The report stated that the official version of Llama 3 will support multimodal processing, while the two small versions released earlier do not have this capability
Local time on Monday, tech media The Information cited a Meta employee as saying that the company plans to launch two small Llama 3 large language models (LLM) next week as a pre-release version of the official Llama 3 set to be launched in the summer.
The release of these two small models is expected to pave the way for the official debut of Llama 3. Meta released Llama 2 in July last year, after which several companies including Google, xAI under Musk, and Mistral have released open-source large language models, intensifying competition.
Llama 3 directly competes with OpenAI's GPT-4, which has become a powerful multimodal model capable of handling longer texts and supporting image inputs.
The report mentioned that the official version of Llama 3 will also support multimodal processing, meaning it can understand and generate both text and images simultaneously; however, the two preliminary versions do not have this capability.
Generally, smaller models have lower costs and run faster, especially given the high costs of running large models, highlighting their value. Smaller models are also easier for developers to use in developing artificial intelligence software on mobile devices.
Meta has previously released three versions of Llama 2, with the largest one having 700 billion parameters, while the other two versions have 130 billion and 70 billion parameters respectively.
According to a previous article by Hard AI, the largest version of Llama 3 may have over 140 billion parameters.
Meta will also improve on the issue of Llama 2 being too conservative in responding to controversial topics in Llama 3. Researchers plan to relax the restrictions of large models in this aspect to allow for more interaction with users, providing background information rather than just refusing to answer