s-nlp/bart-base-detox-svd

Model Overview

It is a SVD-compressed model of original BART-based detoxification model s-nlp/bart-base-detox.

How to use

from transformers import AutoModelForSeq2SeqLM, AutoTokenizer
model = AutoModelForSeq2SeqLM \
    .from_pretrained('s-nlp/bart-base-detox-svd', trust_remote_code=True)
tokenizer = AutoTokenizer.from_pretrained('facebook/bart-base')

toxics = ['that sick fuck is going to be out in 54 years.']
tokens = tokenizer(toxics)
tokens = model.generate(**tokens, num_return_sequences=1, do_sample=False,
                        temperature=1.0, repetition_penalty=10.0,
                        max_length=128, num_beams=5)
neutrals = tokenizer.decode(tokens[0, ...], skip_special_tokens=True)
print(neutrals) # stdout: She is going to be out in 54 years.