Hi, nice work!
1)Could you please provide the final loss (the loss log during the training process) of bliva in the pre-training stage?
2)I find that in epoch>0, cosine_lr_schedule is performed according to epoch. Would it be better to follow the steps here to conduct cosine_lr_schedule?
Hi, nice work! 1)Could you please provide the final loss (the loss log during the training process) of bliva in the pre-training stage? 2)I find that in epoch>0, cosine_lr_schedule is performed according to epoch. Would it be better to follow the steps here to conduct cosine_lr_schedule?