readthedocs yaml configuration

.gitignore

@@ -3,7 +3,7 @@
 *~
 *.swp
 *.log
-deepspeed/git_version_info.py
+deepspeed/git_version_info_installed.py
 
 # Build + installation data
 build/

.readthedocs.yml

@@ -0,0 +1,18 @@
+
+# Required
+version: 2
+
+# Build documentation in the docs/ directory with Sphinx
+sphinx:
+  configuration: docs/code-docs/source/conf.py
+  fail_on_warning: false
+
+# Optionally build your docs in additional formats such as PDF
+formats:
+  - pdf
+
+# Optionally set the version of Python and requirements required to build your docs
+python:
+  version: 3.7
+  install:
+    - requirements: requirements/requirements-readthedocs.txt

deepspeed/git_version_info.py

@@ -0,0 +1,14 @@
+try:
+    #  This is populated by setup.py
+    from .git_version_info_installed import *
+except ModuleNotFoundError:
+    # Will be missing from checkouts that haven't been installed (e.g., readthedocs)
+    version = '0.3.0+[none]'
+    git_hash = '[none]'
+    git_branch = '[none]'
+    installed_ops = {
+        'lamb': False,
+        'transformer': False,
+        'sparse-attn': False,
+        'cpu-adam': False
+    }

docs/_tutorials/pipeline.md

@@ -10,7 +10,7 @@ DeepSpeed's training engine provides hybrid data and pipeline parallelism and
 can be further combined with model parallelism such as
 [Megatron-LM](https://github.com/NVIDIA/Megatron-LM).
 An illustration of
-3D parallelism is shown below. Our latest [results](linklinklink)
+3D parallelism is shown below. Our latest [results]({{ site.press_release_v3 }})
 demonstrate that this 3D parallelism enables training models with over a
 **trillion** parameters.
 

docs/code-docs/source/conf.py

@@ -34,6 +34,7 @@
 extensions = [
     'sphinx.ext.autodoc',
     'sphinx.ext.napoleon',
+    'sphinx.ext.viewcode',
     'recommonmark',
     'sphinx_rtd_theme',
 ]
@@ -78,11 +79,4 @@
 
 autoclass_content = 'both'
 
-autodoc_mock_imports = [
-    "torch",
-    "apex",
-    "mpi4py",
-    "tensorboardX",
-    "deepspeed_transformer_cuda",
-    "deepspeed_stochastic_transformer_cuda",
-]
+autodoc_mock_imports = ["torch", "apex", "mpi4py", "tensorboardX", "numpy"]

docs/code-docs/source/training.rst

@@ -28,7 +28,6 @@ Optimizer Step
 --------------
 .. autofunction:: deepspeed.DeepSpeedEngine.step
 
-
 Gradient Accumulation
 ---------------------
 .. autofunction:: deepspeed.DeepSpeedEngine.is_gradient_accumulation_boundary

install.sh

@@ -137,7 +137,7 @@ rm_if_exist() {
 
 if [ "$no_clean" == "0" ]; then
     # remove deepspeed build files
-    rm_if_exist deepspeed/git_version_info.py
+    rm_if_exist deepspeed/git_version_info_installed.py
     rm_if_exist dist
     rm_if_exist build
     rm_if_exist deepspeed.egg-info

requirements/requirements-dev.txt

@@ -2,3 +2,6 @@ pytest
 pytest-forked
 pre-commit
 clang-format
+sphinx
+recommonmark
+sphinx-rtd-theme

setup.py

@@ -297,7 +297,7 @@ def command_exists(cmd):
     git_hash = "unknown"
     git_branch = "unknown"
 print(f"version={VERSION}+{git_hash}, git_hash={git_hash}, git_branch={git_branch}")
-with open('deepspeed/git_version_info.py', 'w') as fd:
+with open('deepspeed/git_version_info_installed.py', 'w') as fd:
     fd.write(f"version='{VERSION}+{git_hash}'\n")
     fd.write(f"git_hash='{git_hash}'\n")
     fd.write(f"git_branch='{git_branch}'\n")