The LinkBERT model runs as a standalone local CPU service, exposing word-level probability scores over an HTTP endpoint.

sequenceDiagram
    autonumber
    participant App as Main Web App
    participant Service as LinkBERT Service
    participant Model as PyTorch (CPU)

    App->>Service: Send raw text
    activate Service
    Service->>Model: Run Tokenizer & Inference
    activate Model
    Model-->>Service: Output Tensors
    deactivate Model
    Service->>Service: Map Spans & Probabilities
    Service-->>App: JSON (Words, Scores, Character Spans)
    deactivate Service