Multi-SWE-RL-Clean: Filtered Reinforcement Learning Dataset for Code Generation

A filtered subset of the PrimeIntellect/Multi-SWE-RL dataset intended for reinforcement learning training. The dataset was created by running a script to remove the C++ programming language examples and to require a perfect validation reward score. PrimeIntellect authored this dataset, which was last updated on May 15, —.

Use Cases

• Training reinforcement learning agents for code generation based on the filtered, RL-trainable subset.
• Benchmarking code synthesis models based on the requirement for perfect validation rewards.
• Studying the effectiveness of different programming languages in RL training environments based on the removal of C++ examples.

Strengths

• Filtered to require a perfect validation reward of 1.0, suggesting high-quality examples.
• C++ examples were removed after a test found zero passing rows, potentially improving dataset focus.

Limitations

• Description metadata is limited; actual data quality requires manual inspection after download.
• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.

Provenance

Source

PrimeIntellect

Collection Method

Created by filtering the PrimeIntellect/Multi-SWE-RL dataset with a custom Python script.

Time Range

null

Freshness

Last updated 2026-05-15 00:28:31; freshness should be verified.

Geography

null