400 Error thrown due to max token length exceeded returns the request body when streaming

[nsadeh]

Steps to reproduce:

1. Collect a sequence of messages to pass to open AI (of type {:role role :content content}) that exceed the model's token limit. In my case, GPT-3.5 turbo with 4097 tokens
2. Hit the synchronous chat completions endpoint. I use OpenAI without Azure. You get an exception HTTP response with status 400 and the following body

:body
    "{\n  \"error\": {\n    \"message\": \"This model's maximum context length is 4097 tokens. However, your messages resulted in <more>. Please reduce the length of the messages"

3. Repeat the same experiment, this time setting the :stream parameter to true. My setup doesn't have a handler in order to receive a channel back
4. You should see a similar HTTP 400 exception thrown, but in the body you will see your request body rather than the helpful error message.

Attached below is the function I use to invoke the OpenAI endpoint:

(defn ask
 ([questions]
  (ask questions {} {}))
 ([questions payload-params]
  (ask questions payload-params {}))
 ([questions payload-params  opts]
  (let [payload (merge payload-params {:model "gpt-3.5-turbo"
                                       :messages questions
                                       :temperature 0})
        options (assoc opts :api-key openai-token)]
    (api/create-chat-completion payload options))))

In this case it was invoked as (ask messages {:stream true}

Thank you for making this library, it was tremendously helpful in avoiding a lot of really boring work.

[wkok]

Hello!

Thanks, I was able to duplicate, but unfortunately I don't have a solution yet.