Skip to content

[Bugfix] Dtype error with sequence classification model and lora. #24775

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
wants to merge 2 commits into
base: main
Choose a base branch
from
+1 −2
Open

[Bugfix] Dtype error with sequence classification model and lora. #24775

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
e1f9c61
Don't set dtype in classification head, let the dtype context manager…
pb-sameerreddy Sep 12, 2025
0864777
Update adapters.py
pb-sameerreddy Sep 12, 2025
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
3 changes: 1 addition & 2 deletions vllm/model_executor/models/adapters.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -307,7 +307,6 @@ def _init_pooler(self, vllm_config: "VllmConfig", prefix: str = ""):
hidden_size,
config.num_labels,
bias=False,
params_dtype=torch.float32,
Copy link
Collaborator

@jeejeelee jeejeelee Sep 17, 2025 •
edited
Loading

Choose a reason for hiding this comment

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

Maybe we should use params_dtype=vllm_config.model_config.head_dtype

Copy link
Collaborator

Choose a reason for hiding this comment

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

  1. Please add params_dtype=vllm_config.model_config.head_dtype to _create_pooling_model_cls, as reference:
    noooop@c427c9b#diff-d85c8d9a87dd68e2c04e4d9bb9e1c4493646c7fc6ee877ee574dc7805bf6f57aR130-R266

this code wasn’t committed after numerous rebases. at #23810

Copy link
Collaborator

Choose a reason for hiding this comment

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

#24775 (comment)

quant_config=quant_config,
prefix=maybe_prefix(prefix, "score"),
)
Expand Down Expand Up @@ -339,7 +338,7 @@ def _init_pooler(self, vllm_config: "VllmConfig", prefix: str = ""):
})

def _classifier(self, x: torch.Tensor):
x, _ = self.score(x.float())
x, _ = self.score(x)
Copy link
Collaborator

Choose a reason for hiding this comment

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

Suggested change
x, _ = self.score(x)
x = x.to(self.vllm_config.model_config.head_dtype)
x, _ = self.score(x)

return x

def forward(
Expand Down