Environment:Kserve Kserve Python Runtime
| Knowledge Sources | |
|---|---|
| Domains | Runtime, Model Serving |
| Last Updated | 2026-02-13 00:00 GMT |
Overview
Python >= 3.10 runtime environment required by KServe Python SDK, explainer samples, transformer samples, and model packaging code.
Description
The KServe Python SDK and its sample code (explainers, transformers, custom model servers) require a Python runtime environment. The SDK supports Python 3.10 through 3.12, with dependencies on FastAPI, uvicorn, grpcio, numpy, and other scientific computing libraries. Sample code such as AlibiExplainer, AnchorImages/Tabular/Text explainers, Kafka image transformer, PaddlePaddle preprocessor, and custom model servers (DenseNet161, MNIST CNN) all run within this environment.
Usage
Use this environment for Python-based KServe components including custom explainers, transformers, model servers, and any code using the `kserve` Python SDK.
System Requirements
| Category | Requirement | Notes |
|---|---|---|
| Python | >= 3.10, < 3.13 | From kserve pyproject.toml |
| pip | >= 21.0 | For modern dependency resolution |
| OS | Linux (x86_64 or aarch64) | Container images are Linux-based |
Dependencies
Core Python Packages
- `kserve` == 0.16.0
- `numpy` >= 1.26.0, < 3.0.0
- `grpcio` >= 1.64.1, < 2.0.0
- `fastapi` >= 0.115.3
- `uvicorn[standard]` >= 0.30.6, < 1.0.0
- `httpx` >= 0.27.2, < 1.0.0
- `pydantic` >= 2.5.0, < 3.0.0
- `protobuf` >= 5.27.1, < 6.0.0
Explainer-Specific
- `alibi` (for AnchorTabular, AnchorImages, AnchorText)
- `nest-asyncio` (for bridging sync Alibi with async KServe)
- `aif360` (for AI Fairness model explainability)
Transformer-Specific
- `kafka-python` (for Kafka image transformer)
- `boto3` >= 1.29.0 (for S3 image access)
- `Pillow` (for image preprocessing)
- `paddlepaddle` (for Paddle image preprocessor)
Quick Install
# Install the KServe Python SDK
pip install kserve==0.16.0
# For explainer development
pip install kserve[storage]==0.16.0 alibi nest-asyncio
# For transformer development
pip install kserve[storage]==0.16.0 kafka-python boto3 Pillow
Code Evidence
Python version requirement from `python/kserve/pyproject.toml`:
[project]
name = "kserve"
requires-python = ">=3.10,<3.13"
Common Errors
| Error Message | Cause | Solution |
|---|---|---|
| `ModuleNotFoundError: No module named 'kserve'` | kserve SDK not installed | `pip install kserve==0.16.0` |
| `Python version X.X not supported` | Python < 3.10 or >= 3.13 | Use Python 3.10, 3.11, or 3.12 |
| `ImportError: alibi` | Alibi not installed for explainer | `pip install alibi` |
Compatibility Notes
- Container images: KServe provides pre-built images with Python environment included
- Local development: Use virtual environment or conda for isolation
- GPU support: For vLLM-based inference, use the separate Environment:Kserve_Kserve_VLLM_Runtime environment
Related Pages
- Implementation:Kserve_Kserve_AlibiExplainer
- Implementation:Kserve_Kserve_Alibi_Helper
- Implementation:Kserve_Kserve_Alibi_Parser
- Implementation:Kserve_Kserve_AnchorImages_Explainer
- Implementation:Kserve_Kserve_AnchorTabular_Explainer
- Implementation:Kserve_Kserve_AnchorText_Explainer
- Implementation:Kserve_Kserve_DenseNet161_ImageClassifier
- Implementation:Kserve_Kserve_Dex_Auth
- Implementation:Kserve_Kserve_ExplainerWrapper_Base
- Implementation:Kserve_Kserve_GermanCredit_AIFModel
- Implementation:Kserve_Kserve_IAP_Request_Auth
- Implementation:Kserve_Kserve_Kafka_ImageTransformer
- Implementation:Kserve_Kserve_MNIST_CNN_Net
- Implementation:Kserve_Kserve_Paddle_Image_Preprocess
- Implementation:Kserve_Kserve_RFModel_AIX_MNIST
- Implementation:Kserve_Kserve_TorchServe_gRPC_Client