OctoReasoner/verl_codeql1_aggressive_filtered

verl codeql1 aggressive filtered: RL dataset on Hugging Face. Curriculum-filtered subset of OctoReasoner/verl codeql1 aggressive. CoT variant. Models were prompted to think first, then box (max tokens=4096). Kept rows satisfy weak <= 5/8 AND strong = 2/8, where weak/strong = number of correct…