license: apache-2.0 language:

• en
• zh pipeline_tag: visual-question-answering tags:
• multimodal library_name: transformers

简介

Xinyuan-VL-2B 是赛灵格集团推出的面向终端侧的高性能多模态大模型，基于 Qwen/Qwen2-VL-2B-Instruct 微调而成，使用了超过500万条多模态数据及少量纯文本数据。该模型在多个权威基准测试中表现优异。

使用方法

为依托开源社区的蓬勃生态，我们选择对 Qwen/Qwen2-VL-2B-Instruct 进行微调，最终形成 Cylingo/Xinyuan-VL-2B。因此，使用方式与 Qwen/Qwen2-VL-2B-Instruct 完全一致：

from transformers import Qwen2VLForConditionalGeneration, AutoTokenizer, AutoProcessor
from qwen_vl_utils import process_vision_info

# 默认加载至可用设备
model = Qwen2VLForConditionalGeneration.from_pretrained(
    "Cylingo/Xinyuan-VL-2B", torch_dtype="auto", device_map="auto"
)

# 默认处理器
processor = AutoProcessor.from_pretrained("Cylingo/Xinyuan-VL-2B")

messages = [
    {
        "role": "user",
        "content": [
            {
                "type": "image",
                "image": "https://qianwen-res.oss-cn-beijing.aliyuncs.com/Qwen-VL/assets/demo.jpeg",
            },
            {"type": "text", "text": "描述这张图片。"},
        ],
    }
]

# 推理准备
text = processor.apply_chat_template(
    messages, tokenize=False, add_generation_prompt=True
)
image_inputs, video_inputs = process_vision_info(messages)
inputs = processor(
    text=[text],
    images=image_inputs,
    videos=video_inputs,
    padding=True,
    return_tensors="pt",
)
inputs = inputs.to("cuda")

# 推理生成
generated_ids = model.generate(**inputs, max_new_tokens=128)
generated_ids_trimmed = [
    out_ids[len(in_ids) :] for in_ids, out_ids in zip(inputs.input_ids, generated_ids)
]
output_text = processor.batch_decode(
    generated_ids_trimmed, skip_special_tokens=True, clean_up_tokenization_spaces=False
)
print(output_text)

性能评估

我们使用 VLMEvalKit 工具包对 XinYuan-VL-2B 进行了多维度基准测试。结果显示，XinYuan-VL-2B 在多项指标上超越阿里云开源的 Qwen/Qwen2-VL-2B-Instruct 及其他具有影响力的同规模开源模型。

完整结果可查阅 opencompass/open_vlm_leaderboard：