All the model checkpoints here:
RESEARCH
-
Meta Layer Skip Enables 200% Faster Transformer Inference
By
–
Meta presents Layer Skip – up-to 200% fast inference > Applies layer dropout: low rates for early layers, high rates for later layers
> Uses early exit loss with shared exit for all transformer layers Inference: > Increases early exit accuracy without auxiliary layers
> -
Meta FAIR releases Layer Skip for accelerating LLM inference
By
–
We previously shared our research on Layer Skip, an end-to-end solution for accelerating LLMs from researchers at Meta FAIR. It achieves this by executing a subset of an LLM’s layers and utilizing subsequent layers for verification and correction. We’re now releasing inference… pic.twitter.com/gag29HSf6e
— AI at Meta (@AIatMeta) 29 octobre 2024We previously shared our research on Layer Skip, an end-to-end solution for accelerating LLMs from researchers at Meta FAIR. It achieves this by executing a subset of an LLM’s layers and utilizing subsequent layers for verification and correction. We’re now releasing inference
-
Nobel Prize Physics Deep Learning AI Science Panel Discussion
By
–
It was wonderful to be at @Google Zeitgeist and be part of AI+Science panel with @VanJones68 @lilaibrahim James Manyika, on the day that the Nobel Prize in physics was announced for deep learning. Watch our discussions:
-
MCTS Sampling Technique Shows Promise in Language Models
By
–
I wrote it because I was playing with MCTS sampling back then. It was super slow and I didn't run evals to see if it actually performed better, but cool to see it's getting traction now.
-
Meta Llama 3.2 Lightning Fast AI Hackathon $10k Prize
By
–
Want your shot at our $10k prize pool?💰
— SambaNova (@SambaNovaAI) 29 octobre 2024
Build with fast AI inferencing on @AIatMeta's Llama 3.2, where you can achieve speeds of 2470 tokens per sec on 1B and 1566 tokens per sec on 3B.
Join our Lightning Fast #AI #Hackathon ⤵️Want your shot at our $10k prize pool? Build with fast AI inferencing on @AIatMeta
's Llama 3.2, where you can achieve speeds of 2470 tokens per sec on 1B and 1566 tokens per sec on 3B. Join our Lightning Fast #AI #Hackathon -
Tim Berners-Lee Introduces HTML Markup Language in 1991
By
–
#otd in 1991 Tim Berners-Lee first described the HTML mark-up language in a message to the WWW-talk mailing list: https://
bit.ly/2NqpXaV -
Top-k and Top-p Sampling Strategies for LLMs Explained
By
–
There's been a lot of interest around sampling recently with entropix.
— Maxime Labonne @ ICLR (@maximelabonne) 29 octobre 2024
I just uploaded my first LLM blog post to HF, which talked about top-k and top-p sampling with easy-to-understand visualizations and code.
📝 Article: https://t.co/cgoP59fED4 pic.twitter.com/DJ45ToDtPnThere's been a lot of interest around sampling recently with entropix. I just uploaded my first LLM blog post to HF, which talked about top-k and top-p sampling with easy-to-understand visualizations and code. Article: https://
huggingface.co/blog/mlabonne/
decoding-strategies
… -
Agent-as-a-Judge: AI Agents Evaluating Other Agents
By
–
New from @metaai and @KAUST_News
! Introducing Agent-as-a-Judge, a new framework where AI agents evaluate other AI agents, leading to a 97% saving in time AND cost Talk to the team @MingchenZhuge
, @SchmidhuberAI
,
@tydsh
, @zechunliu
, @YoungXiong1
, Yangyang Shi, @vikasc here! -
First Online Message: 55 Years of Internet History
By
–
55 years ago today, UCLA professor Leonard Kleinrock & his team sent the first online message: https://
bit.ly/4fgQQsR Funny enough, it was a typo: "LO" instead of the intended "LOGIN." v/UCLA & Reuters