Skip to content
GitLab
Menu
Projects
Groups
Snippets
Loading...
Help
Help
Support
Community forum
Keyboard shortcuts
?
Submit feedback
Contribute to GitLab
Sign in / Register
Toggle navigation
Menu
Open sidebar
OpenDAS
Megatron-LM
Commits
b7067cbd
Commit
b7067cbd
authored
Feb 22, 2021
by
Deepak Narayanan
Browse files
Fix interleaved schedule assertion
parent
9dc111cc
Changes
2
Hide whitespace changes
Inline
Side-by-side
Showing
2 changed files
with
3 additions
and
3 deletions
+3
-3
megatron/arguments.py
megatron/arguments.py
+0
-3
megatron/training.py
megatron/training.py
+3
-0
No files found.
megatron/arguments.py
View file @
b7067cbd
...
@@ -123,9 +123,6 @@ def parse_args(extra_args_provider=None, defaults={},
...
@@ -123,9 +123,6 @@ def parse_args(extra_args_provider=None, defaults={},
args
.
virtual_pipeline_model_parallel_size
=
\
args
.
virtual_pipeline_model_parallel_size
=
\
(
args
.
num_layers
//
args
.
pipeline_model_parallel_size
)
//
\
(
args
.
num_layers
//
args
.
pipeline_model_parallel_size
)
//
\
args
.
num_layers_per_virtual_pipeline_stage
args
.
num_layers_per_virtual_pipeline_stage
assert
args
.
global_batch_size
%
args
.
pipeline_model_parallel_size
==
0
,
\
'global batch size is not divisible by pipeline parallel size when '
\
'using interleaved schedule'
else
:
else
:
args
.
virtual_pipeline_model_parallel_size
=
None
args
.
virtual_pipeline_model_parallel_size
=
None
...
...
megatron/training.py
View file @
b7067cbd
...
@@ -339,6 +339,9 @@ def train_step(forward_step_func, data_iterator,
...
@@ -339,6 +339,9 @@ def train_step(forward_step_func, data_iterator,
if
mpu
.
get_pipeline_model_parallel_world_size
()
>
1
:
if
mpu
.
get_pipeline_model_parallel_world_size
()
>
1
:
if
args
.
virtual_pipeline_model_parallel_size
is
not
None
:
if
args
.
virtual_pipeline_model_parallel_size
is
not
None
:
forward_backward_func
=
forward_backward_pipelining_with_interleaving
forward_backward_func
=
forward_backward_pipelining_with_interleaving
assert
get_num_microbatches
()
%
args
.
pipeline_model_parallel_size
==
0
,
\
'number of microbatches is not divisible by pipeline-parallel '
\
'size when using interleaved schedule'
else
:
else
:
forward_backward_func
=
forward_backward_pipelining_without_interleaving
forward_backward_func
=
forward_backward_pipelining_without_interleaving
else
:
else
:
...
...
Write
Preview
Markdown
is supported
0%
Try again
or
attach a new file
.
Attach a file
Cancel
You are about to add
0
people
to the discussion. Proceed with caution.
Finish editing this message first!
Cancel
Please
register
or
sign in
to comment