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, 11:15 UTC
[Bugfix] Fix multimodal configuration for Qwen3VL MOE model (#30670)
Signed-off-by: Max Hu <hyoung2991@gmail.com>
Max Hu committed
Dec 15, 2025 at 22:06 UTC
3f175f18a2e5d430ffa17fcb96759a758cc3ec05
1 file changed
+5
vllm/model_executor/models/qwen3_vl_moe.py
+5
@@ -418,6 +418,11 @@ class Qwen3VLMoeForConditionalGeneration(
418
419
self.config = config
420
self.multimodal_config = multimodal_config
421
+ self.use_data_parallel = multimodal_config.mm_encoder_tp_mode == "data"
422
+ self.video_pruning_rate = multimodal_config.video_pruning_rate
423
+ self.is_multimodal_pruning_enabled = (
424
+ multimodal_config.is_multimodal_pruning_enabled()
425
+ )
426
427
if not multimodal_config.get_limit_per_prompt(
428
"image"