youngkingdom/vllm
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

[Model] Support Qwen3-VL Model Series (#24727)

Browse Source 
Signed-off-by: Roger Wang <hey@rogerw.io>
Signed-off-by: Isotr0py <mozf@mail2.sysu.edu.cn>
Co-authored-by: Huang Jie <92386084+JJJYmmm@users.noreply.github.com>
Co-authored-by: 松灵 <26085463+wulipc@users.noreply.github.com>
Co-authored-by: Isotr0py <mozf@mail2.sysu.edu.cn>
This commit is contained in:
Roger Wang
2025-09-16 22:01:04 -07:00
committed by GitHub
parent 5801e49776
commit 0f7acdd73c
13 changed files with 2084 additions and 17 deletions
Download Patch File Download Diff File Expand all files Collapse all files

78
examples/offline_inference/vision_language.py
View File

@ -1437,6 +1437,80 @@ def run_qwen2_5_omni(questions: list[str], modality: str):
)
# Qwen3-VL-Dense
def run_qwen3_vl(questions: list[str], modality: str) -> ModelRequestData:
model_name = "Qwen/Qwen3-VL-4B-Instruct"
engine_args = EngineArgs(
model=model_name,
max_model_len=4096,
max_num_seqs=5,
mm_processor_kwargs={
"min_pixels": 28 * 28,
"max_pixels": 1280 * 28 * 28,
"fps": 1,
},
limit_mm_per_prompt={modality: 1},
)
if modality == "image":
placeholder = "<|image_pad|>"
elif modality == "video":
placeholder = "<|video_pad|>"
prompts = [
(
"<|im_start|>system\nYou are a helpful assistant.<|im_end|>\n"
f"<|im_start|>user\n<|vision_start|>{placeholder}<|vision_end|>"
f"{question}<|im_end|>\n"
"<|im_start|>assistant\n"
)
for question in questions
]
return ModelRequestData(
engine_args=engine_args,
prompts=prompts,
)
# Qwen3-VL-MOE
def run_qwen3_vl_moe(questions: list[str], modality: str) -> ModelRequestData:
model_name = "Qwen/Qwen3-VL-30B-A3B-Instruct"
engine_args = EngineArgs(
model=model_name,
max_model_len=4096,
max_num_seqs=5,
mm_processor_kwargs={
"min_pixels": 28 * 28,
"max_pixels": 1280 * 28 * 28,
"fps": 1,
},
limit_mm_per_prompt={modality: 1},
)
if modality == "image":
placeholder = "<|image_pad|>"
elif modality == "video":
placeholder = "<|video_pad|>"
prompts = [
(
"<|im_start|>system\nYou are a helpful assistant.<|im_end|>\n"
f"<|im_start|>user\n<|vision_start|>{placeholder}<|vision_end|>"
f"{question}<|im_end|>\n"
"<|im_start|>assistant\n"
)
for question in questions
]
return ModelRequestData(
engine_args=engine_args,
prompts=prompts,
)
# R-4B
def run_r_vl(questions: list[str], modality: str) -> ModelRequestData:
assert modality == "image"
@ -1645,6 +1719,8 @@ model_example_map = {
"qwen2_vl": run_qwen2_vl,
"qwen2_5_vl": run_qwen2_5_vl,
"qwen2_5_omni": run_qwen2_5_omni,
"qwen3_vl": run_qwen3_vl,
"qwen3_vl_moe": run_qwen3_vl_moe,
"rvl": run_r_vl,
"skywork_chat": run_skyworkr1v,
"smolvlm": run_smolvlm,
@ -1658,6 +1734,8 @@ MODELS_NEED_VIDEO_METADATA = [
"glm4_1v",
"glm4_5v",
"glm4_5v_fp8",
"qwen3_vl",
"qwen3_vl_moe",
]