Bonsai 27B

Bonsai 27B is a highly compressed variant of the qwen 27B language model, developed by PrismML. It is designed to run on consumer-grade hardware by utilizing advanced compression techniques that reduce memory requirements by approximately 10x compared to the base model.

Key Characteristics

  • Base Model: Derived from qwen 3.6 27B (referred to as Quine 3 27B in some contexts).
  • Memory Efficiency: Achieves ~10x memory reduction, enabling deployment on standard consumer devices.
  • Developer: PrismML.
  • Purpose: Democratizing access to powerful LLMs by lowering the hardware barrier for inference.

Performance Benchmarks

References