Upload checkpoints_vlm_gym_mental_rotation_3d_pad3_by_axis_one_image_lr2e_5_ce_no_mse_ins/checkpoints_vlm_gym_mental_rotation_3d_pad3_by_axis_one_image_lr2e_5_ce_no_mse_ins

checkpoints_vlm_gym_mental_rotation_3d_pad3_by_axis_one_image_lr2e_5_ce_no_mse_ins/checkpoints_vlm_gym_mental_rotation_3d_pad3_by_axis_one_image_lr2e_5_ce_no_mse_ins/wandb/offline-run-20260129_221049-checkpoints_vlm_gym_mental_rotation_3d_pad3_by_axis_one_image_lr2e_5_ce_no_mse_ins-run0/files/output.log

@@ -1,173 +1,3 @@
-FullyShardedDataParallel(
-  (_fsdp_wrapped_module): Bagel(
-    (language_model): Qwen2ForCausalLM(
-      (model): Qwen2Model(
-        (embed_tokens): Embedding(152064, 3584)
-        (layers): ModuleList(
-          (0-27): 28 x FullyShardedDataParallel(
-            (_fsdp_wrapped_module): CheckpointWrapper(
-              (_checkpoint_wrapped_module): Qwen2MoTDecoderLayer(
-                (self_attn): PackedAttentionMoT(
-                  (q_proj): Linear(in_features=3584, out_features=3584, bias=True)
-                  (k_proj): Linear(in_features=3584, out_features=512, bias=True)
-                  (v_proj): Linear(in_features=3584, out_features=512, bias=True)
-                  (o_proj): Linear(in_features=3584, out_features=3584, bias=False)
-                  (q_norm): Qwen2RMSNorm((128,), eps=1e-06)
-                  (k_norm): Qwen2RMSNorm((128,), eps=1e-06)
-                  (q_norm_moe_gen): Qwen2RMSNorm((128,), eps=1e-06)
-                  (k_norm_moe_gen): Qwen2RMSNorm((128,), eps=1e-06)
-                  (q_proj_moe_gen): Linear(in_features=3584, out_features=3584, bias=True)
-                  (k_proj_moe_gen): Linear(in_features=3584, out_features=512, bias=True)
-                  (v_proj_moe_gen): Linear(in_features=3584, out_features=512, bias=True)
-                  (o_proj_moe_gen): Linear(in_features=3584, out_features=3584, bias=False)
-                )
-                (mlp): Qwen2MLP(
-                  (gate_proj): Linear(in_features=3584, out_features=18944, bias=False)
-                  (up_proj): Linear(in_features=3584, out_features=18944, bias=False)
-                  (down_proj): Linear(in_features=18944, out_features=3584, bias=False)
-                  (act_fn): SiLU()
-                )
-                (mlp_moe_gen): Qwen2MLP(
-                  (gate_proj): Linear(in_features=3584, out_features=18944, bias=False)
-                  (up_proj): Linear(in_features=3584, out_features=18944, bias=False)
-                  (down_proj): Linear(in_features=18944, out_features=3584, bias=False)
-                  (act_fn): SiLU()
-                )
-                (input_layernorm): Qwen2RMSNorm((3584,), eps=1e-06)
-                (input_layernorm_moe_gen): Qwen2RMSNorm((3584,), eps=1e-06)
-                (post_attention_layernorm): Qwen2RMSNorm((3584,), eps=1e-06)
-                (post_attention_layernorm_moe_gen): Qwen2RMSNorm((3584,), eps=1e-06)
-              )
-            )
-          )
-        )
-        (norm): Qwen2RMSNorm((3584,), eps=1e-06)
-        (norm_moe_gen): Qwen2RMSNorm((3584,), eps=1e-06)
-        (rotary_emb): Qwen2RotaryEmbedding()
-      )
-      (lm_head): Linear(in_features=3584, out_features=152064, bias=False)
-    )
-    (vit_model): SiglipVisionModel(
-      (vision_model): FullyShardedDataParallel(
-        (_fsdp_wrapped_module): SiglipVisionTransformer(
-          (embeddings): SiglipVisionEmbeddings(
-            (position_embedding): Embedding(4900, 1152)
-            (patch_embedding): Linear(in_features=588, out_features=1152, bias=True)
-          )
-          (encoder): SiglipEncoder(
-            (layers): ModuleList(
-              (0-25): 26 x FullyShardedDataParallel(
-                (_fsdp_wrapped_module): CheckpointWrapper(
-                  (_checkpoint_wrapped_module): SiglipEncoderLayer(
-                    (self_attn): SiglipFlashAttention2(
-                      (k_proj): Linear(in_features=1152, out_features=1152, bias=True)
-                      (v_proj): Linear(in_features=1152, out_features=1152, bias=True)
-                      (q_proj): Linear(in_features=1152, out_features=1152, bias=True)
-                      (out_proj): Linear(in_features=1152, out_features=1152, bias=True)
-                    )
-                    (layer_norm1): LayerNorm((1152,), eps=1e-06, elementwise_affine=True)
-                    (mlp): SiglipMLP(
-                      (activation_fn): PytorchGELUTanh()
