A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 15, 18:30 UTC
[Misc] Avoid unnecessary HF `do_rescale` warning when passing dummy data (#15107)
Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>
Cyrus Leung committed
Mar 19, 2025 at 18:42 UTC
1fe0fd12d33afb345c8e81410f2ebf60462974b6
1 file changed
+2
-2
vllm/multimodal/profiling.py
+2
-2
@@ -73,7 +73,7 @@ class BaseDummyInputsBuilder(ABC, Generic[_I]):
73
height: int,
74
num_images: int,
75
) -> list[Image.Image]:
76
- image = Image.new("RGB", (width, height), color=0)
76
+ image = Image.new("RGB", (width, height), color=255)
77
return [image] * num_images
78
79
def _get_dummy_videos(
@@ -84,7 +84,7 @@ class BaseDummyInputsBuilder(ABC, Generic[_I]):
84
num_frames: int,
85
num_videos: int,
86
) -> list[npt.NDArray]:
87
- video = np.zeros((num_frames, width, height, 3))
87
+ video = np.full((num_frames, width, height, 3), 255)
88
return [video] * num_videos
89
90