I think you are asking something like this: If employee of HDFC Bank (say it is a customer that uses a domain separated instance) queries "How many Casual Leaves do I have?", then the answer is, say, 15 because HDFC Bank gives that many. But, another employee of say L&T (say it is also a customer of same instance but different domain) queries the same, then the answer is, say, 0 and advises to work for 90 hours.
If that is the expected outcome, then you are not looking for 2 separate machine learning models. You are looking for different training data fed into the same model but embeddings are stored in HDFC_input and L&T_input. The ouputs will use these inputs respectively.
On the other hand, if you want to run different Machine Learning models for HDFC and L&T, then that is a completely different conversation.
