LLM of the day: LLAMA 2

Facebook / Meta announced Llama 2 yesterday, which is their updated LLM that’s supposed to be about as good as GPT 3.5. It’s open source and now they’re also allowing it for commercial use for products with up to 700 millions users.

You can try their largest 70B parameter model at HuggingFace or llama2.ai.

It’s clearly not as advanced as GPT 4, but since it’s available for commercial use, lots of companies will be using it to avoid depending on an OpenAI’s or Google’s APIs.

I don’t think there’s any process for people to contribute back to improving Llama, so that’s not why Facebook is open-sourcing it. They might be doing it for goodwill or maybe as a competitive move. Although in a recent interview with Lex Fridman, Mark Zuckerberg does discuss how it would be interesting if crowd-sourcing could be used to improve LLMs.