Code Generation
Producing or completing source code from a description or surrounding context.
Overview
Code generation produces code from a natural language description, or continues the file the cursor is in. It solves a task that language models happen to suit unusually well: code has strict syntax and immediate feedback, so a wrong answer is cheap to detect and often cheap to correct. It exists as one of the first commercially successful applications of large models, and its accuracy depends far more on context than on the model alone.
Technology Graph
How Code Generation connects to the rest of the stack.
Uses and depends on
Applied to
- HumanEval benchmark Benchmarks HumanEval used for Code Generation High ✓ source
- SWE-bench benchmark Benchmarks SWE-bench used for Code Generation High ✓ source