search

wbsg-uni-mannheim / wdc-sotab

4 stars 1 forks source link

How about re-producing the results of Doduo? #1

Closed A11en0 closed 3 months ago

A11en0 commented 6 months ago

Thanks for your wonderful job. I try to re-training a model on SOTAB dataset, but I cannot reach the performance that you report in the paper. Could you please share the code of re-produce of Doduo (or provide more implement details)? Thanks again!

grxiao137 commented 4 months ago

Same problem here :-) Can you please release code for fine-tuning the Doduo model.

kkorini commented 4 months ago

Hello, thank you for checking our work!

I uploaded the code for fine-tuning Doduo in the folder baselines. The hyperparameters used for fine-tuning are batch size set to 32 and learning rate set to 5e-5.