Modelpublic

felipemunoz/prompt-enhancer-safety-deberta-v3

sourceHugging Faceapache-2.0updated 6mo agoView on Hugging Face

0likes

Model Card

Prompt Enhancer Safety Classifier (DeBERTa-v3)

DeBERTa-v3-base fine-tuned as a dual-head safety classifier for LLM prompts. Classifies prompts by risk category and severity. Exported to ONNX for fast CPU inference.

Training Metrics

Parameter	Value
data_path	/home/felipe/hookpromptenhancer/training/data/safety_train.jsonl
data_splits	{'train': 6315, 'val': 701}
hyperparameters	{'learningrate': 2e-05, 'batchsize': 16, 'epochs': 10, 'weightdecay': 0.01, 'warmupratio': 0.1, 'time_budget': 600}
model_name	microsoft/deberta-v3-base
seed	42
stage	safety
timestamp	2026-03-29T11:59:45.399789+00:00
val_macro_f1	0.9768

Usage

python

import onnxruntime as ort
from transformers import AutoTokenizer

tokenizer = AutoTokenizer.from_pretrained("felipemunoz/prompt-enhancer-safety-deberta-v3")
session = ort.InferenceSession("model.onnx")
inputs = tokenizer("Is this prompt safe?", return_tensors="np")
outputs = session.run(None, dict(inputs))

License

Apache-2.0