Upload checkpoints_vlm_gym_mental_rotation_3d_objaverse_pad3_by_axis_one_image_lr2e_5_ce_ins/checkpoints_vlm_gym_mental_rotation_3d_objaverse_pad3_by_axis_one_image_lr2e_5_ce_ins

checkpoints_vlm_gym_mental_rotation_3d_objaverse_pad3_by_axis_one_image_lr2e_5_ce_ins/checkpoints_vlm_gym_mental_rotation_3d_objaverse_pad3_by_axis_one_image_lr2e_5_ce_ins/wandb/offline-run-20260129_221658-checkpoints_vlm_gym_mental_rotation_3d_objaverse_pad3_by_axis_one_image_lr2e_5_ce_ins-run0/files/output.log

@@ -882,27 +882,6 @@ wandb: For more information, check out the docs at: https://weave-docs.wandb.ai/
 [2026-01-30 02:11:40] (step=0000871) Train Loss mse: 0.0635, Train Loss ce: 0.0536, Train Steps/Sec: 0.06,
 [2026-01-30 02:11:56] (step=0000872) Train Loss mse: 0.0721, Train Loss ce: 0.0497, Train Steps/Sec: 0.07,
 [2026-01-30 02:12:11] (step=0000873) Train Loss mse: 0.0725, Train Loss ce: 0.0511, Train Steps/Sec: 0.07,
-[2026-01-30 02:12:26] (step=0000874) Train Loss mse: 0.0770, Train Loss ce: 0.0528, Train Steps/Sec: 0.06,
-[2026-01-30 02:12:42] (step=0000875) Train Loss mse: 0.0661, Train Loss ce: 0.0531, Train Steps/Sec: 0.06,
-[2026-01-30 02:12:57] (step=0000876) Train Loss mse: 0.0701, Train Loss ce: 0.0541, Train Steps/Sec: 0.07,
-[2026-01-30 02:13:12] (step=0000877) Train Loss mse: 0.0701, Train Loss ce: 0.0489, Train Steps/Sec: 0.06,
-[2026-01-30 02:13:29] (step=0000878) Train Loss mse: 0.0672, Train Loss ce: 0.0516, Train Steps/Sec: 0.06,
-[2026-01-30 02:13:45] (step=0000879) Train Loss mse: 0.0585, Train Loss ce: 0.0506, Train Steps/Sec: 0.06,
-[2026-01-30 02:14:00] (step=0000880) Train Loss mse: 0.0618, Train Loss ce: 0.0550, Train Steps/Sec: 0.07,
-[2026-01-30 02:14:15] (step=0000881) Train Loss mse: 0.0649, Train Loss ce: 0.0513, Train Steps/Sec: 0.07,
-[2026-01-30 02:14:30] (step=0000882) Train Loss mse: 0.0751, Train Loss ce: 0.0525, Train Steps/Sec: 0.07,
-[2026-01-30 02:14:46] (step=0000883) Train Loss mse: 0.0483, Train Loss ce: 0.0483, Train Steps/Sec: 0.06,
-[2026-01-30 02:15:01] (step=0000884) Train Loss mse: 0.0715, Train Loss ce: 0.0517, Train Steps/Sec: 0.06,
-[2026-01-30 02:15:18] (step=0000885) Train Loss mse: 0.0637, Train Loss ce: 0.0533, Train Steps/Sec: 0.06,
-[2026-01-30 02:15:33] (step=0000886) Train Loss mse: 0.0739, Train Loss ce: 0.0564, Train Steps/Sec: 0.07,
-[2026-01-30 02:15:50] (step=0000887) Train Loss mse: 0.0569, Train Loss ce: 0.0532, Train Steps/Sec: 0.06,
-[2026-01-30 02:16:04] (step=0000888) Train Loss mse: 0.0588, Train Loss ce: 0.0569, Train Steps/Sec: 0.07,
-[2026-01-30 02:16:19] (step=0000889) Train Loss mse: 0.0650, Train Loss ce: 0.0505, Train Steps/Sec: 0.07,
-[2026-01-30 02:16:34] (step=0000890) Train Loss mse: 0.0644, Train Loss ce: 0.0563, Train Steps/Sec: 0.07,
-[2026-01-30 02:16:49] (step=0000891) Train Loss mse: 0.0654, Train Loss ce: 0.0555, Train Steps/Sec: 0.06,
-[2026-01-30 02:17:04] (step=0000892) Train Loss mse: 0.0543, Train Loss ce: 0.0541, Train Steps/Sec: 0.07,
