I’m in “Paris Match” this week. With a big smile because things are going amazing at Hugging Face 🤗

Paris Match is like the French equivalent of “Life” I guess – AI is really getting mainstream!

here we go again with the usual set of meeting options between SF and Europe – time to disrupt time zones with quantum mechanics or something

This grew quite quickly!

Proof of concept that you can do a lot with low-cost hardware (200$) and a smart robot brain. Is robotics a software problem?

OpenELM: a family of Open-source Efficient Language Models

Welcome Apple Inc. in the family of open-source LLM trainers!

🤯
huggingface.co/collections/ap…

And together with a new library: CoreNet
github.com/apple/corenet

Do you have recommendation on papers for robot navigation in home? Is end-to-end navigation a thing? Is it possible to avoid SLAM or use it only as conditioning/input?

This take on the FineWeb release is one of the most interesting feedback and also a reason FineWeb is very different from even larger datasets like RedPajama-V2 (which is double its size!)

Surprisingly, the size of the dataset of 15T tokens is not very important, what is much…

Most exciting paper of the week? Clearly this one 👇 Finally a successor to the super impressive phi-1.5/2 models – so much looking forward to playing with the weights, come help me encourage the authors to share them in the comments 😅
huggingface.co/papers/2404.14…

🆕 Introducing JAT, the first open-source multi-modal, multi-task multi-domain agent! 🤖 A step toward open generalist agents! 🚀

📰 Blog: huggingface.co/blog/jat

Llama3 was trained on 15 trillion tokens of public data. But where can you find such datasets and recipes??

Here comes the first release of 🍷Fineweb. A high quality large scale filtered web dataset out-performing all current datasets of its scale. We trained 200+ ablation…

Llama3 was trained on 15 trillion tokens of public data. But where can you find such datasets and recipes??

Here comes the first release of 🍷Fineweb. A high quality large scale filtered web dataset out-performing all current datasets of its scale. We trained 200+ ablation…

We have just released 🍷 FineWeb: 15 trillion tokens of high quality web data.
We filtered and deduplicated all CommonCrawl between 2013 and 2024.
Models trained on FineWeb outperform RefinedWeb, C4, DolmaV1.6, The Pile and SlimPajama!

🍿 quite enjoying this recent data-visualization battle where open-source model teams showcase a 2D mapping of the open-source AI models field along interestingly different axes

good data viz skills becoming crucial in any AI model training team these days

and remember

that Llama 3 is in HuggingChat 🔥

running on fast optimized Hugging Face inference

thank you AI at Meta

“I don’t want (AI) to be the property of two US tech companies” Thomas Wolf, co-founder of Hugging Face. Listen to the full conversation:
➝ youtu.be/dGR0vAJAlmI
➝ spoti.fi/49zlhr3
➝ apple.co/3xBZuSk
#scalingtheory

Arena ELO graph updated with new models.

Llama 3 70b looks impressive, but the 8b Instruct version is pure madness: it outperforms GPT-3.5, Claude 2, and Mistral Medium.

High variance at the moment because not a lot of votes, but interesting to see how it evolves.

(Sorry I…

Few bugs but LLama-3 on Huggingchat ios app is amazing to use. System prompt of review:
“You are a hyper-intelligent friendly raccoon that uses first principles based reasoning and system1/system2 thinking to concisely solve every problem in the galaxy while using lots of emojis.