-                      (fc1): Linear(in_features=1152, out_features=4304, bias=True)
-                      (fc2): Linear(in_features=4304, out_features=1152, bias=True)
-                    )
-                    (layer_norm2): LayerNorm((1152,), eps=1e-06, elementwise_affine=True)
-                  )
-                )
-              )
-            )
-          )
-          (post_layernorm): LayerNorm((1152,), eps=1e-06, elementwise_affine=True)
-        )
-      )
-    )
-    (connector): FullyShardedDataParallel(
-      (_fsdp_wrapped_module): CheckpointWrapper(
-        (_checkpoint_wrapped_module): MLPconnector(
-          (activation_fn): PytorchGELUTanh()
-          (fc1): Linear(in_features=1152, out_features=3584, bias=True)
-          (fc2): Linear(in_features=3584, out_features=3584, bias=True)
-        )
-      )
-    )
-    (vit_pos_embed): FullyShardedDataParallel(
-      (_fsdp_wrapped_module): PositionEmbedding()
-    )
-  )
-)
-_flat_param True
-language_model.model.layers.0._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
-language_model.model.layers.1._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
-language_model.model.layers.2._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
-language_model.model.layers.3._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
-language_model.model.layers.4._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
-language_model.model.layers.5._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
-language_model.model.layers.6._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
-language_model.model.layers.7._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
-language_model.model.layers.8._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
-language_model.model.layers.9._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
-language_model.model.layers.10._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
-language_model.model.layers.11._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
-language_model.model.layers.12._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
-language_model.model.layers.13._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
-language_model.model.layers.14._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
-language_model.model.layers.15._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
-language_model.model.layers.16._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
-language_model.model.layers.17._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
-language_model.model.layers.18._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
-language_model.model.layers.19._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
-language_model.model.layers.20._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
-language_model.model.layers.21._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
-language_model.model.layers.22._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
-language_model.model.layers.23._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
-language_model.model.layers.24._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
-language_model.model.layers.25._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
-language_model.model.layers.26._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
-language_model.model.layers.27._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
-vit_model.vision_model._fsdp_wrapped_module._flat_param True
-vit_model.vision_model._fsdp_wrapped_module.encoder.layers.0._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
-vit_model.vision_model._fsdp_wrapped_module.encoder.layers.1._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
-vit_model.vision_model._fsdp_wrapped_module.encoder.layers.2._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
-vit_model.vision_model._fsdp_wrapped_module.encoder.layers.3._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
-vit_model.vision_model._fsdp_wrapped_module.encoder.layers.4._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
-vit_model.vision_model._fsdp_wrapped_module.encoder.layers.5._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
-vit_model.vision_model._fsdp_wrapped_module.encoder.layers.6._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
-vit_model.vision_model._fsdp_wrapped_module.encoder.layers.7._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
