Upload checkpoints_vlm_gym_mental_rotation_2d_one_image_lr2e_5_ce_ins/checkpoints_vlm_gym_mental_rotation_2d_one_image_lr2e_5_ce_ins

checkpoints_vlm_gym_mental_rotation_2d_one_image_lr2e_5_ce_ins/checkpoints_vlm_gym_mental_rotation_2d_one_image_lr2e_5_ce_ins/wandb/offline-run-20260128_050010-checkpoints_vlm_gym_mental_rotation_2d_one_image_lr2e_5_ce_ins-run0/files/output.log

@@ -184,6 +184,13 @@ Preparing Dataset vlm_gym_mental_rotation_2d_celoss_evalonce/vlm_gym_mental_rota
   fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_2d_celoss_evalonce'}]
   fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_2d_celoss_evalonce'}]
 ce_avg: 0.2688937187194824, mse_avg: 0.10026200860738754
+base_dir is /dev/shm/models/checkpoints_vlm_gym_mental_rotation_2d_one_image_lr2e_5_ce_ins/eval_used_rows, step_tag is checkpoints_vlm_gym_mental_rotation_2d_one_image_lr2e_5_ce_ins_step1000
+Preparing Dataset vlm_gym_mental_rotation_2d_celoss_evalonce/vlm_gym_mental_rotation_2d_val
+[eval debug] first 3 batch fingerprints:
+  fp[0]: [{'data_indexes': [0], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_2d_celoss_evalonce'}]
+  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_2d_celoss_evalonce'}]
+  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_2d_celoss_evalonce'}]
+ce_avg: 0.34873995184898376, mse_avg: 0.09575071930885315
 wandb: Detected [huggingface_hub.inference] in use.
 wandb: Use W&B Weave for improved LLM call tracing. Install Weave with `pip install weave` then add `import weave` to the top of your script.
 wandb: For more information, check out the docs at: https://weave-docs.wandb.ai/
@@ -1161,27 +1168,6 @@ wandb: For more information, check out the docs at: https://weave-docs.wandb.ai/
 [2026-01-28 06:49:36] (step=0000964) Train Loss mse: 0.0967, Train Loss ce: 0.2595, Train Steps/Sec: 0.16,
 [2026-01-28 06:49:43] (step=0000965) Train Loss mse: 0.0965, Train Loss ce: 0.2566, Train Steps/Sec: 0.16,
 [2026-01-28 06:49:48] (step=0000966) Train Loss mse: 0.1096, Train Loss ce: 0.2680, Train Steps/Sec: 0.17,
-base_dir is /dev/shm/models/checkpoints_vlm_gym_mental_rotation_2d_one_image_lr2e_5_ce_ins/eval_used_rows, step_tag is checkpoints_vlm_gym_mental_rotation_2d_one_image_lr2e_5_ce_ins_step1000
-Preparing Dataset vlm_gym_mental_rotation_2d_celoss_evalonce/vlm_gym_mental_rotation_2d_val
-[eval debug] first 3 batch fingerprints:
-  fp[0]: [{'data_indexes': [0], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_2d_celoss_evalonce'}]
-  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_2d_celoss_evalonce'}]
-  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_2d_celoss_evalonce'}]
-ce_avg: 0.34873995184898376, mse_avg: 0.09575071930885315
-base_dir is /dev/shm/models/checkpoints_vlm_gym_mental_rotation_2d_one_image_lr2e_5_ce_ins/eval_used_rows, step_tag is checkpoints_vlm_gym_mental_rotation_2d_one_image_lr2e_5_ce_ins_step1500
-Preparing Dataset vlm_gym_mental_rotation_2d_celoss_evalonce/vlm_gym_mental_rotation_2d_val
-[eval debug] first 3 batch fingerprints:
-  fp[0]: [{'data_indexes': [0], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_2d_celoss_evalonce'}]
