Samsung open-sources LittleBit: 13B LLM under 1GB with sub-bit weights and XOR compute
dr_cintas · x · 2026-11-16
Samsung open-sourced LittleBit, a method that shrinks a 13B-parameter LLM to under 1GB. It uses latent factorization to compress weights below 1 bit each, with some setups reaching 0.1 bits per weight.
The compute side is the wild part: at this compression level most floating-point multiplications disappear — the core computation runs via a simple bitwise XOR (sign flips) instead of heavy matrix math.
Reported results: up to 11.6x faster inference than FP16, much lower memory and bandwidth, and it holds up below 0.5 bits where older compression methods completely break down. The author frames it as a roadmap for running large models locally on cheap, low-power hardware.
More from Infra
- Inside SGLang Summit 2026: 23 sessions in one track and a striking open-source turnout — ying11231 · 2026-11-16
- EPYC 32-core + 512GB DDR4 + 44GB VRAM: is a DIY local LLM rig still worth it? — sitytitan · 2026-11-16
- Forensic Analysis: Top Chinese AI Models Fingerprinted to Export-Banned NVIDIA Blackwell GPUs — elie · 2026-11-16
- Samsung to invest $1.9B in Austin to make chips for SpaceX Starlink satellites — elonmusk · 2026-11-16
- Jensen Huang rumored plotting return to Nvidia CEO as Musk, Sweeney back public letter — The Verge AI · 2026-11-16
- Open-source ComfyUI node offloads text encoders to a spare GPU on another machine — randomlyfixed · 2026-11-16