Hiera represents a significant advance over the vision transformer architecture. This work outperforms SOTA while being up to 3.6x faster across a range of image and video tasks — without use of domain specialized modules. Code https://
bit.ly/45OSyhq
@aiatmeta
-
Hiera: Advanced Vision Transformer Architecture with 3.6x Speed Improvement
By
–
-
Meta Lattice: New Model Architecture for Enhanced Ads Performance
By
–
Meta Lattice is a new model architecture that improves Meta’s ads systems performance, AI efficiency and enables faster adaptability to the shifting market landscape.
— AI at Meta (@AIatMeta) 2 juin 2023
More on this new work ⬇️Meta Lattice is a new model architecture that improves Meta’s ads systems performance, AI efficiency and enables faster adaptability to the shifting market landscape. More on this new work
-
Meta Scalable Video Processor: First In-House ASIC for Video
By
–
Meta Scalable Processor is our first in-house designed ASIC for video processing. Alex Mastro & Junqiang Lan share more on how it's supporting some of the unique needs we face at Meta's scale. More on this work https://
bit.ly/3oIvIr1 -
Meta Completes Second Phase of Research SuperCluster Buildout
By
–
We recently completed the second-phase buildout of our Research SuperCluster, making it one of the fastest AI supercomputers in the world.
— AI at Meta (@AIatMeta) 30 mai 2023
📺 Kalyan Saladi shared a look at the scale of RSC by the numbers.
More details ➡️ https://t.co/i1W6ld96Eb pic.twitter.com/FJaZbKvqKqWe recently completed the second-phase buildout of our Research SuperCluster, making it one of the fastest AI supercomputers in the world. Kalyan Saladi shared a look at the scale of RSC by the numbers. More details https://
bit.ly/3N2qHCR -
ImageBind: AI Model Binding Six Sensory Modalities Together
By
–
ImageBind brings machines one step closer to the human ability to bind together information from many different senses. It's the first AI model capable of binding data from six modalities at once.
— AI at Meta (@AIatMeta) 23 mai 2023
Details ➡️ https://t.co/fl5TF5bjev
Demo ➡️ https://t.co/fEsmju1fbc pic.twitter.com/bkI4SxZJqXImageBind brings machines one step closer to the human ability to bind together information from many different senses. It's the first AI model capable of binding data from six modalities at once. Details https://
bit.ly/42rMoRE
Demo https://
bit.ly/3peYLCj -
Meta Scales AI Speech Support to 1,100 Languages
By
–
Today we're sharing new progress on our AI speech work. Our Massively Multilingual Speech (MMS) project has now scaled speech-to-text & text-to-speech to support over 1,100 languages — a 10x increase from previous work.
— AI at Meta (@AIatMeta) 22 mai 2023
Details + access to new pretrained models ⬇️Today we're sharing new progress on our AI speech work. Our Massively Multilingual Speech (MMS) project has now scaled speech-to-text & text-to-speech to support over 1,100 languages — a 10x increase from previous work. Details + access to new pretrained models
-
Meta Reimagines Infrastructure for the AI Age
By
–
This week we shared updates on how we're reimagining Meta's infrastructure for the AI age.
— AI at Meta (@AIatMeta) 19 mai 2023
Want to get caught up?
⏪ Scroll back through the thread
📺 Watch the video below
🔗 Read more ➡ https://t.co/IjsoGucEnD pic.twitter.com/LlsQ2Pg3euThis week we shared updates on how we're reimagining Meta's infrastructure for the AI age. Want to get caught up? Scroll back through the thread Watch the video below Read more https://
bit.ly/3BAFhel -
Next-Generation Data Center Design for Advanced AI Silicon
By
–
Our new silicon is going to need a home as we work to enable future generations of AI. Today we shared more on our next-generation data center design which includes higher density racks, liquid cooling to chips and much more. More on this work
-
Meta Introduces MTIA Custom ASIC for Recommendation Models
By
–
📈 Introducing MTIA — a first-gen, custom-designed ASIC by Meta, built for next-generation recommendation models.
— AI at Meta (@AIatMeta) 18 mai 2023
MTIA is built on TSMC's 7nm process at a 25W TDP, providing 102.4 TOPS at INT8 precision & 51.2 TFLOPS at FP16 precision.
Details ➡️ https://t.co/z9ZauVvVpx pic.twitter.com/Cua2xBgJy3Introducing MTIA — a first-gen, custom-designed ASIC by Meta, built for next-generation recommendation models. MTIA is built on TSMC's 7nm process at a 25W TDP, providing 102.4 TOPS at INT8 precision & 51.2 TFLOPS at FP16 precision. Details https://
bit.ly/3WcrJiO -
Meta Completes Second Phase of Research SuperCluster AI Development
By
–
🖥️ Today we’re sharing that we’ve completed the second-phase buildout of Meta’s Research SuperCluster — one of the world’s fastest AI supercomputers. The immense performance is already enabling us to conduct new research at an unprecedented pace.
— AI at Meta (@AIatMeta) 18 mai 2023
More ➡️ https://t.co/BAEsf0PXgG pic.twitter.com/4Kdf22MCXNToday we’re sharing that we’ve completed the second-phase buildout of Meta’s Research SuperCluster — one of the world’s fastest AI supercomputers. The immense performance is already enabling us to conduct new research at an unprecedented pace. More https://
bit.ly/3BDhvhK