-vit_model.vision_model._fsdp_wrapped_module.encoder.layers.8._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
-vit_model.vision_model._fsdp_wrapped_module.encoder.layers.9._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
-vit_model.vision_model._fsdp_wrapped_module.encoder.layers.10._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
-vit_model.vision_model._fsdp_wrapped_module.encoder.layers.11._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
-vit_model.vision_model._fsdp_wrapped_module.encoder.layers.12._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
-vit_model.vision_model._fsdp_wrapped_module.encoder.layers.13._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
-vit_model.vision_model._fsdp_wrapped_module.encoder.layers.14._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
-vit_model.vision_model._fsdp_wrapped_module.encoder.layers.15._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
-vit_model.vision_model._fsdp_wrapped_module.encoder.layers.16._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
-vit_model.vision_model._fsdp_wrapped_module.encoder.layers.17._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
-vit_model.vision_model._fsdp_wrapped_module.encoder.layers.18._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
-vit_model.vision_model._fsdp_wrapped_module.encoder.layers.19._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
-vit_model.vision_model._fsdp_wrapped_module.encoder.layers.20._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
-vit_model.vision_model._fsdp_wrapped_module.encoder.layers.21._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
-vit_model.vision_model._fsdp_wrapped_module.encoder.layers.22._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
-vit_model.vision_model._fsdp_wrapped_module.encoder.layers.23._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
-vit_model.vision_model._fsdp_wrapped_module.encoder.layers.24._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
-vit_model.vision_model._fsdp_wrapped_module.encoder.layers.25._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
-connector._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
-vit_pos_embed._fsdp_wrapped_module._flat_param False
-Preparing Dataset vlm_gym_mental_rotation_3d_pad3_by_axis_celoss_no_mse/vlm_gym_mental_rotation_3d_pad3_by_axis_train
-base_dir is /dev/shm/models/checkpoints_vlm_gym_mental_rotation_3d_pad3_by_axis_one_image_lr2e_5_ce_no_mse_ins/eval_used_rows, step_tag is checkpoints_vlm_gym_mental_rotation_3d_pad3_by_axis_one_image_lr2e_5_ce_no_mse_ins_step0
-Preparing Dataset vlm_gym_mental_rotation_3d_pad3_by_axis_celoss_no_mse_evalonce/vlm_gym_mental_rotation_3d_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_pad3_by_axis_celoss_no_mse_evalonce'}]
-  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_3d_pad3_by_axis_celoss_no_mse_evalonce'}]
-  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_3d_pad3_by_axis_celoss_no_mse_evalonce'}]
-ce_avg: 0.2727155089378357, mse_avg: 0.0
-base_dir is /dev/shm/models/checkpoints_vlm_gym_mental_rotation_3d_pad3_by_axis_one_image_lr2e_5_ce_no_mse_ins/eval_used_rows, step_tag is checkpoints_vlm_gym_mental_rotation_3d_pad3_by_axis_one_image_lr2e_5_ce_no_mse_ins_step500
-Preparing Dataset vlm_gym_mental_rotation_3d_pad3_by_axis_celoss_no_mse_evalonce/vlm_gym_mental_rotation_3d_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_pad3_by_axis_celoss_no_mse_evalonce'}]
-  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_3d_pad3_by_axis_celoss_no_mse_evalonce'}]
-  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_3d_pad3_by_axis_celoss_no_mse_evalonce'}]
-ce_avg: 0.05276435613632202, mse_avg: 0.0
 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/
@@ -1207,6 +1037,197 @@ wandb: For more information, check out the docs at: https://weave-docs.wandb.ai/
 [2026-01-29 22:53:25] (step=0001026) Train Loss mse: 0.0000, Train Loss ce: 0.0461, Train Steps/Sec: 0.51,
 [2026-01-29 22:53:27] (step=0001027) Train Loss mse: 0.0000, Train Loss ce: 0.0435, Train Steps/Sec: 0.51,
 [2026-01-29 22:53:29] (step=0001028) Train Loss mse: 0.0000, Train Loss ce: 0.0425, Train Steps/Sec: 0.51,
+FullyShardedDataParallel(
+  (_fsdp_wrapped_module): Bagel(
+    (language_model): Qwen2ForCausalLM(
+      (model): Qwen2Model(
+        (embed_tokens): Embedding(152064, 3584)
+        (layers): ModuleList(
+          (0-27): 28 x FullyShardedDataParallel(
+            (_fsdp_wrapped_module): CheckpointWrapper(