-  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_2d_celoss_evalonce'}]
-  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_2d_celoss_evalonce'}]
-ce_avg: 0.459031879901886, mse_avg: 0.09415699541568756
-base_dir is /dev/shm/models/checkpoints_vlm_gym_mental_rotation_2d_one_image_lr2e_5_ce_ins/eval_used_rows, step_tag is checkpoints_vlm_gym_mental_rotation_2d_one_image_lr2e_5_ce_ins_step2000
-Preparing Dataset vlm_gym_mental_rotation_2d_celoss_evalonce/vlm_gym_mental_rotation_2d_val
-[eval debug] first 3 batch fingerprints:
-  fp[0]: [{'data_indexes': [0], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_2d_celoss_evalonce'}]
-  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_2d_celoss_evalonce'}]
-  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_2d_celoss_evalonce'}]
-ce_avg: 1.3050191402435303, mse_avg: 0.09429918974637985
 [2026-01-28 06:49:56] (step=0000967) Train Loss mse: 0.1004, Train Loss ce: 0.2610, Train Steps/Sec: 0.14,
 [2026-01-28 06:50:02] (step=0000968) Train Loss mse: 0.0932, Train Loss ce: 0.2676, Train Steps/Sec: 0.16,
 [2026-01-28 06:50:08] (step=0000969) Train Loss mse: 0.1012, Train Loss ce: 0.2224, Train Steps/Sec: 0.17,
@@ -1224,6 +1210,20 @@ ce_avg: 1.3050191402435303, mse_avg: 0.09429918974637985
 [2026-01-28 06:54:35] (step=0001006) Train Loss mse: 0.1081, Train Loss ce: 0.1910, Train Steps/Sec: 0.15,
 [2026-01-28 06:54:42] (step=0001007) Train Loss mse: 0.0917, Train Loss ce: 0.2918, Train Steps/Sec: 0.16,
 [2026-01-28 06:54:48] (step=0001008) Train Loss mse: 0.1051, Train Loss ce: 0.2713, Train Steps/Sec: 0.15,
+base_dir is /dev/shm/models/checkpoints_vlm_gym_mental_rotation_2d_one_image_lr2e_5_ce_ins/eval_used_rows, step_tag is checkpoints_vlm_gym_mental_rotation_2d_one_image_lr2e_5_ce_ins_step1500
+Preparing Dataset vlm_gym_mental_rotation_2d_celoss_evalonce/vlm_gym_mental_rotation_2d_val
+[eval debug] first 3 batch fingerprints:
+  fp[0]: [{'data_indexes': [0], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_2d_celoss_evalonce'}]
+  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_2d_celoss_evalonce'}]
+  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_2d_celoss_evalonce'}]
+ce_avg: 0.459031879901886, mse_avg: 0.09415699541568756
+base_dir is /dev/shm/models/checkpoints_vlm_gym_mental_rotation_2d_one_image_lr2e_5_ce_ins/eval_used_rows, step_tag is checkpoints_vlm_gym_mental_rotation_2d_one_image_lr2e_5_ce_ins_step2000
+Preparing Dataset vlm_gym_mental_rotation_2d_celoss_evalonce/vlm_gym_mental_rotation_2d_val
+[eval debug] first 3 batch fingerprints:
+  fp[0]: [{'data_indexes': [0], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_2d_celoss_evalonce'}]
+  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_2d_celoss_evalonce'}]
+  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_2d_celoss_evalonce'}]
+ce_avg: 1.3050191402435303, mse_avg: 0.09429918974637985
 [2026-01-28 06:54:55] (step=0001009) Train Loss mse: 0.1093, Train Loss ce: 0.2607, Train Steps/Sec: 0.16,
