VLLM offline inference raise except when using qianfan-vl #25793
Description
Your current environment
vllm 0.10.2
transformers 4.56.2
🐛 Describe the bug
global llm, processor
#engine_args
llm = LLM(
model=./Qianfan-VL-8B",
trust_remote_code=True,
dtype="float16",
tensor_parallel_size=1,
enforce_eager=True,
gpu_memory_utilization=0.9,
hf_overrides={"architectures":["InternVLChatModel"],"model_type":"internvl_chat"}
)
# 初始化处理器
processor = AutoProcessor.from_pretrained("./Qianfan-VL-8B", trust_remote_code=True)
print("✅ 模型加载完成")
def call_local_llm(imgpath,text):
pil_image = Image.open(imgpath)
messages = [
{
"role": "user",
"content": [
{
"type": "image",
"image": pil_image
},
{
"type": "text",
"text": text
}
]
}
]
prompts = processor.apply_chat_template(messages,tokenize=False)
sampling_params = SamplingParams(
n=1,
best_of=1,
presence_penalty=0.0,
frequency_penalty=0.0,
temperature=0,
#top_p=request.top_p,
#top_k=request.top_k,
max_tokens=512,
# stop=request.stop or []
)
outputs = llm.generate(
{
"prompt": prompts,
"multi_modal_data": {"image": pil_image},
},
sampling_params=sampling_params,
)
print("-" * 50)
for o in outputs:
generated_text = o.outputs[0].text
print(generated_text)
print("-" * 50)
call_local_llm("./test.jpg","Please recognize all text in the image")
错误如下：
ile vllm/multimodal/processing.py", line 1817, in apply
prompt_ids, prompt, mm_placeholders = self._maybe_apply_prompt_updates(
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/vllm/multimodal/processing.py", line 1769, in _maybe_apply_prompt_updates
) = self._apply_prompt_updates(
^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "vllm/multimodal/processing.py", line 1691, in _apply_prompt_updates
assert update_idx is not None, (
^^^^^^^^^^^^^^^^^^^^^^
AssertionError: Failed to apply prompt replacement for mm_items['image'][0]
Before submitting a new issue...
- Make sure you already searched for relevant issues, and asked the chatbot living at the bottom right corner of the documentation page, which can answer lots of frequently asked questions.