Meta’s launch of LlaMa2, an advanced open-source LLM (Large Language Model), has created excitement amongst AI enthusiasts for a number of reasons. Meta’s origin as a social media platform notwithstanding, the release of LlaMa2 has solidly positioned it as an AI powerhouse fueling competition against Open AI. It holds a strong potential to mold the future picture of LLMs.
An Overview of the Test Set
A trio of sets featuring different token sizes — 100, 500, and 1000 — made up the assessment set. There were 10 questions assigned under each test set, each of which used a sentence from a product description as the basis for the question. Product labels served as the context for the 100 tokens set. The model was tasked to make a deduction or verdict based on the context provided. The responses from the model were then contrasted against a validation set. Only precise matches were accepted as valid.
The model was directed to acquire values from the context, for a list of characteristics of the 500 and 1000 token sets. On matching it with the validation set, all characteristics were precisely derived before an answer was deemed acceptable.