 [2026-01-28 06:55:02] (step=0001010) Train Loss mse: 0.1076, Train Loss ce: 0.2569, Train Steps/Sec: 0.14,
 [2026-01-28 06:55:09] (step=0001011) Train Loss mse: 0.1154, Train Loss ce: 0.2570, Train Steps/Sec: 0.16,
@@ -2498,20 +2498,6 @@ ce_avg: 1.3050191402435303, mse_avg: 0.09429918974637985
 [2026-01-28 09:10:58] (step=0002280) Train Loss mse: 0.0883, Train Loss ce: 0.2570, Train Steps/Sec: 0.18,
 [2026-01-28 09:11:04] (step=0002281) Train Loss mse: 0.1298, Train Loss ce: 0.2740, Train Steps/Sec: 0.17,
 [2026-01-28 09:11:10] (step=0002282) Train Loss mse: 0.0929, Train Loss ce: 0.2899, Train Steps/Sec: 0.18,
-base_dir is /dev/shm/models/checkpoints_vlm_gym_mental_rotation_2d_one_image_lr2e_5_ce_ins/eval_used_rows, step_tag is checkpoints_vlm_gym_mental_rotation_2d_one_image_lr2e_5_ce_ins_step2500
-Preparing Dataset vlm_gym_mental_rotation_2d_celoss_evalonce/vlm_gym_mental_rotation_2d_val
-[eval debug] first 3 batch fingerprints:
-  fp[0]: [{'data_indexes': [0], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_2d_celoss_evalonce'}]
-  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_2d_celoss_evalonce'}]
-  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_2d_celoss_evalonce'}]
-ce_avg: 2.8753163814544678, mse_avg: 0.09435902535915375
-base_dir is /dev/shm/models/checkpoints_vlm_gym_mental_rotation_2d_one_image_lr2e_5_ce_ins/eval_used_rows, step_tag is checkpoints_vlm_gym_mental_rotation_2d_one_image_lr2e_5_ce_ins_step3000
-Preparing Dataset vlm_gym_mental_rotation_2d_celoss_evalonce/vlm_gym_mental_rotation_2d_val
-[eval debug] first 3 batch fingerprints:
-  fp[0]: [{'data_indexes': [0], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_2d_celoss_evalonce'}]
-  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_2d_celoss_evalonce'}]
-  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_2d_celoss_evalonce'}]
-ce_avg: 0.23429779708385468, mse_avg: 0.09545279294252396
 [2026-01-28 09:11:17] (step=0002283) Train Loss mse: 0.0946, Train Loss ce: 0.1869, Train Steps/Sec: 0.15,
 [2026-01-28 09:11:24] (step=0002284) Train Loss mse: 0.1032, Train Loss ce: 0.2511, Train Steps/Sec: 0.15,
 [2026-01-28 09:11:29] (step=0002285) Train Loss mse: 0.1330, Train Loss ce: 0.2405, Train Steps/Sec: 0.18,
@@ -2681,6 +2667,27 @@ ce_avg: 0.23429779708385468, mse_avg: 0.09545279294252396
 [2026-01-28 09:28:46] (step=0002449) Train Loss mse: 0.0956, Train Loss ce: 0.2226, Train Steps/Sec: 0.16,
 [2026-01-28 09:28:52] (step=0002450) Train Loss mse: 0.1061, Train Loss ce: 0.2766, Train Steps/Sec: 0.19,
 [2026-01-28 09:28:58] (step=0002451) Train Loss mse: 0.1174, Train Loss ce: 0.2472, Train Steps/Sec: 0.15,
+base_dir is /dev/shm/models/checkpoints_vlm_gym_mental_rotation_2d_one_image_lr2e_5_ce_ins/eval_used_rows, step_tag is checkpoints_vlm_gym_mental_rotation_2d_one_image_lr2e_5_ce_ins_step2500
+Preparing Dataset vlm_gym_mental_rotation_2d_celoss_evalonce/vlm_gym_mental_rotation_2d_val
+[eval debug] first 3 batch fingerprints:
+  fp[0]: [{'data_indexes': [0], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_2d_celoss_evalonce'}]
+  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_2d_celoss_evalonce'}]
