xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2025-12-15 06:55:00 +08:00
Code Issues Packages Projects Releases Wiki Activity

[Model] Support Qwen-VL and Qwen-VL-Chat models with text-only inputs (#5710)

Browse Source 
Co-authored-by: Roger Wang <ywang@roblox.com>
This commit is contained in:
Jie Fu (傅杰) 2024-06-22 10:07:08 +08:00 committed by GitHub
parent cf90ae0123
commit 9c62db07ed
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 10 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

10
vllm/model_executor/models/qwen.py
View File

@ -28,6 +28,7 @@ from vllm.model_executor.layers.vocab_parallel_embedding import (
from vllm.model_executor.model_loader.weight_utils import default_weight_loader from vllm.model_executor.model_loader.weight_utils import default_weight_loader
from vllm.model_executor.sampling_metadata import SamplingMetadata from vllm.model_executor.sampling_metadata import SamplingMetadata
from vllm.sequence import SamplerOutput from vllm.sequence import SamplerOutput
from vllm.utils import print_warning_once
class QWenMLP(nn.Module): class QWenMLP(nn.Module):
@ -288,6 +289,15 @@ class QWenLMHeadModel(nn.Module):
# Skip loading extra bias for GPTQ models. # Skip loading extra bias for GPTQ models.
if name.endswith(".bias") and name not in params_dict: if name.endswith(".bias") and name not in params_dict:
continue 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] param = params_dict[name]
weight_loader = getattr(param, "weight_loader", weight_loader = getattr(param, "weight_loader",
default_weight_loader) default_weight_loader)