Sign in

LinkBERT runs as an isolated local service to keep heavy PyTorch weights out of the main web app, serving CPU inference with a generous timeout and character limits to keep latency predictable on large documents.

flowchart LR
    App[Auxy Web App] -->|Send text| LB[LinkBERT Service]
    LB --> Model[(PyTorch Model)]
    Model --> Inference[CPU Inference Engine]
    Inference -->|Words, Probs, Spans| LB
    LB -->|JSON Response| App
    App --> UI[Browser Slider Tuning]

Referenced by