The best Hacker News stories from All from the past day
Latest posts:
Advancing the price-performance frontier with GPT‑5.6
Gemini Robotics 2 brings whole body intelligence to robots
Gemini Robotics 2 brings whole body intelligence to robots
Read this before you buy that TV streaming stick
Read this before you buy that TV streaming stick
AI's top startups are barely publishing their research
UEFA and its national associations will not participate in FIFA competitions
Zig's Incremental Compilation Internals
Delayed Gratification – Proud to Be 'Last to Breaking News'
The coolest use for the Vision Pro
The iPhone Upgrade Program is being replaced by Apple Upgrade
<a href="https://www.apple.com/shop/apple-upgrade" rel="nofollow">https://www.apple.com/shop/apple-upgrade</a>
Document-borne AI worms can self-propagate through Copilot for Word
Kimi K3 Architecture Overview and Notes
Superlogical
<a href="https://mitchellh.com/writing/superlogical" rel="nofollow">https://mitchellh.com/writing/superlogical</a>
Superlogical
<a href="https://mitchellh.com/writing/superlogical" rel="nofollow">https://mitchellh.com/writing/superlogical</a>
Show HN: Open-source engine running Gemma 4 26B in 2 GB RAM on any M-series Mac
Hi HN,<p>I built a specialized inference engine for running 4-bit Gemma 4 26B-A4B-IT on any M-series Mac using about 2 GB of RAM. It is called TurboFieldfare and is written in Swift and Metal.<p>I have always adored on-device AI. It feels like magic that you can run a powerful NN on your Mac or iPhone. So I wanted to push the limits a bit and run a model whose weights don’t fit in memory.<p>The model’s 4-bit quantized weights occupy roughly 14 GB, which makes running it with conventional inference tools almost impossible on an 8 GB or even 16 GB Mac once the OS, applications, and KV cache are included.<p>The trick is to keep the shared part of the model and the KV cache in RAM, then stream only the routed experts needed for each token from SSD. An SSD is way slower than RAM, so the runtime uses a small expert cache and bounded parallel `pread`. While those reads are in flight, the GPU runs the shared part of the layer.<p>I ran more than 100 experiments. Most didn’t work. A few got me here. The experiments are described in the GitHub repo.<p>It currently generates 5–6 tok/s on an 8 GB M2 MacBook Air and 31–35 tok/s on an M5 MacBook Pro.<p>I also added an experimental OpenAI-compatible local server. It supports streaming and tool calls, and reuses one prompt prefix from the KV cache.<p>Try it! The Mac app is easy to install. On the first run, it will download 15 GB of weights from Hugging Face. The model is surprisingly capable.<p>I would love any kind of feedback!
Show HN: Open-source engine running Gemma 4 26B in 2 GB RAM on any M-series Mac
Hi HN,<p>I built a specialized inference engine for running 4-bit Gemma 4 26B-A4B-IT on any M-series Mac using about 2 GB of RAM. It is called TurboFieldfare and is written in Swift and Metal.<p>I have always adored on-device AI. It feels like magic that you can run a powerful NN on your Mac or iPhone. So I wanted to push the limits a bit and run a model whose weights don’t fit in memory.<p>The model’s 4-bit quantized weights occupy roughly 14 GB, which makes running it with conventional inference tools almost impossible on an 8 GB or even 16 GB Mac once the OS, applications, and KV cache are included.<p>The trick is to keep the shared part of the model and the KV cache in RAM, then stream only the routed experts needed for each token from SSD. An SSD is way slower than RAM, so the runtime uses a small expert cache and bounded parallel `pread`. While those reads are in flight, the GPU runs the shared part of the layer.<p>I ran more than 100 experiments. Most didn’t work. A few got me here. The experiments are described in the GitHub repo.<p>It currently generates 5–6 tok/s on an 8 GB M2 MacBook Air and 31–35 tok/s on an M5 MacBook Pro.<p>I also added an experimental OpenAI-compatible local server. It supports streaming and tool calls, and reuses one prompt prefix from the KV cache.<p>Try it! The Mac app is easy to install. On the first run, it will download 15 GB of weights from Hugging Face. The model is surprisingly capable.<p>I would love any kind of feedback!
KOReader
KOReader
A walk through of the DeltaNet family of linear attention variants