Skip to content

fix: vLLM setup race condition inside video captioning stages #1590

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
weijiac0619 merged 2 commits into from
Mar 10, 2026
Merged

fix: vLLM setup race condition inside video captioning stages #1590

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
a19b418
fixed
weijiac0619 Mar 9, 2026
4c52cab
Merge branch 'main' into weijia/fix_concurrent_worker
weijiac0619 Mar 10, 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
9 changes: 7 additions & 2 deletions nemo_curator/stages/video/caption/caption_enhancement.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -67,7 +67,7 @@ def __post_init__(self) -> None:
self.prompt_text,
)

def setup(self, worker_metadata: WorkerMetadata | None = None) -> None: # noqa: ARG002
def _initialize_model(self) -> None:
if self.model_variant == "qwen":
self.model = QwenLM(
model_dir=self.model_dir,
Expand All @@ -81,8 +81,13 @@ def setup(self, worker_metadata: WorkerMetadata | None = None) -> None: # noqa:
self.model.setup()

def setup_on_node(self, node_info: NodeInfo, worker_metadata: WorkerMetadata) -> None: # noqa: ARG002
"""Download the weights for the QwenLM model on the node."""
"""Download weights and initialize vLLM once per node to avoid torch.compile race conditions."""
QwenLM.download_weights_on_node(self.model_dir)
self._initialize_model()

def setup(self, worker_metadata: WorkerMetadata | None = None) -> None: # noqa: ARG002
if not hasattr(self, "model") or self.model is None:
self._initialize_model()

def process(self, task: VideoTask) -> VideoTask:
video = task.data
Expand Down
9 changes: 7 additions & 2 deletions nemo_curator/stages/video/caption/caption_generation.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -51,7 +51,7 @@ def inputs(self) -> tuple[list[str], list[str]]:
def outputs(self) -> tuple[list[str], list[str]]:
return ["data"], ["clips"]

def setup(self, worker_metadata: WorkerMetadata | None = None) -> None: # noqa: ARG002
def _initialize_model(self) -> None:
if self.model_variant == "qwen":
self.model = QwenVL(
model_dir=self.model_dir,
Expand All @@ -68,8 +68,13 @@ def setup(self, worker_metadata: WorkerMetadata | None = None) -> None: # noqa:
self.model.setup()

def setup_on_node(self, node_info: NodeInfo, worker_metadata: WorkerMetadata) -> None: # noqa: ARG002
"""Download the weights for the QwenVL model on the node."""
"""Download weights and initialize vLLM once per node to avoid torch.compile race conditions."""
QwenVL.download_weights_on_node(self.model_dir)
self._initialize_model()

def setup(self, worker_metadata: WorkerMetadata | None = None) -> None: # noqa: ARG002
if not hasattr(self, "model") or self.model is None:
self._initialize_model()

def __post_init__(self) -> None:
self.resources = Resources(gpus=1)
Expand Down
Loading