-[2026-01-30 02:17:19] (step=0000893) Train Loss mse: 0.0614, Train Loss ce: 0.0534, Train Steps/Sec: 0.07,
-[2026-01-30 02:17:36] (step=0000894) Train Loss mse: 0.0646, Train Loss ce: 0.0524, Train Steps/Sec: 0.06,
 FullyShardedDataParallel(
   (_fsdp_wrapped_module): Bagel(
     (language_model): Qwen2ForCausalLM(
@@ -1089,20 +1068,34 @@ Preparing Dataset vlm_gym_mental_rotation_3d_objaverse_pad3_by_axis_celoss_evalo
   fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_3d_objaverse_pad3_by_axis_celoss_evalonce'}]
   fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_3d_objaverse_pad3_by_axis_celoss_evalonce'}]
 ce_avg: 0.056111279875040054, mse_avg: 0.05582037195563316
-base_dir is /dev/shm/models/checkpoints_vlm_gym_mental_rotation_3d_objaverse_pad3_by_axis_one_image_lr2e_5_ce_ins/eval_used_rows, step_tag is checkpoints_vlm_gym_mental_rotation_3d_objaverse_pad3_by_axis_one_image_lr2e_5_ce_ins_step1500
-Preparing Dataset vlm_gym_mental_rotation_3d_objaverse_pad3_by_axis_celoss_evalonce/vlm_gym_mental_rotation_3d_objaverse_pad3_by_axis_val
-[eval debug] first 3 batch fingerprints:
-  fp[0]: [{'data_indexes': [0], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_3d_objaverse_pad3_by_axis_celoss_evalonce'}]
-  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_3d_objaverse_pad3_by_axis_celoss_evalonce'}]
-  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_3d_objaverse_pad3_by_axis_celoss_evalonce'}]
-ce_avg: 0.06184602156281471, mse_avg: 0.05278971791267395
-base_dir is /dev/shm/models/checkpoints_vlm_gym_mental_rotation_3d_objaverse_pad3_by_axis_one_image_lr2e_5_ce_ins/eval_used_rows, step_tag is checkpoints_vlm_gym_mental_rotation_3d_objaverse_pad3_by_axis_one_image_lr2e_5_ce_ins_step2000
+base_dir is /dev/shm/models/checkpoints_vlm_gym_mental_rotation_3d_objaverse_pad3_by_axis_one_image_lr2e_5_ce_ins/eval_used_rows, step_tag is checkpoints_vlm_gym_mental_rotation_3d_objaverse_pad3_by_axis_one_image_lr2e_5_ce_ins_step1000
 Preparing Dataset vlm_gym_mental_rotation_3d_objaverse_pad3_by_axis_celoss_evalonce/vlm_gym_mental_rotation_3d_objaverse_pad3_by_axis_val
 [eval debug] first 3 batch fingerprints:
   fp[0]: [{'data_indexes': [0], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_3d_objaverse_pad3_by_axis_celoss_evalonce'}]
   fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_3d_objaverse_pad3_by_axis_celoss_evalonce'}]
   fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_3d_objaverse_pad3_by_axis_celoss_evalonce'}]
-ce_avg: 0.05899278074502945, mse_avg: 0.052241820842027664
+ce_avg: 0.057540930807590485, mse_avg: 0.054740797728300095
+[2026-01-30 02:12:26] (step=0000874) Train Loss mse: 0.0770, Train Loss ce: 0.0528, Train Steps/Sec: 0.06,
+[2026-01-30 02:12:42] (step=0000875) Train Loss mse: 0.0661, Train Loss ce: 0.0531, Train Steps/Sec: 0.06,
+[2026-01-30 02:12:57] (step=0000876) Train Loss mse: 0.0701, Train Loss ce: 0.0541, Train Steps/Sec: 0.07,
+[2026-01-30 02:13:12] (step=0000877) Train Loss mse: 0.0701, Train Loss ce: 0.0489, Train Steps/Sec: 0.06,
