MENTHOS-rootcause

English

Model Description

MENTHOS-RootCause is a binary classifier fine-tuned from answerdotai/ModernBERT-base to detect whether a log line indicates a root-cause issue. It uses a maximum sequence length of 256.

Intended Use

• Binary classification of logs for root-cause detection.
• Labels: 0/1 from the prepared dataset.

Training Data

Trained on the MENTHOS root-cause dataset.

Benchmark Results

Benchmark results for the MENTHOS evaluation set.

model	samples	accuracy	precision	recall	f1	roc_auc	p50 latency (ms)	throughput (samples/s)
MENTHOS-rootcause	296	1.000000	1.000000	1.000000	1.000000	1.000000	6.0751	164.09

Reference baseline (Morpheus ONNX):

baseline model	accuracy	f1	roc_auc	p50 latency (ms)	throughput (samples/s)
root-cause-binary-bert-20230517.onnx	0.861486	0.839216	0.724343	65.7515	15.13

Benchmark Plots

Limitations

• Perfect benchmark scores may reflect dataset construction and domain overlap.
• Generalization to unseen infrastructures/log formats should be validated separately.

Citation

@misc{borovic_li-dobnik_kranjec_ferme_2026,
  title        = {MENTHOS-rootcause},
  author       = {Borovic, Li Dobnik, Kranjec, Ferme},
  year         = {2026},
  publisher    = {Hugging Face},
  howpublished = {\url{https://huggingface.co/LHRS-UM-FERI/MENTHOS-rootcause}}
}

---

Slovenščina

Opis modela

MENTHOS-RootCause je binarni klasifikator, naučen iz answerdotai/ModernBERT-base, za zaznavanje root-cause dogodkov v dnevniških zapisih. Uporablja maksimalno dolžino zaporedja 256.

Namen uporabe

• Binarna klasifikacija logov za root-cause detekcijo.
• Oznake: 0/1 iz pripravljenega dataseta.

Učni podatki

Učenje je potekalo na MENTHOS root-cause datasetu.

Rezultati benchmarka

model	vzorcev	accuracy	precision	recall	f1	roc_auc	p50 latenca (ms)	prepustnost (vzorcev/s)
MENTHOS-rootcause	296	1.000000	1.000000	1.000000	1.000000	1.000000	6.0751	164.09

Referenčni baseline (Morpheus ONNX):

baseline model	accuracy	f1	roc_auc	p50 latenca (ms)	prepustnost (vzorcev/s)
root-cause-binary-bert-20230517.onnx	0.861486	0.839216	0.724343	65.7515	15.13

Grafi benchmarka

Omejitve

• Popolni rezultati na benchmarku lahko odražajo strukturo podatkov in domensko podobnost.
• Posploševanje na druge sisteme/log formate je treba dodatno validirati.

Citiranje

@misc{borovic_li-dobnik_kranjec_ferme_2026,
  title        = {MENTHOS-rootcause},
  author       = {Borovic, Li Dobnik, Kranjec, Ferme},
  year         = {2026},
  publisher    = {Hugging Face},
  howpublished = {\url{https://huggingface.co/LHRS-UM-FERI/MENTHOS-rootcause}}
}