Upgrade to Pro — share decks privately, control downloads, hide ads and more …

Cloud-Native Generative AI mit Fermyon Serverle...

Thorsten Hans
September 28, 2023

Cloud-Native Generative AI mit Fermyon Serverless AI

Dank generativer künstlicher Intelligenz (Gen AI) können wir Anwendungen intelligenter machen. Aber wie integriere ich KI überhaupt in eine Cloud-Native-Anwendung? Welche Möglichkeiten gibt es? Und warum ist das junge Tech-Startup Fermyon mit Serverless AI gerade so gehyped? Diese und weitere Fragen beantwortet Thorsten Hans in seiner Session und zeigt, wie Sie mit Fermyon Serverless AI und Llama2 einfache Szenarien im Handumdrehen abbilden können.

Thorsten Hans

September 28, 2023
Tweet

More Decks by Thorsten Hans

Other Decks in Technology

Transcript

  1. Cloud-Native Business Applications Day Uhrzeit Titel Sprecher 09:00 – 10:00

    Cloud-Native-all-the-Things: Definition, Praktiken und Patterns Christian Weyer Thorsten Hans 10:30 – 11:30 Containerbasierte Entwicklung für .NET-Entwickler Tobias Fenster 12:00 – 13:00 Cloud-Native Generative AI mit Fermyon Serverless AI Thorsten Hans 15:15 – 16:15 Cloud-Native Microservices, on-premises oder in der Cloud – mit Dapr Christian Weyer 16:45 – 17:45 Mega Mergers: Cloud-Native-Architekturen mit Containern und WebAssembly Thorsten Hans
  2. • Intro • What is Fermyon Spin • Serverless AI

    with Fermyon Cloud • Conclusion Agenda
  3. • Intro • What is Fermyon Spin • Serverless AI

    with Fermyon Cloud • Conclusion Agenda
  4. • Cloud-vendor interest: • They can put more apps on

    a compute resource as today • Wasm and WASI give them a strict security and isolation model • Wasm workloads are way smaller than everything else • They can scale to zero due to super-fast bootstrapping < 1msec Intro Why will Wasm have such a big impact?
  5. • Developer interest: • They can use any language that

    compiles to wasm32_wasi • They can ship just the app • They can reduce cloud spendings • Same workloads will be cheaper because they consume less resources and execute faster Intro Why will Wasm have such a big impact?
  6. Wasm on the server relates to containers in the same

    way containers related to virtual machines 10+ years ago
  7. • Intro • What is Fermyon Spin • Serverless AI

    with Fermyon Cloud • Conclusion Agenda
  8. • Intro • What is Fermyon Spin • Serverless AI

    with Fermyon Cloud • Conclusion Agenda
  9. Fermyon Spin is: • A serverless runtime build using Wasm,

    WASI, and the WebAssembly Component Model (leveraging wasmtime internally) • A collection of SDKs for many popular languages • A super focussed developer tooling Intro Let’s get everybody on track! 🦀
  10. • Intro • What is Fermyon Spin • Serverless AI

    with Fermyon Cloud • Conclusion Agenda
  11. • Intro • What is Fermyon Spin • Serverless AI

    with Fermyon Cloud • Conclusion Agenda
  12. Give application developers sophisticated generative AI capabilities with no-ops and

    maintain developer productivity Serverless AI with Fermyon Cloud
  13. • Fermyon Serverless AI empowers developers to use AI inferencing

    in their apps with no additional setup • Encapsulate inferencing capabilities into two methods • Local developer story (independent from your hardware / operating system) Serverless AI with Fermyon Cloud
  14. • Frictionless AI offering by Fermyon • Execute inferencing against

    LLMs (currently Llama 2 and CodeLlama with 13b parameter variants) with no-ops • Generate sentence embeddings (all-minilm-l6-v2) for your data using a no-ops vector database • Support for additional models coming soon Serverless AI with Fermyon Cloud
  15. • Intro • What is Fermyon Spin • Serverless AI

    with Fermyon Cloud • Conclusion Agenda
  16. • Intro • What is Fermyon Spin • Serverless AI

    with Fermyon Cloud • Conclusion Agenda
  17. • With Spin, Fermyon demonstrates how WebAssembly will change the

    way we build software for the next wave of cloud-computing • Serverless AI allows application developers to add generative AI capabilities to their apps in “no-time” • Although we have access to Llama2 and CodeLlama now, we will see more models in Fermyon Cloud soon Conclusion