Unverified50% confidenceFactExact time

DistilRoBERTa是RoBERTa的蒸馏版本，参数量减少40%但保留97%的性能

1

Sources

50%

Confidence

Long-term

Relevance

8/22/2026

First Seen

Sources

文本态度分析：从情感到认知的NLP模型实战指南

Related Entities

DistilRoBERTatechnology RoBERTatechnology

Related Claims

UnverifiedDistilBERT通过蒸馏将BERT参数量减少40%，推理速度提升60%，性能仅损失约3%75% similar UnverifiedDistilBERT将BERT体积压缩40%同时保留97%性能70% similar Verified模型蒸馏可在保留70-80%原始性能的前提下将参数量压缩至原来的1/1063% similar Verified如果每步准确率为95%，经过20步循环后整体准确率会降至约36%62% similar UnverifiedDistilBERT将参数量压缩至6600万（原版1.1亿），推理速度提升60%，精度损失不足3%61% similar

Cite This Claim

Stable URI

https://kongchang.com/claim/787175

API

curl https://kongchang.com/api/v1/knowledge/claims/787175

MCP

get_claim(id=787175)