+[2026-01-30 02:13:29] (step=0000878) Train Loss mse: 0.0672, Train Loss ce: 0.0516, Train Steps/Sec: 0.06,
+[2026-01-30 02:13:45] (step=0000879) Train Loss mse: 0.0585, Train Loss ce: 0.0506, Train Steps/Sec: 0.06,
+[2026-01-30 02:14:00] (step=0000880) Train Loss mse: 0.0618, Train Loss ce: 0.0550, Train Steps/Sec: 0.07,
+[2026-01-30 02:14:15] (step=0000881) Train Loss mse: 0.0649, Train Loss ce: 0.0513, Train Steps/Sec: 0.07,
+[2026-01-30 02:14:30] (step=0000882) Train Loss mse: 0.0751, Train Loss ce: 0.0525, Train Steps/Sec: 0.07,
+[2026-01-30 02:14:46] (step=0000883) Train Loss mse: 0.0483, Train Loss ce: 0.0483, Train Steps/Sec: 0.06,
+[2026-01-30 02:15:01] (step=0000884) Train Loss mse: 0.0715, Train Loss ce: 0.0517, Train Steps/Sec: 0.06,
+[2026-01-30 02:15:18] (step=0000885) Train Loss mse: 0.0637, Train Loss ce: 0.0533, Train Steps/Sec: 0.06,
+[2026-01-30 02:15:33] (step=0000886) Train Loss mse: 0.0739, Train Loss ce: 0.0564, Train Steps/Sec: 0.07,
+[2026-01-30 02:15:50] (step=0000887) Train Loss mse: 0.0569, Train Loss ce: 0.0532, Train Steps/Sec: 0.06,
+[2026-01-30 02:16:04] (step=0000888) Train Loss mse: 0.0588, Train Loss ce: 0.0569, Train Steps/Sec: 0.07,
+[2026-01-30 02:16:19] (step=0000889) Train Loss mse: 0.0650, Train Loss ce: 0.0505, Train Steps/Sec: 0.07,
+[2026-01-30 02:16:34] (step=0000890) Train Loss mse: 0.0644, Train Loss ce: 0.0563, Train Steps/Sec: 0.07,
+[2026-01-30 02:16:49] (step=0000891) Train Loss mse: 0.0654, Train Loss ce: 0.0555, Train Steps/Sec: 0.06,
+[2026-01-30 02:17:04] (step=0000892) Train Loss mse: 0.0543, Train Loss ce: 0.0541, Train Steps/Sec: 0.07,
+[2026-01-30 02:17:19] (step=0000893) Train Loss mse: 0.0614, Train Loss ce: 0.0534, Train Steps/Sec: 0.07,
+[2026-01-30 02:17:36] (step=0000894) Train Loss mse: 0.0646, Train Loss ce: 0.0524, Train Steps/Sec: 0.06,
 [2026-01-30 02:17:52] (step=0000895) Train Loss mse: 0.0557, Train Loss ce: 0.0519, Train Steps/Sec: 0.06,
 [2026-01-30 02:18:07] (step=0000896) Train Loss mse: 0.0742, Train Loss ce: 0.0534, Train Steps/Sec: 0.06,
 [2026-01-30 02:18:23] (step=0000897) Train Loss mse: 0.0826, Train Loss ce: 0.0551, Train Steps/Sec: 0.06,
@@ -2313,6 +2306,20 @@ ce_avg: 0.05899278074502945, mse_avg: 0.052241820842027664
 [2026-01-30 07:34:20] (step=0002102) Train Loss mse: 0.0760, Train Loss ce: 0.0494, Train Steps/Sec: 0.07,
 [2026-01-30 07:34:35] (step=0002103) Train Loss mse: 0.0668, Train Loss ce: 0.0457, Train Steps/Sec: 0.07,
 [2026-01-30 07:34:50] (step=0002104) Train Loss mse: 0.0554, Train Loss ce: 0.0456, Train Steps/Sec: 0.07,
+base_dir is /dev/shm/models/checkpoints_vlm_gym_mental_rotation_3d_objaverse_pad3_by_axis_one_image_lr2e_5_ce_ins/eval_used_rows, step_tag is checkpoints_vlm_gym_mental_rotation_3d_objaverse_pad3_by_axis_one_image_lr2e_5_ce_ins_step1500
+Preparing Dataset vlm_gym_mental_rotation_3d_objaverse_pad3_by_axis_celoss_evalonce/vlm_gym_mental_rotation_3d_objaverse_pad3_by_axis_val
