Evaluation prompt provided to Judge model to choose winning generation (AI Feedback)

evaluation_prompt.py

prompt = f"""
Compare the quality of the following two instruction-response pairs:


Version 1 ({version1_name}):
Instruction: {version1['instruction']}
Context: {version1['context']}
Response: {version1['response']}


Version 2 ({version2_name}):
Instruction: {version2['instruction']}
Context: {version2['context']}
Response: {version2['response']}


Which version is better in terms of overall quality, coherence, and relevance? Provide a 1-2 sentence, concise explanation of your judgment, highlighting specific strengths and weaknesses of each version.


Your response must be in a valid JSON string with no additional comments, backticks, or spacing:
{{"better_version": "Winning Version Name here- either '{version1_name}' or '{version2_name}'", "explanation": "Your concise explanation here"}}
"""