auren-research/lunaris-sentinel

Lunaris Sentinel

Official Lunaris Sentinel multi-task security encoder (Auren Research).

Fine-tuned from answerdotai/ModernBERT-base on auren-research/astraea-unified-threat.

Heads

Head	Task
cls_head	Malicious / benign (log_classification, phishing)
mitre_head	MITRE technique classification (791 classes)
ner_head	IOC token NER (17 BIO labels)
anomaly_head	Anomaly score regression

Load (training codebase)

from training.model import AstraeaConfig, AstraeaModel
import torch

cfg = AstraeaConfig(
    base_model_name="answerdotai/ModernBERT-base",
    num_mitre_classes=791,
)
model = AstraeaModel(cfg)
state = torch.load("pytorch_model.bin", map_location="cpu", weights_only=True)
model.load_state_dict(state, strict=False)

Package metadata

{
  "model_id": "lunaris-sentinel",
  "base_model": "answerdotai/ModernBERT-base",
  "checkpoint_root": "/root/neobert-astraea-finetune/checkpoints/astraea-modernbert-full",
  "primary_weights": "/root/neobert-astraea-finetune/checkpoints/astraea-modernbert-full/checkpoint-3000",
  "num_mitre_classes": 791,
  "num_ner_labels": 17,
  "checkpoints_included": [
    "checkpoint-2000",
    "checkpoint-2500",
    "checkpoint-3000"
  ]
}