The LinkBERT model runs as an isolated CPU service separate from the main web app, serving scoring and health checks over local HTTP so heavy neural inference stays out of the application process.
Withheld — proprietary.
flowchart LR
A[Auxy Web App] -->|HTTP request| B[LinkBERT Service]
B -->|PyTorch CPU Inference| C[LinkBERT Model]
C -->|Token Spans and Probabilities| B
B -->|JSON Response| A