{"step": 0, "arm": "dppo_pilot", "loss": "dppo", "reward_mean": 0.5992187500000001, "correct_mean": 0.62109375, "n_groups": 32, "n_datums": 256, "executed_updates_per_batch": 2, "mb": [{"n_loss_tokens": 41654.0, "dppo_masked_token_fraction": 0.0, "grpo_clip_fraction": 0.0005281605608104864, "policy_drift_abs_p_q90": 0.009458945132791996, "policy_drift_abs_p_mean": 0.002756078456455495, "mean_ratio": 1.0001731878484286, "nonzero_adv_fraction": 0.942214433187689, "n_datums": 128}, {"n_loss_tokens": 42124.0, "dppo_masked_token_fraction": 0.0, "grpo_clip_fraction": 0.00040357041116703067, "policy_drift_abs_p_q90": 0.010876530781388283, "policy_drift_abs_p_mean": 0.0029764413584677284, "mean_ratio": 1.0000097534533086, "nonzero_adv_fraction": 0.8310939132086221, "n_datums": 128}], "rollout_seconds": 164.32821011543274, "step_seconds": 197.3885521888733} {"step": 1, "arm": "dppo_pilot", "loss": "dppo", "reward_mean": 0.6511718750000001, "correct_mean": 0.66796875, "n_groups": 32, "n_datums": 256, "executed_updates_per_batch": 2, "mb": [{"n_loss_tokens": 38971.0, "dppo_masked_token_fraction": 0.0, "grpo_clip_fraction": 0.00025660106232839805, "policy_drift_abs_p_q90": 0.009069116786122322, "policy_drift_abs_p_mean": 0.0026498684604414793, "mean_ratio": 1.0000335832923226, "nonzero_adv_fraction": 0.7095789176567191, "n_datums": 128}, {"n_loss_tokens": 43497.0, "dppo_masked_token_fraction": 0.0, "grpo_clip_fraction": 0.0003678414603305975, "policy_drift_abs_p_q90": 0.01015632227063179, "policy_drift_abs_p_mean": 0.0028568718395964175, "mean_ratio": 1.0000214995184127, "nonzero_adv_fraction": 0.7777547876865071, "n_datums": 128}], "rollout_seconds": 19.20868730545044, "step_seconds": 32.175128698349} {"step": 2, "arm": "dppo_pilot", "loss": "dppo", "reward_mean": 0.5578125, "correct_mean": 0.5859375, "n_groups": 32, "n_datums": 256, "executed_updates_per_batch": 2, "mb": [{"n_loss_tokens": 45727.0, "dppo_masked_token_fraction": 0.0, "grpo_clip_fraction": 0.00032803376560894, "policy_drift_abs_p_q90": 0.008316918276250362, "policy_drift_abs_p_mean": 0.0025135856295450294, "mean_ratio": 0.9997614892389264, "nonzero_adv_fraction": 0.8304284120978853, "n_datums": 128}, {"n_loss_tokens": 47391.0, "dppo_masked_token_fraction": 0.0, "grpo_clip_fraction": 0.00048532421767846215, "policy_drift_abs_p_q90": 0.011339953169226646, "policy_drift_abs_p_mean": 0.003121381278627259, "mean_ratio": 1.0000116928784417, "nonzero_adv_fraction": 0.8726340444388175, "n_datums": 128}], "rollout_seconds": 11.63692593574524, "step_seconds": 23.314034938812256} {"step": 3, "arm": "dppo_pilot", "loss": "dppo", "reward_mean": 0.576953125, "correct_mean": 0.60546875, "n_groups": 32, "n_datums": 256, "executed_updates_per_batch": 2, "mb": [{"n_loss_tokens": 47388.0, "dppo_masked_token_fraction": 0.0, "grpo_clip_fraction": 0.0004009453870178104, "policy_drift_abs_p_q90": 0.010280854068696499, "policy_drift_abs_p_mean": 0.002867195000944003, "mean_ratio": 1.0002794699002286, "nonzero_adv_fraction": 0.9091964210348611, "n_datums": 128}, {"n_loss_tokens": 41335.0, "dppo_masked_token_fraction": 0.0, "grpo_clip_fraction": 0.00041127373896213864, "policy_drift_abs_p_q90": 0.009953003376722336, "policy_drift_abs_p_mean": 0.0028214049538539915, "mean_ratio": 1.00007863989896, "nonzero_adv_fraction": 0.8139349219789525, "n_datums": 128}], "rollout_seconds": 10.65837287902832, "step_seconds": 19.993511199951172} {"step": 4, "arm": "dppo_pilot", "loss": "dppo", "reward_mean": 0.564453125, "correct_mean": 0.58984375, "n_groups": 32, "n_datums": 256, "executed_updates_per_batch": 2, "mb": [{"n_loss_tokens": 38240.0, "dppo_masked_token_fraction": 0.0, "grpo_clip_fraction": 0.0003399581589958159, "policy_drift_abs_p_q90": 0.011326874606311321, "policy_drift_abs_p_mean": 0.0030352699803810703, "mean_ratio": 0.9997898012112483, "nonzero_adv_fraction": 0.8688023012552302, "n_datums": 128}, {"n_loss_tokens": 43806.0, "dppo_masked_token_fraction": 0.0, "grpo_clip_fraction": 0.0006163539241199836, "policy_drift_abs_p_q90": 0.010256553068757057, "policy_drift_abs_p_mean": 0.002920465316686518, "mean_ratio": 0.9999118550720856, "nonzero_adv_fraction": 0.912546226544309, "n_datums": 128}], "rollout_seconds": 14.723032236099243, "step_seconds": 26.15370202064514}