Same START_ITER values in t*_train.yam and t*_ft.yaml

[mmaaz60]

Hi,

I noticed that the value of START_ITER is the same in all corresponding train and ft configs. For example, it is START_ITER: 18000 in both t2_train.yaml and t2_ft.yaml. As per the script run.sh, the finetuning (t*_ft.yaml) is using the weights generated from the corresponding t*_train.yaml as starting point, so the START_ITER shouldn't be the same in both files. Please guide.

Thanks

[JosephKJ]

Hi Maaz,

The START_ITER that you are referring to is the iteration from which we start the clustering process. Kindly note that we dont do clustering while finetuning. Kindly see this flag.

I am afraid that you are confusing this with MAX_ITER, which controls the number of iterations that the model is trained for.

Kindly let me know whether this clarifies your doubt.

Thanks,
Joseph

[mmaaz60]

Thank you for your reply. It clarifies my doubt.