Filter Low-Quality Synthetic Examples
Beginner
The problem, from first principles
Generating synthetic training data mechanically (templates, model self-generation) produces volume, but not every generated example is actually usable -- some are too short to carry any real signal, and some are degenerate model outputs like a refusal or a disclaimer that leaked into what should have been a clean training example.
The task
Write filter_quality(examples, min_length, banned_phrases) that keeps only examples at least min_length characters long and containing none of banned_phrases, case-insensitively, preserving original order.
Keywords
production-systemssynthetic-data
Python
Run your code to see results
No results yet
Run your code to check it against the examples, or submit to run every test.