Samsung Shrinks 13B LLM to Under 1 GB for Local AI

Samsung developed LittleBit, an open-source compression method that reduces a 13B parameter large language model to under 1 GB of memory. The technique replaces floating-point multiplication with bitwise XOR operations and uses latent factorization to reduce weights to as low as 0.1 bits per weight. It provides an 11.6x inference speedup compared to standard FP16 models and allows state-of-the-art AI to run locally on resource-constrained devices.