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, 04:45 UTC
Fix(models/siglip): Add compatibility for Gemma models quantized by llm-compressor (#19643)
Signed-off-by: Vensenmu <vensenmu@gmail.com>
Vensen committed
Jun 23, 2025 at 11:40 UTC
493c275352bddf7d4877602e19b8bd29d662de63
1 file changed
+1
vllm/model_executor/models/gemma3_mm.py
+1
@@ -479,6 +479,7 @@ class Gemma3ForConditionalGeneration(nn.Module, SupportsMultiModal, SupportsPP,
479
"model.vision_tower.": "vision_tower.",
480
"model.multi_modal_projector.": "multi_modal_projector.",
481
"lm_head.": "language_model.lm_head.",
482
+ "vision_tower.vision_model.": "vision_model.",
483
})
484
485
def __init__(self, *, vllm_config: VllmConfig, prefix: str = ""):