Richard Wei
535 posts

Richard Wei
@rxwei
I lead the Foundation Models framework @ Apple
San Francisco, CA Katılım Eylül 2009
301 Takip Edilen4.1K Takipçiler

Today we are introducing a Python SDK for Mac's on-device LLM! github.com/apple/python-a… apple.github.io/python-apple-f…
English

Introducing AnyLanguageModel: A Swift package that provides a drop-in replacement for Apple's Foundation Models framework with support for custom language model providers.
github.com/mattt/AnyLangu…
Just change your import statement:

English
Richard Wei retweetledi

A list of apps that are using Apple's local models to introduce new features. techcrunch.com/2025/09/19/how…
English

Trying out lots of new apps from App Store today! Love all these delightful features leveraging our on-device LLM 🤩 #iOS26
English
Richard Wei retweetledi

Prime example for how to apply Foundation Models in your app
Devin@JustMeDevin
Loving Foundation Model access on iOS 26, so many possibilities. Here are a few simple features that we’ve added in Crouton so far 😊
English
Richard Wei retweetledi

My @DarkNoiseApp update with the new iOS 26 Liquid Glass design, and a new Magic Mix feature (demoed here) powered by the new Foundation Model API & Apple Intelligence is ready for tomorrow's release!
If you're press writing about app updates, my press release is below 🧵👇
English
Richard Wei retweetledi

9/11だけではありますが、Foundation Model フレームワークのエンジニアも参加してくれることになりました。もし直接エンジニアと話したいって方は9/11にぜひご応募ください。また、9/9,10の枠も、もちろんお待ちしております。
Shun Takeishi@ShunTakeishi
developer.apple.com/events/view/MJ… 新しいイベントのご案内です。Foundation Model フレームワークをアプリに実装するワークショップです。皆さんのプロジェクトを持ってきていただいて、もくもく会のようにそれぞれ作業いただきながら、必要に応じて私たちの方でもアドバイスなどサポートさせていただきます。
日本語

The team that built the Foundation Models framework is hiring! Join us to create the best LLM APIs, both for app developers and for Apple teams.
jobs.apple.com/en-us/details/…
English

@rudrank That’s right — guardrails will no longer be triggered by sensitive/controversial topics that are not fundamentally unsafe. If you ask the model to summarize or rewrite such content, the model will do as told; otherwise, the model may still refuse with an explanation.
English

Has anyone tried this? I am not able to trigger guardrails. Any prompt to experiment with default and permissiveContentTransformations
I tried with summarizing and rewriting a few political ones but it works on both
Rudrank Riyam@rudrank
Biggest change in foundation model framework! There is a new permissiveContentTransformations guardrail!
English
Richard Wei retweetledi

If you missed the WWDC25 recap broadcast, video is now on Youtube! I cover:
- On-device LLM using the Foundation Models framework
- Custom models using CoreML
- Latest research models using open-source MLX
and more! Enjoy!
youtube.com/watch?v=9lO3ds…

YouTube

English

@JPEGuin You need to recompile your app using Xcode 26 beta 5 to run the app on OS 26 beta 5. We made some final API improvements and removed APIs that were deprecated in previous betas.
English
Richard Wei retweetledi
Richard Wei retweetledi
Richard Wei retweetledi

There is a new refusal case with explanation. They explicitly mention that a token is 3-4 characters in English.
I had asked for a property to keep count of it in a feedback but the response was that it is not on the engineering team's roadmap

Rudrank Riyam@rudrank
LanguageModelFeedbackAttachment is now just LanguageModelFeedback
English
Richard Wei retweetledi

@rudrank @NatashaTheRobot @emcro @SwiftyAlex @_julianschiavo +1. An example (prompt + code) that reproduces the issue is the best kind of actionable feedback. We'd like to look into this!
English

Is anyone else having a horrible time with using Tools in the new FoundationModels framework update in Beta 4? The model will not call the tool and will instead insert [Tool X was called] and hallucinate an answer. And that's not when it's failing on calling the tool entirely b/c it can't generate an argument...
English



