Registry
Polyaxon registries organize reusable ML assets across teams. Use them to version and promote models and artifacts, publish components, and preserve lineage back to the runs that produced them.
Register trained models with full lineage back to the training run, parameters, and metrics.
from polyaxon.client import ProjectClient
client = ProjectClient(project="acme/text-classifier")
client.register_model_version(
version="v1.2.0",
description="Fine-tuned on customer feedback dataset",
tags=["production-ready", "pytorch"],
run="<run-uuid>",
artifacts=["model", "env"],
)Version and track any artifact — datasets, preprocessing pipelines, feature stores, and configuration files.
# Register an artifact version
polyaxon artifacts register -p acme/preprocessed-data \
-ver v2.0 \
--run-uid <run-uuid> \
--artifacts processed-data,summary \
--description "Cleaned and normalized training dataset"Share reusable components across teams via the Component Hub. Reference components by name in any operation.
# Use a shared component from the hub
version: 1.1
kind: operation
component_ref:
hub: my-org/train-pytorch
version: latest
params:
model_arch: { value: resnet50 }
dataset: { value: imagenet }
epochs: { value: 90 }Promote models through stages — testing, staging, production, disabled — with audit trails and approval workflows.
from polyaxon.client import ProjectClient, V1Stages
client = ProjectClient(project="acme/text-classifier")
# Promote from staging to production
client.stage_model_version(
version="v1.2.0",
stage=V1Stages.PRODUCTION,
reason="ReleaseReview",
message="Validation passed",
)
# List production models
versions = client.list_model_versions(
query="stage: production",
)Component Hub
Use the Component Hub to manage reusable, versioned components. Components can:
- Have multiple versions.
- Be published publicly or privately within your organization.
- Define team-level permissions.
Model Registry
Use the Model Registry to manage model versions collaboratively. It provides:
- APIs and a UI.
- Team configuration and access controls.
- Logging and lineage through experiment tracking.
- Versioning and lifecycle stages.
- Deployment as an internal tool or test API using services.
- Support for any model packaging format and serving technology.
Artifacts Versioning
Use Artifacts Versioning to promote runs and version reusable assets. It provides:
- APIs and a UI.
- Summaries and metadata for each asset version.
- Lineage and locking for the runs that generate data and artifacts.
- Team-level permissions.