TACO Verified: Programming Solutions Passing All Test Cases

A filtered subset of the TACO dataset, last updated in April 2025, containing only verified programming solutions that pass all test cases. The dataset, created by author likaixin, includes 12,898 problems and 1,043,251 solutions, with a 71.03% correct ratio after removing failing solutions and problems with no correct answer.

Use Cases

• Training code generation models on verified, correct solutions.
• Benchmarking the reliability of AI-generated code based on test case pass rates.
• Analyzing patterns in programming errors by comparing the original and verified datasets.
• Studying the characteristics of problems that have at least one correct solution.

Strengths

• Contains 1,043,251 verified solutions, providing a substantial corpus of correct code.
• Filtered to a 71.03% correct ratio, improving data reliability for training.
• Execution environment details are specified (Intel E5-2620 v3 CPUs, 10-second timeout).

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• The dataset is derived from a specific source (TACO training set), which may introduce inherent bias.

Provenance

Source

huggingface

Collection Method

Derived from the TACO dataset's training set by filtering solutions that pass all test cases and removing problems with no correct solution.

Time Range

null

Freshness

Last updated 2025-04-17 04:00:18; freshness should be verified.

Geography

null