Skip to content

[conversion] fix: skip absent auto mappings across PP #6

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

Draft
bobboli wants to merge 1 commit into from
Draft

[conversion] fix: skip absent auto mappings across PP #6

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
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
9 changes: 7 additions & 2 deletions src/megatron/bridge/models/conversion/param_mapping.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1366,9 +1366,14 @@ def megatron_to_hf(
self._detected_type = self.broadcast_obj_from_pp_rank(self._detected_type, "detected_type")
else:
# Receive from owning rank
self._detected_type = self.broadcast_obj_from_pp_rank(None, "detected_type")
try:
self._detected_type = self.broadcast_obj_from_pp_rank(None, "detected_type")
except ValueError as error:
if str(error) != "Object must exist on at least one PP rank":
raise
self._detected_type = None
if self._detected_type is None:
# PP group likely has 1 member - skipping.
# No PP stage owns this optional parameter.
return {}

self._mapping = self._get_or_create_mapping(self._detected_type)
Expand Down
11 changes: 11 additions & 0 deletions tests/unit_tests/models/test_param_mapping.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -273,6 +273,17 @@ class MyCustomRow(torch.nn.Module):
with pytest.raises(ValueError):
mapping._detect_parallelism_type(torch.nn.Linear(5, 5))

def test_megatron_to_hf_skips_parameter_missing_from_all_pp_stages(self, mock_distributed_env):
_, mock_dist = mock_distributed_env(pp_size=2, pp_rank=0)
mapping = AutoMapping(megatron_param="optional.weight", hf_param="hf.optional.weight")

mock_dist.all_gather_object.side_effect = lambda output, obj, group: output.__setitem__(
slice(None), [False, False]
)

assert mapping.megatron_to_hf(None, None) == {}
mock_dist.broadcast_object_list.assert_not_called()

def test_detect_parallelism_type_dynamic_module(self):
mtq = pytest.importorskip("modelopt.torch.quantization")
DynamicModule = pytest.importorskip("modelopt.torch.opt.dynamic").DynamicModule
Expand Down