Meta makes Llama 2 free for research & commercial use and partners with Microsoft, among other companies, to offer its powerful LLM.
Furthering the cause of open-source AI development, Meta open-sourced the code for Llama 2, one of the most formidable ChatGPT competitors. In quality, Llama outperforms GPT-3 while lacking a little in coding.
The AI model is now source-available after extensive safety evaluations and context distillation. It will also be available for commercial use for applications with less than 700 million monthly active users.
The decision to make Llama 2 available free of charge for research and commercial use stems from Meta’s ambition to “put exploratory research, open source, and collaboration with academic and industry partners at the heart of [their] AI efforts for over a decade.”
Meta has partnered with Microsoft to provide Llama 2 through Microsoft’s Azure platform. The company is also working with Qualcomm to integrate the model into laptops, phones, and headsets from 2024 onward.
Apart from Microsoft Azure, Llama 2 is also going to be available through other providers. Major among these are Amazon Web Services (AWS) and Hugging Face. Perplexity is also running Llama 2. Links to Llama 2 instances:
- Perplexity AI
- Hugging Face (announcement)
- AWS (announcement)
- Azure (announcement)
Committed to building responsibly, Meta has over 50 partnerships with companies using the Llama model. This includes companies like Spotify, Dropbox, IBM, Shopify, Nvidia, Intel, AMD, Accenture, Zoom, Atlassian, and LG.
This will allow companies and startups to deploy Llama 2 for their own uses, one of which can be creating an AI chatbot trained on a large dataset. It’s noteworthy at this point that Llama 2 is not a GPT-4 (ChatGPT Plus) competitor or even a strong contender for GPT-3.5 (free ChatGPT) level output.
There are three available LLMs—7B, 13B, and 70B. There are significant improvements over Llama 1, particularly in the dataset which has 40% more tokens and a significantly larger context length of 4,000 tokens. The fine-tuned models include the Llama 2-Chat, optimized for dialog application using RLHF. According to Hugging Face’s study into the model, the chat models “perform better than most open models and achieve comparable performance to ChatGPT according to human evaluations.”
With the … wisdom of crowds you actually make these systems safer and better and, crucially, you take them out of the … clammy hands of the big tech companies which currently are the only companies that have either the computing power or the vast reservoirs of data to build these models in the first place.
Nick Clegg, President of Global Affairs, Meta told BBC Radio 4’s Today