+  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_2d_celoss_evalonce'}]
+ce_avg: 2.8753163814544678, mse_avg: 0.09435902535915375
+base_dir is /dev/shm/models/checkpoints_vlm_gym_mental_rotation_2d_one_image_lr2e_5_ce_ins/eval_used_rows, step_tag is checkpoints_vlm_gym_mental_rotation_2d_one_image_lr2e_5_ce_ins_step3000
+Preparing Dataset vlm_gym_mental_rotation_2d_celoss_evalonce/vlm_gym_mental_rotation_2d_val
+[eval debug] first 3 batch fingerprints:
+  fp[0]: [{'data_indexes': [0], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_2d_celoss_evalonce'}]
+  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_2d_celoss_evalonce'}]
+  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_2d_celoss_evalonce'}]
+ce_avg: 0.23429779708385468, mse_avg: 0.09545279294252396
+base_dir is /dev/shm/models/checkpoints_vlm_gym_mental_rotation_2d_one_image_lr2e_5_ce_ins/eval_used_rows, step_tag is checkpoints_vlm_gym_mental_rotation_2d_one_image_lr2e_5_ce_ins_step3500
+Preparing Dataset vlm_gym_mental_rotation_2d_celoss_evalonce/vlm_gym_mental_rotation_2d_val
+[eval debug] first 3 batch fingerprints:
+  fp[0]: [{'data_indexes': [0], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_2d_celoss_evalonce'}]
+  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_2d_celoss_evalonce'}]
+  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_2d_celoss_evalonce'}]
+ce_avg: 0.23339946568012238, mse_avg: 0.09179326146841049
 [2026-01-28 09:29:06] (step=0002452) Train Loss mse: 0.1081, Train Loss ce: 0.2334, Train Steps/Sec: 0.13,
 [2026-01-28 09:29:12] (step=0002453) Train Loss mse: 0.1002, Train Loss ce: 0.2645, Train Steps/Sec: 0.16,
 [2026-01-28 09:29:18] (step=0002454) Train Loss mse: 0.0897, Train Loss ce: 0.2104, Train Steps/Sec: 0.16,
@@ -3524,20 +3531,6 @@ ce_avg: 0.23429779708385468, mse_avg: 0.09545279294252396
 [2026-01-28 11:01:58] (step=0003289) Train Loss mse: 0.1045, Train Loss ce: 0.2524, Train Steps/Sec: 0.15,
 [2026-01-28 11:02:04] (step=0003290) Train Loss mse: 0.0882, Train Loss ce: 0.1963, Train Steps/Sec: 0.16,
 [2026-01-28 11:02:11] (step=0003291) Train Loss mse: 0.1076, Train Loss ce: 0.2416, Train Steps/Sec: 0.15,
-base_dir is /dev/shm/models/checkpoints_vlm_gym_mental_rotation_2d_one_image_lr2e_5_ce_ins/eval_used_rows, step_tag is checkpoints_vlm_gym_mental_rotation_2d_one_image_lr2e_5_ce_ins_step3500
-Preparing Dataset vlm_gym_mental_rotation_2d_celoss_evalonce/vlm_gym_mental_rotation_2d_val
-[eval debug] first 3 batch fingerprints:
-  fp[0]: [{'data_indexes': [0], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_2d_celoss_evalonce'}]
-  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_2d_celoss_evalonce'}]
-  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_2d_celoss_evalonce'}]
-ce_avg: 0.23339946568012238, mse_avg: 0.09179326146841049
-base_dir is /dev/shm/models/checkpoints_vlm_gym_mental_rotation_2d_one_image_lr2e_5_ce_ins/eval_used_rows, step_tag is checkpoints_vlm_gym_mental_rotation_2d_one_image_lr2e_5_ce_ins_step4000
