progress: training

training/progress_ball.json

@@ -1,31 +1,31 @@
 {
  "run": "ball-contact-v2",
- "status": "training (heartbeat)",
+ "status": "training",
  "target_steps": 6623999,
  "latest": {
-  "training/sps": 408.23492458961204,
-  "training/entropy_loss": 0.022244084626436234,
-  "training/policy_loss": -0.03546517342329025,
-  "training/total_loss": -0.011253628879785538,
-  "training/v_loss": 0.001967460848391056,
-  "eval/episode_approach": 17.60287857055664,
-  "eval/episode_dist": 832.9267578125,
-  "eval/episode_fell": 0.8828125,
+  "training/sps": 404.6485015725098,
+  "training/entropy_loss": 0.030808379873633385,
+  "training/policy_loss": -0.035319581627845764,
+  "training/total_loss": -0.002890608739107847,
+  "training/v_loss": 0.001620592433027923,
+  "eval/episode_approach": 17.670867919921875,
+  "eval/episode_dist": 823.5367431640625,
+  "eval/episode_fell": 0.890625,
   "eval/episode_hit": 0.0,
-  "eval/episode_reward": 128.203125,
-  "eval/episode_torque": 7.811934471130371,
-  "eval/episode_upright": 76.28021240234375,
-  "eval/episode_approach_std": 3.0936944484710693,
-  "eval/episode_dist_std": 183.4468231201172,
-  "eval/episode_fell_std": 0.32164356112480164,
+  "eval/episode_reward": 128.55641174316406,
+  "eval/episode_torque": 7.54718017578125,
+  "eval/episode_upright": 75.12237548828125,
+  "eval/episode_approach_std": 3.470233678817749,
+  "eval/episode_dist_std": 152.88238525390625,
+  "eval/episode_fell_std": 0.312109112739563,
   "eval/episode_hit_std": 0.0,
-  "eval/episode_reward_std": 29.502559661865234,
-  "eval/episode_torque_std": 1.6987098455429077,
-  "eval/episode_upright_std": 9.35492992401123,
-  "eval/avg_episode_length": 79.21875,
-  "steps": 6225920,
-  "steps_total": 7864320,
-  "wall_s": 18267.126011133194
+  "eval/episode_reward_std": 32.867061614990234,
+  "eval/episode_torque_std": 1.5317257642745972,
+  "eval/episode_upright_std": 8.949020385742188,
+  "eval/avg_episode_length": 77.9609375,
+  "steps": 6553600,
+  "steps_total": 8192000,
+  "wall_s": 19199.09183907509
  },
  "log": [
   {
@@ -522,16 +522,33 @@
    "steps": 6225920,
    "steps_total": 7864320,
    "wall_s": 18267.126011133194
+  },
+  {
+   "training/sps": 404.6485015725098,
+   "training/entropy_loss": 0.030808379873633385,
+   "training/policy_loss": -0.035319581627845764,
+   "training/total_loss": -0.002890608739107847,
+   "training/v_loss": 0.001620592433027923,
+   "eval/episode_approach": 17.670867919921875,
+   "eval/episode_dist": 823.5367431640625,
+   "eval/episode_fell": 0.890625,
+   "eval/episode_hit": 0.0,
+   "eval/episode_reward": 128.55641174316406,
+   "eval/episode_torque": 7.54718017578125,
+   "eval/episode_upright": 75.12237548828125,
+   "eval/episode_approach_std": 3.470233678817749,
+   "eval/episode_dist_std": 152.88238525390625,
+   "eval/episode_fell_std": 0.312109112739563,
+   "eval/episode_hit_std": 0.0,
+   "eval/episode_reward_std": 32.867061614990234,
+   "eval/episode_torque_std": 1.5317257642745972,
+   "eval/episode_upright_std": 8.949020385742188,
+   "eval/avg_episode_length": 77.9609375,
+   "steps": 6553600,
+   "steps_total": 8192000,
+   "wall_s": 19199.09183907509
   }
  ],
- "heartbeat": {
-  "elapsed_train_s": 18934,
-  "last_eval_steps": 6225920,
-  "est_steps_now": 6498424,
-  "steps_total_est": 8136824,
-  "sps": 408.23492458961204,
-  "eta_s": 307
- },
- "wall_s": 19188.206573486328,
- "utc": "2026-09-24T03:49:23Z"
+ "wall_s": 19452.654223680496,
+ "utc": "2026-09-24T03:53:48Z"
 }

training/training_status.txt

@@ -1,26 +1,26 @@
 ===============================================================
   ATH BALL-CONTACT TRAINING STATUS - Kaggle 2x T4 | MJX f32 PPO
 ===============================================================
-  Updated      : 2026-09-24 03:49:23 UTC
+  Updated      : 2026-09-24 03:53:48 UTC
   Run          : ball-contact-v2  |  kernel version 8
-  State        : training (heartbeat)
-  Env steps    : 6,498,424 / 6,623,999  (98.1%)
-  Evals        : 20 / 40
-  Total steps  : 8,136,824 (this session started at 1,638,400)
+  State        : training
+  Env steps    : 6,553,600 / 6,623,999  (98.9%)
+  Evals        : 21 / 40
+  Total steps  : 8,192,000 (this session started at 1,638,400)
 ---------------------------------------------------------------
   Contact rate : 0.000   (f32 eval, 128 eps)
-  Fall rate    : 0.883
-  Reward       : 128.2
-  Ep length    : 79.2 steps (max 172)
-  Policy loss  : -0.0355
-  Value loss   : 0.0020
-  Entropy loss : 0.0222
-  Total loss   : -0.0113
+  Fall rate    : 0.891
+  Reward       : 128.6
+  Ep length    : 78.0 steps (max 172)
+  Policy loss  : -0.0353
+  Value loss   : 0.0016
+  Entropy loss : 0.0308
+  Total loss   : -0.0029
 ---------------------------------------------------------------
   Best so far  : step 2,621,440 | contact 0.016 | fall 0.625
-  Heartbeat    : every 5 min; steps now ~6,498,424 (estimated from measured speed since last eval)
-  Speed        : 408 env steps/s  |  elapsed 320 min
-  ETA          : 5 min to last eval, then float64 oracle (~10 min)
+  Heartbeat    : every 5 min
+  Speed        : 404 env steps/s  |  elapsed 324 min
+  ETA          : 3 min to last eval, then float64 oracle (~10 min)
 ---------------------------------------------------------------
   Physics gate : ell7 p95 12.3mm PASS  ell16 p95 12.3mm PASS  pyr4 p95 1012.2mm fail
   Train sim    : ell7 (MuJoCo 3.3.7 MJX float32, 2 ms, aero drag+Magnus)