Reproduce results on visual7w.

[sleepyshep]

System Info / 系統信息

cuda 11.8
torch 2.3.0

Who can help? / 谁可以帮助到您？

No response

Information / 问题信息

• The official example scripts / 官方的示例脚本
• My own modified scripts / 我自己修改的脚本和任务

Reproduction / 复现过程

之前我成功复现了论文中refcoco的结果，但在evaluate visual7w时遇到了问题。
我参考了shikra在evaluate visual7w时的代码，使用如下的prompt来提示cogvlm
"Please give a brief and direct reply to 'Which item in the photo is the chair at the desk? Candidates: [433,755,512,966] [003,013,053,895] [002,816,438,996] [180,596,397,996] answer in box format.' with the image"

Expected behavior / 期待表现

请问我的prompt有问题吗，能否提供一份您用来做visual7w任务的prompt，感谢！

[sleepyshep]

@1049451037

[1049451037]

The template for visual7w is

f"""{Question} Select from:
A. [[x1,y1,x2,y2]]
B. [[x1,y1,x2,y2]]
C. [[x1,y1,x2,y2]]
D. [[x1,y1,x2,y2]]
Answer:"""