Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

Make pipelined TP comm overlap available with mcore #8005

Merged
merged 2 commits into from
Jan 10, 2024
Merged

Make pipelined TP comm overlap available with mcore #8005

Changes from 1 commit
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
f4721ca
Make pipelined TP comm overlap available with mcore
erhoo82 Dec 9, 2023
ad9b504
remove unnecessary condition
erhoo82 Jan 3, 2024
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
6 changes: 4 additions & 2 deletions nemo/collections/nlp/models/language_modeling/megatron_base_model.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -792,12 +792,14 @@ def _validate_and_override_config(self):
) % vp_size == 0, 'Make sure the number of model chunks is the same across all pipeline stages.'

if self.cfg.get('ub_tp_comm_overlap', False):
if not self.cfg.get('transformer_engine', False) or not self.cfg.get('sequence_parallel', False):
if self.cfg.get('ub_tp_comm_overlap', False) and not self.cfg.get('sequence_parallel', False):
Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

@erhoo82 if self.cfg.get('ub_tp_comm_overlap', False) seems to be redundant in L795 as it already exists in L794.

logging.info(
"Userbuffer tensor-parallel communication overlap is available with both Transformer Engine and sequence-parallelism."
"Pipelined tensor-parallel communication overlap is available with sequence-parallelism."
"Setting `ub_tp_comm_overlap` to False."
)
with open_dict(self.cfg):
self.cfg.ub_tp_comm_overlap = False

if self.cfg.get('fsdp', False):
logging.info(
"Userbuffer tensor-parallel communication overlap is not available with FSDP."
Expand Down
Loading