LLM-based evaluators not always returning a valid JSON

[davidsbatista]

Is your feature request related to a problem? Please describe.

Having an issue with LLM-based evaluators, they do not always return a valid JSON and break the whole evaluation process.

Describe the solution you'd like
One possible (only for ChatGP) is to explicitly ask for JSON-only responses:

• https://platform.openai.com/docs/guides/text-generation/json-mode
• https://community.openai.com/t/how-do-i-use-the-new-json-mode/475890

passing the following parameter to the call response_format={ "type": "json_object" }

Probably a simple change in the LLMEvaluator starting here:

haystack/haystack/components/evaluators/llm_evaluator.py

Line 86 in f14bc53

self.generator = OpenAIGenerator(api_key=api_key)