+[eval debug] first 3 batch fingerprints:
+  fp[0]: [{'data_indexes': [0], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_3d_objaverse_pad3_by_axis_celoss_evalonce'}]
+  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_3d_objaverse_pad3_by_axis_celoss_evalonce'}]
+  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_3d_objaverse_pad3_by_axis_celoss_evalonce'}]
+ce_avg: 0.06184602156281471, mse_avg: 0.05278971791267395
+base_dir is /dev/shm/models/checkpoints_vlm_gym_mental_rotation_3d_objaverse_pad3_by_axis_one_image_lr2e_5_ce_ins/eval_used_rows, step_tag is checkpoints_vlm_gym_mental_rotation_3d_objaverse_pad3_by_axis_one_image_lr2e_5_ce_ins_step2000
+Preparing Dataset vlm_gym_mental_rotation_3d_objaverse_pad3_by_axis_celoss_evalonce/vlm_gym_mental_rotation_3d_objaverse_pad3_by_axis_val
+[eval debug] first 3 batch fingerprints:
+  fp[0]: [{'data_indexes': [0], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_3d_objaverse_pad3_by_axis_celoss_evalonce'}]
+  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_3d_objaverse_pad3_by_axis_celoss_evalonce'}]
+  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_3d_objaverse_pad3_by_axis_celoss_evalonce'}]
+ce_avg: 0.05899278074502945, mse_avg: 0.052241820842027664
 [2026-01-30 07:35:05] (step=0002105) Train Loss mse: 0.0616, Train Loss ce: 0.0466, Train Steps/Sec: 0.07,
 [2026-01-30 07:35:19] (step=0002106) Train Loss mse: 0.0599, Train Loss ce: 0.0531, Train Steps/Sec: 0.07,
 [2026-01-30 07:35:35] (step=0002107) Train Loss mse: 0.0801, Train Loss ce: 0.0488, Train Steps/Sec: 0.06,
@@ -2347,20 +2354,6 @@ ce_avg: 0.05899278074502945, mse_avg: 0.052241820842027664
 [2026-01-30 07:43:09] (step=0002136) Train Loss mse: 0.0754, Train Loss ce: 0.0465, Train Steps/Sec: 0.06,
 [2026-01-30 07:43:26] (step=0002137) Train Loss mse: 0.0662, Train Loss ce: 0.0484, Train Steps/Sec: 0.06,
 [2026-01-30 07:43:41] (step=0002138) Train Loss mse: 0.0688, Train Loss ce: 0.0494, Train Steps/Sec: 0.07,
-base_dir is /dev/shm/models/checkpoints_vlm_gym_mental_rotation_3d_objaverse_pad3_by_axis_one_image_lr2e_5_ce_ins/eval_used_rows, step_tag is checkpoints_vlm_gym_mental_rotation_3d_objaverse_pad3_by_axis_one_image_lr2e_5_ce_ins_step2500
-Preparing Dataset vlm_gym_mental_rotation_3d_objaverse_pad3_by_axis_celoss_evalonce/vlm_gym_mental_rotation_3d_objaverse_pad3_by_axis_val
-[eval debug] first 3 batch fingerprints:
-  fp[0]: [{'data_indexes': [0], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_3d_objaverse_pad3_by_axis_celoss_evalonce'}]
-  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_3d_objaverse_pad3_by_axis_celoss_evalonce'}]
-  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_3d_objaverse_pad3_by_axis_celoss_evalonce'}]
-ce_avg: 0.0595068633556366, mse_avg: 0.05313115939497948
-base_dir is /dev/shm/models/checkpoints_vlm_gym_mental_rotation_3d_objaverse_pad3_by_axis_one_image_lr2e_5_ce_ins/eval_used_rows, step_tag is checkpoints_vlm_gym_mental_rotation_3d_objaverse_pad3_by_axis_one_image_lr2e_5_ce_ins_step3000
