MLOps & Model Serving Bundle
Deployment, inference serving, model registries, and production model monitoring.
Browse all skill bundles
Included Skills (10)
- Defect Image Generator — This tool orchestrates NVIDIA Cosmos AnomalyGen workflows to generate synthetic defect datasets and perform inference for industrial inspection tasks like PCBA and surface analysis.
- Physical AI Infrastructure Manager — This tool assists engineers in deploying, scaling, and hardening resilient NVIDIA AI infrastructure across MicroK8s and Azure AKS for synthetic data generation workflows.
- Video Data Orchestrator — This tool manages end-to-end video data augmentation and auto-labeling workflows on OSMO, assisting engineers with preflight, job submission, monitoring, and output retrieval.
- Megatron Resiliency Manager — This tool enables and manages fault tolerance, straggler detection, and automatic restart features to ensure stable training for Megatron Bridge users.
- LLM Evaluation Pipeline — This tool helps AI developers improve application quality by establishing automated evaluation pipelines, creating golden datasets, and generating actionable insights for performance optimization.
- GPU Memory Optimizer — This tool helps machine learning engineers resolve GPU out-of-memory errors and optimize peak memory usage in Megatron Bridge training environments.
- CUDA Graph Optimizer — This tool helps developers optimize training performance by configuring and validating CUDA graph capture implementations within Megatron Bridge for various model architectures.
- RAG System Architect — This workflow guides developers through building and optimizing RAG systems by streamlining embedding selection, vector database configuration, and retrieval performance evaluation.
- LLM Observability Expert — Assists developers in tracing and monitoring LLM applications using Langtrace, an open-source observability platform built on OpenTelemetry.
- Model Evaluation Manager — This tool helps machine learning engineers automate the integration of benchmark scores and structured evaluation metadata into Hugging Face model cards.