharding the training dataset or scaling the learning rate
-of the optimizer. This can be achieved using the following
-cluster management APIs.
-
-.. automodule:: kungfu.python
-   :members:
-
-TensorFlow operators
---------------------
-
-KungFu provides TensorFlow operators to help you realise
-new distributed training optimizers.
-
-.. automodule:: kungfu.tensorflow.ops
-   :members:
+  apis
 
 Indices and tables
 ==================

From 118e06eac76974d7a3d2f49d28aa0f134499514c Mon Sep 17 00:00:00 2001
From: lg <lg4869@outlook.com>
Date: Fri, 6 Nov 2020 14:17:02 +0000
Subject: [PATCH 2/2] fix build

---
 docs/index.rst | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/docs/index.rst b/docs/index.rst
index 5a8d9da75..c7adb1bb0 100644
--- a/docs/index.rst
+++ b/docs/index.rst
@@ -194,7 +194,7 @@ Here is a full Keras training example: `Keras <https://github.com/lsds/KungFu/bl
 .. toctree::
   :maxdepth: 2
 
-  apis
+  api
 
 Indices and tables
 ==================