Cisco Introduces Model Foundry for Specialized AI Models

Cisco introduced Model Foundry, a service for accessing, serving and fine tuning its specialized AI models, in a company blog post. The service entered Controlled Availability at Partner Summit, with partners receiving access first.

Model Foundry runs within Cisco Cloud Control and provides OpenAI compatible API endpoints. Partners can test models in a playground, create API keys, control access by model and monitor usage by model, key and day. The service also handles training infrastructure and compares base and tuned models before deployment.

The initial catalog includes Deep Network 1.4 for network troubleshooting, Antares-3B for locating vulnerable source files, Decision Model 1.0 for selecting tools and evaluating evidence, and Cisco Time Series Model 1.0 for operational data. It also includes models from the Foundation-Sec, SecureBERT and Antares families.

Cisco tested a tuned version of its 31 billion parameter Deep Network Model with World Wide Technology. In lab tests covering routing, link failures, policy changes and multiple faults, the model recommended the correct remediation in 11 of 12 scenarios.