Unverified50% confidenceBenchmarkExact time
IQ4_XS量化文件大小15.57GB，pp2048为95.4 tok/s，tg128为10.9 tok/s
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
8/21/2026
First Seen
Valid until: 11/19/2026
Sources
Related Claims
UnverifiedQ4_K_M量化文件大小17.77GB，pp2048速度96.8 tok/s，tg128速度11.4 tok/s78% similarUnverified使用IQ4_XS配合-fa 1 -ctk q8_0 -ctv q8_0对KV缓存进行q8_0量化后，可加载完整32k上下文，常驻内存约16.6GB71% similarUnverifiedUD-Q4_K_XL量化版本体积155GB，Top-1精度为96%71% similarUnverifiedIQ2_XS可以将模型压缩到接近2 bit/权重，使原本需要48GB显存的70B模型能塞入24GB消费级显卡70% similarUnverifiedRTX 5090上Qwen3.6 35B-A3B推理速度达220 token/s，显存占用约23GB70% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/781391API
curl https://kongchang.com/api/v1/knowledge/claims/781391MCP
get_claim(id=781391)