Muokkaa

Jaa

Limit Azure OpenAI API token usage

  • Artikkeli

APPLIES TO: Developer | Basic | Basic v2 | Standard | Standard v2 | Premium | Premium v2

The azure-openai-token-limit policy prevents Azure OpenAI Service API usage spikes on a per key basis by limiting consumption of language model tokens to a specified number per minute. When the token usage is exceeded, the caller receives a 429 Too Many Requests response status code.

By relying on token usage metrics returned from the OpenAI endpoint, the policy can accurately monitor and enforce limits in real time. The policy also enables precalculation of prompt tokens by API Management, minimizing unnecessary requests to the OpenAI backend if the limit is already exceeded.

Note

Set the policy's elements and child elements in the order provided in the policy statement. Learn more about how to set or edit API Management policies.

Supported Azure OpenAI Service models

The policy is used with APIs added to API Management from the Azure OpenAI Service of the following types:

API type Supported models
Chat completion gpt-3.5

gpt-4
Completion gpt-3.5-turbo-instruct
Embeddings text-embedding-3-large

text-embedding-3-small

text-embedding-ada-002

For more information, see Azure OpenAI Service models.

Policy statement

<azure-openai-token-limit counter-key="key value"
        tokens-per-minute="number"
        estimate-prompt-tokens="true | false"    
        retry-after-header-name="custom header name, replaces default 'Retry-After'" 
        retry-after-variable-name="policy expression variable name"
        remaining-tokens-header-name="header name"  
        remaining-tokens-variable-name="policy expression variable name"
        tokens-consumed-header-name="header name"
        tokens-consumed-variable-name="policy expression variable name" />

Attributes

Attribute Description Required Default
counter-key The key to use for the token limit policy. For each key value, a single counter is used for all scopes at which the policy is configured. Policy expressions are allowed. Yes N/A
tokens-per-minute The maximum number of tokens consumed by prompt and completion per minute. Yes N/A
estimate-prompt-tokens Boolean value that determines whether to estimate the number of tokens required for a prompt:
- true: estimate the number of tokens based on prompt schema in API; may reduce performance.
- false: don't estimate prompt tokens.

When set to false, the remaining tokens per counter-key are calculated using the actual token usage from the response of the model. This could result in prompts being sent to the model that exceed the token limit. In such case, this will be detected in the response, and all succeeding requests will be blocked by the policy until the token limit frees up again.		 Yes N/A
retry-after-header-name The name of a custom response header whose value is the recommended retry interval in seconds after the specified tokens-per-minute is exceeded. Policy expressions aren't allowed. No Retry-After
retry-after-variable-name The name of a variable that stores the recommended retry interval in seconds after the specified tokens-per-minute is exceeded. Policy expressions aren't allowed. No N/A
remaining-tokens-header-name The name of a response header whose value after each policy execution is the number of remaining tokens allowed for the time interval. Policy expressions aren't allowed. No N/A
remaining-tokens-variable-name The name of a variable that after each policy execution stores the number of remaining tokens allowed for the time interval. Policy expressions aren't allowed. No N/A
tokens-consumed-header-name The name of a response header whose value is the number of tokens consumed by both prompt and completion. The header is added to response only after the response is received from backend. Policy expressions aren't allowed. No N/A
tokens-consumed-variable-name The name of a variable initialized to the estimated number of tokens in the prompt in backend section of pipeline if estimate-prompt-tokens is true and zero otherwise. The variable is updated with the reported count upon receiving the response in outbound section. No N/A

Usage

Usage notes

  • This policy can be used multiple times per policy definition.
  • This policy can optionally be configured when adding an API from the Azure OpenAI Service using the portal.
  • Where available when estimate-prompt-tokens is set to false, values in the usage section of the response from the Azure OpenAI Service API are used to determine token usage.
  • Certain Azure OpenAI endpoints support streaming of responses. When stream is set to true in the API request to enable streaming, prompt tokens are always estimated, regardless of the value of the estimate-prompt-tokens attribute. Completion tokens are also estimated when responses are streamed.
  • API Management uses a single counter for each counter-key value that you specify in the policy. The counter is updated at all scopes at which the policy is configured with that key value. If you want to configure separate counters at different scopes (for example, a specific API or product), specify different key values at the different scopes. For example, append a string that identifies the scope to the value of an expression.

Example

In the following example, the token limit of 5000 per minute is keyed by the caller IP address. The policy doesn't estimate the number of tokens required for a prompt. After each policy execution, the remaining tokens allowed for that caller IP address in the time period are stored in the variable remainingTokens.

<policies>
    <inbound>
        <base />
        <azure-openai-token-limit
            counter-key="@(context.Request.IpAddress)"
            tokens-per-minute="5000" estimate-prompt-tokens="false" remaining-tokens-variable-name="remainingTokens" />
    </inbound>
    <outbound>
        <base />
    </outbound>
</policies>

Related content

For more information about working with policies, see: