Skip to content

Fix sharding < 100 limitation bug #8146

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

Merged
merged 1 commit into from
Apr 16, 2024
Merged

Fix sharding < 100 limitation bug #8146

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
29 changes: 12 additions & 17 deletions paddlenlp/trainer/training_args.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1548,20 +1548,20 @@ def pipeline_parallel_rank(self):
else:
return 0

def _format_name(self, prefix, rank, degree):
size = max(2, len(str(degree)))
return f"{prefix}{rank:0>{size}d}"

@property
def optimizer_name_suffix(self):
if self.use_hybrid_parallel:
name = []
if self.tensor_parallel_degree > 1:
assert self.tensor_parallel_degree < 100, "tensor parallel degree should be less than 100."
name.append(f"tp{self.tensor_parallel_rank:0>2d}")
name.append(self._format_name("tp", self.tensor_parallel_rank, self.tensor_parallel_degree))
if self.pipeline_parallel_degree > 1:
assert self.pipeline_parallel_degree < 100, "pipeline parallel degree should be less than 100."
name.append(f"pp{self.pipeline_parallel_rank:0>2d}")
name.append(self._format_name("pp", self.pipeline_parallel_rank, self.pipeline_parallel_degree))
if self.sharding_parallel_degree > 1:
assert self.sharding_parallel_degree < 100, "sharding parallel degree should be less than 100."
name.append(f"shard{self.sharding_parallel_rank:0>2d}")

name.append(self._format_name("shard", self.sharding_parallel_rank, self.sharding_parallel_degree))
return "_".join(name)
else:
return None
Expand All @@ -1571,11 +1571,9 @@ def weight_name_suffix(self):
if self.use_hybrid_parallel:
name = []
if self.tensor_parallel_degree > 1:
assert self.tensor_parallel_rank < 100, "tensor parallel rank should be less than 100."
name.append(f"tp{self.tensor_parallel_rank:0>2d}")
name.append(self._format_name("tp", self.tensor_parallel_rank, self.tensor_parallel_degree))
if self.pipeline_parallel_degree > 1:
assert self.pipeline_parallel_degree < 100, "tensor parallel rank should be less than 100."
name.append(f"pp{self.pipeline_parallel_rank:0>2d}")
name.append(self._format_name("pp", self.pipeline_parallel_rank, self.pipeline_parallel_degree))
return "_".join(name)

else:
Expand All @@ -1585,20 +1583,17 @@ def sharded_name_suffix(self, shard_id=None, pp_id=None):
if self.use_hybrid_parallel:
name = []
if self.tensor_parallel_degree > 1:
assert self.tensor_parallel_rank < 100, "tensor parallel rank should be less than 100."
name.append(f"tp{self.tensor_parallel_rank:0>2d}")
name.append(self._format_name("tp", self.tensor_parallel_rank, self.tensor_parallel_degree))
if self.pipeline_parallel_degree > 1:
if pp_id is None:
pp_id = self.pipeline_parallel_rank
assert isinstance(pp_id, int)
assert pp_id < 100, "pp_id should be less than 100."
name.append(f"pp{pp_id:0>2d}")
name.append(self._format_name("pp", pp_id, self.pipeline_parallel_degree))
if self.sharding_parallel_degree > 1:
if shard_id is None:
shard_id = self.sharding_parallel_rank
assert isinstance(shard_id, int)
assert shard_id < 100, "shard_id should be less than 100."
name.append(f"shard{shard_id:0>2d}")
name.append(self._format_name("shard", shard_id, self.sharding_parallel_degree))
return "_".join(name)
else:
return None
Expand Down