Model Online / On-Site
Duration
Modules 5 modules

Course Objective & Scope

Handing models over to production via shared folders, chat links or ad-hoc S3 paths inevitably leads to wrong weights in prod, broken input schemas and security vulnerabilities. A Model Registry acts as the formal contract and control plane between experimentation and production serving.

This course covers standardised model packaging (Safetensors, ONNX, enforced signatures), lifecycle promotion via aliases (staging, champion, challenger), webhook-driven validation pipelines, scanning weights for malicious pickle payloads and automated Model Cards across MLflow, W&B and OCI-compliant registries. Participants design a Model Registry governance and promotion workflow tailored to their organisation.

Who Should Attend?

  • MLOps and platform engineers managing model validation, sign-off and production promotion
  • Senior data scientists and ML engineers handing models over to production services
  • Technical specialists involved in model risk management, audit and AI governance
  • Prerequisite: Python, basic Docker familiarity and understanding of the ML lifecycle

Key Takeaways

  • Enforce a packaging standard requiring explicit input/output signatures, locked dependencies and training lineage on every registered model
  • Use model aliases (@champion, @challenger, @archived) to promote or roll back production models without changing application code
  • Trigger automated accuracy, latency and fairness validation gates via registry webhooks and CI workflows
  • Eliminate arbitrary code execution risks in pickle files by adopting Safetensors/ONNX and scanning artefacts with ModelScan
  • Generate auditable Model Cards and approval records capturing known limitations, benchmark slices and sign-off history

Curriculum

From Experiment to Production: Why a Central Model Registry?
  • Dividing responsibilities between raw object storage (S3), OCI Container Registries and an ML Model Registry
  • Registry topologies across multi-team and multi-environment (Dev, Staging, Prod) organisations
Standardised Model Packaging, Signatures and Security
  • Enforcing input/output schemas (ModelSignature), sample payloads and pinned runtime dependencies
  • Mitigating pickle deserialisation risks with Safetensors/ONNX and scanning artefacts with ModelScan
Lifecycle Management: Versions, Tags and Aliases
  • Immutable model version numbers versus mutable routing aliases (@champion, @challenger, @shadow)
  • Decoupling serving code from specific versions via aliases and executing instant rollbacks
Automated Validation Gates and Webhook Integrations
  • Triggering CI/CD validation workflows on model registration or alias-assignment events
  • Automated quality gates checking golden-dataset accuracy, tail latency, memory footprint and slice fairness
Model Cards, Approval Workflows and Governance
  • Auto-generating standardised Model Cards with training lineage, intended use and known failure modes
  • Role-based access control (RBAC), promotion sign-off logs and readiness for internal audit and EU AI Act requirements

Other Courses in This Category

  • Model Serving and High-Performance Inference Architectures

    Online, batch and streaming inference patterns, Triton Inference Server, BentoML, Ray Serve and vLLM, dynamic batching, gRPC/REST, caching and load testing, for teams serving models at low latency and high throughput.

    MLOps & AI Platform Engineering Corporate
    Live Course
    5 modules
    View
  • CI/CD and Continuous Training (CT) for Machine Learning

    Validating data, models and code in GitHub Actions/GitLab CI, automated PR evaluation reports, Continuous Training (CT) triggers with Airflow/Argo/Prefect and GitOps deployment, for MLOps teams.

    MLOps & AI Platform Engineering Corporate
    Live Course
    5 modules
    View
  • Container and Kubernetes Deployment for AI and ML Workloads

    Optimised Docker images for CUDA/PyTorch, Kubernetes NVIDIA GPU Operator, MIG/time-slicing, serverless/autoscaling inference with KServe and distributed training with Kubeflow/Volcano, for platform and MLOps teams.

    MLOps & AI Platform Engineering Corporate
    Live Course
    5 modules
    View
  • GPU Acceleration and Inference Optimization

    GPU memory/compute architecture, profiling with Nsight Systems, TensorRT, FlashAttention, KV cache management, AWQ/GPTQ/FP8 quantization and speculative decoding, for engineers cutting deep learning and LLM inference costs.

    MLOps & AI Platform Engineering Corporate
    Live Course
    5 modules
    View
  • Enterprise MLOps and Model Deployment on Amazon SageMaker

    SageMaker Studio, Processing and Training Jobs, SageMaker Pipelines, Model Registry, Real-Time/Serverless/Async Inference, Model Monitor and IAM/VPC security, for teams running end-to-end ML on AWS.

    MLOps & AI Platform Engineering Corporate
    Live Course
    5 modules
    View
  • Production Model Monitoring and ML Observability

    System and model telemetry with OpenTelemetry, Prometheus and Grafana, prediction logging, delayed ground-truth joining, live performance tracking and incident response, for MLOps and SRE teams.

    MLOps & AI Platform Engineering Corporate
    Live Course
    5 modules
    View

Get in touch