Add requirements and fix imports

[Ankitaghavate]

Summary

This PR adds a requirements.txt file and fixes import-level issues to ensure the dataset, preprocessing, and training pipeline run without errors.

Changes

• Added requirements.txt covering all dependencies used across datasets, models, and training code
• Fixed syntax and import-related issues that caused runtime failures
• Ensured preprocessing, tokenizer creation, and dataset loading work as intended

Notes

• No model logic or data-processing behavior was changed
• Changes are limited to stability, dependency clarity, and reproducibility

Checklist

• Code runs without syntax/import errors
• Dependencies clearly specified
• No breaking changes introduced