Log Anomaly Detection on Thunderbird
99.35F1 ScoreFlogTinyLLM
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| FlogTinyLLMBackbone Architecture=OPT-1.32026.04 | 99.35 | 98.79 | 99.96 | |
| FlogTinyLLMBackbone Architecture=Phi-1.52026.04 | 99.08 | 98.28 | 99.94 | |
| FlogTinyLLMBackbone Architecture=DeepSeek-R12026.04 | 97.97 | 96.15 | 99.96 | |
| Federated LogGPT2026.04 | 97.74 | 99.24 | 96.29 | |
| FlogTinyLLMBackbone Architecture=TinyLlama2026.04 | 96.6 | 93.54 | 99.91 | |
| DeBERTa-v3Category=Fine-Tuned2026.04 | 96.1 | — | — | |
| RoBERTa-baseCategory=Fine-Tuned2026.04 | 95.3 | — | — | |
| BERT-baseCategory=Fine-Tuned2026.04 | 94.7 | — | — | |
| GPT-4 + SLCP (5-shot)Category=LLM (Prompt), Shot-count=5, Prompting strategy (SLCP)=true2026.04 | 89.7 | — | — | |
| Drain + RFCategory=Traditional2026.04 | 88.6 | — | — | |
| Drain + RF (best)Category=Traditional2026.04 | 88.6 | — | — | |
| GPT-4 (5-shot)Category=LLM (Prompt), Shot-count=5, Prompting strategy (SLCP)=false2026.04 | 87.2 | — | — | |
| GPT-4 + SLCP (zero)Category=LLM (Prompt), Shot-count=0, Prompting strategy (SLCP)=true2026.04 | 86.4 | — | — | |
| Drain + LRCategory=Traditional2026.04 | 85.3 | — | — | |
| Spell + SVMCategory=Traditional2026.04 | 84.9 | — | — | |
| Federated LogBERT2026.04 | 84.77 | 96.25 | 75.74 | |
| GPT-4 (zero-shot)Category=LLM (Prompt), Shot-count=0, Prompting strategy (SLCP)=false2026.04 | 83.1 | — | — | |
| LogDeep(FL)2026.04 | 81.91 | 100 | 79.35 | |
| LLaMA-3 (zero-shot)Category=LLM (Prompt), Shot-count=0, Prompting strategy (SLCP)=false2026.04 | 78.6 | — | — | |
| GPT-3.5 (zero-shot)Category=LLM (Prompt), Shot-count=0, Prompting strategy (SLCP)=false2026.04 | 76.3 | — | — |