Artificial Intelligence Rockstars
Facebook/Meta Storms The AI Charts With LLAMA-2
Facebook’s New Open-Source Language Model Available Now
And they keep coming: we’ve seen the release of some truly amazing open-source language models here and here, but now it’s Facebook/Meta in the spotlight again.
This time with a rockstar release, they catapulted their new large language model (LLM) straight to pole position on the Huggingface Open LLM charts.
Ladies and gentlemen say hello to LLAMA-2, a collection of pre-trained and fine-tuned generative text models coming with a full commercial license!
Here is everything you need to know about LLAMA-2 and how you can start using it today!
What Is LLAMA-2?
LLAMA-2 is not a single language model but rather a suite of several pre-trained and fine-tuned large language models (LLMs) in different sizes (ranging from 7 to 70 billion parameters).
Some of the fine-tuned versions, referred to as LLAMA-2-Chat, are specifically enhanced for chatbot-style interfaces and apps like ChatGPT.
And, LLAMA-2 models rock. They are already dominating the so-called Open LLM Leaderboard at Huggingface, which means that they perform better than any other open-source large language model currently out there:
Interestingly, LLAMA-2 models not only demonstrate superior performance compared with other open-source models across a range of benchmarks: but researchers also found promising results in terms of safety and utility, suggesting that LLAMA-2 models (coming with full commercial licenses) might serve as an alternative to proprietary models in many cases.
Take a look at this comparison of how different models can be forced to produce hate speech, unqualified advice, or criminal activities:
But wait, there’s more.
In another creative experiment, the researchers conducted “games” in which they let GPT-4 decide which open source model provides the best answers and thus wins.
Watch how LLAMA-2 beats some of the best models so far, such as Falcon, Vicuna, MPT, and PaLM-Bison, and even almost wins against the old ChatGPT-0301 model:
(For further information about the training process and the evaluation of the models, I recommend you read the official release paper here)
How Does LLAMA-2 Compare To ChatGPT?
Despite the really impressive results, there is still a lot of room for growth for open-source models like LLAMA-2 to reach the quality of closed models like OpenAI’s GPT-3.5 or GPT-4.
The official LLAMA-2 research paper is very transparent about this:
However, we are only at the beginning of this new AI paradigm, and given recent discoveries about the nature of models like GPT-4 or the leaked agenda of OpenAI, we may be surprised by a future where open-source models can catch up.
How To Use LLAMA-2?
If you want to try LLAMA-2, you can do so with the generous demo apps at Hugging Face. Here are links to the 7B, 13B, and 70B chatbot versions:
LLAMA-2 7B-CHAT
is optimized for ChatGPT-style conversations. Link: https://huggingface.co/spaces/huggingface-projects/llama-2-7b-chat
LLAMA-2 13B-CHAT
is another chatbot optimized model but with almost twice as much parameters as the previous 7B version. Link: https://huggingface.co/spaces/huggingface-projects/llama-2-13b-chat
LLAMA-2 70B-CHAT
comes with 70 billion parameters. Also chatbot-optimized. Link: https://huggingface.co/spaces/ysharma/Explore_llamav2_with_TGI
(If you want to run LLAMA-2 privately, you can follow this guide on how to run your own private instance of one of the hundreds of models provided by Huggingface)
Further Reading & Links
- Official LLAMA-2 Github repository: https://github.com/facebookresearch/llama
- LLAMA-2 Research paper (LINK)
➡️ Follow me to stay up to date on “AI &Creativity”. If you want to support my work, become a Medium member using my referral link and get full access to all my articles (140+ and growing) and those of thousands of other writers. 🙏
➡️ If you like my content, why not leave a “clap” at the end of this article, so more people can see it?
This story is published on Generative AI. Connect with us on LinkedIn to get the latest AI stories and insights right in your feed. Let’s shape the future of AI together!
