feat: HuggingFace cybersec model integration — pretrained classifier on ingested entries #9
Labels
No labels
compliance
demo
deployment
docs
enhancement
parser
patterns
performance
security
ux
No project
No assignees
1 participant
Notifications
Due date
No due date set.
Dependencies
No dependencies set.
Reference: Circuit-Forge/turnstone#9
Loading…
Reference in a new issue
No description provided.
Delete branch "%!s()"
Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?
Integrate a pretrained cybersec/log-anomaly model from HuggingFace (candidates:
markusbayer/log-anomaly-detection, datasets likeSecBench,BETH, or LANL auth logs models). Run inference on newly ingested entries, persist aml_scoreandml_labelcolumn. Local inference via Ollama or direct transformers.Closing — cybersec zero-shot scoring pipeline is complete and verified.
First clean run: 320 entries scored, 0 detections (as expected for a fresh baseline),
last_error: null, 38.6 min CPU inference time. The blocking "database is locked" contention (163,975 errors) was resolved in the same session by removing per-flush FTS sync fromwatcher.py(see commitccc9a9e). The security alerts UI for displaying detections was shipped in commita33d983.Model:
MoritzLaurer/DeBERTa-v3-base-mnli-fever-anli, threshold: 0.60, device: cpu.