From 4aed0ca6a2de8c56838fe850f8d8f8559074744a Mon Sep 17 00:00:00 2001
From: Jintao <huangjintao@mail.ustc.edu.cn>
Date: Thu, 10 Apr 2025 12:30:05 +0800
Subject: [PATCH] [bugfix] Avoid the time consumption caused by creating dummy
 videos. (#16371)

---
 vllm/multimodal/profiling.py | 6 ++++++
 1 file changed, 6 insertions(+)

diff --git a/vllm/multimodal/profiling.py b/vllm/multimodal/profiling.py
index 9a733d3bb44e8..ec3625f2f4265 100644
--- a/vllm/multimodal/profiling.py
+++ b/vllm/multimodal/profiling.py
@@ -78,6 +78,8 @@ class BaseDummyInputsBuilder(ABC, Generic[_I]):
         length: int,
         num_audios: int,
     ) -> list[npt.NDArray]:
+        if num_audios == 0:
+            return []
         audio = np.zeros((length, ))
         return [audio] * num_audios
 
@@ -88,6 +90,8 @@ class BaseDummyInputsBuilder(ABC, Generic[_I]):
         height: int,
         num_images: int,
     ) -> list[Image.Image]:
+        if num_images == 0:
+            return []
         image = Image.new("RGB", (width, height), color=255)
         return [image] * num_images
 
@@ -99,6 +103,8 @@ class BaseDummyInputsBuilder(ABC, Generic[_I]):
         num_frames: int,
         num_videos: int,
     ) -> list[npt.NDArray]:
+        if num_videos == 0:
+            return []
         video = np.full((num_frames, width, height, 3), 255)
         return [video] * num_videos