+              (_checkpoint_wrapped_module): Qwen2MoTDecoderLayer(
+                (self_attn): PackedAttentionMoT(
+                  (q_proj): Linear(in_features=3584, out_features=3584, bias=True)
+                  (k_proj): Linear(in_features=3584, out_features=512, bias=True)
+                  (v_proj): Linear(in_features=3584, out_features=512, bias=True)
+                  (o_proj): Linear(in_features=3584, out_features=3584, bias=False)
+                  (q_norm): Qwen2RMSNorm((128,), eps=1e-06)
+                  (k_norm): Qwen2RMSNorm((128,), eps=1e-06)
+                  (q_norm_moe_gen): Qwen2RMSNorm((128,), eps=1e-06)
+                  (k_norm_moe_gen): Qwen2RMSNorm((128,), eps=1e-06)
+                  (q_proj_moe_gen): Linear(in_features=3584, out_features=3584, bias=True)
+                  (k_proj_moe_gen): Linear(in_features=3584, out_features=512, bias=True)
+                  (v_proj_moe_gen): Linear(in_features=3584, out_features=512, bias=True)
+                  (o_proj_moe_gen): Linear(in_features=3584, out_features=3584, bias=False)
+                )
+                (mlp): Qwen2MLP(
+                  (gate_proj): Linear(in_features=3584, out_features=18944, bias=False)
+                  (up_proj): Linear(in_features=3584, out_features=18944, bias=False)
+                  (down_proj): Linear(in_features=18944, out_features=3584, bias=False)
+                  (act_fn): SiLU()
+                )
+                (mlp_moe_gen): Qwen2MLP(
+                  (gate_proj): Linear(in_features=3584, out_features=18944, bias=False)
+                  (up_proj): Linear(in_features=3584, out_features=18944, bias=False)
+                  (down_proj): Linear(in_features=18944, out_features=3584, bias=False)
+                  (act_fn): SiLU()
+                )
+                (input_layernorm): Qwen2RMSNorm((3584,), eps=1e-06)
+                (input_layernorm_moe_gen): Qwen2RMSNorm((3584,), eps=1e-06)
+                (post_attention_layernorm): Qwen2RMSNorm((3584,), eps=1e-06)
+                (post_attention_layernorm_moe_gen): Qwen2RMSNorm((3584,), eps=1e-06)
+              )
+            )
+          )
+        )
+        (norm): Qwen2RMSNorm((3584,), eps=1e-06)
+        (norm_moe_gen): Qwen2RMSNorm((3584,), eps=1e-06)
+        (rotary_emb): Qwen2RotaryEmbedding()
+      )
+      (lm_head): Linear(in_features=3584, out_features=152064, bias=False)
+    )
+    (vit_model): SiglipVisionModel(
+      (vision_model): FullyShardedDataParallel(
+        (_fsdp_wrapped_module): SiglipVisionTransformer(
+          (embeddings): SiglipVisionEmbeddings(
+            (position_embedding): Embedding(4900, 1152)
+            (patch_embedding): Linear(in_features=588, out_features=1152, bias=True)
+          )
+          (encoder): SiglipEncoder(
+            (layers): ModuleList(
+              (0-25): 26 x FullyShardedDataParallel(
+                (_fsdp_wrapped_module): CheckpointWrapper(
+                  (_checkpoint_wrapped_module): SiglipEncoderLayer(
+                    (self_attn): SiglipFlashAttention2(
+                      (k_proj): Linear(in_features=1152, out_features=1152, bias=True)
+                      (v_proj): Linear(in_features=1152, out_features=1152, bias=True)
+                      (q_proj): Linear(in_features=1152, out_features=1152, bias=True)
+                      (out_proj): Linear(in_features=1152, out_features=1152, bias=True)
+                    )
+                    (layer_norm1): LayerNorm((1152,), eps=1e-06, elementwise_affine=True)
+                    (mlp): SiglipMLP(
+                      (activation_fn): PytorchGELUTanh()
+                      (fc1): Linear(in_features=1152, out_features=4304, bias=True)
+                      (fc2): Linear(in_features=4304, out_features=1152, bias=True)
+                    )
+                    (layer_norm2): LayerNorm((1152,), eps=1e-06, elementwise_affine=True)
+                  )
+                )
+              )
+            )
+          )
+          (post_layernorm): LayerNorm((1152,), eps=1e-06, elementwise_affine=True)
+        )
+      )
+    )
+    (connector): FullyShardedDataParallel(
+      (_fsdp_wrapped_module): CheckpointWrapper(
+        (_checkpoint_wrapped_module): MLPconnector(
+          (activation_fn): PytorchGELUTanh()
+          (fc1): Linear(in_features=1152, out_features=3584, bias=True)
+          (fc2): Linear(in_features=3584, out_features=3584, bias=True)
