I've encountered significant performance gaps while reproducing your work. I understand that your work is based on code-davinci-002, and ViperGPT's performance on ChatGPT 3.5 seems far from what is mentioned in your paper. Could you offer prompts designed for ChatGPT (such as Refcoco dataset)? This is crucial for my work, and I appreciate your assistance!
I've encountered significant performance gaps while reproducing your work. I understand that your work is based on code-davinci-002, and ViperGPT's performance on ChatGPT 3.5 seems far from what is mentioned in your paper. Could you offer prompts designed for ChatGPT (such as Refcoco dataset)? This is crucial for my work, and I appreciate your assistance!