| # ============================================================ |
| # SIMPLE EXAMPLE: How to Use Your Trained Model |
| # ============================================================ |
|
|
| ```python |
| from transformers import AutoTokenizer, AutoModelForSequenceClassification |
| import torch |
| |
| # Step 1: Load the model and tokenizer from the local directory |
| # (This assumes you ran Cell 18 earlier to save the model) |
| model_path = "optimized-bert-model" |
| model = AutoModelForSequenceClassification.from_pretrained(model_path) |
| tokenizer = AutoTokenizer.from_pretrained(model_path) |
| |
| # Step 2: Put model in evaluation mode |
| model.eval() |
| |
| # Step 3: Test on simple examples using a helper function |
| |
| def predict_paraphrase(sentence1, sentence2): |
| """ |
| Predicts whether two sentences are paraphrases and returns prediction and confidence. |
| """ |
| inputs = tokenizer(sentence1, sentence2, return_tensors="pt", |
| truncation=True, padding=True, max_length=128) |
| with torch.no_grad(): |
| outputs = model(**inputs) |
| logits = outputs.logits |
| prediction = torch.argmax(logits, dim=1).item() |
| confidence = torch.softmax(logits, dim=1)[0].max().item() |
| return prediction, confidence |
| |
| def display_result(example_idx, sentence1, sentence2): |
| prediction, confidence = predict_paraphrase(sentence1, sentence2) |
| print("="*60) |
| print(f"EXAMPLE {example_idx} - Are these paraphrases?") |
| print("="*60) |
| print(f"Sentence 1: {sentence1}") |
| print(f"Sentence 2: {sentence2}") |
| print(f"Prediction: {'YES (paraphrases)' if prediction == 1 else 'NO (not paraphrases)'}") |
| print(f"Confidence: {confidence:.4f}") |
| print() |
| |
| # Example 1: Two sentences that ARE paraphrases |
| sentence1_1 = "The cat is sleeping on the mat" |
| sentence2_1 = "The cat is napping on the mat" |
| |
| display_result(1, sentence1_1, sentence2_1) |
| |
| # Example 2: Two sentences that are NOT paraphrases |
| sentence1_2 = "The dog is barking loudly" |
| sentence2_2 = "I love eating pizza" |
| |
| display_result(2, sentence1_2, sentence2_2) |
| |
| print("="*60) |
| |
| # ----------------------- |
| # Try your own examples! |
| # ----------------------- |
| # Uncomment and edit the sentences below to test your own custom examples: |
| # user_sentence1 = "Your first sentence here." |
| # user_sentence2 = "Your second sentence here." |
| # display_result("USER", user_sentence1, user_sentence2) |
| ``` |
|
|
| # ------------------------------------------------------------ |
| # How to call/use this model: |
| # ------------------------------------------------------------ |
| # 1. Make sure you have the saved model files in the directory 'optimized-bert-model' |
| # 2. Run this script in your Python environment (with 'transformers' and 'torch' installed) |
| # 3. Change the example sentences inside the code block above to your own inputs to test paraphrase detection |
| # 4. The script prints whether the sentences are paraphrases and gives a confidence score |
|
|
| # Sample Output: |
| # ============================================================ |
| # EXAMPLE 1 - Are these paraphrases? |
| # ============================================================ |
| # Sentence 1: The cat is sleeping on the mat |
| # Sentence 2: The cat is napping on the mat |
| # Prediction: YES (paraphrases) |
| # Confidence: 0.9998 |
| # |
| # ============================================================ |
| # EXAMPLE 2 - Are these paraphrases? |
| # ============================================================ |
| # Sentence 1: The dog is barking loudly |
| # Sentence 2: I love eating pizza |
| # Prediction: NO (not paraphrases) |
| # Confidence: 0.9584 |
| # ============================================================ |
