Unverified50% confidenceFactExact time
GPT-2引入了Pre-LN（前置层归一化）设计，相比原始Transformer的Post-LN，将LayerNorm置于残差块内部的子层之前，改善了深层网络的梯度流动稳定性
1
Sources
50%
Confidence
Long-term
Relevance
7/15/2026
First Seen
Sources
Related Claims
UnverifiedTransformer 取代了此前主流的 RNN/LSTM 序列模型，成为 GPT、LLaMA、Claude 等主流大模型的基础结构72% similarVerifiedGPT stands for 'Generative Pre-trained Transformer'.68% similarUnverifiedModern LLMs (such as the GPT series, Claude, and Llama) are all based on the Transformer architecture, and their decoding process is essentially a Markov chain.67% similarPartially VerifiedNex N2's output style is highly similar to GPT models, suggesting that GPT-style outputs were distilled into it during the post-training stage.67% similarUnverified早期GLM采用了自回归填空（Autoregressive Blank Infilling）架构，与GPT系列的纯自回归架构不同65% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/518116API
curl https://kongchang.com/api/v1/knowledge/claims/518116MCP
get_claim(id=518116)