KodCode Light RL 10K Synthetic Programming Tasks

KodCode is a fully-synthetic open-source dataset providing verifiable solutions and tests for coding tasks. It contains 12 distinct subsets spanning domains from algorithmic to package-specific knowledge and difficulty levels from basic exercises to competitive programming. The dataset was created by KodCode and last updated in April 2025.

Use Cases

• Supervised fine-tuning of code models using the provided coding tasks and their verifiable solutions.
• Reinforcement learning tuning for code generation by leveraging the dataset's challenge levels and test cases.
• Benchmarking model performance across the 12 distinct subsets covering different programming domains.
• Training models on package-specific knowledge tasks contained within the dataset.

Strengths

• Contains 12 distinct subsets covering various programming domains.
• Includes tasks across multiple difficulty levels, from basic exercises to competitive programming.
• Provides verifiable solutions and tests for coding tasks.

Limitations

• Dataset size, row count, and specific column information are unknown.
• Data is fully synthetic, which may not fully capture the nuance and patterns of real-world code repositories.
• The 'Light RL 10K' variant suggests a potentially smaller or sampled version of a larger collection.

Provenance

Source

KodCode organization via Hugging Face.

Collection Method

Fully synthetic generation.

Time Range

null

Freshness

Last updated on 2025-04-02.

Geography

null