CoolFace
Modelpublic

jainsatyam26/mamba-shield-v2

sourceHugging Faceapache-2.0updated 5mo agoView on Hugging Face
0likes
Model Card

MambaShield v2 (Balanced)

Mamba-2 SSM dual-head classifier for content moderation and prompt injection detection. Trained with per-class balancing (cap=30K) + WeightedRandomSampler for improved rare-class detection.

Performance

MetricScore
Safety Accuracy83.1%
Category Macro F10.482
Category Micro F10.514

Architecture

  • —Backbone: Mamba-2 SSD × 4 layers, dmodel=256, nheads=8
  • —Parameters: 9.6M
  • —Tokenizer: bert-base-uncased
  • —Heads: Safety binary (sigmoid) + 11-class multi-label (sigmoid)

Categories (11)

CategoryTest F1
benign0.41
childsexualexploitation0.35
hateandharassment0.47
indiscriminate_weapons0.32
misinformationandspecialized_advice0.38
nonviolentcrimes0.68
piandjailbreak0.60
privacy0.55
sexual_content0.44
suicideandself_harm0.49
violent_crimes0.61

Training

  • —Dataset: 285K samples after 40:60 resampling + per-class cap (30K max per class)
  • —WeightedRandomSampler for rare class oversampling
  • —BF16 mixed precision, AdamW + cosine LR, 5 epochs
  • —NVIDIA L4 24GB