v0.0.1

What's Changed

• Pod Lifecycle(AI): Pod startup speed optimization, cold-start, sleep mode, and offloading.
• DRA updates: NVIDIA GPU Operator and DRA Driver, NRI
• Workload solutions(P/D disaggregation): LWS, SGLang RBG, AIBrix StormService, Kthena, KServe, Dynamo, vllm Production Stackm OME.
• KV Cache comparison: NIXL, LMCache, Mooncake
• Scheduling: Volcano, NVIDIA Grove, Kueue, Godel, Koordinator, HAMI, KAI Scheduler.
• Gateway: Envoy AI Gateway, Semantic Router, KGateway, Kong.
• Performance testing and benchmarking tools
• Community Update: AI Conformance, Kubernetes workgroups and CNCF tags/initiatives.

More

• Large Scale Experts (MoE)
• AIConfigurator
• Observability
• Training on Kubernetes: Kubeflow Trainer V2 and ArgoCD ; GPU checkpoint/restore
• Serverless, Knative
• AI workload isolation
• parallelism
• pre-training

Full Changelog: https://github.com/pacoxu/AI-Infra/commits/v0.0.1

初步成形

• 目前缺少一些基础的模型AI知识
• 另外训练内容可能相对较少
• 缺少中文
• landscape 比较粗糙

但是

• AI workloads 编排管理
• 现有相关的项目（不包含更高一层的agent 内容）
• 云原生方向为主

基本还是覆盖了的