Skip to content

Dataset · KodCode

Synthetic coding problems with test suites

KodCode/KodCode-V1

A large synthetic code task corpus with verifiable solutions and tests for SFT and RL.

What's inside

KodCode is a fully synthetic open dataset featuring verifiable solutions and tests for programming tasks. It includes 12 subsets ranging from algorithms to package-specific knowledge, and from basic exercises to interviews and competitive programming. The data is in English, with a size between 100k and 1M examples; modalities are tables and text.

What you can do with it

  • Fine-tune a code model on SFT with solution verification
  • Perform RL tuning on tasks with tests
  • Evaluate code generation across various difficulty levels
Languages
English
Size
100K–1M rows
Downloads this month
31K
Likes
121

Models trained on it

How to load it

from datasets import load_dataset

ds = load_dataset("KodCode/KodCode-V1")