Paddle Serving roadmap

General Serving

• Integrate with Paddle seamlessly, and make most models trained with Paddle deployable with Serving framework.
• Support key-value features.
• Support several applications, including bert-as-a-service, image-semantic-vector-service, LAC, IMDB, CTR on criteo
• Benchmark on serveral tasks.

Single GPU Card with Multiple Models

• Load multiple models with GPU, and limit the GPU memory.

Deployment with EDL on Kubernetes

• deploy with docker on Kubernetes and support elastic scheduling.

Model Serving Cloud

• Release lots of model service that can be easy-to-use.