skills.rest
arbazkhan971
arbazkhan971COMMUNITY
Explore all skills by this author →
Website𝕏@arb5zGitHub

mlops

Deploy trained ML models to serving infrastructure with canary rollback.

26|8|Updated Mar 19, 2026
Search Tags:#monitoring#mlops#drift-detection#canary#serving#model-deployment#retraining
One-click install
npx skills add https://github.com/arbazkhan971/godmode --skill mlops-arbazkhan971
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: mlops
Source: https://github.com/arbazkhan971/godmode/tree/main/skills/mlops
Command: npx skills add https://github.com/arbazkhan971/godmode --skill mlops-arbazkhan971

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

This Skill removes the friction of taking trained models from experimental artifacts into safe, production-ready services by enforcing readiness checks, guardrails, and automated rollback to prevent regressions and downtime.

Core Features & Use Cases

  • Readiness checklist: evaluation metrics, bias checks, artifacts and I/O schema verification, latency and size thresholds.
  • Serving infrastructure guidance: recommendations for TF Serving, Triton, SageMaker, FastAPI, and Ray Serve with trade-offs.
  • Inference optimization: quantization, FP16/INT8, ONNX conversion, TensorRT, batching strategies, and distillation guidance.
  • Versioning & traffic control: staged lifecycle (staged->canary->champion), canary splits, routing strategies, and rollback preparation.
  • Drift detection & retraining: PSI thresholds, performance alerts, scheduled or trigger-based retraining with human validation gates.
  • Monitoring & logging: p50/p95/p99 latency, error rates, primary metric rolling windows, and TSV logging for results and audits.
  • Use Case: Safely deploy a new image-classification model behind a Triton endpoint with a 10% canary, validate p99 latency and accuracy, then ramp or revert automatically.

Quick Start

Deploy the trained model my-model:v1 to Triton as a 10% canary, monitor p99 latency and primary metric, and promote if all guardrails pass.

Frequently Asked Questions about mlops

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I safely deploy a trained machine learning model to production serving?▼

Production model deployment applies readiness checks, serving infrastructure selection, and automated rollback to safely operationalize trained machine learning models. It enforces latency targets, bias checks, and canary traffic control to prevent downtime before promoting new model versions.

What is model drift detection and when do I need to trigger retraining?▼

Model drift detection monitors PSI thresholds and primary metric rolling windows to identify performance degradation in production models. You trigger scheduled or performance-alert-based retraining when drift is detected, requiring human validation gates before redeploying the updated model.

How do I run a canary A/B test for a new model version in serving infrastructure?▼

Canary A/B testing routes a controlled traffic percentage, such as 10%, to a new model version within serving infrastructure like Triton or TF Serving. The deployment monitors p99 latency and primary metrics, automatically reverting if guardrails fail before promoting the model.

Does this model deployment workflow support inference optimization with ONNX conversion and TensorRT?▼

Model deployment workflows support inference optimization through ONNX conversion, TensorRT integration, quantization, FP16/INT8 precision reduction, batching strategies, and model distillation. These techniques reduce inference latency and model size to meet strict production serving constraints.

What is the best way to monitor machine learning model latency and error rates in production?▼

Production model monitoring tracks p50, p95, and p99 latency alongside error rates and primary metric rolling windows. It implements TSV logging for results and audits, generating performance alerts when metrics breach predefined thresholds to ensure reliable serving.

Can I use Triton or SageMaker for reliable production model serving?▼

Production model serving supports Triton, SageMaker, TF Serving, FastAPI, and Ray Serve infrastructure. Each platform offers distinct trade-offs for latency, scaling, and deployment workflows, allowing you to select the optimal environment for your model size and traffic requirements.

Related Skills

Explore More Data & Analytics Skills→
rocklambrosrocklambros
2

packaging-model-for-deployment

Package trained machine learning models into self-describing deployment artifacts.

Community
Advanced
DuqueOMDuqueOM
5

model-retrain

Automate model retraining workflows with quality gates and MLflow promotion.

Community
Advanced
EanLeeEanLee

istio-traffic-management

Configure Istio VirtualServices, DestinationRules, and Gateways for traffic routing.

Community
Advanced
Ryno-Crypto-Mining-ServicesRyno-Crypto-Mining-Services

devops-role-skill

Create CI/CD pipelines with GitHub Actions, Terraform, and Prometheus.

Official
Advanced