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

Fix missing parallelisms #9725

Merged
merged 2 commits into from
Jul 17, 2024
Merged

Fix missing parallelisms #9725

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
999efcc
pass cp and ep cfg to init mp
maanug-nv Jul 13, 2024
047421f
update test
maanug-nv Jul 15, 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: 5 additions & 1 deletion nemo/lightning/_strategy_lib.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -61,12 +61,14 @@ def init_parallel_ranks(
global_rank=init_global_rank,
local_rank=init_local_rank,
tensor_model_parallel_size=parallel_config.tensor_model_parallel_size,
expert_model_parallel_size=parallel_config.expert_model_parallel_size,
pipeline_model_parallel_size=parallel_config.pipeline_model_parallel_size,
virtual_pipeline_model_parallel_size=parallel_config.virtual_pipeline_model_parallel_size,
context_parallel_size=parallel_config.context_parallel_size,
seed=seed,
pipeline_model_parallel_split_rank=getattr(parallel_config, "pipeline_model_parallel_split_rank", None),
use_fp8=fp8,
init_mpi_proc_group=getattr(parallel_config, "ub_tp_comm_overlap", False),
init_mpi_proc_group=getattr(parallel_config, "tp_comm_overlap", False),
# apex_transformer_log_level=self.cfg.get('apex_transformer_log_level', 30),
)

Expand All @@ -92,6 +94,8 @@ def init_model_parallel(model: Optional[nn.Module] = None) -> None:
pipeline_model_parallel_size=app_state.pipeline_model_parallel_size,
virtual_pipeline_model_parallel_size=app_state.virtual_pipeline_model_parallel_size,
pipeline_model_parallel_split_rank=app_state.pipeline_model_parallel_split_rank,
context_parallel_size=app_state.context_parallel_size,
expert_model_parallel_size=app_state.expert_model_parallel_size,
)

# assert that fake tp and pp rank match after model parallel init
Expand Down
19 changes: 17 additions & 2 deletions tests/lightning/test_strategy_lib.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -23,18 +23,27 @@ def test_init_parallel_ranks(mock_initialize_model_parallel) -> None:

app_state.tensor_model_parallel_size = 2
app_state.pipeline_model_parallel_size = 3
app_state.context_parallel_size = 2
app_state.expert_model_parallel_size = 2
app_state.global_rank = 1
app_state.local_rank = 0

mock_parallel_config = MagicMock()
mock_parallel_config.tensor_model_parallel_size = 2
mock_parallel_config.pipeline_model_parallel_size = 3
mock_parallel_config.virtual_pipeline_model_parallel_size = 4
mock_parallel_config.ub_tp_comm_overlap = False
mock_parallel_config.context_parallel_size = 2
mock_parallel_config.expert_model_parallel_size = 2
mock_parallel_config.tp_comm_overlap = False
mock_parallel_config.pipeline_model_parallel_split_rank = None

_strategy_lib.init_parallel_ranks(
world_size=2, global_rank=1, local_rank=0, parallel_config=mock_parallel_config, seed=1234, fp8=False,
world_size=2,
global_rank=1,
local_rank=0,
parallel_config=mock_parallel_config,
seed=1234,
fp8=False,
)
mock_initialize_model_parallel.assert_called_once_with(
world_size=2,
Expand All @@ -43,6 +52,8 @@ def test_init_parallel_ranks(mock_initialize_model_parallel) -> None:
tensor_model_parallel_size=2,
pipeline_model_parallel_size=3,
virtual_pipeline_model_parallel_size=4,
context_parallel_size=2,
expert_model_parallel_size=2,
seed=1234,
pipeline_model_parallel_split_rank=None,
use_fp8=False,
Expand All @@ -60,6 +71,8 @@ def test_init_model_parallel(mock_mpu, *args):
app_state.tensor_model_parallel_size = 2
app_state.pipeline_model_parallel_size = 1
app_state.pipeline_model_parallel_split_rank = None
app_state.context_parallel_size = 2
app_state.expert_model_parallel_size = 2
app_state.init_mpi_proc_group = False
app_state.tensor_model_parallel_rank = 2
app_state.pipeline_model_parallel_rank = 0
Expand All @@ -72,6 +85,8 @@ def test_init_model_parallel(mock_mpu, *args):
pipeline_model_parallel_size=1,
virtual_pipeline_model_parallel_size=None,
pipeline_model_parallel_split_rank=None,
context_parallel_size=2,
expert_model_parallel_size=2,
)


Expand Down
Loading