Fine-Tune a Sequence-to-Sequence Model for Code-Doc Generation
Overview
What this challenge is about.
Take a small base model (CodeT5+ or a distilled CodeLlama-Instruct). Build the dataset by mining around 8,000 high-quality function-docstring pairs from permissively-licensed Python repos (high test coverage, mature projects). Fine-tune. Evaluate automatically with BLEU, ROUGE-L, and CodeBLEU. Run a 20-developer study scoring docstring quality on a 5-point Likert across 30 sample functions, fine-tuned vs base. Write a 4-page memo with a ship/no-ship recommendation.
The Brief
What you'll do, and what you'll demonstrate.
Fine-tune a seq2seq model for Python docstring generation that wins both automated metrics and a 20-developer user study.
Earning criteria — what you'll demonstrate
- Build a high-quality dataset from open-source code
- Fine-tune a seq2seq code-LM with parameter-efficient methods
- Evaluate code-generation with both automated and human metrics
- Translate evaluation into a product decision
Program Fit
Where this fits in your program.
Sharpens the same skills your degree expects you to demonstrate.
Skills
Skills you'll demonstrate.
Each one shows up on your verified credential.
Careers
Roles this prepares you for.
Real titles. Real skill bridges. Pick the one closest to your trajectory.
NLP Engineer
Fine-tuning code-LMs with paired human evaluation is the NLP-engineer work that developer-tools companies invest in to differentiate.
This challenge sharpens
- seq2seq
- transformers
- code-nlp
Applied AI Scientist
Balancing automated metrics with developer studies and making a ship/no-ship call is exactly the applied-AI work AI startups need on their first hire.
This challenge sharpens
- lora-fine-tuning
- user-study-design
- code-nlp
AI Engineer
Owning the dataset build + fine-tune + eval loop end to end is the AI-engineer skill set developer-tools startups recruit for.
This challenge sharpens
- pytorch
- lora-fine-tuning
- transformers