language: en tags:
- text2text-generation
- text-style-transfer
- detoxification
- bart model-index:
- name: AgenticDetox-BART
results:
- task:
type: text-style-transfer
name: Text Detoxification
dataset:
name: DetoxLLM Benchmark
type: custom
metrics:
- type: joint_score value: 63.95 name: Joint Score
- task:
type: text-style-transfer
name: Text Detoxification
dataset:
name: DetoxLLM Benchmark
type: custom
metrics:
AgenticDetox-BART
This model is a fine-tuned version of facebook/bart-large designed for text detoxification (text style transfer). It rewrites harmful, aggressive, or offensive text into neutral, non-toxic language while preserving the original meaning and intent.
Model Description
Model type: Sequence-to-Sequence (Seq2Seq) Transformer
Base model: facebook/bart-large
Language:English
Paper: AgenticDetox: Recursive Agentic Reasoning Framework with RAG for Detoxification
Usage
You can use this model directly with the transformers library using a pipeline for text-to-text generation.
from transformers import pipeline
# Load the model
detoxifier = pipeline("text2text-generation", model="RAJESHKANNA21/bart-large-toxic-detoxification")
# Test with a toxic input
toxic_text = "It's all good. They made a really stupid comment that would piss any reasonable person off."
detoxified_text = detoxifier(toxic_text, max_length=50)
print(detoxified_text[0]['generated_text'])
# Expected Output: "It's all good. They made a really thoughtless comment that would upset a reasonable person."
---