Skip to content

[models] fix: accept Gemma4 router input ids #3835

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
yaoyu-33 merged 3 commits into from
May 15, 2026
Merged

[models] fix: accept Gemma4 router input ids #3835

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
f6b9e0c
chore(beep boop 🤖): Bump `uv.lock` (main, mcore-dev) (2026-05-14)
dimapihtar May 14, 2026
c36259c
[models] fix: accept Gemma4 router input ids
yaoyu-33 May 14, 2026
e4d4802
[build] fix: keep MCore dev bump metadata only
yaoyu-33 May 15, 2026
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
2 changes: 1 addition & 1 deletion .dev.commit
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1 +1 @@
85bced0ae6ab46f61a0fd774074a3273daf6ae02
cf081d5df0b34b665d214ff936f27489d7396876
4 changes: 2 additions & 2 deletions src/megatron/bridge/models/gemma/gemma4_provider.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -260,9 +260,9 @@ def __init__(self, config, **kwargs):
torch.ones(config.hidden_size, dtype=config.params_dtype),
)

def routing(self, logits, padding_mask=None):
def routing(self, logits, padding_mask=None, input_ids=None):
"""Apply standard routing, then renormalize and scale by per_expert_scale."""
routing_probs, routing_map = super().routing(logits, padding_mask=padding_mask)
routing_probs, routing_map = super().routing(logits, padding_mask=padding_mask, input_ids=input_ids)
# routing_probs: [num_tokens, num_experts] sparse — non-zero at selected experts
# routing_map: [num_tokens, num_experts] boolean mask
#
Expand Down
Loading