Manage Cloud Native LLM Workloads Across Edge and Cloud Seamlessly Using KubeEdge and WasmEdge
Talkat Cloud Native + Kubernetes AI Day Europe 2025 · on Tuesday, 2025-04-01· at 11:15· Level 1 | Hall Entrance N10 | Room G·
summary
Edge nodes run tiny models locally and automatically escalate to cloud models when accuracy drops, using WasmEdge runtime (<50MB) for portable, dependency‑free inference and KubeEdge + Sedna orchestration to monitor confidence and route requests. Demo runs a ~1.05B LLM on CPU edge, forwards low‑confidence helmet detections to a larger cloud model, and shows Rust workloads compiled to Rust→Wasm binaries that deploy unchanged across devices. Solution is production‑tested in aerospace and bank branches.
AI-written summary · about
chapters (12)
- 0:00Cloud-native LLM workloads
- 1:04Inference workloads at edge
- 3:15Edge AI inference challenges
- 5:01KubeEdge and Wasm runtime
- 6:24Wasm runtime deep dive
- 10:37KubeEdge architecture
- 12:04Sedna for cloud-edge AI
- 13:33Cloud-edge joint inference
- 15:43LLM workloads with KubeEdge
- 16:25Demo: Deploy LLM to edge
- 20:28Demo: Cloud-edge joint inference
- 24:39Q&A
related talks
33mEfficient and Cross-Platform LLM Inference in the Heterogenous Cloud
32mFrom Ground to Space: Cloud-Native Edge Machine-Learning Case Studies with KubeEdge-Sedna
25mLeveraging Wasm for Portable AI Inference Across GPUs, CPUs, OS & Cloud-Nativ...
21mKubeEdge Powered Edge-Devices With Next Generation Cloud Native Runtime
55mFast And Portable LLM Inference With WebAssembly And Rust
48mWorkshop: Efficient and Portable AI / LLM Inference on the Edge Cloud
33mWebAssembly based AI as a Service with Kubernetes
32mWebAssembly Based AI as a Service on the Edge with Kubernetes