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 8, 14:13 UTC
[Misc] allow specify is_mm_prefix_lm in hf_config (#34215)
Phúc H. Lê Khắc committed
Feb 10, 2026 at 18:16 UTC
94de871546e8da687c08ed8a7e0a26531500d4bd
1 file changed
+3
vllm/config/model.py
+3
@@ -1119,6 +1119,9 @@ class ModelConfig:
1119
@cached_property
1120
def is_mm_prefix_lm(self) -> bool:
1121
"""Whether to use bidirectional attention for mm positions."""
1122
+ if hasattr(self.hf_config, "is_mm_prefix_lm"):
1123
+ return bool(self.hf_config.is_mm_prefix_lm)
1124
+ # fallback to list of known models
1125
MM_PREFIX_LM_MODELS = (
1126
"gemma3",
1127
"molmo2",