-Preparing Dataset vlm_gym_mental_rotation_3d_objaverse_pad3_by_axis_celoss_evalonce/vlm_gym_mental_rotation_3d_objaverse_pad3_by_axis_val
-[eval debug] first 3 batch fingerprints:
-  fp[0]: [{'data_indexes': [0], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_3d_objaverse_pad3_by_axis_celoss_evalonce'}]
-  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_3d_objaverse_pad3_by_axis_celoss_evalonce'}]
-  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_3d_objaverse_pad3_by_axis_celoss_evalonce'}]
-ce_avg: 0.04585418477654457, mse_avg: 0.050133317708969116
 [2026-01-30 07:43:56] (step=0002139) Train Loss mse: 0.0552, Train Loss ce: 0.0488, Train Steps/Sec: 0.07,
 [2026-01-30 07:44:12] (step=0002140) Train Loss mse: 0.0666, Train Loss ce: 0.0472, Train Steps/Sec: 0.06,
 [2026-01-30 07:44:27] (step=0002141) Train Loss mse: 0.0669, Train Loss ce: 0.0468, Train Steps/Sec: 0.07,
@@ -3206,6 +3199,34 @@ ce_avg: 0.04585418477654457, mse_avg: 0.050133317708969116
 [2026-01-30 11:26:18] (step=0002978) Train Loss mse: 0.0600, Train Loss ce: 0.0457, Train Steps/Sec: 0.06,
 [2026-01-30 11:26:33] (step=0002979) Train Loss mse: 0.0572, Train Loss ce: 0.0449, Train Steps/Sec: 0.06,
 [2026-01-30 11:26:48] (step=0002980) Train Loss mse: 0.0565, Train Loss ce: 0.0443, Train Steps/Sec: 0.07,
+base_dir is /dev/shm/models/checkpoints_vlm_gym_mental_rotation_3d_objaverse_pad3_by_axis_one_image_lr2e_5_ce_ins/eval_used_rows, step_tag is checkpoints_vlm_gym_mental_rotation_3d_objaverse_pad3_by_axis_one_image_lr2e_5_ce_ins_step2500
+Preparing Dataset vlm_gym_mental_rotation_3d_objaverse_pad3_by_axis_celoss_evalonce/vlm_gym_mental_rotation_3d_objaverse_pad3_by_axis_val
+[eval debug] first 3 batch fingerprints:
+  fp[0]: [{'data_indexes': [0], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_3d_objaverse_pad3_by_axis_celoss_evalonce'}]
+  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_3d_objaverse_pad3_by_axis_celoss_evalonce'}]
+  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_3d_objaverse_pad3_by_axis_celoss_evalonce'}]
+ce_avg: 0.0595068633556366, mse_avg: 0.05313115939497948
+base_dir is /dev/shm/models/checkpoints_vlm_gym_mental_rotation_3d_objaverse_pad3_by_axis_one_image_lr2e_5_ce_ins/eval_used_rows, step_tag is checkpoints_vlm_gym_mental_rotation_3d_objaverse_pad3_by_axis_one_image_lr2e_5_ce_ins_step3000
+Preparing Dataset vlm_gym_mental_rotation_3d_objaverse_pad3_by_axis_celoss_evalonce/vlm_gym_mental_rotation_3d_objaverse_pad3_by_axis_val
+[eval debug] first 3 batch fingerprints:
+  fp[0]: [{'data_indexes': [0], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_3d_objaverse_pad3_by_axis_celoss_evalonce'}]
+  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_3d_objaverse_pad3_by_axis_celoss_evalonce'}]
+  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_3d_objaverse_pad3_by_axis_celoss_evalonce'}]
+ce_avg: 0.04585418477654457, mse_avg: 0.050133317708969116