+        )
+      )
+    )
+    (vit_pos_embed): FullyShardedDataParallel(
+      (_fsdp_wrapped_module): PositionEmbedding()
+    )
+  )
+)
+_flat_param True
+language_model.model.layers.0._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+language_model.model.layers.1._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+language_model.model.layers.2._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+language_model.model.layers.3._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+language_model.model.layers.4._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+language_model.model.layers.5._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+language_model.model.layers.6._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+language_model.model.layers.7._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+language_model.model.layers.8._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+language_model.model.layers.9._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+language_model.model.layers.10._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+language_model.model.layers.11._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+language_model.model.layers.12._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+language_model.model.layers.13._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+language_model.model.layers.14._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+language_model.model.layers.15._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+language_model.model.layers.16._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+language_model.model.layers.17._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+language_model.model.layers.18._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+language_model.model.layers.19._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+language_model.model.layers.20._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+language_model.model.layers.21._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+language_model.model.layers.22._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+language_model.model.layers.23._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+language_model.model.layers.24._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+language_model.model.layers.25._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+language_model.model.layers.26._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+language_model.model.layers.27._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+vit_model.vision_model._fsdp_wrapped_module._flat_param True
+vit_model.vision_model._fsdp_wrapped_module.encoder.layers.0._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+vit_model.vision_model._fsdp_wrapped_module.encoder.layers.1._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+vit_model.vision_model._fsdp_wrapped_module.encoder.layers.2._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+vit_model.vision_model._fsdp_wrapped_module.encoder.layers.3._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+vit_model.vision_model._fsdp_wrapped_module.encoder.layers.4._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+vit_model.vision_model._fsdp_wrapped_module.encoder.layers.5._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+vit_model.vision_model._fsdp_wrapped_module.encoder.layers.6._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+vit_model.vision_model._fsdp_wrapped_module.encoder.layers.7._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+vit_model.vision_model._fsdp_wrapped_module.encoder.layers.8._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+vit_model.vision_model._fsdp_wrapped_module.encoder.layers.9._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+vit_model.vision_model._fsdp_wrapped_module.encoder.layers.10._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+vit_model.vision_model._fsdp_wrapped_module.encoder.layers.11._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+vit_model.vision_model._fsdp_wrapped_module.encoder.layers.12._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+vit_model.vision_model._fsdp_wrapped_module.encoder.layers.13._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+vit_model.vision_model._fsdp_wrapped_module.encoder.layers.14._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+vit_model.vision_model._fsdp_wrapped_module.encoder.layers.15._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