Keynote: How WebAssembly can Power the New Wave of Serverless AI

on YouTube ↗08m

" summary

WebAssembly components compile inferencing runtimes down to 2.3 megabyte Wasm components that start in microseconds, avoiding container cold starts that commonly take 30 seconds to minutes. The stack shown combines Spin and Fermion Cloud plus a model-loader API (Waznn) to load large models near GPUs, enabling fractional GPU allocation just-in-time and the same artifact to run on M1, CPU, CUDA GPU, or cluster. Fermion ships a serverless AI beta and open-source Spin examples.

AI-written summary · about

" chapters (6)

1. 0:00Introduction to the speaker
2. 1:32The state of serverless AI
3. 3:04Challenges with containers
4. 4:02WebAssembly efficiency
5. 5:46Hardware-agnostic runtime
6. 6:23Implementation and tools

" related talks

• 25mBuilding Serverless AI Apps with Spin and WebAssemblyRadu Matei, Matt Butcher · Cloud Native AI Day Europe 2024
• 31m"Serverless AI Inferencing Using Python and WebAssembly"Shivay Lamba · PyOhio 2024
• 24mA Greener, Cost Effective Cloud with Serverless WebAssemblyKate Goldenring · DevOpsDays Seattle 2024
• 01h 34mTutorial: Getting Started with Serverless WebAssembly and SpinMikkel Mørk Hegnhøj · Open Source Summit Europe 2023
• 44mA Greener, Cost-Effective Cloud with Serverless WebAssemblyKate Goldenring · SustainabilityCon @ Open Source Summit North America 2024
• 41mRevolutionizing Cloud Native: WebAssembly's Role in Serverless...Kate Goldenring, MacKenzie Adam · ContainerCon @ Open Source Summit North America 2024
• 30mA Greener, Cost-Effective Cloud with Serverless WebAssemblySohan Maheshwar · KubeDay India 2023
• 01h 59mCreate a First AI Program with TypeScript and Serverless WebAssemblyCaleb Schoepp, Justin Pflueger · KCD Los Angeles 2024