How nguynking/cs224n works
Project type and purpose
• This repository is primarily a student’s collection of solutions for the Stanford CS224N assignments (A1–A5), plus a “minBERT” final project under project/minbert-default-final-project-main.
• The “minBERT” subfolder is a standalone NLP tool: a small‐scale BERT-style model supporting tokenization, pretraining, fine-tuning, and multitask classification on several datasets.
File structure
• Top-level:
├ .gitignore
├ README.md (course overview)
├ requirements.txt
├ assignment1 … assignment5 (each a self-contained homework)
└ project/minbert-default-final-project-main (the final project)
• In the minBERT project folder:
– Configuration & setup: config.py, setup.sh, run.sh, prepare_submit.py
– Data & preprocessing: data/*.csv, tokenizer.py, datasets.py
– Model definitions: base_bert.py, bert.py, classifier.py, multitask_classifier.py
– Training loop: optimizer.py, evaluation.py, sanity_check.py
– Utilities: utils.py
– Predictions output directory
README for additional insights
• The high-level README emphasizes CS224N course solutions; the “Future works” note calls out completion of minBERT.
• The project subfolder has its own README.md and STRUCTURE.md (not shown), which likely describe the training/evaluation phases.
Detailed elements to capture
– datasets.py and tokenizer.py grouped as “Data Prep”
– base_bert.py, bert.py, classifier.py, multitask_classifier.py grouped as “Model Library”
– optimizer.py + run script as “Training Pipeline”
– evaluation.py + predictions folder as “Evaluation/Submission”
– config.py centrally feeding into all components