TensorFlow Serving
E457353
TensorFlow ecosystem component
machine learning infrastructure software
model serving system
open-source software
TensorFlow Serving is a flexible, high-performance system for deploying and serving machine learning models in production, particularly those built with TensorFlow.
Statements (51)
| Predicate | Object |
|---|---|
| instanceOf |
TensorFlow ecosystem component
ⓘ
machine learning infrastructure software ⓘ model serving system ⓘ open-source software ⓘ |
| architecture | servable-based modular architecture ⓘ |
| component |
Loader
ⓘ
Manager ⓘ ModelServer NERFINISHED ⓘ Servable ⓘ Source ⓘ |
| deploymentModel |
Docker container
ⓘ
Kubernetes NERFINISHED ⓘ cloud virtual machines ⓘ on-premises servers ⓘ |
| developer | Google ⓘ |
| documentation | https://www.tensorflow.org/tfx/guide/serving ⓘ |
| feature |
A/B testing support via multiple model versions
ⓘ
CPU-only serving support ⓘ GPU acceleration support ⓘ batching of inference requests ⓘ canary model deployment ⓘ dynamic model configuration ⓘ high-performance inference serving ⓘ hot model swapping without downtime ⓘ model lifecycle management ⓘ model rollback ⓘ monitoring hooks via custom servables ⓘ multi-model serving ⓘ production model deployment ⓘ versioned model management ⓘ |
| goal | provide flexible, high-performance serving of machine learning models in production ⓘ |
| license | Apache License 2.0 ⓘ |
| optimizedFor | TensorFlow models in SavedModel format ⓘ |
| partOf | TensorFlow NERFINISHED ⓘ |
| programmingLanguage |
C++
ⓘ
Python ⓘ |
| repository | https://github.com/tensorflow/serving ⓘ |
| supportsFramework |
Keras
NERFINISHED
ⓘ
TensorFlow NERFINISHED ⓘ |
| supportsLanguageBinding |
REST
ⓘ
gRPC NERFINISHED ⓘ |
| supportsModelFormat |
SavedModel
NERFINISHED
ⓘ
TensorFlow Hub module (via SavedModel) NERFINISHED ⓘ |
| supportsPlatform |
Docker-compatible platforms
ⓘ
Linux ⓘ |
| supportsProtocol |
HTTP/JSON
ⓘ
gRPC binary protocol ⓘ |
| useCase |
large-scale production ML systems
ⓘ
microservice-based ML APIs ⓘ online prediction ⓘ real-time inference ⓘ |
Referenced by (1)
Full triples — surface form annotated when it differs from this entity's canonical label.