Unverified50% confidenceFactExact time
SwiGLU将Swish激活函数与门控线性单元（GLU）结合，Google在PaLM模型中首先大规模应用，实验表明在相同参数量下比ReLU、GeLU等激活函数带来约8%的性能提升
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
UnverifiedIn the LLM field, GELU (Gaussian Error Linear Unit) and SwiGLU are becoming the mainstream activation function choices.64% similarUnverifiedSGLang 引入了 RadixAttention 技术，在某些基准测试中推理速度可比 vLLM 等主流推理框架提升 2-5 倍61% similarUnverifiedThe industry best practice is to combine fine-tuning with RAG—first fine-tuning to give the model domain understanding capability, then supplementing the latest dynamic knowledge through RAG.59% similarUnverifiedjlens-gguf supports both Dense models and Mixture-of-Experts (MoE) models.58% similarUnverifiedApproaches that use LLMs for post-correction of ASR outputs have already shown significant performance improvements in experiments.57% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/502354API
curl https://kongchang.com/api/v1/knowledge/claims/502354MCP
get_claim(id=502354)