Is this limited to specific machine learning frameworks?

No, it uses standard Kubernetes patterns and Python requests, making it compatible with any containerized workload regardless of the framework.

What GPUs does this skill support?

It includes pre-defined configurations for NVIDIA A100 (80GB and NVLink 8-way), H100 SXM5, and L40 cards, and can be easily extended for others.

Does it handle node affinity automatically?

Yes, the skill includes helper functions to generate the correct nodeSelector and matchExpressions for specific CoreWeave GPU classes.

Can I use this for KServe deployments?

Yes, these patterns are specifically designed to be compatible with KServe and standard Kubernetes deployments on the CoreWeave platform.

CoreWeave SDK Patterns

Name: CoreWeave SDK Patterns
Author: jeremylongshore

byjeremylongshore

•

2,083

•

Infraestructura en la Nube

Manages production-grade CoreWeave GPU workloads using Kubernetes-native Python patterns and deployment templates.

This skill provides specialized patterns for programmatically managing GPU-intensive workloads on CoreWeave's cloud infrastructure. It includes helper functions for GPU node affinity configuration, resource requests for various NVIDIA cards (A100, H100, L40), and Python wrappers for building robust inference clients. Whether you are automating model deployments or creating reusable infrastructure templates, these patterns bridge the gap between Python logic and Kubernetes-native GPU orchestration, ensuring optimal resource allocation and high-performance model serving.

Características Principales

01Pre-configured GPU catalog for NVIDIA A100, H100, and L40 series

02Standardized error handling for common GPU orchestration issues

032,083 GitHub stars

04Dynamic Kubernetes node affinity and resource limit generation

05Automated YAML deployment template generation for K8s workloads

06Production-ready Python inference client wrappers with health checks

Casos de Uso

01Automating the deployment of LLM inference services on H100 clusters

02Programmatically scaling GPU resources based on workload demands

03Building custom Python clients for CoreWeave-hosted model endpoints

Características Principales

01Pre-configured GPU catalog for NVIDIA A100, H100, and L40 series

02Standardized error handling for common GPU orchestration issues

032,083 GitHub stars

04Dynamic Kubernetes node affinity and resource limit generation

05Automated YAML deployment template generation for K8s workloads

06Production-ready Python inference client wrappers with health checks

Casos de Uso

01Automating the deployment of LLM inference services on H100 clusters

02Programmatically scaling GPU resources based on workload demands

03Building custom Python clients for CoreWeave-hosted model endpoints