-
Notifications
You must be signed in to change notification settings - Fork 3
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
the code in generator and critic #2
Comments
Just remove the ```json and ``` in your |
I try to run the code in my environment,the 2wiki result shows very low. the below is a example of my result .The results from hotpot are excellent, but the results from 2wiki are not. Is there anything special about them besides the evaluation method? |
It seems there is no problem with the operation of this case. You can try a few more cases and take a look. |
I have checked my code and that seems there is no mistake. Moreover, the results on the hotpotqa dataset are mostly equal to the results in your paper. After additional checking, we found that the 2wiki dataset has many versions. I wonder which version did you use in your experiments and the source link is appreciate. |
Due to a bit of time, my original data file can no longer be found. In my impression, you can try using the version from my FlashRAG library. |
I noticed that the two fine-tuned expert models in your code are fine-tuned on the hotpot dataset. Did you fine-tune an expert model on the 2wiki dataset during your research? Will this improve the performance of the 2wiki dataset? |
Sorry for delayed response. We don't fine-tune any models in this work. |
In both the generator and critic parts, the json-like string produced by LLM is parsed, but in my case, the eval function reports an error, as shown below:
I try to solve this by removing the beginning and end of the string like
x=output.split('```json') if len(x)==2: output=x[1] x=output.split('```') if len(x)==2: output=x[0]
By the way, 你的代码风格真的可读性巨棒!!!
The text was updated successfully, but these errors were encountered: