GPT-4 content_filter_error

API
, ,
1

I am trying to do function calling from a script, which worked fine yesterday but today returns an empty completion (see below). When I switch the model to 3.5-turbo, everything runs fine again. Why is this happening? The completion also says there are 1192 completion tokens, yet the contents are empty.

ChatCompletion(id=‘chatcmpl-8frbPcK0ETp32JV4’, choices=[Choice(finish_reason=‘stop’, index=0, logprobs=None, message=ChatCompletionMessage(content=None, role=‘assistant’, function_call=None, tool_calls=None), content_filter_result={‘error’: {‘code’: ‘content_filter_error’, ‘message’: ‘The contents are not filtered’}}, content_filter_results={})], created=1704988187, model=‘gpt-4’, object=‘chat.completion’, system_fingerprint=None, usage=CompletionUsage(completion_tokens=1192, prompt_tokens=1392, total_tokens=2584), prompt_filter_results=[{‘prompt_index’: 0, ‘content_filter_results’: {}}])

2

I ran into the same issue, when removing the tool call it worked but the JSON was not properly formatted in a robust way across all the queries.

3

Looks like some other folks had a similar issue.

Which model are you using? Could you try tweaking your prompt slightly? It may be erroneously

mindful that the error message seems to suggest there is no filtering

4

Ah thanks for the link. It is indeed the exact same issue I am having. I believe everyone is experiencing the problem on the 1106 models. Also seems to be part of the Azure OpenAI Service.

5

I’m experimenting the same error since 4 days. Function call works, but tool calls have been removed from Microsoft Azure OpenAI service.