YAML-BERT — structural understanding of Kubernetes YAML

Code: github.com/vimalk78/yaml-bert · Trained with MLM + reconstruction on 276K K8s manifests · 18,414,480 params

This Space includes 3 demos — pick a tab below, or use the tiles on the Overview tab.

Demos in this Space

🧩 Missing-field suggester

Paste a YAML manifest; the model predicts which structural fields it expects but you didn't include, ranked by confidence.

🌌 Manifest galaxy

10,000 K8s manifests projected to 2D from their doc_vec embeddings. Kinds cluster spontaneously — the model was never told what kind is.

🔬 Structural probes

Preset manifest sets that test whether the model has learned specific structural distinctions, shown as a 2D layout. Add your own YAML to see where it lands.