Skip to content

[https://nvbugs/5154414][fix] Balanced layer to PP rank assignment #3827

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

Closed
wants to merge 1 commit into from
Closed

[https://nvbugs/5154414][fix] Balanced layer to PP rank assignment #3827

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
15 changes: 6 additions & 9 deletions tensorrt_llm/mapping.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -14,6 +14,8 @@
# limitations under the License.
from typing import List

import torch


class Mapping(object):
'''
Expand Down Expand Up @@ -419,16 +421,11 @@ def has_moe_tp(self):
def has_moe_ep(self):
return self.moe_ep_size > 1

# TODO: Add support of uneven/arbitrary layer segmentation
def pp_layers(self, num_layers: int) -> List[int]:
layers_per_pipeline_stage = num_layers // self.pp_size
if self.pp_rank == self.pp_size - 1:
layers_range = range(self.pp_rank * layers_per_pipeline_stage,
num_layers)
else:
layers_range = range(self.pp_rank * layers_per_pipeline_stage,
(self.pp_rank + 1) * layers_per_pipeline_stage)
return list(layers_range)
assert self.pp_size <= num_layers, "pp size cannot be larger than num_hidden_layers"
# If num_layers % pp_size = n != 0, first n ranks get one extra layer
return torch.tensor_split(torch.arange(num_layers),
self.pp_size)[self.pp_rank].tolist()

def ep_experts(self, num_experts: int) -> List[int]:
assert self.cp_size == 1
Expand Down