perf: update llm starter script to allow distributed inference

[EntroSanity]

This pull request updates validateVram function in llm starter script to check for total available vram when user intends to deploy large models across multiple gpus; it comes with a usage example in README