-
Notifications
You must be signed in to change notification settings - Fork 131
New issue
Have a question about this project? # for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “#”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? # to your account
情景问答对 #12
Comments
可以的,可惜我们的这一版生成的训练数据可能初始法律清洗的不够干净,有些分点的法律没有归纳好。如果用我们后来整理的9k条核心法律一对一的生成数据(我们在测试集上这样做的),可以确保至少生成的法律依据百分百没有问题。 |
如果是一对一的生成数据,那应该生成的问答对只是适用于当前用于生成的法律条款的,但是我看见例子里面还涉及到了其他的法律条款,请问这种是怎么处理的呢? |
之前生成数据集的时候考虑到很多法条有前后文补充关系,所以都是取了临近的几条法律生成样本。如果需要单条法条的,可以参考我们的测试集,在data/核心法条9k_测试5k里。 |
分点的规纳, 只有靠规则了。 |
这个生成的指令, 可分享吗? |
# for free
to join this conversation on GitHub.
Already have an account?
# to comment
利用ChatGPT联想生成具体的情景问答,能够百分百确定生成的法律依据没有错误吗
The text was updated successfully, but these errors were encountered: