act_pusht_dec7_bs8_800k

Batch-8 equal-sample-budget run (800k steps = 6.4M samples): 159.1 avg_sum_imputed_reward, 60.2% success (n=5000). Matches batch-64 at 100k steps to one decimal.

Part of a 6-week robot-learning training block (https://github.com/JaHeRoth/robot-learning). Evaluated on gym-pusht at n_action_steps=16; avg_sum_imputed_reward imputes 0.95/step to horizon 300 after success.