Unverified50% confidenceFactExact time
GPT-2采用纯解码器的Transformer堆叠结构，去掉了编码器-解码器的交叉注意力部分，专注于单向自回归语言建模
1
Sources
50%
Confidence
Long-term
Relevance
7/15/2026
First Seen
Sources
Related Claims
UnverifiedTransformer架构分化为BERT类（Encoder-only，适合理解任务）和GPT类（Decoder-only，适合生成任务）两大路线78% similarUnverified现代大语言模型（如GPT系列）几乎全部采用仅解码器（Decoder-only）架构，通过掩码自注意力确保自回归生成不泄露未来token信息77% similarUnverified现代LLM普遍采用的GPT架构本质上是仅使用Transformer解码器（Decoder-only）部分的自回归语言模型77% similarVerifiedGPT系列采用Transformer的解码器（Decoder-only）变体，通过自回归方式逐Token预测下一个词76% similarPartially VerifiedGPT系列采用了仅解码器（Decoder-only）架构74% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/514341API
curl https://kongchang.com/api/v1/knowledge/claims/514341MCP
get_claim(id=514341)