Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

💡 [REQUEST] - <few shot learning(in-context learning)评估脚本或评估设置> #681

Open
ColorDavid opened this issue Nov 26, 2024 · 0 comments
Labels
question Further information is requested

Comments

@ColorDavid
Copy link

起始日期 | Start Date

No response

实现PR | Implementation PR

No response

相关Issues | Reference Issues

No response

摘要 | Summary

请问是否有in-context learning或者few shot learning相关的评估脚本或者评估设置?
评估设置主要包括:数据集对应的instruction(类似于'Answer in a word or phrase.'这种)、数据集的评估metric,以okvqa为例,是否与VQA数据集metric一致(https://github.com/GT-Vision-Lab/VQA/blob/master/README.md)、**model状态**,README中提到(We evaluate the pretraining ckpt without SFT.)是否代表您是使用未发布的,不带instruction tuning的model ckpt版本的model进行测试的(我使用huggingface的int4量化版本ckpt并不能复现结果,zero-shot性能已经和README中4-shot性能接近)、示例选取方法(手工的示例、RICES或者是其他示例选取方法)
进一步地,如果您使用不带instruction tuning的model ckpt版本的model进行测试,zero-shot的性能较差的前提是否因为输出的为一个句子而不是一个单词或短语而导致的ACC比较低?

基本示例 | Basic Example

image

缺陷 | Drawbacks

需要复现README中few-shot结果

未解决问题 | Unresolved questions

No response

@ColorDavid ColorDavid added the question Further information is requested label Nov 26, 2024
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
question Further information is requested
Projects
None yet
Development

No branches or pull requests

1 participant