Trainer integration

CVLabel Python SDK

Preflight and sync lossless datasets, publish structured evaluations, and push versioned models from headless trainers—with restart-safe upload checkpoints.

v0.3.0 Python 3.9+ macOS · Linux · Windows
pip install
python3 -m pip install https://api.cvlabel.cloud/downloads/sdk/v0.3.0/cvlabel-0.3.0-py3-none-any.whl

The immutable universal wheel installs into any Python 3.9+ environment. Use a virtual environment for isolated trainer deployments.

Release artifacts

One SDK, packaged for every delivery path.

Install the wheel directly, retain the source archive for controlled builds, or download the complete offline bundle. Every public file is immutable and checksum-published.

SHA256SUMS AppPorter release
v0.3.0

Universal wheel

Recommended · Python 3.9+ · all platforms

v0.3.0

Source archive

TAR.GZ · build from source

v0.3.0

Offline bundle

ZIP · wheel, source, checksums, and guide

v0.3.0

macOS disk image

DMG · AppPorter-ready offline distribution

01 / DATASETS

Lossless push, pull, and sync

Infer ImageFolder classes, inspect a dry-run report, preserve parent lineage, and resume direct or tar-shard uploads.

02 / LINEAGE

Record every training run

Link parameters, metrics, and cvlabel.evaluation.v1 results to the precise dataset version used by the trainer.

03 / MODELS

Publish append-only versions

Resume signed artifact staging, publish atomically, and retain the dataset-to-run-to-model reproduction chain.

Ready for headless trainers.

Authenticate with a workspace-bound service app; no browser or local bridge is required.

Create workspace