First, thank you for your insightful work in CLoT. I'm currently reproducing your experiments and have a question regarding the prompt design for Qwen-14B's answer rewriting process.
In 4.1. Associable Instruction Tuning and 5.1. Evaluation Questions and Metrics, you mention using Qwen-14B to rewrite answers to form the choice question. I tried several prompts (e.g., "rewrite this in a relatively bland tone") to reduce humor or stylistic bias in the responses, but the rewritten answers still retained noticeable humor or stylistic variations. Could you please clarify:
- The exact prompt template used for answer rewriting?
- Any special parameters (temperature, top-p, etc.) or formatting instructions?
- Whether any few-shot examples were provided in the prompt?
This information would be incredibly helpful for ensuring faithful reproduction of your results. If possible, could you share the complete prompt or any implementation details?
Thank you for your time and consideration. Looking forward to your response.
First, thank you for your insightful work in CLoT. I'm currently reproducing your experiments and have a question regarding the prompt design for Qwen-14B's answer rewriting process.
In 4.1. Associable Instruction Tuning and 5.1. Evaluation Questions and Metrics, you mention using Qwen-14B to rewrite answers to form the choice question. I tried several prompts (e.g., "rewrite this in a relatively bland tone") to reduce humor or stylistic bias in the responses, but the rewritten answers still retained noticeable humor or stylistic variations. Could you please clarify:
This information would be incredibly helpful for ensuring faithful reproduction of your results. If possible, could you share the complete prompt or any implementation details?
Thank you for your time and consideration. Looking forward to your response.