Privacy preferences

Strictly necessary

Always on

Remembers your choice on this page. Stored in your browser as gb_consent for up to 6 months. If you switch the site to dark mode, that choice is kept as gb_theme. Nothing is sent to a server.

Anonymous analytics

Helps us count visits. At most once every 24 hours we send one ping to our own telemetry server, using the same anonymous approach as the WeightsLab package. No advertising, no third-party trackers, no cross-site tracking.

Exactly what is sent
uuid
A random ID created in your browser (gb_tid). It is not linked to your name or email.
event, version
website_visit and website, so website visits can be told apart from WeightsLab package pings.
os
Operating-system family only, for example Windows, Darwin, Linux or Android.
tz
Your time zone, for example Europe/Paris.
On our server
Your IP address is used to estimate country and city. Only a salted hash of it is kept, never the raw address.
GitHub star count
With analytics on, your browser also asks api.github.com for the public star count of our repository, so GitHub sees your IP address. With analytics off, the page shows a fixed number and makes no request to GitHub.

Turning analytics off deletes gb_tid and stops all pings.

Cookie policy · Privacy policy

Where it matters

Four areas we focus on.

WeightsLab applies wherever data quality limits the model. These use cases are examples of applications.

01 · Physical AI

Find the thin scenarios before they cause downtime.

Robots, vehicles and perception systems seldom fail on the average scene; they fail on an unusual one that the dataset covers poorly. WeightsLab shows which scenarios are thin and which labels are wrong, and lets you rebalance during the run.

  • See coverage for each scenario next to the aggregate score
  • Trace a failure back to the frames, labels or edits behind it
  • Works across images, video and LiDAR, on your own infrastructure
  • Robot manipulation (VLA)
  • Driving perception
  • LiDAR & 3D
  • Smart cities

Watch the robot-manipulation demo Watch the LiDAR demo

02 · LLM training & fine-tuning

Clean up your data while it trains.

Instruction and domain datasets often contain near-duplicates, contradictory answers and noisy labels, and each one costs GPU hours. WeightsLab shows per-sample loss while the run is live, so you can prune or rewrite examples and carry on from the same checkpoint.

  • Spot duplicated, contradictory or truncated examples by their loss trajectory
  • Slice by tag, length or any custom signal you define
  • Fewer restarts when something looks off at step 2,000
  • Supervised fine-tuning
  • Instruction data
  • Domain adaptation

03 · Medical imaging

Rare findings are often the least covered.

In imaging, the cases a model sees least often matter most: a small lesion, an unusual presentation, an artefact that looks like pathology. WeightsLab shows which of them are thin or probably mislabelled in your data, and lets you fix them while the model trains. It runs on your infrastructure, so your images stay on your site.

  • Surface likely mislabels and hard cases from live per-sample loss
  • Slice by finding, scanner or site with tags and your own signals
  • Fix the data, then resume from the same checkpoint
  • Radiology
  • Pathology
  • Segmentation
  • Classification

04 · AI Training

Find label errors and weak classes in your training data.

Models on images, tabular data and other inputs often stall because of annotation errors and class imbalance. See which samples the model cannot learn, find those problems, and fix them in the same session.

  • Mislabels, outliers and imbalance show up in the live loss signals
  • Browse samples with predictions and ground truth overlaid
  • Pause, edit, resume; checkpoints preserved
  • Classification
  • Detection
  • Segmentation
  • Tabular data
  • Clustering
  • Satellite imagery

Watch the image demo

weightslab · data exploration
WeightsLab data exploration view: a grid of driving images with segmentation overlays, each labelled with its per-sample loss and sorted from highest to lowest

See it on one of your runs.

Try the sandbox in your browser, or book 30 minutes with us and bring a real run.

Book a demo Try the sandbox

Join the early-access programme

Tell us what you train and we'll get back to you with a demo and licence terms.