Add offline evaluation results

eval_results.json

@@ -0,0 +1,33 @@
+{
+  "checkpoint": "/home/sky/outputs/train/diffusion_chess_300/checkpoints/last/pretrained_model",
+  "dataset_repo": "AnonymousMouse404/chess_300",
+  "n_eval_episodes": 20,
+  "eval_episodes": [
+    0,
+    20,
+    40,
+    60,
+    80,
+    100,
+    120,
+    140,
+    160,
+    180,
+    200,
+    220,
+    240,
+    260,
+    280,
+    300,
+    320,
+    340,
+    360,
+    380
+  ],
+  "n_eval_batches": 214,
+  "n_eval_frames": 6819,
+  "diffusion_loss": 0.0041590746192337194,
+  "action_mse": 11.68516247761228,
+  "action_l1": 1.2478218859618562,
+  "note": "Offline action-prediction metrics on dataset frames (not closed-loop task success). Evaluated within the trained episode prefix, so this reflects training-set fit rather than held-out generalization. action_mse / action_l1 are in the raw action space (joint positions, degrees)."
+}