vgel/

SYNTHETIC-2-RL-fn_name-Qwen3-32B-Hard

Subsetted and cleaned version of PrimeIntellect/SYNTHETIC-2-RL.

Subsetted to just prime_rl_code_* problems with a fn_name field (dropping the stdin/stdout problems) Prompts run through ftfy to correct mojibake (~3.5%) Answer formatting instructions stripped ("Write Python code to solve the problem...") so you can append your own Verification info unwrapped to fn_name (str) and test_cases (json) columns Problems with test_cases json longer than 1000 bytes (90 total) were dropped, to filter… See the full description on the dataset page: https://huggingface.co/datasets/vgel/SYNTHETIC-2-RL-fn_name-Qwen3-32B-Hard.