FALCON dual-encoder — SNORT / `e5-base-v2`

Contrastive encoder fine-tuned to map CTI text and SNORT rules into a shared embedding space. Backbone: intfloat/e5-base-v2.

Test-set metrics

split	recall@1	F1	threshold	diag mean	off-diag mean
pretrained	0.4738	0.2576	0.7030	0.8503	0.8149
run_0	0.9489	0.9065	0.6707	0.8161	0.0253
run_1	0.9526	0.9122	0.6849	0.8765	0.0660
run_2	0.9526	0.9248	0.6838	0.8613	0.0308
run_3	0.9539	0.9351	0.6925	0.8980	0.0355
run_4	0.9551	0.9440	0.7001	0.9503	0.0068

Training

Symmetric InfoNCE / NT-Xent over in-batch negatives. Best checkpoint selected by validation loss.

Run 0 — batch=16, epochs=5, lr=2e-05, schedule=constant, T=0.05
Run 1 — batch=50, epochs=10, lr=2e-05, schedule=constant, T=0.05
Run 2 — batch=70, epochs=30, lr=2e-05, schedule=constant, T=0.05
Run 3 — batch=128, epochs=30, lr=5e-05, schedule=warmup_cosine, T=0.05
Run 4 — batch=70, epochs=50, lr=2e-05, schedule=constant, T=0.07

Loading

from transformers import AutoModel, AutoTokenizer
tok   = AutoTokenizer.from_pretrained("shaswatamitra/falcon-snort-dual-e5-base-v2", subfolder='rule')
model = AutoModel.from_pretrained("shaswatamitra/falcon-snort-dual-e5-base-v2", subfolder='rule')

Dual-encoder layout: this repo has rule/ (encodes SNORT rules) and cti/ (encodes CTI text) subfolders. Load each with subfolder=....

Citation

@article{mitra2025falcon,
  title={FALCON: Autonomous Cyber Threat Intelligence Mining with LLMs for IDS Rule Generation},
  author={Mitra, Shaswata and Bazarov, Azim and Duclos, Martin and Mittal, Sudip and Piplai, Aritran and Rahman, Md Rayhanur and Zieglar, Edward and Rahimi, Shahram},
  journal={arXiv preprint arXiv:2508.18684},
  year={2025}
}