多图推理报错 You set image=1 (or defaulted to 1) in --limit-mm-per-prompt, but found 2 items in the same prompt.

[Li-Jicheng]

我尝试用swift做多图推理，并整理成了所需的jsonl格式
{"query": "根据图片，***", "images": [ Path1, Path2]}

推理指令如下
CUDA_VISIBLE_DEVICES=0 swift infer
--model_type internvl2_5
--model ${MODEL_PATH}
--val_dataset ${VAL_DATA_PATH}
--infer_backend vllm
--temperature 0
--streaming true

但是遇到了如下问题

File "/lib/python3.10/site-packages/vllm/multimodal/registry.py", line 121, in map_input
raise ValueError(
ValueError: You set image=1 (or defaulted to 1) in --limit-mm-per-prompt, but found 2 items in the same prompt.

[Li-Jicheng]

版本信息
Name: ms-swift
Version: 3.0.2.post1

Name: vllm
Version: 0.6.1.post2+cu118

[Jintao-Huang]

--limit_mm_per_prompt参数设置一下

--limit_mm_per_prompt '{"image": 5}'

[Li-Jicheng]

谢谢，可以再问一下，infer如何设置verbose=true吗?

CUDA_VISIBLE_DEVICES=0 swift infer
--model_type internvl2_5
--model ${MODEL_PATH}
--val_dataset ${VAL_DATA_PATH}
--infer_backend vllm
--limit_mm_per_prompt '{"image": 2}'
--temperature 0
--verbose true
--streaming true

报错
3.10/site-packages/swift/llm/base.py", line 32, in _parse_args
raise ValueError(f'remaining_argv: {remaining_argv}')
ValueError: remaining_argv: ['--verbose', 'true']