← Back to AIBOM index

Cisco · local · 8B

Foundation-Sec-8B-Reasoning

License
Llama 3.1 Community License
Provider
Cisco
Parameters
8B

Chain assignments

Which inference chains this model serves in production. Source: policy.toml.

  • secret-scan

Most-recent measurement

Cyber benchmark composite = outperforms Llama 70B baseline

Bench: CyberMetric (Cisco internal) · Measured: 2025-12

Provenance

Weights source: https://huggingface.co/fdtn-ai/Foundation-Sec-8B-Reasoning

Notes

  • Self-hosted on RTX 4090. $0 marginal cost per inference.
  • Primary on the secret-scan chain — cyber-pretrained base outperforms general-purpose models on credential pattern detection.
  • Llama 3.1 base is dual-tracked for defense/finance/sovereign accounts where Llama license is flagged.
Source of truth: app/_data/model-cards.ts · Updated via PR + reviewable in git history.