+vit_model.vision_model._fsdp_wrapped_module.encoder.layers.16._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+vit_model.vision_model._fsdp_wrapped_module.encoder.layers.17._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+vit_model.vision_model._fsdp_wrapped_module.encoder.layers.18._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+vit_model.vision_model._fsdp_wrapped_module.encoder.layers.19._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+vit_model.vision_model._fsdp_wrapped_module.encoder.layers.20._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+vit_model.vision_model._fsdp_wrapped_module.encoder.layers.21._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+vit_model.vision_model._fsdp_wrapped_module.encoder.layers.22._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+vit_model.vision_model._fsdp_wrapped_module.encoder.layers.23._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+vit_model.vision_model._fsdp_wrapped_module.encoder.layers.24._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+vit_model.vision_model._fsdp_wrapped_module.encoder.layers.25._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+connector._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+vit_pos_embed._fsdp_wrapped_module._flat_param False
+Preparing Dataset vlm_gym_mental_rotation_3d_pad3_by_axis_celoss_no_mse/vlm_gym_mental_rotation_3d_pad3_by_axis_train
+base_dir is /dev/shm/models/checkpoints_vlm_gym_mental_rotation_3d_pad3_by_axis_one_image_lr2e_5_ce_no_mse_ins/eval_used_rows, step_tag is checkpoints_vlm_gym_mental_rotation_3d_pad3_by_axis_one_image_lr2e_5_ce_no_mse_ins_step0
+Preparing Dataset vlm_gym_mental_rotation_3d_pad3_by_axis_celoss_no_mse_evalonce/vlm_gym_mental_rotation_3d_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_pad3_by_axis_celoss_no_mse_evalonce'}]
+  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_3d_pad3_by_axis_celoss_no_mse_evalonce'}]
+  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_3d_pad3_by_axis_celoss_no_mse_evalonce'}]
+ce_avg: 0.2727155089378357, mse_avg: 0.0
+base_dir is /dev/shm/models/checkpoints_vlm_gym_mental_rotation_3d_pad3_by_axis_one_image_lr2e_5_ce_no_mse_ins/eval_used_rows, step_tag is checkpoints_vlm_gym_mental_rotation_3d_pad3_by_axis_one_image_lr2e_5_ce_no_mse_ins_step500
+Preparing Dataset vlm_gym_mental_rotation_3d_pad3_by_axis_celoss_no_mse_evalonce/vlm_gym_mental_rotation_3d_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_pad3_by_axis_celoss_no_mse_evalonce'}]
+  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_3d_pad3_by_axis_celoss_no_mse_evalonce'}]
+  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_3d_pad3_by_axis_celoss_no_mse_evalonce'}]
+ce_avg: 0.05276435613632202, mse_avg: 0.0
+base_dir is /dev/shm/models/checkpoints_vlm_gym_mental_rotation_3d_pad3_by_axis_one_image_lr2e_5_ce_no_mse_ins/eval_used_rows, step_tag is checkpoints_vlm_gym_mental_rotation_3d_pad3_by_axis_one_image_lr2e_5_ce_no_mse_ins_step1000
+Preparing Dataset vlm_gym_mental_rotation_3d_pad3_by_axis_celoss_no_mse_evalonce/vlm_gym_mental_rotation_3d_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_pad3_by_axis_celoss_no_mse_evalonce'}]
+  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_3d_pad3_by_axis_celoss_no_mse_evalonce'}]
+  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_3d_pad3_by_axis_celoss_no_mse_evalonce'}]
+ce_avg: 0.04682447761297226, mse_avg: 0.0
+base_dir is /dev/shm/models/checkpoints_vlm_gym_mental_rotation_3d_pad3_by_axis_one_image_lr2e_5_ce_no_mse_ins/eval_used_rows, step_tag is checkpoints_vlm_gym_mental_rotation_3d_pad3_by_axis_one_image_lr2e_5_ce_no_mse_ins_step1500
+Preparing Dataset vlm_gym_mental_rotation_3d_pad3_by_axis_celoss_no_mse_evalonce/vlm_gym_mental_rotation_3d_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_pad3_by_axis_celoss_no_mse_evalonce'}]
+  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_3d_pad3_by_axis_celoss_no_mse_evalonce'}]
+  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_3d_pad3_by_axis_celoss_no_mse_evalonce'}]
+ce_avg: 0.04674335569143295, mse_avg: 0.0
+base_dir is /dev/shm/models/checkpoints_vlm_gym_mental_rotation_3d_pad3_by_axis_one_image_lr2e_5_ce_no_mse_ins/eval_used_rows, step_tag is checkpoints_vlm_gym_mental_rotation_3d_pad3_by_axis_one_image_lr2e_5_ce_no_mse_ins_step2000
+Preparing Dataset vlm_gym_mental_rotation_3d_pad3_by_axis_celoss_no_mse_evalonce/vlm_gym_mental_rotation_3d_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_pad3_by_axis_celoss_no_mse_evalonce'}]
