I built a tiny (332MB) CPU-friendly model for document sorting that actually knows when to say "none fits" (BeeNara)

A developer created BeeNara, a 332MB CPU-friendly model for document sorting, addressing a common issue with smaller LLMs like Qwen3.5-4B. These models often struggle to admit when a document doesn't fit a category, leading to hallucination and incorrect placement. BeeNara is designed to recognize when no category applies, preventing misfiling and offering a more efficient solution for local document organization without requiring a powerful GPU.

Unlike smaller models such as Qwen3.5-4B, BeeNara is specifically designed to identify when a document does not fit any category, preventing misclassification.