+base_dir is /dev/shm/models/checkpoints_vlm_gym_mental_rotation_3d_objaverse_pad3_by_axis_one_image_lr2e_5_ce_ins/eval_used_rows, step_tag is checkpoints_vlm_gym_mental_rotation_3d_objaverse_pad3_by_axis_one_image_lr2e_5_ce_ins_step3500
+Preparing Dataset vlm_gym_mental_rotation_3d_objaverse_pad3_by_axis_celoss_evalonce/vlm_gym_mental_rotation_3d_objaverse_pad3_by_axis_val
+[eval debug] first 3 batch fingerprints:
+  fp[0]: [{'data_indexes': [0], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_3d_objaverse_pad3_by_axis_celoss_evalonce'}]
+  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_3d_objaverse_pad3_by_axis_celoss_evalonce'}]
+  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_3d_objaverse_pad3_by_axis_celoss_evalonce'}]
+ce_avg: 0.0454382449388504, mse_avg: 0.050951480865478516
+base_dir is /dev/shm/models/checkpoints_vlm_gym_mental_rotation_3d_objaverse_pad3_by_axis_one_image_lr2e_5_ce_ins/eval_used_rows, step_tag is checkpoints_vlm_gym_mental_rotation_3d_objaverse_pad3_by_axis_one_image_lr2e_5_ce_ins_step4000
+Preparing Dataset vlm_gym_mental_rotation_3d_objaverse_pad3_by_axis_celoss_evalonce/vlm_gym_mental_rotation_3d_objaverse_pad3_by_axis_val
+[eval debug] first 3 batch fingerprints:
+  fp[0]: [{'data_indexes': [0], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_3d_objaverse_pad3_by_axis_celoss_evalonce'}]
+  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_3d_objaverse_pad3_by_axis_celoss_evalonce'}]
+  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_3d_objaverse_pad3_by_axis_celoss_evalonce'}]
+ce_avg: 0.04481803998351097, mse_avg: 0.051003698259592056
 [2026-01-30 11:27:04] (step=0002981) Train Loss mse: 0.0733, Train Loss ce: 0.0434, Train Steps/Sec: 0.07,
 [2026-01-30 11:27:20] (step=0002982) Train Loss mse: 0.0724, Train Loss ce: 0.0553, Train Steps/Sec: 0.06,
 [2026-01-30 11:27:36] (step=0002983) Train Loss mse: 0.0568, Train Loss ce: 0.0460, Train Steps/Sec: 0.06,
@@ -3441,20 +3462,6 @@ ce_avg: 0.04585418477654457, mse_avg: 0.050133317708969116
 [2026-01-30 12:28:00] (step=0003213) Train Loss mse: 0.0647, Train Loss ce: 0.0521, Train Steps/Sec: 0.06,
 [2026-01-30 12:28:16] (step=0003214) Train Loss mse: 0.0573, Train Loss ce: 0.0436, Train Steps/Sec: 0.06,
 [2026-01-30 12:28:31] (step=0003215) Train Loss mse: 0.0523, Train Loss ce: 0.0444, Train Steps/Sec: 0.07,
-[2026-01-30 12:28:47
-base_dir is /dev/shm/models/checkpoints_vlm_gym_mental_rotation_3d_objaverse_pad3_by_axis_one_image_lr2e_5_ce_ins/eval_used_rows, step_tag is checkpoints_vlm_gym_mental_rotation_3d_objaverse_pad3_by_axis_one_image_lr2e_5_ce_ins_step3500
-Preparing Dataset vlm_gym_mental_rotation_3d_objaverse_pad3_by_axis_celoss_evalonce/vlm_gym_mental_rotation_3d_objaverse_pad3_by_axis_val
-[eval debug] first 3 batch fingerprints:
-  fp[0]: [{'data_indexes': [0], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_3d_objaverse_pad3_by_axis_celoss_evalonce'}]
-  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_3d_objaverse_pad3_by_axis_celoss_evalonce'}]
