17 April 2023 sylvainsylvestre.com

Facebook’s LLM

LLaMA (Large Language Model Meta AI)

Since Facebook’s LLM (Large Language Model) code was made open source, the world of artificial intelligence has been vibrating with anticipation. Allowing, Researcher and scientific to study the functionality of Large Language Model.

Interestingly, everything took place in less than a month, which surprised many people who thought Llama’s technology would always be too big to operate locally on a computer to help them in their daily lives. Someone, however, rebuilt Facebook’s code in C++ and published LlaMaCPP, allowing anybody to run Llama’s models in only one week.

"By giving people the power to share, we’re making the world more transparent."

Mark Zuckerberg

What makes LlaMaCPP so useful is that it allows users to construct ultra-optimized versions of GPT-3 for certain needs. Companies, for example, can give GPT-3 with the complete project documentation, and the language model will become an expert in the project and offer users with exact answers to their concerns. Previously, this was difficult since specialists had to be payed, their models were expensive, and they did not operate as well, because of many restriction. People can now run LlaMaCPP locally without needing a large Nvidia card because LlamaCPP runs on the CPU.

LlamaCPP’s performance, results, and relevancy are outstanding. It is a 7 billion parameter model which is 25 times larger than Facebook LLM that outperforms GPT-3, in certain aspects. We also need to remember that this version is the smallest of them. It’s quite interesting that people may now have a local GPT-type assistance that runs on their computer and has a relatively rapid response time. The negative aspects of having a local GPT-type assistance are that filters no longer exist, and controlling what it says is difficult. However, the model may be improved, and there are larger and better models available.

LLM’s Bright Technological Future: Reshaping Virtual Assistance and Shifting Perspectives

LLM’s future is really bright. Consider having a virtual assistant on your computer; it’s not only amazing, but it’s also really useful for particular jobs. People can even create super-specialized GPT-3 variants for certain tasks. This is wonderful for enterprises and startups because it allows them to create their own virtual assistants without having to rely on other firms or spend a lot of money. People’s perceptions of Llama’s technology have shifted dramatically in a short period of time, affecting how they use and interact with technology.

Adaptability will decide whether or not an organization succeeds. The fast-paced environment in which we live is rapidly changing. Be open to fresh ideas. There will be significant challenges ahead.

Update

On July 18, 2023, Meta and Microsoft released the Llama 2 model family, a collection of open-source LLMs that are free to use for research and business purposes. Llama 2 marks an extension of Microsoft and Meta’s collaboration, including availability via the Azure AI model collection and optimization for operating locally on Windows. Llama 2 is also accessible via Amazon Web Services, Hugging Face, and other sources. Meta made Llama 2 available with the help of a diverse group of organizations and academics from IT, academia, and policy that believe in open creation of AI technology.