Which model is recommended for zero-shot classification tasks?

[midofalasol]

Hello!

Thank you for your outstanding work.

However, I am sorry that my English is poor and I could not understand whether this code warehouse can realize the task of zero-sample image classification in a short time.

If I may, I would like to ask which model has the best performance on the zero-sample image classification task?

Since my device with GPU can't connect to the network, I have to download the model offline and then do the inference, is this feasible? Do I need to make any changes after downloading checkpoint (.safetensors)?

Last question, from the other questions, some people said that CUDA 12.1 must be used, so I would like to ask what is the essential configuration in the environment?