Unverified50% confidenceFactExact time
ChatGPT、DeepSeek、通义千问、豆包、文心、Claude、Gemini等所有主流大模型均基于Transformer训练而来
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
7/20/2026
First Seen
Valid until: 10/18/2026
Sources
Related Claims
UnverifiedTransformer的核心创新是自注意力机制，GPT、Claude、Gemini等主流大模型均基于Transformer构建76% similarUnverifiedBERT、GPT系列、T5等预训练大模型均以Transformer为基础架构，形成了预训练+微调的主流范式74% similarUnverifiedGPT系列、LLaMA、Gemini等主流模型均基于Transformer的解码器（Decoder-only）变体71% similarVerified现代大模型（如GPT系列、Qwen、DeepSeek）普遍采用Transformer的Decoder-only变体71% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/571310API
curl https://kongchang.com/api/v1/knowledge/claims/571310MCP
get_claim(id=571310)