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 11, 16:15 UTC
Ensure minimum frames for GLM 4.6V compatibility (#30285)
Signed-off-by: Wayne Ferguson <wayneferguson@gmail.com>
gh-wf committed
Dec 11, 2025 at 00:26 UTC
36c9ce25543b4f48194d7adc4ba3d17f5b6102be
1 file changed
+1
vllm/model_executor/models/glm4_1v.py
+1
@@ -1257,6 +1257,7 @@ class Glm4vDummyInputsBuilder(BaseDummyInputsBuilder[Glm4vProcessingInfo]):
1257
)
1258
height = min(height, overrides.height)
1259
1260
+ num_frames = max(num_frames, 2) # GLM 4.6V requires 2 frames
1261
video = np.full((num_frames, width, height, 3), 255, dtype=np.uint8)
1262
video_items = []
1263
for i in range(num_videos):