when I finetune CLIP_ViT_L_14 model , Logit Scale is decrease from 100.0 to 95. and keep going , is right?

[Johnson-yue]

Hi, everybody, when I finetune ViT_L_14 model with my own dataset, the log is :

1. Logit Scale is decrease from 100.0 to 95 (at epoch 1) and keep going.
2. loss is about 0.0xxx
3. | Eval Epoch: 1
   image_to_text_mean_rank: 2.3887
   image_to_text_median_rank: 2.0000
   image_to_text_R@1: 0.4132
   image_to_text_R@5: 0.9666
   image_to_text_R@10: 0.9917
   text_to_image_mean_rank: 2.6540
   text_to_image_median_rank: 2.0000
   text_to_image_R@1: 0.4230
   text_to_image_R@5: 0.9745
   text_to_image_R@10: 0.9935
   clip_val_loss: 0.0076
4. train log:

5) val log : should I shut it down ?