Feed Entities Podcasts Deep Dives Developers About

[KongchangAI]

Real-time AI knowledge base. Every claim is traceable to its source, with data freshness under 24 hours.

Powered by 240+ first-hand AI sources

Knowledge

Feed
Hot Entities
Podcasts

Developers

CLI
MCP Server
REST API
About

© 2026 KongchangAI. The freshest, most accurate AI knowledge

Unverified50% confidenceSolutionExact time

量化感知训练(QAT)让模型保留全精度权重副本，每次前向传播练习舍入操作，学会在极端压缩下的权重分布

1

Sources

50%

Confidence

Long-term

Relevance

7/21/2026

First Seen

Sources

27B大模型跑进iPhone：1-bit量化如何实现极限压缩

Related Claims

VerifiedQLoRA将基础模型量化为4位精度存储，同时在量化后的模型上应用LoRA适配器进行训练80% similar Unverified三值量化训练需引入量化感知训练(QAT)，因{-1,0,+1}离散化不可微，通常采用直通估计器(STE)在反向传播中近似梯度77% similar UnverifiedQLoRA论文提出的NF4针对预训练权重服从正态分布的特性设计了最优量化格点74% similar UnverifiedQwen3采用四阶段训练流程，在后期专门针对混合推理能力进行对齐微调74% similar Unverified量化感知训练在前向传播时插入伪量化节点模拟量化噪声，反向传播时因取整操作不可微而使用STE近似梯度73% similar

Cite This Claim

Stable URI

https://kongchang.com/claim/577363

API

curl https://kongchang.com/api/v1/knowledge/claims/577363

MCP

get_claim(id=577363)