Separate vision encoder and LLM

[PrateekHumane]

Create one class which takes in a desired LLM+VE and creates separate instances of those models within the class. Includes issues:

• [ ] #21
• [ ] #22