Skip to content
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

[Bugfix] Fix dtype mismatch in PaliGemma #6367

Merged
merged 6 commits into from
Jul 12, 2024
Merged

[Bugfix] Fix dtype mismatch in PaliGemma #6367

Show file tree
Hide file tree
Changes from 2 commits
Commits
Show all changes
6 commits
Select commit Hold shift + click to select a range
b21e498
Fix paligemma
DarkLight1337 Jul 12, 2024
2b3e7cd
yapf
DarkLight1337 Jul 12, 2024
7296d9c
Test both float and half dtypes
DarkLight1337 Jul 12, 2024
f577b89
Fix missing arg to language model
DarkLight1337 Jul 12, 2024
3626a39
Fix missing arg to language model
DarkLight1337 Jul 12, 2024
c73887e
yapf
DarkLight1337 Jul 12, 2024
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 tests/models/test_paligemma.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -129,7 +129,7 @@ def run_test(
[0.25, 0.5, 1.0],
],
)
@pytest.mark.parametrize("dtype", ["float"])
@pytest.mark.parametrize("dtype", ["half"])
@pytest.mark.parametrize("max_tokens", [128])
@pytest.mark.parametrize("num_logprobs", [5])
def test_models(hf_runner, vllm_runner, image_assets, model, size_factors,
Expand Down
6 changes: 4 additions & 2 deletions vllm/model_executor/models/paligemma.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -111,7 +111,7 @@ def input_processor_for_paligemma(ctx: InputContext, llm_inputs: LLMInputs):
orig_prompt = llm_inputs.get("prompt")
orig_prompt_ids = llm_inputs.get("prompt_token_ids")

if image_token_str in orig_prompt:
if orig_prompt is not None and image_token_str in orig_prompt:
Copy link
Member Author

Choose a reason for hiding this comment

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

My IDE flagged this so I fixed it along the way.

logger.warning(
"The image token '%s' was detected in the prompt and "
"will be removed. Please follow the proper prompt format"
Expand Down Expand Up @@ -214,7 +214,9 @@ def _parse_and_validate_image_input(
def _image_pixels_to_features(self, vision_tower: SiglipVisionModel,
pixel_values: torch.Tensor) -> torch.Tensor:

image_outputs = vision_tower(pixel_values, output_hidden_states=True)
target_dtype = vision_tower.get_input_embeddings().weight.dtype
image_outputs = vision_tower(pixel_values.to(dtype=target_dtype),
output_hidden_states=True)

selected_image_features = image_outputs.last_hidden_state

Expand Down
Loading