-Preparing Dataset vlm_gym_mental_rotation_2d_celoss_evalonce/vlm_gym_mental_rotation_2d_val
-[eval debug] first 3 batch fingerprints:
-  fp[0]: [{'data_indexes': [0], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_2d_celoss_evalonce'}]
-  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_2d_celoss_evalonce'}]
-  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_2d_celoss_evalonce'}]
-ce_avg: 0.23216108977794647, mse_avg: 0.09232720732688904
 [2026-01-28 11:02:18] (step=0003292) Train Loss mse: 0.1043, Train Loss ce: 0.2579, Train Steps/Sec: 0.15,
 [2026-01-28 11:02:24] (step=0003293) Train Loss mse: 0.1035, Train Loss ce: 0.2152, Train Steps/Sec: 0.17,
 [2026-01-28 11:02:30] (step=0003294) Train Loss mse: 0.0917, Train Loss ce: 0.2382, Train Steps/Sec: 0.16,
@@ -3661,6 +3654,20 @@ ce_avg: 0.23216108977794647, mse_avg: 0.09232720732688904
 [2026-01-28 11:15:08] (step=0003412) Train Loss mse: 0.0918, Train Loss ce: 0.2324, Train Steps/Sec: 0.15,
 [2026-01-28 11:15:14] (step=0003413) Train Loss mse: 0.1073, Train Loss ce: 0.2467, Train Steps/Sec: 0.16,
 [2026-01-28 11:15:21] (step=0003414) Train Loss mse: 0.0914, Train Loss ce: 0.2380, Train Steps/Sec: 0.14,
+base_dir is /dev/shm/models/checkpoints_vlm_gym_mental_rotation_2d_one_image_lr2e_5_ce_ins/eval_used_rows, step_tag is checkpoints_vlm_gym_mental_rotation_2d_one_image_lr2e_5_ce_ins_step4000
+Preparing Dataset vlm_gym_mental_rotation_2d_celoss_evalonce/vlm_gym_mental_rotation_2d_val
+[eval debug] first 3 batch fingerprints:
+  fp[0]: [{'data_indexes': [0], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_2d_celoss_evalonce'}]
+  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_2d_celoss_evalonce'}]
+  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_2d_celoss_evalonce'}]
+ce_avg: 0.23216108977794647, mse_avg: 0.09232720732688904
+base_dir is /dev/shm/models/checkpoints_vlm_gym_mental_rotation_2d_one_image_lr2e_5_ce_ins/eval_used_rows, step_tag is checkpoints_vlm_gym_mental_rotation_2d_one_image_lr2e_5_ce_ins_step4500
+Preparing Dataset vlm_gym_mental_rotation_2d_celoss_evalonce/vlm_gym_mental_rotation_2d_val
+[eval debug] first 3 batch fingerprints:
+  fp[0]: [{'data_indexes': [0], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_2d_celoss_evalonce'}]
+  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_2d_celoss_evalonce'}]
+  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_2d_celoss_evalonce'}]
+ce_avg: 0.23146067559719086, mse_avg: 0.09148821234703064
 [2026-01-28 11:15:28] (step=0003415) Train Loss mse: 0.1021, Train Loss ce: 0.2517, Train Steps/Sec: 0.14,
 [2026-01-28 11:15:34] (step=0003416) Train Loss mse: 0.1034, Train Loss ce: 0.2328, Train Steps/Sec: 0.17,
 [2026-01-28 11:15:40] (step=0003417) Train Loss mse: 0.0965, Train Loss ce: 0.2408, Train Steps/Sec: 0.16,
@@ -4947,20 +4954,6 @@ ce_avg: 0.23216108977794647, mse_avg: 0.09232720732688904
 [2026-01-28 13:32:55] (step=0004698) Train Loss mse: 0.1010, Train Loss ce: 0.2219, Train Steps/Sec: 0.14,
 [2026-01-28 13:33:02] (step=0004699) Train Loss mse: 0.1133, Train Loss ce: 0.2032, Train Steps/Sec: 0.14,
