LogLLM — Reproduction
Reproducing LLM-based log anomaly detection, with reproducibility and robustness fixes.
Reproduction of LogLLM (arXiv:2411.08561), a framework for log anomaly detection built on large language models. The contribution here is code-only: reproducibility seeding, replacing hard-coded paths with argparse, a bugfix in robust answer parsing, and optional quantization / CPU execution so the pipeline runs without a high-end GPU.
Original implementation: guanwei49/LogLLM.
Stack: Python · Transformers