Unverified50% confidenceBenchmarkExact time
文章称IQ3_S档位通常能将准确率保留提升到90%以上、体积约13-15GB，Q4_K_M能保留95%以上准确率、体积约16-18GB
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
8/25/2026
First Seen
Valid until: 11/23/2026
Sources
Related Entities
Related Claims
UnverifiedUD-Q4_K_XL量化版本体积155GB，Top-1精度为96%76% similarUnverifiedIQ4_XS量化文件大小15.57GB，pp2048为95.4 tok/s，tg128为10.9 tok/s75% similarUnverifiedMeta Llama 3的405B参数版本以BF16精度加载需约810GB内存，量化至INT4精度需约200GB75% similarUnverifiedLlama 3 70B量化版（Q4精度）仍需约35GB内存71% similarUnverifiedThe FP16 version of Gemma 3 12B requires approximately 24 GB of VRAM, while the Q4_K_M version reduces this to approximately 7–8 GB, with inference speed improving by approximately 2–3x and structured output accuracy losing only approximately 2–5%.70% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/798072API
curl https://kongchang.com/api/v1/knowledge/claims/798072MCP
get_claim(id=798072)