 [2026-01-28 13:33:09] (step=0004700) Train Loss mse: 0.1043, Train Loss ce: 0.2092, Train Steps/Sec: 0.16,
-base_dir is /dev/shm/models/checkpoints_vlm_gym_mental_rotation_2d_one_image_lr2e_5_ce_ins/eval_used_rows, step_tag is checkpoints_vlm_gym_mental_rotation_2d_one_image_lr2e_5_ce_ins_step4500
-Preparing Dataset vlm_gym_mental_rotation_2d_celoss_evalonce/vlm_gym_mental_rotation_2d_val
-[eval debug] first 3 batch fingerprints:
-  fp[0]: [{'data_indexes': [0], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_2d_celoss_evalonce'}]
-  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_2d_celoss_evalonce'}]
-  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_2d_celoss_evalonce'}]
-ce_avg: 0.23146067559719086, mse_avg: 0.09148821234703064
-base_dir is /dev/shm/models/checkpoints_vlm_gym_mental_rotation_2d_one_image_lr2e_5_ce_ins/eval_used_rows, step_tag is checkpoints_vlm_gym_mental_rotation_2d_one_image_lr2e_5_ce_ins_step5000
-Preparing Dataset vlm_gym_mental_rotation_2d_celoss_evalonce/vlm_gym_mental_rotation_2d_val
-[eval debug] first 3 batch fingerprints:
-  fp[0]: [{'data_indexes': [0], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_2d_celoss_evalonce'}]
-  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_2d_celoss_evalonce'}]
-  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_2d_celoss_evalonce'}]
-ce_avg: 0.2311139553785324, mse_avg: 0.09259119629859924
 [2026-01-28 13:33:15] (step=0004701) Train Loss mse: 0.1018, Train Loss ce: 0.2420, Train Steps/Sec: 0.15,
 [2026-01-28 13:33:22] (step=0004702) Train Loss mse: 0.0966, Train Loss ce: 0.2338, Train Steps/Sec: 0.14,
 [2026-01-28 13:33:29] (step=0004703) Train Loss mse: 0.1212, Train Loss ce: 0.2302, Train Steps/Sec: 0.16,
@@ -5175,6 +5168,13 @@ ce_avg: 0.2311139553785324, mse_avg: 0.09259119629859924
 [2026-01-28 13:55:45] (step=0004912) Train Loss mse: 0.1120, Train Loss ce: 0.2248, Train Steps/Sec: 0.16,
 [2026-01-28 13:55:52] (step=0004913) Train Loss mse: 0.1049, Train Loss ce: 0.2181, Train Steps/Sec: 0.17,
 [2026-01-28 13:55:58] (step=0004914) Train Loss mse: 0.0920, Train Loss ce: 0.2414, Train Steps/Sec: 0.16,
+base_dir is /dev/shm/models/checkpoints_vlm_gym_mental_rotation_2d_one_image_lr2e_5_ce_ins/eval_used_rows, step_tag is checkpoints_vlm_gym_mental_rotation_2d_one_image_lr2e_5_ce_ins_step5000
+Preparing Dataset vlm_gym_mental_rotation_2d_celoss_evalonce/vlm_gym_mental_rotation_2d_val
+[eval debug] first 3 batch fingerprints:
+  fp[0]: [{'data_indexes': [0], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_2d_celoss_evalonce'}]
+  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_2d_celoss_evalonce'}]
+  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_2d_celoss_evalonce'}]
+ce_avg: 0.2311139553785324, mse_avg: 0.09259119629859924
 [2026-01-28 13:56:05] (step=0004915) Train Loss mse: 0.0982, Train Loss ce: 0.2438, Train Steps/Sec: 0.15,
 [2026-01-28 13:56:11] (step=0004916) Train Loss mse: 0.1100, Train Loss ce: 0.2129, Train Steps/Sec: 0.15,
 [2026-01-28 13:56:17] (step=0004917) Train Loss mse: 0.1248, Train Loss ce: 0.2591, Train Steps/Sec: 0.18,