Unverified50% confidenceFactExact time
Bonsai将27B模型从全精度的54GB压缩至3.9GB
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
7/20/2026
First Seen
Valid until: 10/18/2026
Sources
Related Claims
Unverified原版QW3.6 27B模型文件体积约为54GB，Prismo通过数学压缩技术压缩至不到4GB，压缩比超过13倍，且声称不影响模型性能80% similarUnverifiedBonsai 27B是首个可在手机端运行的27B级大模型，通过量化技术将54GB模型压缩至3.8GB，体积缩减约93%，保留约90%智能水平，可在iPhone 17 Pro上本地运行78% similarUnverifiedQ3_K_M将每个权重压缩至约3bit，27B模型权重体积可降至约12～13GB75% similarUnverified70B模型在INT4精度下可压缩至约35GB72% similarUnverifiedBonsai三元版为1.71 bits约6GB，保留基座模型约97%质量；1-bit版为3.9GB，保留约90%质量71% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/569975API
curl https://kongchang.com/api/v1/knowledge/claims/569975MCP
get_claim(id=569975)