LLM TinyMMLU Dataset Implementation

[farook-edev]

As described in the parent issue, implementation of TinyMMLU is quite complete. The only things I think might need to be done are:

• Load data sample by sample as needed instead of all at once. (not necessary)
• Save output tokens in processing step and do all accuracy related calculations on the accuracy side. (needs discussion)
• handle instances where no answer letter is provided. (likely fail query) (currently being done implicitly)
• properly build zero-shot prompt rather than cutting input_formatted