+  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_3d_pad3_by_axis_celoss_no_mse_evalonce'}]
+  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_3d_pad3_by_axis_celoss_no_mse_evalonce'}]
+ce_avg: 0.04835177958011627, mse_avg: 0.0
 [2026-01-29 22:53:31] (step=0001029) Train Loss mse: 0.0000, Train Loss ce: 0.0456, Train Steps/Sec: 0.51,
 [2026-01-29 22:53:33] (step=0001030) Train Loss mse: 0.0000, Train Loss ce: 0.0425, Train Steps/Sec: 0.46,
 [2026-01-29 22:53:35] (step=0001031) Train Loss mse: 0.0000, Train Loss ce: 0.0421, Train Steps/Sec: 0.51,
@@ -1333,27 +1354,6 @@ wandb: For more information, check out the docs at: https://weave-docs.wandb.ai/
 [2026-01-29 22:57:50] (step=0001152) Train Loss mse: 0.0000, Train Loss ce: 0.0421, Train Steps/Sec: 0.51,
 [2026-01-29 22:57:52] (step=0001153) Train Loss mse: 0.0000, Train Loss ce: 0.0400, Train Steps/Sec: 0.39,
 [2026-01-29 22:57:54] (step=0001154) Train Loss mse: 0.0000, Train Loss ce: 0.0450, Train Steps/Sec: 0.42,
-base_dir is /dev/shm/models/checkpoints_vlm_gym_mental_rotation_3d_pad3_by_axis_one_image_lr2e_5_ce_no_mse_ins/eval_used_rows, step_tag is checkpoints_vlm_gym_mental_rotation_3d_pad3_by_axis_one_image_lr2e_5_ce_no_mse_ins_step1500
-Preparing Dataset vlm_gym_mental_rotation_3d_pad3_by_axis_celoss_no_mse_evalonce/vlm_gym_mental_rotation_3d_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_pad3_by_axis_celoss_no_mse_evalonce'}]
-  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_3d_pad3_by_axis_celoss_no_mse_evalonce'}]
-  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_3d_pad3_by_axis_celoss_no_mse_evalonce'}]
-ce_avg: 0.04674335569143295, mse_avg: 0.0
-base_dir is /dev/shm/models/checkpoints_vlm_gym_mental_rotation_3d_pad3_by_axis_one_image_lr2e_5_ce_no_mse_ins/eval_used_rows, step_tag is checkpoints_vlm_gym_mental_rotation_3d_pad3_by_axis_one_image_lr2e_5_ce_no_mse_ins_step2000
-Preparing Dataset vlm_gym_mental_rotation_3d_pad3_by_axis_celoss_no_mse_evalonce/vlm_gym_mental_rotation_3d_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_pad3_by_axis_celoss_no_mse_evalonce'}]
-  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_3d_pad3_by_axis_celoss_no_mse_evalonce'}]
-  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_3d_pad3_by_axis_celoss_no_mse_evalonce'}]
-ce_avg: 0.04835177958011627, mse_avg: 0.0
-base_dir is /dev/shm/models/checkpoints_vlm_gym_mental_rotation_3d_pad3_by_axis_one_image_lr2e_5_ce_no_mse_ins/eval_used_rows, step_tag is checkpoints_vlm_gym_mental_rotation_3d_pad3_by_axis_one_image_lr2e_5_ce_no_mse_ins_step2500
-Preparing Dataset vlm_gym_mental_rotation_3d_pad3_by_axis_celoss_no_mse_evalonce/vlm_gym_mental_rotation_3d_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_pad3_by_axis_celoss_no_mse_evalonce'}]
-  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_3d_pad3_by_axis_celoss_no_mse_evalonce'}]
-  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_3d_pad3_by_axis_celoss_no_mse_evalonce'}]
-ce_avg: 0.05139823630452156, mse_avg: 0.0
 [2026-01-29 22:57:57] (step=0001155) Train Loss mse: 0.0000, Train Loss ce: 0.0432, Train Steps/Sec: 0.46,
 [2026-01-29 22:57:59] (step=0001156) Train Loss mse: 0.0000, Train Loss ce: 0.0424, Train Steps/Sec: 0.52,
 [2026-01-29 22:58:01] (step=0001157) Train Loss mse: 0.0000, Train Loss ce: 0.0411, Train Steps/Sec: 0.51,
@@ -2766,20 +2766,6 @@ ce_avg: 0.05139823630452156, mse_avg: 0.0
 [2026-01-29 23:48:06] (step=0002564) Train Loss mse: 0.0000, Train Loss ce: 0.0328, Train Steps/Sec: 0.51,
 [2026-01-29 23:48:09] (step=0002565) Train Loss mse: 0.0000, Train Loss ce: 0.0326, Train Steps/Sec: 0.40,
 [2026-01-29 23:48:11] (step=0002566) Train Loss mse: 0.0000, Train Loss ce: 0.0364, Train Steps/Sec: 0.47,
-[2026-01-29 23:48:13
-base_dir is /dev/shm/models/checkpoints_vlm_gym_mental_rotation_3d_pad3_by_axis_one_image_lr2e_5_ce_no_mse_ins/eval_used_rows, step_tag is checkpoints_vlm_gym_mental_rotation_3d_pad3_by_axis_one_image_lr2e_5_ce_no_mse_ins_step3000
-Preparing Dataset vlm_gym_mental_rotation_3d_pad3_by_axis_celoss_no_mse_evalonce/vlm_gym_mental_rotation_3d_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_pad3_by_axis_celoss_no_mse_evalonce'}]
-  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_3d_pad3_by_axis_celoss_no_mse_evalonce'}]
-  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_3d_pad3_by_axis_celoss_no_mse_evalonce'}]
