Load balancing

Load balancing is a feature that allows you to balance the request load across different models or deployments. You could specify weights for each model/deployment based on their rate limit and your preference.

See all supported params here.

​

Load balancing between models

You could specify the load balancing weights for different models. This is useful when you want to balance the load between different models from different providers.

{
// you don't need to specify the model parameter, otherwise, the model parameter will overwrite the load balance group
    "messages": [
        {
            "role": "user",
            "content": "Hi, how are you?"
        }
    ],
    "load_balance_group": {
        "group_id":"THE_GROUP_ID" // from Load balancing page
    }
}

{
  "load_balance_group": {
      "group_id":"THE_GROUP_ID", // from Load balancing page
      "models": [
        {
          "model": "azure/gpt-35-turbo",
          "weight": 1
        },
        {
          "model": "azure/gpt-4",
          "credentials": { // add your own credentials if you want to use your own Azure credentials or custom model name
              "api_base": "Your own Azure api_base",
              "api_version": "Your own Azure api_version",
              "api_key": "Your own Azure api_key"
          },
          "weight": 1
        } 
      ]
  }
}

You could also set up fallback models to avoid errors. It will fall back to the list of models you specified in the fallback field once have any outages. Check out the Fallbacks section for more information.

​

Load balancing between deployments

​

In the platform

You could go to the platform and add multiple deployments for the same provider. You could specify the load balancing weights for each deployment, which could be helpful when you want to enhance rate limits for a single provider.

​

In the codebase

You could also load balance between deployments in your codebase. You can add different deployments in the customer_credentials field and specify the weight for each deployment.

Example:

{
  "customer_credentials": [
    {
        "credentials": {
            "openai": {
                "api_key": "YOUR_OPENAI_API_KEY",
            }
        },
        "weight": 1.0 // The weight of the deployment
    },
    {
        "credentials": {
            "openai": {
                "api_key": "YOUR_OPENAI_API_KEY", // Another deployment
            }
        },
        "weight": 1.0 // The weight of the deployment
    },
  ],
}

In this example, requests to OpenAI models will be evenly distributed between the two deployments based on their specified weights.

​

Specify available models

You could also specify the available models for load balancing. This is useful when you want to specify the models you want to load balance. For example, if you only want to use gpt-3.5-turbo in an OpenAI deployment, you could specify it in the available_models field or do it in the platform.

Learn more about how to specify available models in the platform here.

Example code:

{
  "customer_credentials": [
    {
        "credentials": {
            "openai": {
                "api_key": "YOUR_OPENAI_API_KEY",
            }
        },
        "weight": 1.0, // The weight of the deployment
        "available_models": ["gpt-3.5-turbo"],
        "exclude_models": ["gpt-4"] // Exclude gpt-4 from this deployment
    },
    {
        "credentials": {
            "openai": {
                "api_key": "YOUR_OPENAI_API_KEY", // Another deployment
            }
        },
        "weight": 1.0, // The weight of the deployment
    },
  ],
}

In this example, requests to OpenAI models will be distributed between the two deployments, with the first deployment only handling gpt-3.5-turbo requests and explicitly excluding gpt-4, while the second deployment can handle requests for any OpenAI model.

Based on the deployment weights and model configurations:

• GPT-3.5-turbo requests are evenly split (50/50) between both deployments
• GPT-4 requests are routed exclusively to the second deployment since it’s excluded from the first
• All other model requests are distributed evenly between deployments according to their weights

​

Deprecated params

The loadbalance_models parameter is deprecated. You should use the load_balance_group parameter instead.

{
    // ...other parameters...
    "loadbalance_models": [
        {
            "model": "claude-3-5-sonnet-20240620",
            "weight": 34,
            "credentials": { // Your own Anthropic API key, optional for team plan and above
                "api_key": "Your own Anthropic API key"
            }
        },
        {
            "model": "azure/gpt-35-turbo",
            "weight": 34,
            "credentials": { // Your own Azure credentials, optional for team plan and above
                "api_base": "Your own Azure api_base",
                "api_version": "Your own Azure api_version",
                "api_key": "Your own Azure api_key"
            }
        }
    ]
}