Ok, has anyone actually been successful at getting a LLM running fully native on an Apple iOS device with CoreML? Is that a thing? #llm #swiftDev #machinelearning #coreml

🧑‍🏫
Wolltet ihr schon immer die Grundlagen künstlicher Intelligenz aus der Perspektive eines Swift Entwicklers sehen? Ein eigenes neuronales Netz schreiben?

Dann besucht den Talk von Michal Harakal auf der #Macoun23.
Tickets: https://macoun.de/anmelden

#NeuralNetwork #KI #Artificialintelligence #Swift #AI #CoreML

Hugging Face、Appleデバイス上で大規模言語モデルを実行するためのSwiftパッケージとDemoアプリを公開。
https://applech2.com/archives/20230809-swift-transformers-run-on-device-llms-in-apple-devices.html
#applech2 #apple_silicon #chatgpt #coreml #mac #macappstore #oss #アプリ #オープンソース

Sono di ritorno da un breve viaggio di lavoro/piacere (ho approfittato per andare a trovare il lato veneto della famiglia) e ho usato #ChatGPT al posto di Siri come assistente (tramite #shortcut) promosso a pieni voti su tante piccole esigenze, tra cui il prepararmi una veloce lista puntata delle cose da portare nel bagagliaio a mano. Spero che #Apple sveli presto i suoi piani per sfruttare a dovere tutto quel potenziale di #CoreML che ci portiamo, inutilizzato, in tasca

Apple recently released Core ML versions of Stable Diffusion models optimised for running on Apple Silicon. Here's how to run them from python (instead of Swift) https://latent.differentialist.info/2023/02/15/diffusers-coreml.html #StableDiffusion #CoreML

CoreML Player: Freier Mac-Videoplayer für maschinelles Lernen

Das Open-Source-Projekt CoreML Player ist mehr Machbarkeitsstudie denn produktives Alltagswerkzeug, nichts desto trotz aber ein spannendes Softwareprojekt, das einen einfachen Zugang beziehungsweise die einfache Nutzung der von Apple angebotenen Core-ML-Modelle ermöglicht. Maschinelles Lernen mit Core-ML-Modellen Der Multimedia-Player kann Fotos und Videos anzeigen und auf diese kompatible Core-ML-Modelle anwenden. Bei Core ML selbst handelt es sich […]

https://www.ifun.de/coreml-player-freier-mac-videoplayer-fuer-maschinelles-lernen-204333/

#Apps #CoreML #Mac #MachineLearning #Videos

@f Does #CoreML work on Intel Mac's or is it exclusive to Apple Silicon?

I've been coding a version of SD that someone ported to TensorFlow which does use Apple Metal with Intel Macs.

Is there someone with solid #KI / #MachineLearning (#Apple #CoreML) experience here? I'm asking for our company.

It's about a very specific task. Shortly:
Documents (PDFs), which are more or less in a specific format, are to be read in and dedicated information is to be extracted from them.

If you feel addressed, or know someone, or know someone who knows someone.... → DM. 😃

Please share this vigorously. Thanks. 🙏🏼

Translated with www.DeepL.com/Translator (free version)

Ist hier vielleicht jemand mit solider #KI / #MachineLearning (#Apple #CoreML) Erfahrung? Ich frage fuer unsere Firma.

Es geht um eine ganz konkrete Aufgabe. Kurz nur soviel:
Es sollen Dokumente (PDFs, die mehr oder weniger in einem bestimmten Format vorliegen, eingelesen und daraus dezidierte Informationen ausgelesen werden.

Wenn du dich angesprochen fuehlst, oder jemanden kennst, oder jemanden kennst, der/die jemanden kennt... → DM. 😃

Bitte teilt das kraeftig. Danke. 🙏🏼

Is there someone with solid #KI/#MachineLearning (#Apple #CoreML) experience here? I'm asking for our company.

It's about a very specific task. Shortly:
Documents (PDFs), which are more or less in a specific format, are to be read in and dedicated information is to be extracted from them.

If you feel addressed, or know someone, or know someone who knows someone.... → DM. 😃

Please share this vigorously. Thanks. 🙏🏼

Translated with www.DeepL.com/Translator (free version)

Ist hier vielleicht jemand mit solider #KI/#MachineLearning (#Apple #CoreML) Erfahrung? Ich frage fuer unsere Firma.

