Exciting to see Llama 3.2 running on Arm!
OPEN SOURCE
-
Meta Supports Open Source AI Development with Llama
By
–
Open source is how we build AI technology that benefits everyone and we're proud to keep supporting that push with Llama.
-
Anticipation for Hugging Face Community’s work with new models
By
–
Can't wait to see the Hugging Face Community do what they do best with these new models!
-
Llama 3.2 Integration Across Amazon Ecosystem Announced
By
–
We're incredibly excited to see Llama 3.2 light up across the Amazon ecosystem!
-
Llama 3.2 Released: Major Gains in Code and Spanish
By
–
Llama 3.2 is here! @scale_AI partnered closely to push the limits: SEAL Evals: Llama 3.2 90B shows big gains— +3 in IF +5 in Coding
It’s even beating 405B in Spanish! Enterprise partnership for custom Llama models Data Foundry partnership for RLHF & SFT -
AI2 Releases Multimodal Open-Source Models
By
–
congrats to AI2 for also releasing (what look like) great multimodal open-source models today!!! https://t.co/hgcVwy4BER
— Soumith Chintala (@soumithchintala) 25 septembre 2024congrats to AI2 for also releasing (what look like) great multimodal open-source models today!!!
-
Llama 3.2 Small Models Enable Mobile Development on Snapdragon
By
–
These models may be small, but this news is big! Excited to see what mobile developers will be able to do with Llama 3.2 1B & 3B on Snapdragon!
-
Llama 3.2 Vision Models Compete with Leading Closed Models
By
–
Evaluating performance on extensive human evaluations and benchmarks, results suggest that Llama 3.2 vision models are competitive with leading closed models on image recognition + a range of visual understanding tasks.
-
New Llama Vision Models Updates and Details Released
By
–
This is just a small slice of what’s new with our Llama vision models. You can find more details in the full model card published on GitHub
-
Llama 3.2 Multimodal Models with Improved Image Understanding
By
–
By training adapter weights without updating the language-model parameters, Llama 3.2 11B & 90B retain their text-only performance while outperforming on image understanding tasks vs closed models. Enabling developers to use these new models as drop-in replacements for Llama 3.1. pic.twitter.com/TQo4IIyqRH
— AI at Meta (@AIatMeta) 25 septembre 2024By training adapter weights without updating the language-model parameters, Llama 3.2 11B & 90B retain their text-only performance while outperforming on image understanding tasks vs closed models. Enabling developers to use these new models as drop-in replacements for Llama 3.1.