-  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_3d_objaverse_pad3_by_axis_celoss_evalonce'}]
-ce_avg: 0.0454382449388504, mse_avg: 0.050951480865478516
-base_dir is /dev/shm/models/checkpoints_vlm_gym_mental_rotation_3d_objaverse_pad3_by_axis_one_image_lr2e_5_ce_ins/eval_used_rows, step_tag is checkpoints_vlm_gym_mental_rotation_3d_objaverse_pad3_by_axis_one_image_lr2e_5_ce_ins_step4000
-Preparing Dataset vlm_gym_mental_rotation_3d_objaverse_pad3_by_axis_celoss_evalonce/vlm_gym_mental_rotation_3d_objaverse_pad3_by_axis_val
-[eval debug] first 3 batch fingerprints:
-  fp[0]: [{'data_indexes': [0], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_3d_objaverse_pad3_by_axis_celoss_evalonce'}]
-  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_3d_objaverse_pad3_by_axis_celoss_evalonce'}]
-  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_3d_objaverse_pad3_by_axis_celoss_evalonce'}]
 [2026-01-30 12:28:47] (step=0003216) Train Loss mse: 0.0666, Train Loss ce: 0.0429, Train Steps/Sec: 0.06,
 [2026-01-30 12:29:02] (step=0003217) Train Loss mse: 0.0587, Train Loss ce: 0.0472, Train Steps/Sec: 0.06,
 [2026-01-30 12:29:17] (step=0003218) Train Loss mse: 0.0663, Train Loss ce: 0.0479, Train Steps/Sec: 0.07,
@@ -4540,6 +4547,20 @@ Preparing Dataset vlm_gym_mental_rotation_3d_objaverse_pad3_by_axis_celoss_evalo
 [2026-01-30 17:12:18] (step=0004298) Train Loss mse: 0.0616, Train Loss ce: 0.0438, Train Steps/Sec: 0.06,
 [2026-01-30 17:12:34] (step=0004299) Train Loss mse: 0.0613, Train Loss ce: 0.0470, Train Steps/Sec: 0.06,
 [2026-01-30 17:12:50] (step=0004300) Train Loss mse: 0.0674, Train Loss ce: 0.0424, Train Steps/Sec: 0.06,
+base_dir is /dev/shm/models/checkpoints_vlm_gym_mental_rotation_3d_objaverse_pad3_by_axis_one_image_lr2e_5_ce_ins/eval_used_rows, step_tag is checkpoints_vlm_gym_mental_rotation_3d_objaverse_pad3_by_axis_one_image_lr2e_5_ce_ins_step4500
+Preparing Dataset vlm_gym_mental_rotation_3d_objaverse_pad3_by_axis_celoss_evalonce/vlm_gym_mental_rotation_3d_objaverse_pad3_by_axis_val
+[eval debug] first 3 batch fingerprints:
+  fp[0]: [{'data_indexes': [0], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_3d_objaverse_pad3_by_axis_celoss_evalonce'}]
+  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_3d_objaverse_pad3_by_axis_celoss_evalonce'}]
+  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_3d_objaverse_pad3_by_axis_celoss_evalonce'}]
+ce_avg: 0.04478031024336815, mse_avg: 0.04991578683257103
+base_dir is /dev/shm/models/checkpoints_vlm_gym_mental_rotation_3d_objaverse_pad3_by_axis_one_image_lr2e_5_ce_ins/eval_used_rows, step_tag is checkpoints_vlm_gym_mental_rotation_3d_objaverse_pad3_by_axis_one_image_lr2e_5_ce_ins_step5000
+Preparing Dataset vlm_gym_mental_rotation_3d_objaverse_pad3_by_axis_celoss_evalonce/vlm_gym_mental_rotation_3d_objaverse_pad3_by_axis_val
+[eval debug] first 3 batch fingerprints:
+  fp[0]: [{'data_indexes': [0], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_3d_objaverse_pad3_by_axis_celoss_evalonce'}]
