Skip to main content

Welcome to KServe

Deploy and scale AI models effortlessly - from cutting-edge Large Language Models to traditional ML models - with enterprise-grade reliability across any cloud or on-premises environment.

Why KServe?​

KServe eliminates the complexity of productionizing AI models. Whether you're a data scientist wanting to deploy your latest LLM experiment, a DevOps engineer building scalable ML infrastructure, or a decision maker evaluating AI platforms, KServe provides a unified solution that works across clouds and scales with your needs.

🚀 From Experiment to Production in Minutes - Deploy GenAI services and ML models with simple YAML configurations, no complex infrastructure setup required.

☁️ Cloud-Agnostic by Design - Run anywhere: AWS, Azure, GCP, on-premises, or hybrid environments with consistent behavior.

📈 Enterprise-Scale Ready - Automatically handle traffic spikes, scale to zero when idle, and manage hundreds of models efficiently.

What Makes KServe Different?​

GenAI-First Platform​

Deploy Large Language Models with OpenAI-compatible APIs out of the box. Chat completions, streaming responses, embeddings - all just work with your existing tools and SDKs.

Universal ML Support​

Beyond GenAI, serve any ML framework: TensorFlow, PyTorch, Scikit-Learn, XGBoost, and more. One platform for all your AI workloads.

Zero Infrastructure Overhead​

Focus on your models, not infrastructure. KServe handles load balancing, autoscaling, canary deployments, and monitoring automatically.

Production-Ready Security​

Enterprise authentication, network policies, and compliance features built-in. Deploy with confidence in regulated environments.

Key Benefits​

Generative Inference Benefits​

✅ LLM Multi-framework Support - Deploy LLMs from Hugging Face, vLLM, and custom generative models
✅ OpenAI-Compatible APIs - Chat completion, completion, streaming, and embedding endpoints
✅ LocalModelCache for LLMs - Cache large models locally to reduce startup time from 15-20 minutes to ~1 minute
✅ KV Cache Offloading - Optimized memory management for long conversations and large contexts
✅ Multi-node Inference - Distributed LLM serving ✅ Envoy AI Gateway Integration - Enterprise-grade API management and routing for AI workloads
✅ Metric-based Autoscaling - Scale based on token throughput, queue depth, and GPU utilization
✅ Advanced Generative Deployments - Canary rollouts and A/B testing for LLM experiments

Predictive Inference Benefits​

✅ Multi-framework Model Serving - Deploy models from TensorFlow, PyTorch, Scikit-Learn, XGBoost, and more
✅ InferenceGraph for Model Ensembles - Chain and ensemble multiple models together for complex workflows
✅ Batch Prediction Support - Efficient processing of large datasets with batch inference capabilities
✅ Preprocessing & Postprocessing - Built-in data transformation pipelines and feature engineering
✅ Real-time Scoring - Low-latency prediction serving for real-time applications
✅ Production ML Monitoring - Comprehensive observability, drift detection, and explainability
✅ Standard Inference Protocols - Support for Open Inference Protocol (V1/V2) across frameworks

Universal Benefits (Both Inference Types)​

✅ Serverless Inference Workloads - Automatic scaling including scale-to-zero on both CPU and GPU
✅ High Scalability & Density - Intelligent routing and density packing using ModelMesh
✅ Enterprise-Ready Operations - Production monitoring, logging, and observability out of the box

How It Works​

Simple Deployment Model: Describe your model requirements in a simple configuration file, and KServe handles the rest - from container orchestration to load balancing.

Built on Kubernetes: KServe extends Kubernetes with custom resources specifically designed for AI/ML workloads, providing cloud-native scalability and reliability while abstracting away the complexity.

Pluggable Architecture: Support for multiple serving runtimes means you can use the best inference engine for your specific model type - whether that's vLLM for LLMs, TorchServe for PyTorch models, or custom containers for specialized needs.

Architecture Overview​

KServe consists of two main components:

Control Plane​

  • InferenceService CRD - Manages model serving lifecycle
  • InferenceGraph CRD - Orchestrates model ensembles and chaining workflows
  • Serving Runtime - Pluggable model runtime implementations
  • ClusterServingRuntime: Define cluster-wide model runtimes
  • LocalModelCache CRD - Cache large models locally on nodes for faster startup and scaling
  • Model Storage - Support for various storage systems (Huggingface, S3, GCS, Azure, PVC, etc.)

Data Plane​

  • Predictor - Serves model predictions
  • Transformer - Pre/post processing logic
  • Explainer - Model explanations and interpretability

Quick Start​

Ready to deploy your first model? Choose your path:

🚀 Get Started with KServe​

Deploy your first GenAI service with Qwen LLM in minutes

🏗️ Installation Guide​

Set up KServe on your Kubernetes cluster

📚 Model Serving Guide​

Learn about different serving patterns and frameworks

Supported Model Frameworks​

Predictive Inference​

  • Scikit-Learn - Python-based ML models
  • XGBoost - Gradient boosting framework
  • TensorFlow - Deep learning models
  • PyTorch - PyTorch models via Triton
  • ONNX - Open Neural Network Exchange models
  • TRT - TensorRT optimized models
  • Hugging Face - Transformers and NLP models
  • MLflow - MLflow packaged models
  • Custom Runtimes - Bring your own serving logic

Generative Inference​

  • Large Language Models (LLMs) - Text generation via vLLM
  • Hugging Face Transformers - Text2Text generation
  • OpenAI-Compatible APIs - Chat completions, embeddings, and more

Multi-Framework Support​

  • NVIDIA Triton - High-performance inference server
  • AMD - Optimized inference on AMD hardware

What's Next?​

Community & Support​

Learning Path​

  1. Start Here: Deploy your first GenAI model or Deploy your first predictive model
  2. Learn the Basics: Model Serving Guide
  3. API Reference: Complete API documentation
  4. Join the Community: See who's using KServe

Need Help?​

KServe is a CNCF incubating project and part of the Kubeflow ecosystem.