Unverified60% confidenceFactExact time
V4 Flash的缓存大小只有V3.2的7%
2
Sources
60%
Confidence
Long-term
Relevance
9/11/2026
First Seen
Sources
Related Entities
Related Claims
Unverified经过压缩后，DeepSeek V4 Flash模型体积从接近500GB降至80.8GB73% similarUnverified从V1到V4.1 Flash，单Token缓存从389000字节压缩到890字节，总共压缩了437倍71% similarPartially VerifiedQwen3 Next Flash的4-bit量化版本（Q4K XL）约111GB70% similarUnverifiedDeepSeek V4 Flash的UD-Q8_K_XL量化版本文件体积为162GB，采用MXFP4+BF16混合精度，达到完全无损69% similarUnverified在双Spark系统上，DeepSeek V4 Flash（FP8量化）峰值约每秒53个Token69% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/900397API
curl https://kongchang.com/api/v1/knowledge/claims/900397MCP
get_claim(id=900397)