biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

[VLM] Update compatibility with transformers 4.49

Browse Source
This commit is contained in:
Cyrus Leung
2025-02-06 11:09:45 +08:00
committed by GitHub
parent bf3b79efb8
commit 75404d041b
9 changed files with 48 additions and 57 deletions
Download Patch File Download Diff File Expand all files Collapse all files

29
vllm/model_executor/models/llava.py
View File

@@ -293,16 +293,29 @@ class PixtralHFMultiModalProcessor(
pixel_values = processed_outputs.get("pixel_values")
if pixel_values is not None:
images = mm_data["images"]
assert isinstance(images, list)
# Before/after https://github.com/huggingface/transformers/pull/35122
if Version(TRANSFORMERS_VERSION) <= Version("4.48.2"):
images = mm_data["images"]
assert isinstance(images, list)
# Original output: (1, num_images, C, H, W)
# New output: (num_images, C, H, W)
assert (isinstance(pixel_values, list) and len(pixel_values) == 1)
assert (isinstance(pixel_values[0], list)
and len(pixel_values[0]) == len(images))
# Original output: (1, num_images, C, H, W)
# New output: (num_images, C, H, W)
assert (isinstance(pixel_values, list)
and len(pixel_values) == 1)
assert (isinstance(pixel_values[0], list)
and len(pixel_values[0]) == len(images))
processed_outputs["pixel_values"] = pixel_values[0]
processed_outputs["pixel_values"] = pixel_values[0]
else:
# Avoid padding since we need the output for each image to be
# independent of other images for the cache to work correctly
image_sizes = processed_outputs["image_sizes"]
assert len(pixel_values) == len(image_sizes)
processed_outputs["pixel_values"] = [
p[:, :h, :w]
for p, (h, w) in zip(pixel_values, image_sizes)
]
return processed_outputs