At Graybx we build developer tools for ML engineers who train computer vision models on image, video & LiDAR data — bringing visibility, debuggability & control to the PyTorch training loop. Weightslab is our first product.

ML engineers spend too much time debugging blind — we build tools to fix that.

⭐ Star weightslab · Quickstart · ⏰ Early Access

PyTorch dataset debugger for computer vision — pause training, mine live loss signals to surface mislabels, class imbalance & outliers, then curate your image, video & LiDAR data without restarting.

• SDK — Wrap your training script to capture live, per-sample signals from your model, optimizer, and loss functions.

• Studio — Visual interface to inspect those signals, surface problematic samples, and curate your image, video & LiDAR datasets — mid-training, without restarting.

• Documentation – unified docs for all Graybx projects (coming soon)
• Slack Community – ask questions, share feedback, talk shop
• Graybx Academy – free courses on Graybx products (coming soon)
• Website – product updates and deep dives