mia-project-2025/

bert-base-uncased-finetuned-quora-question-pairs

Training Data

Preprocessing

• Extracted the two questions per pair:
  • question1 = questions.text[0]
  • question2 = questions.text[1]
• Converted the is_duplicate field to binary labels (0 = not duplicate, 1 = duplicate).
• Tokenized using tomaarsen/bert-base-nq-prompts tokenizer.
• Maximum sequence length: 128

Results

Evaluation Metrics

Training Metrics