Trained Models

[xinsu626]

Hello, sorry, I have one more question. I was wondering if it's possible that you could release trained checkpoints, especially FiD models trained on the easy version of the dataset.

[wenhuchen]

It's been a while, I probably won't be able to find the checkpoint. But the original FiD checkpoint is there, which doesn't take long to finetune.

[xinsu626]

It's been a while, I probably won't be able to find the checkpoint. But the original FiD checkpoint is there, which doesn't take long to finetune.

Hi @wenhuchen, thanks for your reply. Sorry for commenting on a closed issue. Do you mind sharing the detailed hyperparameters used for training the FiD model on the easy version of the dataset? I was able to get a similar performance on the dev.hard dataset, but the best performance on easy.dev I got used the default settings in config.yaml is ('exact', 54.319761668321746), ('f1', 63.80402672963455).