Changelog

Notable changes for OneOpen ML Studio.

1.0.0 — first PyPI release

Platform

  • Active learning: unlabeled, lowest-confidence, entropy, margin; persisted batches

  • Kubernetes starter manifests (API probes, worker, Postgres, Redis, PVC)

  • Enterprise audit events + SSO configuration stubs

  • Plugin manager with on-demand install of training / data-prep extras

  • Famous classification & segmentation train catalogs (YOLO-Cls/Seg + Torchvision)

  • Data cleaning (preview/apply); pipeline data_cleaning step

  • Sample quality / readiness for text, tabular, audio, video, LLM

  • Visual pipeline DAG editor on the project Pipelines tab

Hierarchy & Team / Enterprise

  • Platform → Organization → Workspace → Project end-to-end (UI + APIs)

  • Deployment mode from ONEOPEN_DEPLOYMENT_MODE only (no editions simulator)

  • First-time /setup wizard for Team / Enterprise (org + workspace + admin)

  • Org / workspace member management, custom roles with selectable capabilities

  • Workspace picker when creating or importing projects

Security (Team / Enterprise)

  • Auth enforcement on data-plane APIs; HTML login redirect when auth is on

  • Project membership checks (no open-project bypass when auth is on)

  • Org IDOR hardening; scoped /api/users and roles listing

  • Zip-slip and path-traversal protections on imports / exports / images

  • Role-assignment rank checks (no privilege escalation)

  • Stronger passwords (8+ chars) when auth is enforced

  • Secure session cookies for Enterprise; warnings on default secrets

  • Safer model registration paths and torch.load(weights_only=True) when auth is on

Docs

  • docs/deploy/docker.md, docs/deploy/enterprise.md

  • docs/user-guide/data-prep-and-training.md, cleaning.md, editions-and-plugins.md

  • Install from PyPI: pip install oneopen-ml-studio

0.9.0

  • Audio / video / LLM modalities and project types

  • JSONL import/export; transcript, preference, instruction, RAG annotate UIs

  • Canonical schema for non-image samples

  • Guide: docs/modalities/audio-video-llm.md

0.8.0

  • Text / tabular project types and sample storage

  • CSV / JSONL import-export; modality-aware canonical schema

  • Annotate UIs for classification, NER spans, tabular values

  • sklearn + Hugging Face adapters (prepare; sklearn train)

  • Optional extras [tabular] and [nlp]

  • Guide: docs/modalities/text-tabular.md

0.7.0

  • Plugin registry with entry-point discovery (oneopen.plugins)

  • Pipeline schema: cycle detection, YAML helpers, topo sort

  • Pipeline CRUD + validate / YAML import-export

  • Runner: validation, stratified split, YOLO export, training enqueue

  • Project Pipelines tab

  • Guide: docs/pipelines.md

0.6.0

  • Celery worker for training (oneopen worker) with local thread fallback

  • MinIO/S3 artifact upload for exports and weights

  • Docker Compose (healthchecks, minio-init, worker)

  • /api/health and /api/ready probes

  • Guide: docs/deploy/docker.md

0.5.0

  • Local users, password auth, sessions, API tokens

  • Organizations, workspaces, project memberships + role capabilities

  • Sample locks in annotate UI; annotation revisions

  • Assignments and review queue

  • Team tab, login/register/account pages; oneopen user create-admin

0.4.0

  • Stratified + seeded splits; version quality snapshots

  • Quality checks: blur, near-duplicates, exact dup/leakage

  • Training readiness gate on train start

  • Configurable quality checks (bypass, toggles, thresholds)

  • Canonical sample schema export API

  • Experiment compare API/UI; model lifecycle controls

  • Failed training jobs persist as experiments

0.3.0

  • Dataset profiling, quality engine, training readiness, experiments, adapters

  • Collaboration foundations (locks, revisions, roles)

  • Docker Compose + Dockerfile

  • Plugin registry + pipeline schema/API

  • Multi-modal project type foundations

  • Active learning selection starters + Kubernetes manifests

  • Quality + Experiments tabs

0.2.0

  • Product packaged as OneOpen ML Studio (oneopen-ml-studio / oneopen_ml_studio)

  • CLI: oneopen start (alias serve) and oneopen-ml-studio

  • Product framework docs and hosted docs site

0.1.1

  • Corrected project URLs to GitHub org repo

  • Sphinx documentation under docs/

0.1.0

Initial public release:

  • Local FastAPI UI for projects, annotation, import/export

  • YOLO Label Assist and optional SAM

  • Dataset versions, preprocess/augment pipeline

  • Local Ultralytics / LibreYOLO training with live metrics

  • Apache-2.0 packaging