Sean13/grpo_nocurriculum_Qwen3-1.7B-100step

Condition A — GRPO alone (100 RL steps, no curriculum)

Final-step (step 100) policy from Condition A of the 2×2 ablation in Cao 2026 disentangling MaxRL and curriculum learning.

Base	Qwen/Qwen3-1.7B-Base
Algorithm	GRPO (actor_rollout_ref.algorithm.adv_estimator=grpo)
Curriculum	OFF (uniform sampling over POLARIS-53K)
Steps	100
Hardware	4×H200 on ou_sloan_gpu (MIT ORCD)
Training-time val (step 100)	MATH P@32 = 0.892, AIME P@32 = 0.247

Reproduction

• Code: https://github.com/XiaoyangCao1113/maxrl-curriculum
• Recipe: scripts/06_train_A_*.sbatch + scripts/_train_inner_AB.sh