How do I extract only code content from chat gpt response?

I use chatGpt to generate SQL query using openai api(/v1/chat/completions) and gpt-3.5-turbo as the model.

But I am facing difficulty in extracting SQL query from the response. Because sometime chatGpt will provide some explanation for query sometimes not. I have tried with regex expressions, but it is not reliable.

regex = r"SELECT .*?;"
match = re.search(regex, result)
if match:
   sql_query = match.group()
   print(sql_query)

Is there any other approach to extract only the code section from the response?

I added "don't include any explanations in your responses" and got only the code as output.

I added "don't include the question in your response" and got only the code as output.

Try to write a good prompt; if it still doesn't give you the output in the required format, try to provide examples to try few-shot prompting.

See the below request body,

{
    "model": "gpt-3.5-turbo",
    "messages": [
        {
            "role": "system",
            "content": "You are a SQL query generation assistant. You will be given a description of the table and a text description on what needs to be extracted or updated. Respond with only the query without explanation"
        },
        {
            "role": "user",
            "content": "There is a MySQL database table 'person' with four columns 'id', 'first_name', 'last_name' and 'address'. Write a query to return all entries of the table"
        },
        {
            "role": "assistant",
            "content": "SELECT * FROM person;"
        },
        {
            "role": "user",
            "content": "Write a query to update the address of the user with the id 103"
        }
    ],
    "temperature": 0.8
}

Here, I am giving a simple example response from the model as the context to my actual prompt. If the required output format is complex, you might need to give a few examples.

Below is the response I received for the above request.

{
    "id": "chatcmpl-7698KLZnVNmPUHMXx7R4",
    "object": "chat.completion",
    "created": 1687816228,
    "model": "gpt-3.5-turbo-0301",
    "choices": [
        {
            "index": 0,
            "message": {
                "role": "assistant",
                "content": "UPDATE person SET address = 'New Address' WHERE id = 103;"
            },
            "finish_reason": "stop"
        }
    ],
    "usage": {
        "prompt_tokens": 117,
        "completion_tokens": 15,
        "total_tokens": 132
    }
}

I struggle with this endlessly. I found this works reasonably reliably..
-- Return only the code in the completion. I don't want any other comments. Don't say "here is your code" or similar remarks.