Skip to content

fix: resolve tensor file overwrite between target and draft models #21694

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
Kangyan-Zhou merged 12 commits into from
Apr 28, 2026
Merged

fix: resolve tensor file overwrite between target and draft models #21694

Changes from 3 commits
Commits
Show all changes
12 commits
Select commit Hold shift + click to select a range
2cb370c
fix: resolve tensor file overwrite between target and draft models
Mar 30, 2026
5b8728d
In Eagle mode, tensor files generated by the target and draft models …
Mar 30, 2026
761039d
chore: trigger CI rerun
Mar 31, 2026
f5b3b6a
docs: clarify tensor dump output behavior in Eagle mode
Mar 31, 2026
3c2829d
Merge branch 'main' into fix_eagle_fileoverwrite
kpham-sgl Apr 7, 2026
a89b336
Merge branch 'main' into fix_eagle_fileoverwrite
kpham-sgl Apr 9, 2026
3a3807e
Merge branch 'main' into fix_eagle_fileoverwrite
kpham-sgl Apr 10, 2026
36c862f
Merge branch 'main' into fix_eagle_fileoverwrite
kpham-sgl Apr 10, 2026
6f7aa29
Merge branch 'main' into fix_eagle_fileoverwrite
kpham-sgl Apr 14, 2026
518210d
Merge branch 'main' into fix_eagle_fileoverwrite
kpham-sgl Apr 15, 2026
a201957
Merge branch 'main' into fix_eagle_fileoverwrite
kpham-sgl Apr 16, 2026
23479a4
Merge branch 'main' into fix_eagle_fileoverwrite
kpham-sgl Apr 22, 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
6 changes: 5 additions & 1 deletion python/sglang/srt/model_executor/model_runner.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1192,9 +1192,13 @@ def load_model(self):
f"mem usage={self.weight_load_mem_usage:.2f} GB."
)
if self.server_args.debug_tensor_dump_output_folder is not None:
dump_folder = self.server_args.debug_tensor_dump_output_folder

@kpham-sgl kpham-sgl Mar 31, 2026

Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

nit: possibly worth to document this behavior in self.server_args.debug_tensor_dump_output_folder help docstring

@yaya159456 yaya159456 Mar 31, 2026

Copy link
Copy Markdown
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Good suggestion, thanks!
I've added clarification to the help docstring regarding the behavior in Eagle mode.
In addition, I submitted a PR to update the documentation in sgl-project.github.io to make this behavior more explicit:
sgl-project/sgl-project.github.io#26

if self.spec_algorithm.is_eagle():
role = "draft" if self.is_draft_worker else "target"
dump_folder = os.path.join(dump_folder, role)
register_forward_hook_for_model(
self.model,
self.server_args.debug_tensor_dump_output_folder,
dump_folder,
self.server_args.debug_tensor_dump_layers,
self.tp_size,
self.tp_rank,
Expand Down
Loading