Unverified50% confidenceFactExact time

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.

1

Sources

50%

Confidence

Medium-term (~90 days)

Relevance

9/17/2026

First Seen

Valid until: 12/16/2026

Sources

Bonsai 2 27B: Exploring Near-Lossless Model Compression at 9x Smaller Footprint

Related Entities

Bonsai 2 27Bproduct

Related Claims

Unverified1-bit Bonsai的智能密度得分为0.53/GB，是全精度模型的10倍以上，是最优低比特方案的约2.7倍65% similar VerifiedINT8 quantization can compress model size to 1/2 to 1/4 of the original.65% similar Unverifiedllama.cpp通过INT4/INT8量化将模型权重压缩至原始大小的25%以下62% similar Unverified压缩技术LLMLingua研究表明可在压缩80%内容的情况下保持模型性能基本稳定60% similar

Cite This Claim

Stable URI

https://kongchang.com/claim/930398

API

curl https://kongchang.com/api/v1/knowledge/claims/930398

MCP

get_claim(id=930398)