Meta Launches Llama 2 AI Based Language Model

By Aarti Dhapte , 08 September, 2023

Meta newly launched an open-source family of artificial intelligence (AI) language models, Llama 2, in July 2023; it also contains the license for allowing integration with commercial products. With the growing era of artificial intelligence and generative AI, they play a crucial role in enhancing various opportunities for entrepreneurs, digital creators, tech individuals, and others. The economic and social development in the technological field is improving due to artificial intelligence.

The structure of Llama 2 models ranges from 7-70 billion parameters, making the model alarming in the AI world. Meta has conducted research whose outcomes claim that Llama 2 models can outperform open-source chat models mostly on every benchmark considered. BoolQ and Hellaswag scored 81.7 and 80.7 points for the Llama 2 13B model, among the many benchmarks. Similarly, the Llama 2 7B model scored 77.4 points for BoolQ, followed by 77.2 points for the HellaSwag. These two benchmarks have achieved the highest points compared to others for the Llama 2 model.

The Llama 2 model comes in two forms- pre-trained and fine-tuned; here, the former ones are trained with two trillion tokens and a context window of 4,096 tokens allowing to process many contents. The fine-tuned later model is designed explicitly for chat applications or chatbots like ChatGPT and is trained to boost the language processing power. However, the newly launched model is a great help as an open-source model but does not rival OpenAI's GPT-4.