Show HN: Cactus Needle 3: 8-29MB automation models can match DeepSeek V4 Flash
Thread
Unofficial Hacker News client; not affiliated with Y Combinator.
Show HN: Cactus Needle 3: 8-29MB automation models can match DeepSeek V4 Flash
Unofficial Hacker News client; not affiliated with Y Combinator.
Tsarp · · focus · HN ↗
Are there perhaps some industrial or agri use cases?
HenryNdubuaku · · focus · HN ↗
So an 8-27B on a LAN box wins for generic tasks on hardware that can hold it. Needle is for hardware that can't, like plain ARMv7, MIPS32 (the Ingenic chips in cheap IP cameras), RISC-V and watches. Also, we found cost to not really be the lever for on-device models, but availability and latency.
ash_091 · · focus · HN ↗
All of the demo setups (smart home, robot vacuum, watch, etc) could easily have access to a bigger model running on a more capable device either locally or via the internet.