在随机模式下，如何通过URL连接指定所使用的AI服务？

[aaro-n]

遇到的问题

在simple-one-api代理百度千帆、讯飞星火、Google gemini和CF 的AI，每次访问都会随机到上述AI的一种，有没有方法能通过URL指定所使用的后台AI ？我查找过文档，指定过models，但没用。

config.json

{
  "log_level": "prod",
  "server_port": ":9090",
  "api_key": "sk-dfKrzhtPruv4OQDG********c4dDdAbF0",
  "load_balancing": "random",
  "proxy": {},
  "services": {
    "qianfan": [
      {
        "models": ["ERNIE-Lite-8K"],
        "enabled": true,
        "model_map": {
          "baidu": "ERNIE-Lite-8K"
        },
        "model_redirect": {
          "gpt-3.5-turbo": "ERNIE-Lite-8K"
        },
        "limit": {
          "qps": 300
        },
        "credentials": {
          "api_key": "lku54*****NseB",
          "secret_key": "IG*****mXzfo"
        }
      }
    ],
    "xinghuo": [
      {
        "models": ["spark-lite"],
        "enabled": true,
        "model_map": {
          "xinghuo": "spark-lite"
        },
        "model_redirect": {
          "gpt-3.5-turbo": "spark-lite"
        },
        "limit": {},
        "credentials": {
          "appid": "866*****",
          "api_key": "a6******c8a5",
          "api_secret": "N******M0"
        }
      }
    ],
    "openai": [
      {
        "models": ["@cf/meta/llama-3-8b-instruct"],
        "enabled": true,
        "model_map": {
          "cf": "@cf/meta/llama-3-8b-instruct"
        },
        "model_redirect": {
          "gpt-3.5-turbo": "@cf/meta/llama-3-8b-instruct"
        },
        "limit": {},
        "credentials": {
          "api_key": "gEbUL******ZKxL"
        },
        "server_url": "https://api.cloudflare.com/client/v4/accounts/7d********138/ai/v1"
      }
    ],
    "gemini": [
      {
        "models": ["gemini-1.5-flash"],
        "enabled": true,
        "model_map": {
          "google": "gemini-1.5-flash"
        },
        "model_redirect": {
          "gpt-3.5-turbo": "gemini-1.5-flash"
        },
        "limit": {
          "qps": 15
        },
        "credentials": {
          "api_key": "AI******Os"
        }
      }
    ]
  }
}

测试的连接

http://104.3*6.4*7.2*4:19090/v1/google
http://104.3*6.4*7.2*4:19090/v1/baidu
http://104.3*6.4*7.2*4:19090/v1/gemini-1.5-flash

以上连接都无法指定后端所使用的Ai模型
我该如何做？

[lopins]

？

启用哪个模型是由你的客户端决定的

[fruitbars]

模型是由消息中的model来决定的。可以参考下openai的文档https://platform.openai.com/docs/guides/text-generation

curl https://api.openai.com/v1/chat/completions
-H "Content-Type: application/json"
-H "Authorization: Bearer $OPENAI_API_KEY"
-d '{
        "model": "gpt-4o",
        "messages": [
            {"role": "user", "content": "write a haiku about ai"}
        ]
    }'

[aaro-n]

@fruitbars @lopins 我希望simple-one-api可以像one-api那样指定simple-one-api所运行的后端AI，例如one-api密钥为sk-dfKrzhtPruv4OQDG********c4dDdAbF0，则sk-dfKrzhtPruv4OQDG********c4dDdAbF0-1为Google ai，sk-dfKrzhtPruv4OQDG********c4dDdAbF0-2为百度AI，我以为simple-one-api也可以使用类似的功能，只是要在URL中指定。