Update curriculum-learning.md

docs/_tutorials/curriculum-learning.md

@@ -130,7 +130,7 @@ In our [paper](https://arxiv.org/abs/2108.06084) section 5.4 we demonstrate that
 
 ### 2.3 Token-based training termination
 
-Because curriculum learning changes length of each sequence/sample during training, it is very hard/impossible to use number of steps/samples to terminate the training exactly at the desired number of tokens. Thus, we add a `--train-tokens` config for accurate token-based termination. We recommend increasing your original `--train-samples` or `--train-iters` to a large enough number (e.g., 3X of what you used for baseline), and set `--train-tokens` at the exact desired number of training tokens.
+Because curriculum learning changes the length of each sequence/sample during training, it is very hard/impossible to use  a number of steps/samples to terminate the training exactly at the desired number of tokens. Thus, we add a `--train-tokens` config for accurate token-based termination. We recommend increasing your original `--train-samples` or `--train-iters` to a large enough number (e.g., 3X of what you used for baseline), and set `--train-tokens` at the exact desired number of training tokens.
 
 ### 2.4 Token-based LR decay
 

setup.py

@@ -4,10 +4,10 @@
 DeepSpeed library
 
 To build wheel on Windows:
-    1. Install pytorch, such as pytorch 1.12 + cuda 11.6
-    2. Install visual cpp build tool
-    3. Include cuda toolkit
-    4. Launch cmd console with Administrator privilege for creating required symlink folders
+    1. Install pytorch, such as pytorch 1.12 + cuda 11.6.
+    2. Install visual cpp build tool.
+    3. Include cuda toolkit.
+    4. Launch cmd console with Administrator privilege for creating required symlink folders.
 
 Create a new wheel via the following command:
     build_win.bat
@@ -34,7 +34,7 @@
 from op_builder.all_ops import ALL_OPS
 from op_builder.builder import installed_cuda_version
 
-# fetch rocm state
+# Fetch rocm state.
 is_rocm_pytorch = OpBuilder.is_rocm_pytorch()
 rocm_version = OpBuilder.installed_rocm_version()
 
@@ -66,12 +66,12 @@ def fetch_requirements(path):
     'sd': fetch_requirements('requirements/requirements-sd.txt')
 }
 
-# Add specific cupy version to both onebit extension variants
+# Add specific cupy version to both onebit extension variants.
 if torch_available and torch.cuda.is_available():
     cupy = None
     if is_rocm_pytorch:
         rocm_major, rocm_minor = rocm_version
-        # XXX cupy support for rocm 5 is not available yet
+        # XXX cupy support for rocm 5 is not available yet.
         if rocm_major <= 4:
             cupy = f"cupy-rocm-{rocm_major}-{rocm_minor}"
     else:
@@ -80,7 +80,7 @@ def fetch_requirements(path):
         extras_require['1bit'].append(cupy)
         extras_require['1bit_mpi'].append(cupy)
 
-# Make an [all] extra that installs all needed dependencies
+# Make an [all] extra that installs all needed dependencies.
 all_extras = set()
 for extra in extras_require.items():
     for req in extra[1]:
@@ -89,7 +89,7 @@ def fetch_requirements(path):
 
 cmdclass = {}
 
-# For any pre-installed ops force disable ninja
+# For any pre-installed ops force disable ninja.
 if torch_available:
     from accelerator import get_accelerator
     cmdclass['build_ext'] = get_accelerator().build_extension().with_options(
@@ -103,7 +103,7 @@ def fetch_requirements(path):
     TORCH_MINOR = "0"
 
 if torch_available and not torch.cuda.is_available():
-    # Fix to allow docker builds, similar to https://github.com/NVIDIA/apex/issues/486
+    # Fix to allow docker builds, similar to https://github.com/NVIDIA/apex/issues/486.
     print(
         "[WARNING] Torch did not find cuda available, if cross-compiling or running with cpu only "
         "you can ignore this message. Adding compute capability for Pascal, Volta, and Turing "
@@ -148,14 +148,14 @@ def op_enabled(op_name):
     op_compatible = builder.is_compatible()
     compatible_ops[op_name] = op_compatible
 
-    # If op is requested but not available, throw an error
+    # If op is requested but not available, throw an error.
     if op_enabled(op_name) and not op_compatible:
         env_var = op_envvar(op_name)
         if env_var not in os.environ:
             builder.warning(f"One can disable {op_name} with {env_var}=0")
         abort(f"Unable to pre-compile {op_name}")
 
-    # if op is compatible but install is not enabled (JIT mode)
+    # If op is compatible but install is not enabled (JIT mode)
     if is_rocm_pytorch and op_compatible and not op_enabled(op_name):
         builder.hipify_extension()
 
@@ -167,7 +167,7 @@ def op_enabled(op_name):
 
 print(f'Install Ops={install_ops}')
 
-# Write out version/git info
+# Write out version/git info.
 git_hash_cmd = "git rev-parse --short HEAD"
 git_branch_cmd = "git rev-parse --abbrev-ref HEAD"
 if command_exists('git') and 'DS_BUILD_STRING' not in os.environ:
@@ -200,38 +200,38 @@ def create_dir_symlink(src, dest):
     create_dir_symlink('..\\accelerator', '.\\deepspeed\\accelerator')
     egg_info.manifest_maker.template = 'MANIFEST_win.in'
 
-# Parse the DeepSpeed version string from version.txt
+# Parse the DeepSpeed version string from version.txt.
 version_str = open('version.txt', 'r').read().strip()
 
 # Build specifiers like .devX can be added at install time. Otherwise, add the git hash.
-# example: DS_BUILD_STRING=".dev20201022" python setup.py sdist bdist_wheel
+# Example: DS_BUILD_STRING=".dev20201022" python setup.py sdist bdist_wheel.
 
-# Building wheel for distribution, update version file
+# Building wheel for distribution, update version file.
 if 'DS_BUILD_STRING' in os.environ:
-    # Build string env specified, probably building for distribution
+    # Build string env specified, probably building for distribution.
     with open('build.txt', 'w') as fd:
         fd.write(os.environ.get('DS_BUILD_STRING'))
     version_str += os.environ.get('DS_BUILD_STRING')
 elif os.path.isfile('build.txt'):
-    # build.txt exists, probably installing from distribution
+    # build.txt exists, probably installing from distribution.
     with open('build.txt', 'r') as fd:
         version_str += fd.read().strip()
 else:
-    # None of the above, probably installing from source
+    # None of the above, probably installing from source.
     version_str += f'+{git_hash}'
 
 torch_version = ".".join([TORCH_MAJOR, TORCH_MINOR])
 bf16_support = False
-# Set cuda_version to 0.0 if cpu-only
+# Set cuda_version to 0.0 if cpu-only.
 cuda_version = "0.0"
 nccl_version = "0.0"
-# Set hip_version to 0.0 if cpu-only
+# Set hip_version to 0.0 if cpu-only.
 hip_version = "0.0"
 if torch_available and torch.version.cuda is not None:
     cuda_version = ".".join(torch.version.cuda.split('.')[:2])
     if sys.platform != "win32":
         if isinstance(torch.cuda.nccl.version(), int):
-            # This will break if minor version > 9
+            # This will break if minor version > 9.
             nccl_version = ".".join(str(torch.cuda.nccl.version())[:2])
         else:
             nccl_version = ".".join(map(str, torch.cuda.nccl.version()[:2]))