Strategii de deployment al modelelor

MLOps Deployment and Life Cycling

Nemanja Radojkovic

Senior Machine Learning Engineer

Deployment reușit!

  • Aplicația ML rulează
  • API-ul gestionează mii de cereri/oră
  • O lună mai târziu:
    • Caracteristici noi și mai bune identificate
    • Lot nou de date de antrenare colectat
    • Pipeline-ul de construire a modelului rulat
    • Pachet nou de model creat
MLOps Deployment and Life Cycling

înlocuire simplă

MLOps Deployment and Life Cycling

predicție offline

MLOps Deployment and Life Cycling

fereastră de modificare

MLOps Deployment and Life Cycling

întrerupere în timp real

MLOps Deployment and Life Cycling

întrerupere costisitoare

MLOps Deployment and Life Cycling

configurație blue-green

MLOps Deployment and Life Cycling

un singur clic

MLOps Deployment and Life Cycling

termenul blue-green

MLOps Deployment and Life Cycling

avantaje și dezavantaje

MLOps Deployment and Life Cycling

rollback

MLOps Deployment and Life Cycling

deployment canary

MLOps Deployment and Life Cycling

divizarea cererilor

MLOps Deployment and Life Cycling

etapa a doua

MLOps Deployment and Life Cycling

rezultat final

MLOps Deployment and Life Cycling

MLOps Deployment and Life Cycling

validare

MLOps Deployment and Life Cycling

dezavantaj

MLOps Deployment and Life Cycling

reducerea riscurilor

MLOps Deployment and Life Cycling

Să exersăm!

MLOps Deployment and Life Cycling

Preparing Video For Download...