vLLM tuning should not stop at checking whether the API returns an answer. TTFT, TPOT, throughput, KV Cache usage, queue length, and tail latency form the real feedback loop.
vLLM 调优不能只看服务是否返回答案，而要围绕 TTFT、TPOT、吞吐量、KV Cache 使用率和尾延迟建立监控闭环。本文用本地脚本串起启动、测量、压测和指标采集。
Speculative decoding does not let the model guess blindly. A cheap draft model proposes tokens, while the target model verifies them in parallel. The real question is whether acceptance rate and batch size make it worthwhile.
推测解码不是让大模型乱猜，而是用草稿模型先给候选，再由目标模型一次性验证。理解 Medusa、EAGLE 和大 Batch 的关系，才能判断它在生产高并发里是否真的划算。
LuminaRAG is not positioned as another RAG demo. Its real value is treating enterprise AI search as governed, auditable knowledge infrastructure.
LuminaRAG 不是一个只追求问答效果的 RAG Demo，而是把文档审核、权限隔离、混合检索、引用来源和内网部署放在一起思考的企业知识库开源模板。
Many Chinese companies accept the need for digital transformation, yet execution keeps stalling. The real friction is not technology, but process ownership, transparency, and organizational risk.
国内企业拥抱数字化转型的口号很多，真正执行却常常迟缓。我的判断是，矛盾不在技术方案，而在流程权责、经营透明度和组织风险重新分配。
In high-concurrency LLM serving, model weights are rarely the only memory problem. KV Cache grows with requests and tokens, so PagedAttention and Continuous Batching are the real keys to vLLM throughput.
高并发大模型服务真正容易爆掉的不是模型参数，而是随请求增长的 KV Cache。理解 PagedAttention 和 Continuous Batching，才能看懂 vLLM 为什么能把吞吐做上去。
