One of the most awaited open-source LLM is here now…LLAMA-2 outperforms rest of them.
Meta has released LLAMA -2, which is an open-source large language model & it is currently available through Microsoft Azure, Amazon Web Services (AWS), Hugging Face & few others. Check out how you can get started with it.
You get a Llama! And you!
— Tech at Meta (@techatfacebook) July 18, 2023
Meta is releasing Llama 2, the next generation of Llama. With Llama 2 now commercially available, anyone can leverage the large language model to build their own applications. https://t.co/wCTEp4qE81 pic.twitter.com/MBen3FHjzZ
LLAMA-2 is actually a result of expanded AI partnership between Meta & Microsoft. It is available for FREE for both commercial & research purposes.
Let’s highlight the features of LLAMA-2:
- LLAMA-2 is available in 3 parameter sizes, 7B, 13B & 70B.
- All model sizes are pre-trained with reinforced learning from human feedback.
- That is the same technique used to train ChatGPT & GPT 4 from OpenAI.
- Context Length is doubled from 2048 to 4090.
- LLAMA-2 outperforms all opensource models which are currently available by a wide margin.
- For the parameter size of 7B, LLAMA 2 had double the benchmarks figures in comparison to MPT-7B & Falcon 7B.
- Since, both Falcon & MPT, don’t come with a 13B parameter size, they could only compete with LLAMA-2’s 70B type.
- Naturally, it outperformed both of them.
- You can checkout the table below as presented by Meta.
How to use LLAMA-2 right now?
If you want to access LLAMA-2, then first you’ll have to accept Meta’s terms and conditions. It is a simple form that will grant access to your email-address within 24-hrs.
After that, it is up to you to choose the provider you want. As mentioned above, LLAMA-2 is available via Azure, AWS & HuggingFace. I’ll add more providers to this list as I find them.
If you just want to checkout how LLAMA-2 performs then you can do so here without needing any access from Meta.
Above link doesn’t require any sign-up or agreement. So, do check it out.
How to use LLAMA-2 with HuggingFace?
If you want to build apps using LLAMA-2 then you must have access to the weights & model card.
So, after you have submitted the Meta’s agreement form, wait till you receive a congrats email from them regarding LLAMA-2’s access.
After that, head over to Hugging Face LLAMA-2 .
Above link has all the details about the model itself & all available types.
In resources section of this article (at the bottom), you’ll find individual link to LLAMA-2 7B, 13B & 70B parameter sizes.
You will need accept a mandatory model agreement from HuggingFace as well. It takes few hours for them to check your profile & approve you.
After that you can simply download all the necessary weights & files from HuggingFace to use to build your AI App.
Once I have an access to the model, I’ll upload the source code of a simple AI ChatBot based on LLAMA-7B & 13B for you to checkout & expand further.
You can checkout the video below to learn more about LLAMA-2. You can find all the related link in the resources section below as well.
I was granted access within 1Hr, so the wait-time isn’t too long. 😊