KURE-sparse-nano

0.1B Sparse Model Beats 0.6B: Local RAG Gets Lighter

A 0.1B SPLADE model outperformed Qwen3-Embedding-0.6B in a developer test, using under 1.5GB of VRAM.