37
Mark Zuckerberg open sources 3 new LLMs
(youtu.be)
All in all pretty decent sorry I attached a 35 min video but didn't wanna link to twitter and wanted to comment on this...pretty cool tho not a huge fan of mark but I prefer this over what the rest are doing...
The open source AI model that you can fine-tune, distill and deploy anywhere. It is available in 8B, 70B and 405B versions.
Benchmarks
Sure. It's big. I think they linked some cloud services where you can run it. Like HuggingFace(?), Azure, Amazon, ... And we have some services available like runpod.io where you also can rent a Linux machine by the minute. With several datacenter NVidia cards with 80GB VRAM each.
It won't run on a normal high-end gaming PC at that size. I think a Mac Studio with lots of RAM can do it. Or you'd need to buy several of the very expensive NVidia cards. But I think that's exactly why they gave us the other variants with less parameters.
I'm happy that they released it anyways. Before that it was just a game for the big players and nobody could participate. Now we have it and no one can take it away. It is certainly possible to run it. Albeit not easy to run at home. But it's like that with lots of things in life. Sometimes the professional tools or expensive infrastructure aren't affordable for private people. But we can share and rent such things.