Secure Delivery of Trained LLM for Client Demo

[humza-sami]

I have recently trained a LLM based on llama-2 using a private dataset for a client. They require the model for a demo on their machines. Unfortunately, I don't have the option to host the model and provide them with an endpoint. The model size is approximately 68GB, and it's stored in SafeTensors. Additionally, I have developed a binary for inference+RAG pipeline. I am using vLLM for inference. The model is located within a folder . The client needs to test the demo on their local machines.
I am seeking advice on the best possible secure method to deliver the LLM model to the client while ensuring that the model files are encrypted to prevent misuse. Given the sensitivity of the model and its potential misuse, encryption is crucial to maintain data security.