🇺🇸

Llama 3.1 is here. 8B, 70B and 405B. Download: github.com/meta-llama/lla… Paper: ai.meta.com/research/publi… Agentic System Examples: github.com/meta-llama/lla… Blog post: ai.meta.com/blog/meta-llam…

A short time ago in 10 timezones from California away... While Llama 3.1 is (rightfully) all the rage, some weirdos are making progress on generating all tokens at once with flow matching (a diffusion family process), and testing on the hardest task to get exactly right: codegen!

New research paper from Meta FAIR – Transfusion: Predict the Next Token and Diffuse Images with One Multi-Modal Model. Chunting Zhou, Lili Yu and team introduce this recipe for training a multi-modal model over discrete and continuous data. Transfusion combines next token

"Llama is approaching 350M downloads on Hugging Face More than 10x compared to this time last year"

True

New research from Meta FAIR: MoMa — Efficient Early-Fusion Pre-training with Mixture of Modality-Aware Experts ➡️ go.fb.me/kz3b0c This paper introduces modality-aware sparse architectures for early fusion, mixed-modality foundation models and opens up several promising

With Llama 3.2 we released our first-ever lightweight Llama models: 1B & 3B. These models empower developers to build personalized, on-device agentic applications with capabilities like summarization, tool use and RAG where data never leaves the device.

1B and 3B: Smaller and more performant models informed by powerful teacher models.

A personal favorite of mine from yesterday: Meta AI translations in your own voice in Reels. A direct result of speech and voice research we published last year, now making it into product!

Announcing Meta Movie Gen: new foundational media generation models We introduce 4 new capabilities: (1) text-to-video (2) video editing, (3) personalize images of yourself within video (4) video-to-audio w/ optional text prompts. Website: ai.meta.com/research/movie… Blog:

Now HEAR this (not just watch) - We've got audio covered for generated videos 🔊 Introducing Movie Gen Audio, which adds 48kHz synced SFX and aligned music to amazing videos from Movie Gen Video (and other sources!) Super honored to work with this amazing team! More to come🔥🔥

Going from SF early stage startups to Meta has been another culture shock. This is one of the most functional places I’ve worked, with the East coast culture of my high school & college years. Operations are clean. Dads go blueberry picking with their kids on weekends. No

🚨 New drop of models, research and datasets from FAIR

We're at NeurIPS Conference this week showcasing some our latest research across GenAI, FAIR, Reality Labs at Meta Research and more. This year researchers from across Meta had 47+ publications accepted, are taking part in 7+ different talks/workshops/panels and we'll be showcasing a number of

A collection of new AI research released out of FAIR today. Check out the models, papers, code and datasets: ai.meta.com/blog/meta-fair…

Say 👋 goodbye to tokens with 🥪 BLT: Byte Latent Transformers. Paper: dl.fbaipublicfiles.com/blt/BLT__Patch… Code: github.com/facebookresear…

don’t leave anything for later. later, the coffee gets cold. later, you lose interest. later, the day turns into night. later, people grow up. later, people grow old. later, life goes by. later, you regret not doing something… and you had the chance.