Skip to content

[Tiny Fix] Disable BCG when inner layer_model unresolved #25021

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
ispobock merged 4 commits into from
May 12, 2026
Merged

[Tiny Fix] Disable BCG when inner layer_model unresolved #25021

Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
b64a9cb
breakable cuda graph: warn when layer_model fallback fires
Oasis-Git May 12, 2026
544af2f
breakable cuda graph: disable BCG when layer_model unresolved
Oasis-Git May 12, 2026
77b6393
breakable cuda graph: trim verbose comment and warning text
Oasis-Git May 12, 2026
b58837c
Merge branch 'main' into bcg-fix
Oasis-Git May 12, 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
19 changes: 13 additions & 6 deletions python/sglang/srt/model_executor/breakable_cuda_graph_runner.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -122,12 +122,17 @@ def __init__(self, model_runner: ModelRunner):
language_model = getattr(
model_runner.model, "language_model", model_runner.model
)
self.layer_model = (
language_model.model
if hasattr(language_model, "model")
and hasattr(language_model.model, "layers")
else language_model
)
if hasattr(language_model, "model") and hasattr(language_model.model, "layers"):
self.layer_model = language_model.model
else:
# If we can't find the inner layer_model, disable BCG.
self.layer_model = None
logger.warning(
"[BCG] Could not resolve inner layer_model on %s. BCG is "
"disabled for this model; prefill will fall back to eager.",
type(language_model).__name__,
)
return

# Memory pool
if get_global_graph_memory_pool() is None:
Expand Down Expand Up @@ -330,6 +335,8 @@ def _capture_all(self):
self.output_buffers[num_tokens] = output

def can_run(self, forward_batch: "ForwardBatch"):
if self.layer_model is None:
return False
if forward_batch.forward_mode.is_target_verify():
return False
if forward_batch.input_embeds is not None:
Expand Down
Loading