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, 13:30 UTC
[Bugfix] Fix QWen2 VL multimodal mapping (#17240)
Signed-off-by: Jee Jee Li <pandaleefree@gmail.com>
Jee Jee Li committed
Apr 27, 2025 at 13:53 UTC
4283a28c2fb7595c5d72d3edc7ad356c87273d94
2 files changed
+6
-4
vllm/model_executor/models/qwen2_5_vl.py
+3
-2
@@ -1131,5 +1131,6 @@ class Qwen2_5_VLForConditionalGeneration(nn.Module, SupportsMultiModal,
1131
"""
1132
return MultiModelKeys.from_string_field(
1133
language_model="language_model",
1134
- connector="visual.",
1135
- tower_model="visual.merger.")
1134
+ connector="visual.merger.",
1135
+ tower_model="visual.",
1136
+ )
vllm/model_executor/models/qwen2_vl.py
+3
-2
@@ -1404,5 +1404,6 @@ class Qwen2VLForConditionalGeneration(nn.Module, SupportsMultiModal,
1404
"""
1405
return MultiModelKeys.from_string_field(
1406
language_model="language_model",
1407
- connector="visual.",
1408
- tower_model="visual.merger.")
1407
+ connector="visual.merger.",
1408
+ tower_model="visual.",
1409
+ )