---
license: mit
datasets:
- tdavidson/hate_speech_offensive
base_model:
- FacebookAI/roberta-large
pipeline_tag: text-classification
library_name: transformers
---
# Davidson RoBERTa Hate Speech Classifier

- Model: roberta-large fine-tuned for 3-way classification (toxic, neutral, non-toxic).  
- Dataset: tdavidson/hate_speech_offensive (Twitter), split into train/val/test locally.  
- Metrics (test): paste from metrics.json.  
- Intended use: content moderation research/demos; not for deployment without bias/fairness review.  
- Limitations/risks: social bias, dataset age/domain mismatch; errors possible on slang/irony.  
- How to use:
## Usage
from transformers import AutoModelForSequenceClassification, AutoTokenizer, pipeline

mid = "Yash22CSU192/davidson-roberta-hatespeech"

# Load tokenizer and model
tok = AutoTokenizer.from_pretrained(mid)
mdl = AutoModelForSequenceClassification.from_pretrained(mid)

# Create a text-classification pipeline
clf = pipeline("text-classification", model=mdl, tokenizer=tok, return_all_scores=True)

# Test the classifier
print(clf("Have a nice day."))



## Files
- model.safetensors, config.json, tokenizer.json, tokenizer_config.json, vocab.json, merges.txt, special_tokens_map.json
- training_args.bin (Trainer settings), metrics.json (evaluation summary)