
Pinned Tweet
New Apple models are here, super exciting stuff that we built with the team! Especially for the AFM 3 Core Advanced architecture it’s worth a read.
Technical report is coming later this summer!
machinelearning.apple.com/re…
Come work with us to improve multilinguality for AFMs!
jobs.apple.com/en-us/details…
Robin M. Schmidt retweeted
Apple's on-device models give us free, private and low-latency inference in Swift. But their performance is limited by device constraints (memory, compute, thermal considerations), and the context window size (8k as of iOS 27 beta 3).
Learn how to overcome these limits with a hybrid AI router based on Apple's Dynamic Profiles:
peterfriese.dev/blog/2026/hy…
Robin M. Schmidt retweeted
We did Shampoo so we kill all the variants of Adam. Now we call Shampoo b2=0.0 as Muon, then created 10s of variants thats marginally better.
Heading to #EMNLP2025 in Suzhou, reach out if you want to grab a drink!
2
381
If you were impacted by the Meta layoffs yesterday and wan't to work on Multilingual LLMs / Machine Translation @Apple, feel free to send me a DM or apply:
jobs.apple.com/en-us/details…
3
1,184
New on-device improvements!
At WWDC we introduce a new generation of LLMs developed to enhance the Apple Intelligence features. We also introduce the new Foundation Models framework, which gives app developers direct access to the on-device foundation language model.
machinelearning.apple.com/re…
1
455
It’s extremely upsetting when I write an extensive review for a paper submission and the authors just withdraw the paper without any rebuttal whatsoever
2
420
Exciting times to join Apple for AI/ML!
machinelearning.apple.com/re…
Check out #WWDC24 for more info what we do in ML @ Apple!
1
273
you get priority boarding because you fly business, I get it because of extra security screening, we’re not the same
1
288
the Starbucks honor pickup system is the only thing holding our society together at this point
1
190
why have a blog post when you can just drop a torrent link to the model weights without any context
magnet:?xt=urn:btih:5546272da9065eddeb6fcd7ffddeef5b75be79a7&dn=mixtral-8x7b-32kseqlen&tr=udp%3A%2F%2Fopentracker.i2p.rocks%3A6969%2Fannounce&tr=http%3A%2F%https://xitter.gq/t.co/g0m9cEUz0T%3A80%2Fannounce
RELEASE a6bbd9affe0c2725c1b7410d66833e24
405
Show everyone that your optimizer is truly better than AdamW and claim the price money 👇🏻💰
After 3 years of hard work, our unprecedented neural network training algorithm competition is finally open! The exciting part starts now, seeing what the community can create. 🏆Submit, become the next Adam, and bag $50,000 in prizes!
mlcommons.org/2023/11/mlc-al…
Is it bad that the grid size of 1 (as is standard) used for hyperparameter tuning is reminding me of optimizer papers 🤔
Presenting this tomorrow 9:00 — 10:30am in Frontenac Ballroom / Queen’s Quay, drop by everyone and say hi !! #ACL2023NLP
Learning Language-Specific Layers for Multilingual Machine Translation
abs: arxiv.org/abs/2305.02665
paper page: huggingface.co/papers/2305.0…











