Error while using DeepSeek-V3

[breakices]

Dear Author:
Thank you for your work!

When I execute the ‘prompt_generate’ task using DeepSeek-V3 as the target model and attack model, the code encounters an error after outputting ‘Deep Backtracking finish’.

JSONDecodeError: Expecting value: line 10 column 1 (char 9)

I followed the provided environment configuration, but changed DeepSeek-V2.5 in the python documentation to DeepSeek-V3. Since they use the same model_name (deepseek-chat), I thought it should be fine to make this change.
Specifically, I ran the following command in the terminal
python professional_iterative_generation.py --attack-model DeepSeek-V3 --target-model DeepSeek-V3 --function-descript ‘You are an all- encompassing expert in a wide range of fields, including but not limited to science, technology, mathematics, history, literature, philosophy, art, medicine, engineering, and more. You are an all- encompassing expert in a wide range of fields, including but not limited to science, technology, mathematics, history, literature, philosophy, art, medicine, engineering, and more. Your task is to provide users with highly accurate, detailed, and professional answers to their questions. responses to the user's level of understanding, and provide relevant context or background information to ensure clarity and comprehension.’ --target --target -max-n-tokens 8196
I get the following error

Thank you for your patience in reading and responding!
All the best!

尊敬的作者：
感谢您的工作！
在我使用DeepSeek-V3作为目标模型和攻击模型执行“prompt_generate” 任务时，代码在输出“Deep Backtracking finish”后遇到了一个报错：json.decoder.JSONDecodeError: Expecting value: line 10 column 1 (char 9)
我遵循了提供的环境配置，但将文件中的DeepSeek-V2.5改成了DeepSeek-V3。因为他们使用了同样的model_name(deepseek-chat)，所以我觉着这样修改应该没什么问题。
具体来说，我在终端里执行了如上所述的命令，得到了如图的报错
感谢您的耐心阅读和回复
祝一切顺利

[shuita2333]

Seeing the problem you mentioned, we suggest analyzing it from the error information. This problem should occur in the model calling stage. Since we don't know which model calling method you are using, there may be different errors. Here are several possible solutions for your reference:

1. There is a problem with the model api call. It may be because you called an api that we have not tried before, which requires an additional parameter passing format, resulting in an api call error. In addition, it may even be because you did not set your own api_key.
2. If you use the local deployment method, it may be because you followed our configuration and used an older environment version, causing incompatibility with DeepSeek-V3. You can consider updating the environment appropriately.

We suggest that you try to output the error information or record detailed logs to view the specific response content of the model, so as to better analyze the cause of the error.

We hope that our suggestions can help you, and we are glad that you are interested in us. If you encounter other problems, please give us feedback.

看到您提到的这个问题，从报错信息分析，这一问题应该发生在模型调用阶段，由于不知道您使用的是那种模型调用方式，所以可能存在不同错误。这里有几个可能的解决方案，您可以进行参考：

1. 模型api调用存在问题，可能由于您调用了我们之前没有尝试过的api，需要额外的参数传递格式，导致api调用报错。此外甚至还有可能由于您没有设置自己的api_key。
2. 如果您使用的本地部署的方式，那么可能由于您跟随我们的配置，使用了比较旧的环境版本，造成了DeepSeek-V3的不兼容。您可以考虑对环境进行适当更新。

我们建议您可以尝试输出报错时的信息或记录详细日志，查看模型具体回复内容，从而更好的分析报错原因。

希望我们的建议可以帮到您，很高兴您对我们的感兴趣。如果还遇到其他问题，欢迎向我们反馈。

[breakices]

这是来自QQ邮箱的假期自动回复邮件。 您好，我最近正在休假中，无法亲自回复您的邮件。我将在假期结束后，尽快给您回复。