Es geht um eine ganz konkrete Aufgabe. Kurz nur soviel:
Es sollen Dokumente (PDFs, die mehr oder weniger in einem bestimmten Format vorliegen, eingelesen und daraus dezidierte Informationen ausgelesen werden.

Wenn du dich angesprochen fuehlst, oder jemanden kennst, oder jemanden kennst, der/die jemanden kennt... → DM. 😃

Bitte teilt das kraeftig. Danke. 🙏🏼

Hey friends. I've been working on a Whisper port to #Swift / #CoreML / #vDSP and hit a road block, and am looking for some help. If we can cross a final hurdle this will be an INSANELY fast, energy efficient port of Whisper for macOS, iOS devices.

Here's the current PR - any brave souls want to join me in a challenge?

https://github.com/tanmayb123/OpenAI-Whisper-CoreML/pull/2

Not a huge deal/achievement since I just copied the code for my #CoreML #StableDiffusion UI to do this, but I do have a native #macOS app running my Python code for Stable Diffusion image generation now.

It's not quite where I'd like it to be functionality/usability-wise, but it works. But the #Flutter app I did sometime back behaves better than this one 😛

Interesting (or maybe not) fact: if you are on #macOS and use #StableDiffusion (or anything using #MPS, probably) you get a bunch of folders/files created under the /private/var/folders/{random name}/{random name}/{single letter}/com.apple.MetalPerformanceShadersGraph folder.

For SD, the disk space taken up is negligible but I got over 300 folder created for generating 3 images. For #CoreML StableDiffusion, a single run takes up about 1.5GB sometimes!

#Apple #DeepLearning

In September, I gave a talk at #FrenchKit called “How Not to Develop an Umm Detector Using Create ML”.

In the talk I show how I trained a audio classifier for #CoreML and improved it incrementally.

The recording has been posted online now:

https://async.techconnection.io/talks/frenchkit-2022/yono-mittlefehldt-how-not-to-build-an-umm-detector-using-create-ml/

If you have any feedback for me, I’m all ears! It’s the best way to improve 😀

Over the course of the past two weeks or so, not including non-coding projects and work-related projects, I've worked on at least five coding projects:

1. An all-in-one #StableDiffusion CLI generator in Python
2. Updates to a #Swift #macOS Mastodon client
3. Started from scratch work on a pure #JS Mastodon client
4. A simple web server to host multiple instances of HTML/JS (for the JS Masto client)
5. Work on the Hugging Face #CoreML Stable Diffusion UI

1/2

If you have an #Apple Silicon Mac and like to generate images with #StableDiffusion, then I've just released the 1.1 version of my #CoreML macOS app based on the initial Hugging Face code 🙂

https://github.com/FahimF/swift-coreml-diffusers/releases/tag/v1.1

The binary is notarized and so should hopefully run without any issues for everybody. If you run into issues, please let me know.

This adds most of the features that I wanted for the first release. Now to start adding smaller features and/or more complex ones ...

I am not sure this is worth talking about — a part of me feels that I should just shut up and move on 😛

I've been working on the #Swift #CoreML #StabeDiffusion app a lot over the weekend and I noticed someone releasing the same features as I did. (I won't go into who ...)

Took a closer look and they are copying my code and "releasing" new builds, no attribution or anything.

I've had this happen before in writing circles, but first time (I know of) on GitHub. Just feels weird ... 1/

I've been working on the #CoreML #StableDiffusion app over the weekend and it's at a fairly good point right now 🙂

Negative prompts, support for multiple models (you have to download them), multiple schedulers, a gallery of images generated etc. So looking pretty good if I say so myself 😛

https://github.com/FahimF/swift-coreml-diffusers/

A few more things I want to fix but then it's time to see if I can add more (any?) features to the generator itself ...

#macOS #Swift

Feels like it's been forever, but I've pushed out a new beta update for #GiveMeASniglet today in #TestFlight for #iOS and #iPadOS, which brings a brand-new table UI for iPads. Still a way to go before 1.2 releases, though...

#iOSAppDev #SwiftUI #Swift #CoreML #linguistics #conlang #sniglet