XNews

Cactus Needle 2 Packs 45M-Parameter Model into 14MB for On-Device AI

Cactus Needle 2 Packs 45M-Parameter Model into 14MB for On-Device AI

Cactus Needle 2 compresses a 45M-parameter model into a 14MB file using 28MB memory. It handles tool instructions, device operation, and structured data extraction. Output is syntax-constrained, and confidence scores route uncertain cases to cloud models. It supports Python install, function description tool calling, and lightweight fine-tuning.

Source posts