Portland, OR
- Tweets 81
- Following 73
- Followers 34,400
- Likes 2,882
Mattt retweeted
Alert: Apple just dropped a new model on Hugging Face.
It's a Qwen3.5-9B finetune that turns long documents into small page images to save tokens, then pulls up the full text of only the pages relevant to your question 💡
huggingface.co/apple/LensVLM…
75
278
3,464
278,450
Mattt retweeted
What I like about Jev: for years, large generalist models sucked almost all the oxygen out of AI.
But there are huge opportunities in much more specialized and customized models, built for specific tasks and languages and as a result, orders of magnitude cheaper, faster and more optimized. There are 3 million of them publicly available on @huggingface.
Let's build a much more diverse AI ecosystem!
83
97
1,064
57,360
When my coding agent starts a long job, I usually end up scrolling Hacker News. I wanted a smaller distraction, so I built trifle.day
It's daily programming trivia, served by your assistant over MCP while it works. Like Wordle, everyone gets the same question.
11
5
73
8,992
Back in the day, I ran @NSHipster pub quizzes at meetups and a few years of WWDC. I'd been going to weekly pub trivia at the Ginger Man in Austin (RIP), and wished they'd ask about obscure corners of Cocoa and Objective-C instead.
Trifle is that quiz, one question a day.
3
6
1,041
iMCP 1.5.0 is here, with a Phone service for call history, per-tool toggles, fresher Messages results, and automatic updates — so this is the last release you'll have to download by hand 🥲
github.com/mattt/iMCP/releas…
ALT iMCP banner showing icons for all of the provided services.
4
11
135
17,213
Mattt retweeted
this is super easy to run
install llama binary: curl -LsSf llama.app/install.sh | sh
run: llama serve -hf meta-models/muse-glimmer-30b --spec-type draft-dflash -fa on --jinja
please spread the word
Muse Glimmer 30B is shipped with DFlash drafter which speeds-up generation 2-4x at little memory cost 🔥
we support this in llama.cpp and transformers, see below how it looks like in the wild (llama webui) ⤵️
6
13
103
10,793
...which ironically would have been especially useful during the outage sosumi.ai had yesterday in the process of shipping the CLI 🫠
github.com/NSHipster/sosumi.…
2
1
11
6,370
Apologies to anyone impacted by yesterday's outage. I'm taking it as an opportunity to shore up the infrastructure — adding the kind of things you'd expect for a production deployment (synthetics, alerting, etc.).
It's a curious thing about software, and really any shared endeavor: as something becomes depended upon by more people, the nature of what it demands from you changes. It's silly for a hobby project nobody uses to have a lot of process. But at some point neglect stops being cool-kid pragmatism and becomes stubborn recklessness. What made you fast early on becomes a liability.
It's a lot like growing up. You do it reluctantly, and usually because you learned something the hard way.
12
4,505
Mattt retweeted
Today ggml.ai joins Hugging Face
Together we will continue to build ggml, make llama.cpp more accessible and empower the open-source community. Our joint mission is to make local AI easy and efficient to use by everyone on their own hardware.
I've started a company: ggml.ai
From a fun side project just a few months ago, ggml has now become a useful library and framework for machine learning with a great open-source community
142
233
1,646
313,497
By popular demand, sosumi.ai now supports externally-hosted Swift-DocC packages!
swift-argument-parser (GitHub Pages):
sosumi.ai/external/apple.git…
swift-composable-architecture (@SwiftPackages): sosumi.ai/external/swiftpack…
5
12
66
7,018
We want to be good citizens of the ecosystem, so sosumi.ai respects robots.txt on every external host. Site owners can opt out by disallowing the sosumi-ai user-agent.
Full crawler policy and contact info: sosumi.ai/bot
1
1
5
2,397
AnyLanguageModel 0.7.0 is here, with guided generation for all models — Core ML, MLX, llama.cpp, Ollama, OpenAI, Anthropic, Gemini — plus, a new Open Responses model, and a ToolExecutionDelegate for observing and controlling tool calls.
Check it out!
github.com/mattt/AnyLanguage…
ALT @Generable(description: "Basic profile information about a cat") struct CatProfile { // A guide isn't necessary for basic fields. var name: String @Guide(description: "The age of the cat", .range(0...20)) var age: Int @Guide(description: "A one sentence profile about the cat's personality") var profile: String } let session = LanguageModelSession(model: model) let response = try await session.respond( to: "Generate a cute rescue cat", generating: CatProfile.self ) print(response.content)
2
3
33
3,372
Next up: Improving guided generation with XGrammar.
This release uses a hand-rolled implementation of constrained generation, but we do much better.
I wrote a Swift wrapper around the C++ library and published it here: github.com/mattt/swift-xgram…
PR here: github.com/mattt/AnyLanguage…
1
3
1,656