Skip to content

Releases: GrayboxTech/weightslab

Release v1.2.5

17 Jun 14:59

Choose a tag to compare

Weightslab

Follow us on LinkedIn | Graybx


Welcome

Welcome to the v1.2.5 release of WeightsLab. This release was made with the idea to improve how you Inspect Data, Edit and Evolve Neural Networks, all by Graybx:

  • #207 v1.2.5 — 2026-06-17

    Fix EMA Sync. from Ultralytics trainer and evaluate mode

One platform. Unmatched flexibility. Ultimate control.
Make sure to install our last version on PyPI

Happy Training!


What's Changed precisely:


Thank you!

Release v1.2.4

17 Jun 10:51

Choose a tag to compare

Weightslab

Follow us on LinkedIn | Graybx


Welcome

Welcome to the v1.2.4 release of WeightsLab. This release was made with the idea to improve how you Inspect Data, Edit and Evolve Neural Networks, all by Graybx:

  • #203 v1.2.4 — 2026-06-17

    • Fix the evaluation mode issue - Disable the watchdog for threads by default - Fix code quality issues

One platform. Unmatched flexibility. Ultimate control.
Make sure to install our last version on PyPI

Happy Training!


What's Changed precisely:


Thank you!

Release v1.2.3

16 Jun 12:00

Choose a tag to compare

Weightslab

Follow us on LinkedIn | Graybx


Welcome

Welcome to the v1.2.3 release of WeightsLab. This release was made with the idea to improve how you Inspect Data, Edit and Evolve Neural Networks, all by Graybx:

  • #201 v1.2.3 - Fixes — 2026-06-16

    • Fixed some bugs around metadata fetching from the UI - Fixed logger spam and TLS format - Add new materials in the weights lab around experiment running inside Docker (DinD, and DoutD)

One platform. Unmatched flexibility. Ultimate control.
Make sure to install our last version on PyPI

Happy Training!


What's Changed precisely:


Thank you!

Dev Release v1.2.3.dev0

16 Jun 13:26

Choose a tag to compare

Pre-release

Weightslab

Follow us on LinkedIn | Graybx


Welcome

Welcome to the v1.2.3.dev0 release of WeightsLab. This release was made with the idea to improve how you Inspect Data, Edit and Evolve Neural Networks, all by Graybx:

(no pull requests found since last release)

One platform. Unmatched flexibility. Ultimate control.
Make sure to install our last version on TestPyPI

Happy Training!


What's Changed precisely:

  • 117bb4e remove useless dir
  • 59a45bc Fix main process issues with windows and logs; and fix ultralytics integration issues with evaluation mode
  • 6c3e1d5 add auto start in examples
  • 3b88c59 maximum resolution set to 360p by default
  • 72d682e fix logger prints format and spam
  • e38c4f4 add docker initialization materials
  • 8deb810 update gitignore and add debug print exc
  • 67f90a5 remove useless certs
  • 1bbd971 Python version tests in workflows
  • 19f2f16 Exclude heavy blob columns from histogram/metadata sweeps
  • 73fca0c Fix 4 ui_docker_bridge unit tests broken by --certs flag refactor
  • 8076c25 fix utests
  • 1e23d37 Fix code quality issue
  • 0927cce Add missing source files skipped by data/ gitignore rule
  • 6ed796f fix code quality
  • 5432688 fix code quality
  • 57c8313 Integrate 2d 3d point cloud detection usecase with Lidar (#199)
  • 12f5130 fix cmds tests
  • 8695667 fix exec bits from exec files
  • 0c7f1c0 fix permissions issues and docker v1
  • 579d8cc Refactor wl examples syntax
  • af6846c Fix certs issues; change cls config; and fix issues with hist and agent status
  • 912c03c Fix certs issues and related utests
  • 2e4778d fix onboarding certs issue and update doc and readme
  • f40792f add torchmetrics deps for cls example

Thank you!

