Attention mechanism optimization (RoPE, Flash Attention)
Discussed in 1 analyzed podcast episode across 1 show
Discussed On
Episodes
张小珺Jùn｜商业访谈录 · Jul 28, 2026
148. 对游凯超3小时访谈：开源Infra、和模型Co-design 、“如果vLLM失败，我们会后悔一辈子”
vLLM architecture and PagedAttention algorithmModel and infrastructure co-design principlesOpen-source project governance and community managementInference optimization techniques (speculative decoding, continuous batching)Mixture-of-Experts (MoE) model architecture
View Analysis