From bc4ae911649ca1c7d84ab3f67423297ece1922db Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?Jie=20Fu=20=28=E5=82=85=E6=9D=B0=29?= Date: Sat, 22 Jun 2024 10:07:08 +0800 Subject: [PATCH] [Model] Support Qwen-VL and Qwen-VL-Chat models with text-only inputs (#5710) Co-authored-by: Roger Wang --- vllm/model_executor/models/qwen.py | 10 ++++++++++ 1 file changed, 10 insertions(+) diff --git a/vllm/model_executor/models/qwen.py b/vllm/model_executor/models/qwen.py index d22ea6b79de0f..b6ea6ab396642 100644 --- a/vllm/model_executor/models/qwen.py +++ b/vllm/model_executor/models/qwen.py @@ -28,6 +28,7 @@ from vllm.model_executor.model_loader.weight_utils import default_weight_loader from vllm.model_executor.sampling_metadata import SamplingMetadata from vllm.sequence import SamplerOutput +from vllm.utils import print_warning_once class QWenMLP(nn.Module): @@ -288,6 +289,15 @@ def load_weights(self, weights: Iterable[Tuple[str, torch.Tensor]]): # Skip loading extra bias for GPTQ models. if name.endswith(".bias") and name not in params_dict: continue + # Skip loading visual weights to support Qwen-VL models + # in cases with text-only inputs + # TODO: add support for Qwen-VL + if (name not in params_dict + and name.startswith("transformer.visual.")): + print_warning_once( + "Only text inputs are allowed. Images won't be handled " + "until Qwen-VL models are fully supported.") + continue param = params_dict[name] weight_loader = getattr(param, "weight_loader", default_weight_loader)