Unverified50% confidenceBenchmarkExact time
在128K上下文下,Q2_0输出65.1 tok/s、IQ2_XS输出52.0 tok/s、IQ3_XXS输出44.8 tok/s
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
9/24/2026
First Seen
Valid until: 12/23/2026
Sources
Related Entities
Related Claims
UnverifiedIQ2_XXS以约2.06位的平均比特率编码权重,IQ3_S约3.44位,Q4_K_M约4.83位72% similarUnverifiedUD-Q4_K_XL体积155GB精度96%,UD-Q3_K_XL体积129GB精度87%,UD-IQ3_S体积117GB精度83%,UD-Q2_K_XL体积97GB精度79%68% similarUnverifiedQwen2.5-Max-0902最大输出为13万Token,推理模式下上限可达26万Token63% similarUnverifiedQ3_XXS模型搭配F16 KV缓存needle测试得分约80%,而Q3_K_XL配合kvarn3 KV缓存得分接近100%62% similarUnverifiedQwen2.5-72B 在多项基准测试中接近 Llama-3.1-405B 的表现62% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/948583API
curl https://kongchang.com/api/v1/knowledge/claims/948583MCP
get_claim(id=948583)