Support rate limiting via exponential backoff

[hempels]

All Open AI APIs potentially impose rate limiting. Ideally, any library designed to abstract the APIs should support exponential backoff.

https://beta.openai.com/docs/guides/production-best-practices/managing-rate-limits-and-latency

[gotmike]

@hempels -- do you have some code that would perform this task?

[yungd1plomat]

There are different limits for different users
https://platform.openai.com/docs/guides/rate-limits/overview

[pacarrier]

I believe at the very least, the ApiResultBase should include usage as shown in https://platform.openai.com/docs/api-reference/completions/create. Knowing the number of token used can help set our own limits and prevent request that will be rejected.

[pacarrier]

Thank you for adding usage information! This helps a lot.

[Baklap4]

Polly is a great library to help with exponential backoff and it plays nicely with the dotnet HTTPClient: https://github.com/App-vNext/Polly it has also functionality for Retrying, circuit braker and ratelimiting: https://github.com/App-vNext/Polly#resilience-policies

[StefH]

I get this exception when calling a embeddings.

Error at embeddings (https://api.openai.com/v1/embeddings) with HTTP status code: TooManyRequests. Content: {
    "error": {
        "message": "Rate limit reached for default-global-with-image-limits in organization org-*** on requests per min. Limit: 60 / min. Please try again in 1s. Contact support@openai.com if you continue to have issues. Please add a payment method to your account to increase your rate limit. Visit https://platform.openai.com/account/billing to add a payment method.",
        "type": "requests",
        "param": null,
        "code": null
    }
}

So implementing Polly would be great.

[StefH]

@hempels
@gotmike
@pacarrier
@Baklap4
@yungd1plomat

I've created a small NuGet package which can be used to handle rate limits.

See https://www.nuget.org/packages/OpenAI.Polly/0.0.1-preview-01

OpenAI.Polly

Can be used to handle exceptions like:

Unhandled exception. System.Net.Http.HttpRequestException: Error at embeddings (https://api.openai.com/v1/embeddings) with HTTP status code: TooManyRequests. Content: {
    "error": {
        "message": "Rate limit reached for default-global-with-image-limits in organization org-*** on requests per min. Limit: 60 / min. Please try again in 1s. Contact support@openai.com if you continue to have issues. Please add a payment method to your account to increase your rate limit. Visit https://platform.openai.com/account/billing to add a payment method.",
        "type": "requests",
        "param": null,
        "code": null
    }
}

Polly

Polly is used to handle the TooManyRequests exceptions.

Usage

IOpenAIAPI openAiAPI = new OpenAIAPI();
float[] embeddings = await openAiAPI.WithRetry(api => api.Embeddings.GetEmbeddingsAsync("What is a cat?"));

Extension Methods

There are 3 extension methods that can be used to handle TooManyRequests Exceptions:

• WithRetry which returns a Task<TResult>
• WithRetry which returns a Task
• WithRetry which returns nothing (void)

[Ruud-cb]

@StefH Does this also cover the ServiceUnavailable/Internal server error? Unfortunately the docs for a 503 say:

Cause: Our servers are experiencing high traffic.
Solution: Please retry your requests after a brief wait.

Just occured to me, nothing was reported on their status page. Didn't keep sending request to see how long I had to wait.

[StefH]

@Ruud-cb
When the error message contains Please retry your request, the internal logic should retry.