Sec-Llama: a Compact Fine-Tuned LLM for Network Intrusion Detection in Kubernetes Clusters
Bibliographic record
Abstract
In today’s interconnected world, cyber threats have grown both pervasive and sophisticated, especially in the Artificial Intelligence (AI) era, where digital systems face unprecedented risks. As companies and enterprises increasingly rely on Kubernetes to orchestrate an enormous number of microservices and deliver high-quality services to users, a serious security threat endangers the application flow. In particular, recent advances in generative AI have increased the pace, scale, and level of cyberattacks on microservice-based systems. At the network level, traditional defensive tools fail to detect these threats accurately and timely, increasing the average latencies of detection and remediation and hence, risking worse damages on the services and system levels. Due to their excessive computation and memory requirements, large language models (LLM) have not been considered viable network intrusion detection systems (NIDS) or tools within the Kubernetes clusters. We developed Sec-Llama, a compact LLM for intrusion detection. We demonstrated a case study where we applied a memory-efficient data-driven technique incorporating Byte-based transformation of the raw network flow, to optimize the model’s training and inference processes, on resource-wise, while maintaining a 96% threat detection F1-score. To facilitate the model’s training and deployment processes for users, we have developed an application to monitor, train, and launch our Sec-Llama for real-time inferences that took, on average, 21.1ms per inference while occupying only 172 MB.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.001 | 0.007 |
| Meta-epidemiology (narrow) | 0.001 | 0.001 |
| Meta-epidemiology (broad) | 0.000 | 0.001 |
| Bibliometrics | 0.001 | 0.000 |
| Science and technology studies | 0.001 | 0.001 |
| Scholarly communication | 0.001 | 0.003 |
| Open science | 0.003 | 0.002 |
| Research integrity | 0.001 | 0.002 |
| Insufficient payload (model declined to judge) | 0.006 | 0.002 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".