Meta Teases Launch of Small Llama 3 Language Models Next Week

Meta, the technology company, is reportedly set to release two smaller versions of its Llama 3 language model (LLM) next week, serving as a precursor to the official launch of Llama 3 in the summer. The largest variant of Llama 3 may boast over 140 billion parameters.

The imminent release of these two smaller models is expected to pave the way for the official debut of Llama 3. Meta had previously unveiled Llama 2 in July last year. Since then, several companies, including Google, Musk's xAI, and Mistral, have introduced open source large language models, intensifying the competition in this space.

Llama 3 will directly compete with OpenAI's GPT-4, which has evolved into a robust multi-modal model capable of handling extensive texts and supporting image inputs.

Reports suggest that the official version of Llama 3 will also incorporate multi-modal processing, enabling it to comprehend and generate both text and images simultaneously. However, the initial two smaller versions being launched lack this capability.

Smaller models are generally more cost-effective and faster to run, particularly during a time when running large models can be expensive. Moreover, these compact models facilitate the development of AI software for mobile devices, making it more accessible for developers.

Previously, Meta released three versions of Llama 2, with the largest variant featuring 70 billion parameters, while the other two had 13 billion and 7 billion parameters respectively.

In Llama 3, Meta aims to address the issue of Llama 2 being overly cautious when responding to controversial topics. Researchers plan to relax the constraints of the larger model in this regard, allowing it to engage more with users, provide context, and offer responses beyond mere refusal to answer.