# MLOps i LLMOps

EN: **MLOps and LLMOps** · id `mlops`

Path: [AI Atlas](https://aiatlasgraph.com/llms.txt) › [Inferencja i infrastruktura](https://aiatlasgraph.com/t/systems.md) › MLOps i LLMOps

## Description

- PL: Utrzymywanie powtarzalnych, obserwowalnych wdrożeń.
- EN: Keeping deployments repeatable and observable.

## In one sentence

- PL: To organizowanie treningu, wersji i wdrożeń modeli tak, żeby dało się je śledzić, porównywać i utrzymywać.
- EN: This organises model training, versions and deployments so they can be tracked, compared and maintained.

## Problem it solves

- PL: Nie wiesz, skąd pochodzi działający model, dlaczego zmieniły się wyniki ani jak bezpiecznie wrócić do poprzedniej wersji.
- EN: You cannot trace the running model, explain changed results or safely return to an earlier version.

## What it does

- PL: Rejestry łączą dane, eksperymenty i wersje, a testowane potoki oraz wspólne interfejsy porządkują wdrażanie i obsługę zmian.
- EN: Registries connect data, experiments and versions, while tested pipelines and shared interfaces organise deployment and change management.

## Everyday example

- PL: Po gorszych odpowiedziach asystenta ustalasz wersję modelu i instrukcji, porównujesz testy oraz przywracasz sprawdzony wariant.
- EN: After an assistant's responses worsen, you identify model and prompt versions, compare tests and restore a verified variant.

## This concerns you if you…

- PL: utrzymujesz model poza pojedynczym eksperymentem
- PL: potrzebujesz powtarzalnych wdrożeń i historii zmian
- EN: maintain a model beyond a single experiment
- EN: need repeatable deployments and change history

## Keywords

MLflow · CI/CD · model registries · experiment tracking · artifact lineage · continuous training · feature stores · prompt registries · model monitoring · version promotion

## How it works

- PL: Wersjonowane dane, kod i konfiguracja tworzą powtarzalne potoki treningu oraz wdrożenia. Rejestry artefaktów, ewaluacje i monitoring wiążą stan produkcji z jego pochodzeniem, a kontrolowane promocje oraz wycofania zarządzają zmianami.
- EN: Versioned data, code and configuration form reproducible training and deployment pipelines. Artifact registries, evaluations and monitoring link production state to provenance, while controlled promotion and rollback manage changes.

## Limits

- PL: Powtarzalność wymaga uchwycenia zależności i źródeł losowości. Zmiana danych może pogorszyć model bez awarii usługi; wycofanie artefaktu nie cofa automatycznie cech ani zewnętrznych skutków.
- EN: Reproducibility requires capturing dependencies and randomness. Data changes can degrade models without service failures; artifact rollback does not automatically revert features or external effects.

## Key terms

- PL: **artefakt modelu**: Wersjonowany zestaw wag, konfiguracji i materiałów potrzebnych do wykonania.
- PL: **promocja wersji**: Przeniesienie sprawdzonego artefaktu do wybranego etapu wdrożenia.
- EN: **model artifact**: Versioned weights, configuration and materials needed for execution.
- EN: **version promotion**: Moving an assessed artifact to a selected deployment stage.

## Know first

- [Wersjonowanie zbiorów danych (Dataset versioning)](https://aiatlasgraph.com/t/dataversioning.md)
- [Monitoring produkcyjny (Monitoring and observability)](https://aiatlasgraph.com/t/monitoring.md)

## Examples

- PL: Rejestr modeli · CI/CD · monitoring
- EN: Model registry · CI/CD · monitoring

## Subtopics (6 direct, 6 in total)

- [Automatyczne retrenowanie i CI/CD dla ML (Continuous training and CI/CD for ML)](https://aiatlasgraph.com/t/continuoustraining.md): Automating an ML pipeline: retraining and redeploying a model after a defined trigger such as a schedule, new data, quality degradation or detected drift, while testing and deploying the pipeline itself like code.
- [Bramka LLM: jedno API do wielu modeli (LLM gateway (AI gateway))](https://aiatlasgraph.com/t/llmgateway.md): An intermediate layer in front of many providers and self-hosted models: one unified API, central keys and access control, rate limits and budgets, fallbacks, load balancing and cost logging.
- [Magazyn cech (feature store) (Feature store)](https://aiatlasgraph.com/t/featurestore.md): A shared feature platform for training and production: an offline store with feature history for building training sets and an online store serving current features at low latency. Shared definitions and point-in-time joins help limit training-serving skew and leakage from the future.
- [Rejestr i wersjonowanie modeli (Model registry and versioning)](https://aiatlasgraph.com/t/modelregistry.md): A central registry of model versions with metadata and lineage to experiment runs. Mutable aliases such as champion or production can decouple production-version selection from deployment code and make promotion or rollback easier.
- [Wersjonowanie i rejestr promptów (Prompt management and versioning)](https://aiatlasgraph.com/t/promptversioning.md): Storing prompts as immutable, numbered versions with metadata and a production alias. Linking versions to traces and evaluations supports comparison; when an application loads prompts dynamically through an alias, rollback is possible without redeploying code.
- [Śledzenie eksperymentów (parametry, metryki, artefakty) (Experiment tracking)](https://aiatlasgraph.com/t/experimenttracking.md): Logging training runs — hyperparameters, metrics, artefacts, and code and data versions — to support result comparison, lineage and reconstruction of experiment conditions.

## Related topics

Untyped, symmetric "see also" links.

- [Monitoring produkcyjny (Monitoring and observability)](https://aiatlasgraph.com/t/monitoring.md)
- [Wdrożenie w cieniu i stopniowe (canary) (Shadow and canary rollouts)](https://aiatlasgraph.com/t/progressiverollout.md)
- [Wersjonowanie zbiorów danych (Dataset versioning)](https://aiatlasgraph.com/t/dataversioning.md)

## Sources

- [Stanford CS336 · Language Modeling from Scratch (2025)](<https://cs336.stanford.edu/spring2025/>) `cs336` · reference: Budowa modeli, dane, trening, systemy, inferencja i post-training.
- [Full Stack Deep Learning (2022, archiwum)](<https://fullstackdeeplearning.com/course/2022/>) `fsdl` · reference: Pełny cykl systemu ML: dane, testy, wdrażanie, monitoring, zespoły i etyka.
- [FSDL · LLM Bootcamp (2023, archiwum)](<https://fullstackdeeplearning.com/llm-bootcamp/spring-2023/>) `bootcamp` · reference: Aplikacje LLM, prompting, retrieval, UX i LLMOps.
- [DeepLearning.AI — course catalogue](<https://www.deeplearning.ai/courses>) `dlai` · course catalogue: 42 topic labels with counts, about 130 courses.

---

AI Atlas revision 122 (2026-10-30) · HTML: https://aiatlasgraph.com/t/mlops · index: https://aiatlasgraph.com/llms.txt · everything: https://aiatlasgraph.com/llms-full.txt · updated 2026-10-30