内网环境下怎么判断模型真假?

Linus Torvalds 2026-07-21 16:01 1

公司接入了大模型,但是内网才能访问.


现有模型有 openai 的 Gpt 5.5 但是只能是 /v1/chat/completions 接口. 这对吗?


请问下,怎么鉴定 5.5 真假?


我让他出个随机数,试了 10 几次,大部分是 47, 结果有 73,57


在 codex 中问他 juice 值, 已经是 xhigh 了, juice 还是 二十几


使用 大发展有限公司官网 和 给主人留下些什么吧 在 api 调用下, “stream”: false 和 true 的结果也不一样



最新回复 (7)
  • Linus Torvalds 楼主 07-21 16:06
    1

  • ShiroEirin 07-21 16:08
    2

    思维链回复里面有没有什么彩票 菠菜,赌博之类的

  • 博丽灵梦 07-21 16:20
    3

    这个问题5.3-codex出现的最多,5.4大幅减少,5.5已经不会出现这个问题了

  • Linus Torvalds 楼主 07-21 17:14
    4


    会出现 <thinking>tool call</thinking> 这对吗?


    我用站里的中转 codex GPT模型 重来没见过

  • Linus Torvalds 楼主 07-21 19:00
    5

    windows 上报错:


    网络工具和文件系统工具当前不可用。请检查您的网络连接或工具配置,然后重试。</think>Looks like the shell commands are failing. Let me try a different approach. The user wants me to check if the README is
    consistent with the current codebase. Let me try to read the README file first, then compare with the actual project structure.

    Let me try a simpler command first to see if the shell is working
  • 过期绿坝粮 07-21 19:04
    6

    用Hub鉴鉴)


    (LinuxDo Hub)

  • Linus Torvalds 楼主 07-21 19:17
    7

* 帖子来源Linux.do
返回