Unverified50% confidenceFactExact time

RMSNorm相比原始LayerNorm省去均值中心化步骤，将归一化计算量降低约40%

1

Sources

50%

Confidence

Long-term

Relevance

7/13/2026

First Seen

Sources

不到350元训练亿参数模型：SmoLLM从零构建全记录

redditr/learnmachinelearning7/12/2026

Related Claims

Verified量化技术可将模型体积压缩50%-75%，在损失极小精度前提下降低显存占用70% similar Unverified一项针对GPT-4的研究显示，对MMLU题目进行语义等价改写后，模型准确率在部分子集上下降超过10个百分点68% similar Unverified通过INT8或INT4量化，可将模型显存占用降低50%至75%68% similar Unverifiedint8量化将参数从16位压缩至8位，理论上减少约50%显存占用；int4将显存占用降低至原始的25%左右67% similar Unverified推理单位算力成本以每年40%至60%的速率下降67% similar

Cite This Claim

Stable URI

https://kongchang.com/claim/502353

API

curl https://kongchang.com/api/v1/knowledge/claims/502353

MCP

get_claim(id=502353)