Unverified50% confidenceFactExact time
Q4_K_M is approximately 1/4 the size of the original model and is the mainstream choice for balancing quality and size.
1
Sources
50%
Confidence
Long-term
Relevance
9/2/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedThe Q4_K_M quantization scheme averages approximately 4.8 bits per weight, while the Q8_0 quantization scheme uses 8 bits per weight.79% similarVerifiedQ4_K_M中的K代表k-quants方案，M代表中等质量级别，是llama.cpp引入的混合量化策略75% similarVerifiedQ4量化将每个权重参数从16位压缩到4位，理论上可将模型体积缩小约75%74% similarUnverifiedQ4量化可将模型体积缩减至原来的约四分之一，MLX原生支持分组量化（Group Quantization）72% similarUnverifiedQ4_K方案使用4位量化，将权重分为超级块（256个权重）和子块，每个子块有独立缩放因子，超级块拥有更高精度的全局缩放因子72% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/845780API
curl https://kongchang.com/api/v1/knowledge/claims/845780MCP
get_claim(id=845780)