API error message mixes up object and string

API Bugs
1

I’m using the Python SDK. I’m first doing an audio translation, whose result (Translation) I was mistakenly passing to the method completions.create instead of the string Translation.text. This resulted in the following error:

openai.BadRequestError: Error code: 400 - {‘error’: {‘message’: “Invalid type for ‘messages[1].content[0]’: expected an object, but got a string instead.”, ‘type’: ‘invalid_request_error’, ‘param’: ‘messages[1].content[0]’, ‘code’: ‘invalid_type’}}

Completions requires the messages to have the content key, with a value that is either a string literal or an Iterable of ChatCompletionContentPartParam. When I used Translation.text for the content of a message instead of the Translation object, the error was resolved.

This leads me to believe that this part of the error message has swapped the words ‘object’ and ‘string’:

expected an object, but got a string instead.

It could also be that something weirder is going on, but the message is very confusing nonetheless.

1 Like
2

You are looking too deep into the API library. The types you are examining are only for validating the return from the API.

Let’s look at the error:

‘messages[1].content[0]’

In a list of messages, the second message. In a list of contents, the first “content” must be an object.

This is correct. You must either send a single string as content, not a list, or if using a list (array) to send multi-part content (such as images), each shall be a JSON object with the type.

Since you only want to translate text, your messages would look like:

messages = [
{"role": "system", "content": "You are a language expert"},
{"role": "user", "content": f"Improve quality:\n{my_text}"}
]

You need to get a string out of the return object from the first API call.

3

In your example, you’re sending strings under content, but the error message says

expected an object

Please read my post again. I think you’re missing what the problem is.

1 Like
4

It seems the second message gets this form:

{‘role’: ‘user’, ‘content’: [‘text’]}

This makes more sense. I don’t know if the error message can be made more descriptive. Inspecting the actual request by setting the logging level to DEBUG revealed the culprit.

5

The problem is likely that you are not fully extracting a string from the prior return.

Here’s an example of using the transcriptions endpoint (translation is similar, it just always produces English):

# Open the audio file
with open(input_file_path, "rb") as audio_file:
    # Create a transcription using OpenAI API
    try:
        transcription = client.audio.transcriptions.create(
            file=audio_file, # a flac, mp3, mp4, mpeg, mpga, m4a, ogg, wav, webm
            language="en",   # ISO code
            model="whisper-1",
            prompt="Welcome to our radio show.",  # lead-up to input audio
            response_format="json",  # also text, srt, verbose_json, or vtt
            temperature=0.2)
    except Exception as e:
        print(f"An API error occurred: {e}")

# get just the transcribed text out of the response
transcribed_text = transcription.text
6

Yep. I said this in the initial post:

When I used Translation.text for the content of a message instead of the Translation object, the error was resolved.

I was confused by the error message, that’s all.