Spaces:
Running
Running
| from transformers import pipeline | |
| # Load once (slow only first time) | |
| nli_pipeline = pipeline( | |
| "text-classification", | |
| model="roberta-large-mnli", | |
| device=-1 # CPU | |
| ) | |
| def nli_contradiction(text1, text2, threshold=0.8): | |
| """ | |
| Returns True if NLI model strongly predicts contradiction | |
| """ | |
| input_text = f"{text1} </s></s> {text2}" | |
| result = nli_pipeline(input_text)[0] | |
| return ( | |
| result["label"] == "CONTRADICTION" and | |
| result["score"] >= threshold | |
| ) | |