Training step

SymptomCauseFix
Exception right at start — active experiment ID does not matchThe trainer calls mlflow.set_experiment() itselfRemove the call, or pass os.environ["MLFLOW_EXPERIMENT_NAME"] as is
No curves on the training screenMetrics are only printed, or start_run() is called twice and logs to a different runRecord with mlflow.log_metrics(..., step=epoch), and use only one run
No progress barThe param epochs or the metric step is missingAdd mlflow.log_param("epochs", N)
Training completes but model registration is skippedOnly the weight file was uploaded as an artifact and log_model() was not calledLog with mlflow.pyfunc.log_model(...)
Fails in the training step, with data-related messages in the logThe dataset does not match the label shape the trainer expects (for example the MNIST example expects meta.label)Change it to read manifest.json and the annotation files — Paths and environment variables
Progress logs appear late, all at onceOutput bufferingENV PYTHONUNBUFFERED=1 in the Dockerfile
Checkpoints · reports disappearSaved only in /geo/workUpload them with mlflow.log_artifact()

Serving step (fine during training)

SymptomCauseFix
Every inference request failsMissing signature → inputs are converted to float64Build a signature with infer_signature() and pass it
The serving container does not startpredictor.py imports training code, or system packages are missingMake predictor.py self-contained; list metadata.apt_packages
The checkpoint does not loadpip_requirements versions differ from the training environmentPin exactly the training environment's versions; verify with --env-manager virtualenv

Import · run preparation

SymptomCauseFix
Image import rejectedSaved without a tag, architecture mismatch, several images in the archiveSave again with docker save <name:tag>, build for linux/amd64, one image per file
Cannot be chosen in the variant list ("No image (이미지 없음)")The image the variant points to is not in the registryCheck that the import is complete and the address is correct
Submission rejected — 데이터셋 스테이저 변형이 없습니다 ("no dataset stager variant")The tenant has no image to stage dataAsk the operator to publish (or share) a stager variant
Submission rejected — the runtime variant is for another frameworkA variant with a framework set was chosen for training in another frameworkClear the external variant's framework (unrestricted) or choose a matching variant
Prepare training environment (학습 환경 준비) step fails — namespaces "tenant-…" not foundThe tenant namespace does not exist in the clusterThe operator creates the tenant-<tenant code in lower case> namespace
Training runs on an unexpected imageThe framework has no default variant, so it fell back to the "Unrestricted (제한 없음)" default variant (an external image)Keep a default variant per framework, or set a framework on the external variant — Make it selectable in the wizard

Written for the platform as of 2026-09-21.

© Geo-MLOps