  • AlexGrayBox
  • Alexandru Rotaru
  • Guillaume
  • GuillaumePELLUET

Release v1.2.2

12 Jun 17:54

Choose a tag to compare

Weightslab

Follow us on LinkedIn | Graybx


Welcome

Welcome to the v1.2.2 release of WeightsLab. This release was made with the idea to improve how you Inspect Data, Edit and Evolve Neural Networks, all by Graybx:

  • #197 v1.2.2 Lidar point cloud integration; Onboarding experience; and fixes — 2026-06-12

    Fix some issues with the onboarding experience: - Weightslab certs generation and management - Weightslab example starting from the command line (e.g., weightslab start example --seg) - Update Readme and documentations - Fix permissions issue from sh files (related to a) bits exec do not manage by git with Windows when committing; or b) chmod needed) - Integrate 2d and 3d point cloud detection for Lidar

  • #194 v1.2.1 - Small fixes around v1.2.0 — 2026-06-09

    Fixes around Certs generation with Windows. And the ultralytics aspect ratio issue

  • #190 v1.2.0 - Data Granularity and UL integration — 2026-06-09

    Main features added are: - Data granularity, i.e., we now save data at the instance level, i.e., 0 for a one-label per sample, or n for n bounding boxes per sample. - Ultralytics integration with WeightsLab (v1) - Audit file during experiment, saving user interactions - Logger history per instance - New user functions to export tabular data

  • #182 Remove leftover conflict markers in test_data_loader_interface.py — 2026-05-29

    Problem main has unresolved (nested) git conflict markers committed in weightslab/tests/backend/test_data_loader_interface.py (a botched dev->main merge). This breaks linting/CI for anything based on or merging into main. ## Fix Strip the 9 marker lines only — no logic change. The conflict's only real content was two test methods (test_epoch_exhausted_flag_behavior, test_multiple_sequential_epochs_with_auto_reset), which are preserved. ## Notes - git diff is purely deletions (9 lines); fil...

  • #178 v1.1.8 Fixes — 2026-05-20

    • Add new hyperparameter modifications - Fix audit mode after restarting exp. from chkpt - Fix bugs with audit mode, backward function, and Ultralytics
  • #176 v1.1.7 - Fix some bugs from v1.1.6 — 2026-05-19

    Fixed some bugs: - Audit mode - DataLoader bugs with next -> iter -> next raising StopIteration without being reset - Ultralytics model save and load weights bug

  • #174 v1.1.6 Detection — 2026-05-18

    • Fixes some bugs - Improve bounding boxes rendering on the UI side
  • #173 v1.1.4 — 2026-05-15

    Fix Envoy issue with TLS and frontend communication

  • #172 v1.1.2 fixes — 2026-05-15

    Add some renaming and fixes from v1.1.2

  • #170 v1.1.2 - Ultralytics integration with detection usecase — 2026-05-13

    • Integrate Ultralytics use case for detection and the YOLO model with bounding_boxes rendering - Fix some UI bugs
  • #162 v1.0.8.1 fix — 2026-04-28

    Fix bugs with the UI, flickering plots, grid right-click menu, ..etc

  • #160 v1.0.8 - Robustness and new features — 2026-04-28

    WL Updates - WL/WS - Secured communication (TLS and gRPC auth) - WL/WS - Evaluate mode (Feature) - WL/WS - Agent Initialization from UI and CLI (Feature) - WS - UI (Improvements) - WL - Robustness (Improvements) ### Bug Fixes - Sorting numeric uids was interpreted as sorting string values - Fix discard and tag features with numeric uids ### Related https://github.com/GrayboxTech/weights_studio/pull/80

  • #150 v1.0.7 — 2026-04-10

    Fix the weightslab Docker command for easy onboarding

  • #149 Fix: move UI docker configs into weightslab.ui package — 2026-04-10

    Summary - Moves docker-compose.yml and envoy.yaml from ui/ root folder into weightslab/ui/ Python package - Adds weightslab/ui/__init__.py so importlib.resources.files("weightslab.ui") can resolve the files - Removes the root ui/ folder (source-env scripts and envoy template were unused) - Fixes ModuleNotFoundError: No module named 'weightslab.ui' on weightslab ui launch ## Test plan - [ ] weightslab ui launch starts without errors - [ ] weightslab ui stop / weightslab ui drop...

One platform. Unmatched flexibility. Ultimate control.
Make sure to install our last version on PyPI

Happy Training!


What's Changed precisely:


Thank you!

