Kimi VL A3B Thinking 2506

🚀 Kimi-VL-A3B-Thinking-2506

Kimi-VL-A3B-Thinking-2506 是 Kimi-VL-A3B-Thinking 的升级版，在多模态推理、视觉感知与理解、视频场景处理等方面有显著提升，同时支持更高分辨率图像，且能在消耗更少令牌的情况下实现更智能的思考。

基础信息

⚠️ 重要提示

这是 Kimi-VL-A3B-Thinking 的改进版本，请考虑使用此更新后的模型，而非之前的版本。

💡 使用建议

请访问我们的技术博客，获取此模型的推荐推理方案：Kimi-VL-A3B-Thinking-2506: A Quick Navigation

✨ 主要特性

这是 Kimi-VL-A3B-Thinking 的更新版本，具有以下改进能力：

• 更智能思考，更少令牌消耗：2506 版本在多模态推理基准测试中达到了更好的准确率，如 MathVision 达到 56.9（提升 20.1）、MathVista 达到 80.1（提升 8.4）、MMMU-Pro 达到 46.3（提升 3.3）、MMMU 达到 64.0（提升 2.1），同时平均所需思考长度减少 20%。
• 思考助力，视觉更清晰：与专注于思考任务的前一版本不同，2506 版本在一般视觉感知和理解方面也能达到相同甚至更好的能力，如 MMBench-EN-v1.1（84.4）、MMStar（70.4）、RealWorldQA（70.0）、MMVet（78.4），超越或匹配了我们非思考模型（Kimi-VL-A3B-Instruct）的能力。
• 拓展至视频场景：新的 2506 版本在视频推理和理解基准测试中也有所改进。它在 VideoMMMU 上为开源模型设定了新的技术水平（65.2），同时在一般视频理解方面也保留了良好的能力（Video-MME 达到 71.9，与 Kimi-VL-A3B-Instruct 相当）。
• 支持更高分辨率：新的 2506 版本支持单张图像 320 万总像素，是前一版本的 4 倍。这在高分辨率感知和 OS 代理接地基准测试中带来了显著改进，如 V* Benchmark（无额外工具）达到 83.2、ScreenSpot-Pro 达到 52.8、OSWorld-G（完整集，含拒绝）达到 52.5。

📈 性能表现

与高效模型及 Kimi-VL 前两个版本的比较

与 30B - 70B 开源模型的比较

💻 使用示例

基础用法

使用 VLLM 进行推理（推荐）

作为一个长解码模型，最多可生成 32K 令牌，我们推荐使用 VLLM 进行推理，它已经支持 Kimi-VL 系列。

MAX_JOBS=4 pip install vllm==0.9.1 blobfile flash-attn --no-build-isolation

⚠️ 重要提示

务必显式安装 flash-attn 以避免 CUDA 内存不足。

from transformers import AutoProcessor
from vllm import LLM, SamplingParams

model_path = "moonshotai/Kimi-VL-A3B-Thinking-2506"
llm = LLM(
    model_path,
    trust_remote_code=True,
    max_num_seqs=8,
    max_model_len=131072,
    limit_mm_per_prompt={"image": 256}
)

processor = AutoProcessor.from_pretrained(model_path, trust_remote_code=True)

sampling_params = SamplingParams(max_tokens=32768, temperature=0.8)


import requests
from PIL import Image

def extract_thinking_and_summary(text: str, bot: str = "◁think▷", eot: str = "◁/think▷") -> str:
    if bot in text and eot not in text:
        return ""
    if eot in text:
        return text[text.index(bot) + len(bot):text.index(eot)].strip(), text[text.index(eot) + len(eot) :].strip()
    return "", text

OUTPUT_FORMAT = "--------Thinking--------\n{thinking}\n\n--------Summary--------\n{summary}"

url = "https://huggingface.co/spaces/moonshotai/Kimi-VL-A3B-Thinking/resolve/main/images/demo6.jpeg"
image = Image.open(requests.get(url,stream=True).raw)

messages = [
    {"role": "user", "content": [{"type": "image", "image": ""}, {"type": "text", "text": "What kind of cat is this? Answer with one word."}]}
]
text = processor.apply_chat_template(messages, add_generation_prompt=True, return_tensors="pt")

outputs = llm.generate([{"prompt": text, "multi_modal_data": {"image": image}}], sampling_params=sampling_params)
generated_text = outputs[0].outputs[0].text

thinking, summary = extract_thinking_and_summary(generated_text)
print(OUTPUT_FORMAT.format(thinking=thinking, summary=summary))

