Bonsai 2 27B
A compressed large language model with 27 billion parameters that claims to achieve a 9x storage compression ratio through a combination of techniques such as quantization and pruning, while maintaining near-lossless model capability. It is designed for low-resource deployment scenarios.
Timeline (last 90 days)
The Bonsai 2 27B non-thinking configuration achieves 12/16, and the Medium thinking configuration achieves 15/16 (both substantive and strict), which is the most impressive result.
The core claim of the Bonsai 2 27B project is to compress the model size to one-ninth of the original while keeping the model's capabilities nearly intact.
9x compression means the storage and memory footprint of the model is reduced to approximately 11% of its original size.
The publicly available information on Bonsai 2 27B is currently quite limited, remaining mainly at the level of core claims, and lacking detailed technical documentation, evaluation data, and independent community reproduction.
All Facts (3)
The Bonsai 2 27B non-thinking configuration achieves 12/16, and the Medium thinking configuration achieves 15/16 (both substantive and strict), which is the most impressive result.
50%UnverifiedThe core claim of the Bonsai 2 27B project is to compress the model size to one-ninth of the original while keeping the model's capabilities nearly intact.
50%Unverified9x compression means the storage and memory footprint of the model is reduced to approximately 11% of its original size.
50%