+  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_3d_objaverse_pad3_by_axis_celoss_evalonce'}]
+  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_3d_objaverse_pad3_by_axis_celoss_evalonce'}]
+ce_avg: 0.04477126523852348, mse_avg: 0.05164656043052673
 [2026-01-30 17:13:06] (step=0004301) Train Loss mse: 0.0708, Train Loss ce: 0.0447, Train Steps/Sec: 0.06,
 [2026-01-30 17:13:23] (step=0004302) Train Loss mse: 0.0632, Train Loss ce: 0.0413, Train Steps/Sec: 0.06,
 [2026-01-30 17:13:38] (step=0004303) Train Loss mse: 0.0644, Train Loss ce: 0.0458, Train Steps/Sec: 0.07,
@@ -4560,20 +4581,6 @@ Preparing Dataset vlm_gym_mental_rotation_3d_objaverse_pad3_by_axis_celoss_evalo
 [2026-01-30 17:17:35] (step=0004318) Train Loss mse: 0.0564, Train Loss ce: 0.0434, Train Steps/Sec: 0.06,
 [2026-01-30 17:17:51] (step=0004319) Train Loss mse: 0.0597, Train Loss ce: 0.0463, Train Steps/Sec: 0.06,
 [2026-01-30 17:18:07] (step=0004320) Train Loss mse: 0.0718, Train Loss ce: 0.0441, Train Steps/Sec: 0.06,
-base_dir is /dev/shm/models/checkpoints_vlm_gym_mental_rotation_3d_objaverse_pad3_by_axis_one_image_lr2e_5_ce_ins/eval_used_rows, step_tag is checkpoints_vlm_gym_mental_rotation_3d_objaverse_pad3_by_axis_one_image_lr2e_5_ce_ins_step4500
-Preparing Dataset vlm_gym_mental_rotation_3d_objaverse_pad3_by_axis_celoss_evalonce/vlm_gym_mental_rotation_3d_objaverse_pad3_by_axis_val
-[eval debug] first 3 batch fingerprints:
-  fp[0]: [{'data_indexes': [0], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_3d_objaverse_pad3_by_axis_celoss_evalonce'}]
-  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_3d_objaverse_pad3_by_axis_celoss_evalonce'}]
-  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_3d_objaverse_pad3_by_axis_celoss_evalonce'}]
-ce_avg: 0.04478031024336815, mse_avg: 0.04991578683257103
-base_dir is /dev/shm/models/checkpoints_vlm_gym_mental_rotation_3d_objaverse_pad3_by_axis_one_image_lr2e_5_ce_ins/eval_used_rows, step_tag is checkpoints_vlm_gym_mental_rotation_3d_objaverse_pad3_by_axis_one_image_lr2e_5_ce_ins_step5000
-Preparing Dataset vlm_gym_mental_rotation_3d_objaverse_pad3_by_axis_celoss_evalonce/vlm_gym_mental_rotation_3d_objaverse_pad3_by_axis_val
-[eval debug] first 3 batch fingerprints:
-  fp[0]: [{'data_indexes': [0], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_3d_objaverse_pad3_by_axis_celoss_evalonce'}]
-  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_3d_objaverse_pad3_by_axis_celoss_evalonce'}]
-  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_3d_objaverse_pad3_by_axis_celoss_evalonce'}]
-ce_avg: 0.04477126523852348, mse_avg: 0.05164656043052673
 [2026-01-30 17:18:22] (step=0004321) Train Loss mse: 0.0604, Train Loss ce: 0.0477, Train Steps/Sec: 0.07,
 [2026-01-30 17:18:39] (step=0004322) Train Loss mse: 0.0561, Train Loss ce: 0.0457, Train Steps/Sec: 0.06,
 [2026-01-30 17:18:54] (step=0004323) Train Loss mse: 0.0657, Train Loss ce: 0.0452, Train Steps/Sec: 0.07,