on-device AI
6 articles tagged with this topic
Meta Puts a 5.3B-Parameter Model Into 3GB of Phone Memory—For Research Only
Meta releases MobileMoE: 5.3B total parameters, under 1B active, and under 3GB at INT4—but its FAIR NC license bars commercial use.
Qwen 4B Reasoning Jumps 17% With Almost No Size Increase
ByteOtter's QLAB method lifts Qwen 3.5 4B reasoning from 46.875 to 54.688 at IQ2_XS—a 16.67% gain with only 0.4% size increase.
Horizon J6 Runs Full YOLOv5 — Auto-Grade AI Chip Toolchain Closes Its Last Gap
Horizon's Journey J6 now has a complete YOLOv5 deployment tutorial — signaling domestic auto-grade AI chip toolchains are closing the usability gap
Meta Crams AI Models Into Your Phone — But Is On-Device Really Worth It?
Meta open-sourced two phone-runnable models, Muse Glimmer and Muse Spark 1.2. We ask: is local inference actually cheaper than cloud APIs?
Liquid's 2.6B Model Pushes Local AI Closer to the Mainstream
Liquid AI ships a 2.6B parameter model hitting 260 tokens/sec on an RTX 3090. Limited capability, but enough for daily chores — and it lowers the bar
手机本地跑 AI 不再需要联网—— 一个开源安卓应用正在把这件事变得可操作
Pocket LLM v 1.4.0 shrinks to ~200MB, lets users download models on demand and run AI fully offline on Android.