Toxic Comment Detector - Aurora - islamamro
Flags toxic vs non-toxic online comments.
Fine-tuned from distilbert-base-uncased on SetFit/toxic_conversations. Built, trained and published end-to-end via the Aurora Research Portal by islamamro on an NVIDIA RTX 3090.
- Task: text classification (2 classes)
- Held-out accuracy: 0.89
Use it
from transformers import pipeline
clf = pipeline("text-classification", model="Islamamro/toxic-conversations-aurora-islamamro")
clf("your text here")
Note
Trained on a 1,400-example demo subset - a proof of the Aurora build->train->publish pipeline, not a production model. Fine-tune on full data for real use.
- Downloads last month
- 19