Skip to content

Override moe_token_dispatcher_type to "alltoall" when export megatron… #2658

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

Open
jaeminh wants to merge 1 commit into
base: main
Choose a base branch
from
+6 −1
Open

Override moe_token_dispatcher_type to "alltoall" when export megatron… #2658

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
7 changes: 6 additions & 1 deletion src/megatron/bridge/models/conversion/auto_bridge.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -835,7 +835,12 @@ def export_ckpt(
# Export ckpt performs on CPU
with temporary_distributed_context(backend="gloo"):
# Load the Megatron model
megatron_model = self.load_megatron_model(megatron_path, wrap_with_ddp=False)
# 'flex' dispatcher requires TPxEP > 1; fall back to 'alltoall'
megatron_model = self.load_megatron_model(
megatron_path,
wrap_with_ddp=False,
mp_overrides={"moe_token_dispatcher_type": "alltoall"},
)

# Save in HuggingFace format
self.save_hf_pretrained(
Expand Down