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 14, 23:45 UTC
[Model] Fix a check for None but the return value was empty list in Gemma3 MM vision_embeddings (#21479)
Signed-off-by: Hongmin Fan <fanhongmin@google.com>
hfan committed
Jul 25, 2025 at 01:46 UTC
fde60ee775de6406477d467ab5782f6a70141734
1 file changed
+1
-1
vllm/model_executor/models/gemma3_mm.py
+1
-1
@@ -627,7 +627,7 @@ class Gemma3ForConditionalGeneration(nn.Module, SupportsMultiModal, SupportsPP,
627
628
inputs_embeds = self.get_input_embeddings(input_ids,
629
vision_embeddings)
630
- if vision_embeddings is not None:
630
+ if (vision_embeddings is not None) and len(vision_embeddings) != 0:
631
kwargs = self.prepare_attn_masks(
632
input_ids,
633
positions,