Modelpublic

jainsatyam26/xlmr-shield

sourceHugging Faceapache-2.0updated 5mo agoView on Hugging Face

0likes9downloads

Model Card

XLMRShield

XLM-RoBERTa-base dual-head classifier for content moderation and prompt injection detection. Fine-tuned with 40:60 resampling + per-class cap (30K) + WeightedRandomSampler.

Test Results

Metric	Score
Safety Accuracy	0.8561
Safety F1	0.5321
Category Macro F1	0.6116
Category Micro F1	0.6502

Per-Class F1

Category	F1
benign	0.5339
childsexualexploitation	0.4491
hateandharassment	0.6302
indiscriminate_weapons	0.5208
misinformationandspecialized_advice	0.4748
nonviolentcrimes	0.7812
piandjailbreak	0.6587
privacy	0.6972
sexual_content	0.6889
suicideandself_harm	0.5734
violent_crimes	0.7195

Architecture

—Backbone: xlm-roberta-base (278M params)
—Head 1: Safety binary (sigmoid)
—Head 2: 11-class multi-label (sigmoid)
—Tokenizer: xlm-roberta-base

Training

—Dataset: satyamsaf3ai/mergedcontentmoderationandpromptinjectionnew
—Balancing: 40:60 resampling + per-class cap 30K + WeightedRandomSampler
—BF16 mixed precision, AdamW + cosine LR, 4 epochs
—maxlength=128, batchsize=32, lr=2e-05