Unverified80% confidenceFactTime unknown
SVDQuant中低秩分支的参数量通常只占总参数的几个百分点
1
Sources
80%
Confidence
Long-term
Relevance
6/1/2026
First Seen
Sources
SVDQuant：4-bit量化让扩散模型在消费级GPU上高效运行
githubnunchaku-ai
Related Entities
Related Claims
Unverified单条推理通常只能占用GPU SM利用率的5%～15%，而合理的批次大小可将利用率提升至60%～90%64% similarUnverified轻量版模型单次推理的算力消耗往往只有旗舰版本的十分之一甚至更低63% similarUnverified马尔可夫头仅需记录前一个token状态即可推导下一token修正概率，参数量通常不超过原模型的0.1%62% similarUnverifiedPER中重要性采样权重的β参数从较小值（如0.4）线性退火到161% similarUnverified轻量换来的代价是耐用性和抗穿刺性下降，口碑中扎胎和慢撒气的反馈明显多于标准版SV1561% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/28970API
curl https://kongchang.com/api/v1/knowledge/claims/28970MCP
get_claim(id=28970)