Dev Release v1.2.1.dev0

10 Jun 14:53

Choose a tag to compare

Pre-release

Weightslab

Follow us on LinkedIn | Graybx


Welcome

Welcome to the v1.2.1.dev0 release of WeightsLab. This release was made with the idea to improve how you Inspect Data, Edit and Evolve Neural Networks, all by Graybx:

  • #191 Add weightslab.integrations.ultralytics SDK — 2026-06-09

    Summary Introduces weightslab.integrations.ultralytics — the first framework integration package, exposing a two-name public surface (WLAwareTrainer, WLAwareDataset) modeled on Determined AI / wandb.integrations.*. ## What this adds python from ultralytics import YOLO from weightslab.integrations.ultralytics import WLAwareTrainer YOLO("yolo11n.pt").train( trainer=WLAwareTrainer, data="data.yaml", imgsz=640, epochs=1000, batch=4, ) No bridge code in user-land. WLAwareTrainer is a `Dete...

  • #189 188 logger granularity and structure — 2026-06-05

    • Add to logger history per instance - Add new logger functions and tests - Add a user function to write history to a path with filters as json or tabular data (csv): wl.write_history - Add a user function to write the df view to a path with filters as json or tabular data (csv): wl.write_dataframe - Add and fix new utests - Update detection example with those functions and the custom signals function also

One platform. Unmatched flexibility. Ultimate control.
Make sure to install our last version on TestPyPI

Happy Training!


What's Changed precisely:

  • 314c60c fix ui docker bridge and simplify certs
  • 6a09909 Add weightslab example start to run the bundled classification example
  • e985a66 Keep a valid certs bind-mount in unsecured/--no-certs launches
  • bea6314 Make WEIGHTSLAB_CERTS_DIR the single source of truth for TLS/auth
  • aae4b65 Make CI run the real test suite and fix/triage dev test failures
  • a9a2165 Rework UI onboarding CLI: secure-by-default launch, cleanup, banner help
  • ed2d64b Fix failing unit tests and dataloader infinite loop
  • 12c91ef Revert "Add geometry test for val rect-mode opt-in"
  • 3459058 Add geometry test for val rect-mode opt-in
  • f3d5128 fix quotes issue
  • ca8e38b fix ul deps version
  • 901da04 small fixes around certs. and windows; and add ul deps in wl
  • 1974d3b Source val rect from ledger config; default False
  • c5f0807 Force val to letterbox to (imgsz, imgsz) like train
  • 031a887 remove test from CI for now
  • 700c665 remove test from the release
  • 8923862 fix tests
  • 72420b3 Update changelog date and version
  • 77aa398 v1.2.0 - Data Granularity (#190)
  • 6cc624b fix useless imports and vars: code quality issues
  • f411225 Add weightslab.integrations.ultralytics SDK (#191)
  • 456ffb1 fix merged conflicts function name
  • 10a65ff Remove hardcoded value from src ModelInterface
  • 685c88f fix merged conflicts error on fct name
  • 3383035 fix iou metric name

Thank you!

Release v1.2.1

09 Jun 17:12

Choose a tag to compare

Weightslab

Follow us on LinkedIn | Graybx


Welcome

Welcome to the v1.2.1 release of WeightsLab. This release was made with the idea to improve how you Inspect Data, Edit and Evolve Neural Networks, all by Graybx:

  • #194 v1.2.1 - Small fixes around v1.2.0 — 2026-06-09

    Fixes around Certs generation with Windows. And the ultralytics aspect ratio issue

  • #190 v1.2.0 - Data Granularity and UL integration — 2026-06-09

    Main features added are: - Data granularity, i.e., we now save data at the instance level, i.e., 0 for a one-label per sample, or n for n bounding boxes per sample. - Ultralytics integration with WeightsLab (v1) - Audit file during experiment, saving user interactions - Logger history per instance - New user functions to export tabular data

One platform. Unmatched flexibility. Ultimate control.
Make sure to install our last version on PyPI

Happy Training!


What's Changed precisely:


Thank you!

Release v1.2.0

09 Jun 15:53

Choose a tag to compare

Weightslab

Follow us on LinkedIn | Graybx


Welcome

Welcome to the v1.2.0 release of WeightsLab. This release was made with the idea to improve how you Inspect Data, Edit and Evolve Neural Networks, all by Graybx:

  • #190 v1.2.0 - Data Granularity and UL integration— 2026-06-09

One platform. Unmatched flexibility. Ultimate control.
Make sure to install our last version on PyPI

Happy Training!


What's Changed precisely:


Thank you!