About making inferences

[yusufani]

Firstly, congratulations for your amazing work. In order to test the CLIP model in Turkish, I trained 2M of sentences data in Turkish with distil-bert-multi-cased, as in the instructions. Also, I use precomputed embeddings which you provided.
When Loss values ​​are around 0.02, I thought that I can test :D

When I examined the code to test the model, I realized that it was loading pickle files of about 2m in size as weights. I did not understand how to use the 2GB TF model which I have trained with this code. Could you give more information about this?