entailment.py

from transformers import pipeline
import numpy as np

def analyze_entailment(original_sentence, paraphrased_sentences, threshold):
    # Load the entailment model using pipeline
    entailment_pipe = pipeline("text-classification", model="ynie/roberta-large-snli_mnli_fever_anli_R1_R2_R3-nli")

    # Function to perform entailment
    def check_entailment(premise, hypothesis):
        results = entailment_pipe(f"{premise} [SEP] {hypothesis}", return_all_scores=True)
        return results[0]

    all_sentences = {}
    selected_sentences = {}
    discarded_sentences = {}

    # Check entailment for each paraphrased sentence
    for paraphrased_sentence in paraphrased_sentences:
        entailment_results = check_entailment(original_sentence, paraphrased_sentence)
        entailment_score = next(result['score'] for result in entailment_results if result['label'] == 'entailment')
        
        all_sentences[paraphrased_sentence] = entailment_score
        
        if entailment_score >= threshold:
            selected_sentences[paraphrased_sentence] = entailment_score
        else:
            discarded_sentences[paraphrased_sentence] = entailment_score

    return all_sentences, selected_sentences, discarded_sentences

print(analyze_entailment("I love you", ["You're being loved by me"], 0.7))