Contradiction Detector using an HF roberta project by ynie @ FB/Meta

Contradiction_test1.py

from transformers import AutoTokenizer, AutoModelForSequenceClassification # 4.0.1
import torch # 1.7

# use "contra activate dual" on laptop
# https://github.com/facebookresearch/ParlAI/issues/3391
# https://github.com/facebookresearch/ParlAI/issues/3665
# https://arxiv.org/abs/2012.13391
# https://huggingface.co/ynie/roberta-large_conv_contradiction_detector_v0

if __name__ == '__main__':
    max_length = 256

    hg_model_hub_name = "ynie/roberta-large_conv_contradiction_detector_v0"

    tokenizer = AutoTokenizer.from_pretrained(hg_model_hub_name)
    model = AutoModelForSequenceClassification.from_pretrained(hg_model_hub_name)
    
    premiseList =["I'm an Spanish teacher.",
    "I am at home.",
    "I love my job.",
    "I have no siblings",
    "I found the treasure.",
    "I am alone.",
    "I am single.",
    "I know python."]
    
    # premise = "I'm an Spanish teacher."
    while(True):

        hypothesis = input("Hypo:") # "I don't know how to speak Spanish.","I only speak English","I am married",...
        for premise in premiseList:


            tokenized_input_seq_pair = tokenizer.encode_plus(premise, hypothesis,
                                                             max_length=max_length,
                                                             return_token_type_ids=True, truncation=True)

            input_ids = torch.Tensor(tokenized_input_seq_pair['input_ids']).long().unsqueeze(0)
            # remember bart doesn't have 'token_type_ids', remove the line below if you are using bart.
            token_type_ids = torch.Tensor(tokenized_input_seq_pair['token_type_ids']).long().unsqueeze(0)
            attention_mask = torch.Tensor(tokenized_input_seq_pair['attention_mask']).long().unsqueeze(0)

            outputs = model(input_ids,
                            attention_mask=attention_mask,
                            token_type_ids=token_type_ids,
                            labels=None)

            predicted_probability = torch.softmax(outputs[0], dim=1)[0].tolist()  # batch_size only one

            #print("Premise:", premise)
            # print("Hypothesis:", hypothesis)
            # print("Non contradiction:", predicted_probability[0])
            # print("Contradiction:", predicted_probability[1])
            thresh = 0.7
            if ( predicted_probability[1]>predicted_probability[0]):
                if (predicted_probability[1]> thresh):
                    print("    |--->:",premise, predicted_probability[1],predicted_probability[0])
                else:
                    print("    |    :",premise, predicted_probability[1],predicted_probability[0])
                 
            else:
                if (predicted_probability[0]> thresh):
                    print("    |   *:",premise, predicted_probability[1],predicted_probability[0])
                else:
                    print("    |    :",premise, predicted_probability[1],predicted_probability[0])