the effect of different values of 'alpha' in loss

[justin hall]

Hello!
The pseudo-code of your fine-tuning process shows that 'the total loss =loss fine + alpha * loss_coarse'. Have you ever studied the effect of different values of this 'alpha' parameter?
thank you!
Best wishes!

[Beichen Zhang]

We made some ablations on alpha. The result is similar for alpha=0.1 or alpha=0.2.

[justin hall]

We made some ablations on alpha. The result is similar for alpha=0.1 or alpha=0.2.

Thank you very much!