[AINews] Talaria: Apple's new MLOps Superweapon
This is AI News! an MVP of a service that goes thru all AI discords/Twitters/reddits and summarizes what people are talking about, so that you can keep up without the fatigue. Signing up here opts you in to the real thing when we launch it 🔜
AI News for 6/7/2024-6/10/2024. We checked 7 subreddits, 384 Twitters and 30 Discords (411 channels, and 7641 messages) for you. Estimated reading time saved (at 200wpm): 816 minutes.
With Apple Intelligence, Apple has claimed to leapfrog Google Gemma, Mistral Mixtral, Microsoft Phi, and Mosaic DBRX in one go, with a small "Apple On-Device" model (~3b parameters) and a "larger" Apple Server model (available with Private Cloud Compute running on Apple Silicon).
The Apple ML blogpost also briefly mentioned two other models - an Xcode code-focused model, and a diffusion model for Genmoji.
For on-device inference, we use low-bit palletization, a critical optimization technique that achieves the necessary memory, power, and performance requirements. To maintain model quality, we developed a new framework using LoRA adapters that incorporates a mixed 2-bit and 4-bit configuration strategy — averaging 3.5 bits-per-weight — to achieve the same accuracy as the uncompressed models.