使用 🤗 Hugging Face Transformers 进行推理

我们介绍如何使用 transformers 库在推理阶段使用我们的模型。建议使用 python=3.10、torch>=2.1.0 和 transformers=4.48.2 作为开发环境。

from PIL import Image
from transformers import AutoModelForCausalLM, AutoProcessor

def extract_thinking_and_summary(text: str, bot: str = "◁think▷", eot: str = "◁/think▷") -> str:
    if bot in text and eot not in text:
        return ""
    if eot in text:
        return text[text.index(bot) + len(bot):text.index(eot)].strip(), text[text.index(eot) + len(eot) :].strip()
    return "", text

OUTPUT_FORMAT = "--------Thinking--------\n{thinking}\n\n--------Summary--------\n{summary}"

url = "https://huggingface.co/spaces/moonshotai/Kimi-VL-A3B-Thinking/resolve/main/images/demo6.jpeg"

model_path = "moonshotai/Kimi-VL-A3B-Thinking-2506"
model = AutoModelForCausalLM.from_pretrained(
    model_path,
    torch_dtype="auto",
    device_map="auto",
    trust_remote_code=True,
)
processor = AutoProcessor.from_pretrained(model_path, trust_remote_code=True)

image_paths = ["url"]
images = [Image.open(path) for path in image_paths]
messages = [
    {
        "role": "user",
        "content": [
            {"type": "image", "image": image_path} for image_path in image_paths
        ] + [{"type": "text", "text": "What kind of cat is this? Answer with one word."}],
    },
]
text = processor.apply_chat_template(messages, add_generation_prompt=True, return_tensors="pt")
inputs = processor(images=images, text=text, return_tensors="pt", padding=True, truncation=True).to(model.device)
generated_ids = model.generate(**inputs, max_new_tokens=32768, temperature=0.8)
generated_ids_trimmed = [
    out_ids[len(in_ids) :] for in_ids, out_ids in zip(inputs.input_ids, generated_ids)
]
response = processor.batch_decode(
    generated_ids_trimmed, skip_special_tokens=True, clean_up_tokenization_spaces=False
)[0]
print(response)

📚 引用信息

@misc{kimiteam2025kimivltechnicalreport,
      title={{Kimi-VL} Technical Report}, 
      author={Kimi Team and Angang Du and Bohong Yin and Bowei Xing and Bowen Qu and Bowen Wang and Cheng Chen and Chenlin Zhang and Chenzhuang Du and Chu Wei and Congcong Wang and Dehao Zhang and Dikang Du and Dongliang Wang and Enming Yuan and Enzhe Lu and Fang Li and Flood Sung and Guangda Wei and Guokun Lai and Han Zhu and Hao Ding and Hao Hu and Hao Yang and Hao Zhang and Haoning Wu and Haotian Yao and Haoyu Lu and Heng Wang and Hongcheng Gao and Huabin Zheng and Jiaming Li and Jianlin Su and Jianzhou Wang and Jiaqi Deng and Jiezhong Qiu and Jin Xie and Jinhong Wang and Jingyuan Liu and Junjie Yan and Kun Ouyang and Liang Chen and Lin Sui and Longhui Yu and Mengfan Dong and Mengnan Dong and Nuo Xu and Pengyu Cheng and Qizheng Gu and Runjie Zhou and Shaowei Liu and Sihan Cao and Tao Yu and Tianhui Song and Tongtong Bai and Wei Song and Weiran He and Weixiao Huang and Weixin Xu and Xiaokun Yuan and Xingcheng Yao and Xingzhe Wu and Xinxing Zu and Xinyu Zhou and Xinyuan Wang and Y. Charles and Yan Zhong and Yang Li and Yangyang Hu and Yanru Chen and Yejie Wang and Yibo Liu and Yibo Miao and Yidao Qin and Yimin Chen and Yiping Bao and Yiqin Wang and Yongsheng Kang and Yuanxin Liu and Yulun Du and Yuxin Wu and Yuzhi Wang and Yuzi Yan and Zaida Zhou and Zhaowei Li and Zhejun Jiang and Zheng Zhang and Zhilin Yang and Zhiqi Huang and Zihao Huang and Zijia Zhao and Ziwei Chen},
      year={2025},
      eprint={2504.07491},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2504.07491}, 
}

📄 许可证

本项目采用 MIT 许可证。