-ce_avg: 0.054898910224437714, mse_avg: 0.0
-base_dir is /dev/shm/models/checkpoints_vlm_gym_mental_rotation_3d_pad3_by_axis_one_image_lr2e_5_ce_no_mse_ins/eval_used_rows, step_tag is checkpoints_vlm_gym_mental_rotation_3d_pad3_by_axis_one_image_lr2e_5_ce_no_mse_ins_step3500
-Preparing Dataset vlm_gym_mental_rotation_3d_pad3_by_axis_celoss_no_mse_evalonce/vlm_gym_mental_rotation_3d_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_pad3_by_axis_celoss_no_mse_evalonce'}]
-  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_3d_pad3_by_axis_celoss_no_mse_evalonce'}]
-  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_3d_pad3_by_axis_celoss_no_mse_evalonce'}]
 [2026-01-29 23:48:13] (step=0002567) Train Loss mse: 0.0000, Train Loss ce: 0.0333, Train Steps/Sec: 0.46,
 [2026-01-29 23:48:15] (step=0002568) Train Loss mse: 0.0000, Train Loss ce: 0.0331, Train Steps/Sec: 0.51,
 [2026-01-29 23:48:17] (step=0002569) Train Loss mse: 0.0000, Train Loss ce: 0.0392, Train Steps/Sec: 0.51,
@@ -2872,6 +2858,27 @@ Preparing Dataset vlm_gym_mental_rotation_3d_pad3_by_axis_celoss_no_mse_evalonce
 [2026-01-29 23:51:23] (step=0002656) Train Loss mse: 0.0000, Train Loss ce: 0.0317, Train Steps/Sec: 0.51,
 [2026-01-29 23:51:25] (step=0002657) Train Loss mse: 0.0000, Train Loss ce: 0.0354, Train Steps/Sec: 0.47,
 [2026-01-29 23:51:27] (step=0002658) Train Loss mse: 0.0000, Train Loss ce: 0.0380, Train Steps/Sec: 0.51,
+base_dir is /dev/shm/models/checkpoints_vlm_gym_mental_rotation_3d_pad3_by_axis_one_image_lr2e_5_ce_no_mse_ins/eval_used_rows, step_tag is checkpoints_vlm_gym_mental_rotation_3d_pad3_by_axis_one_image_lr2e_5_ce_no_mse_ins_step2500
+Preparing Dataset vlm_gym_mental_rotation_3d_pad3_by_axis_celoss_no_mse_evalonce/vlm_gym_mental_rotation_3d_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_pad3_by_axis_celoss_no_mse_evalonce'}]
+  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_3d_pad3_by_axis_celoss_no_mse_evalonce'}]
+  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_3d_pad3_by_axis_celoss_no_mse_evalonce'}]
+ce_avg: 0.05139823630452156, mse_avg: 0.0
+base_dir is /dev/shm/models/checkpoints_vlm_gym_mental_rotation_3d_pad3_by_axis_one_image_lr2e_5_ce_no_mse_ins/eval_used_rows, step_tag is checkpoints_vlm_gym_mental_rotation_3d_pad3_by_axis_one_image_lr2e_5_ce_no_mse_ins_step3000
+Preparing Dataset vlm_gym_mental_rotation_3d_pad3_by_axis_celoss_no_mse_evalonce/vlm_gym_mental_rotation_3d_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_pad3_by_axis_celoss_no_mse_evalonce'}]
+  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_3d_pad3_by_axis_celoss_no_mse_evalonce'}]
+  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_3d_pad3_by_axis_celoss_no_mse_evalonce'}]
+ce_avg: 0.054898910224437714, mse_avg: 0.0
+base_dir is /dev/shm/models/checkpoints_vlm_gym_mental_rotation_3d_pad3_by_axis_one_image_lr2e_5_ce_no_mse_ins/eval_used_rows, step_tag is checkpoints_vlm_gym_mental_rotation_3d_pad3_by_axis_one_image_lr2e_5_ce_no_mse_ins_step3500
+Preparing Dataset vlm_gym_mental_rotation_3d_pad3_by_axis_celoss_no_mse_evalonce/vlm_gym_mental_rotation_3d_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_pad3_by_axis_celoss_no_mse_evalonce'}]
+  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_3d_pad3_by_axis_celoss_no_mse_evalonce'}]
+  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_3d_pad3_by_axis_celoss_no_mse_evalonce'}]
+ce_avg: 0.05028936266899109, mse_avg: 0.0
 [2026-01-29 23:51:29] (step=0002659) Train Loss mse: 0.0000, Train Loss ce: 0.0363, Train Steps/Sec: 0.46,
 [2026-01-29 23:51:31] (step=0002660) Train Loss mse: 0.0000, Train Loss ce: 0.0347, Train Steps/Sec: 0.46,
 [2026-01-29 23:51:33] (step=0002661) Train Loss mse: 0.0000, Train Loss ce: 0.0356, Train Steps/Sec: 0.51,
@@ -3900,6 +3907,27 @@ Preparing Dataset vlm_gym_mental_rotation_3d_pad3_by_axis_celoss_no_mse_evalonce
 [2026-01-30 00:27:52] (step=0003684) Train Loss mse: 0.0000, Train Loss ce: 0.0313, Train Steps/Sec: 0.52,
 [2026-01-30 00:27:54] (step=0003685) Train Loss mse: 0.0000, Train Loss ce: 0.0327, Train Steps/Sec: 0.52,
 [2026-01-30 00:27:56] (step=0003686) Train Loss mse: 0.0000, Train Loss ce: 0.0292, Train Steps/Sec: 0.45,
+base_dir is /dev/shm/models/checkpoints_vlm_gym_mental_rotation_3d_pad3_by_axis_one_image_lr2e_5_ce_no_mse_ins/eval_used_rows, step_tag is checkpoints_vlm_gym_mental_rotation_3d_pad3_by_axis_one_image_lr2e_5_ce_no_mse_ins_step4000
+Preparing Dataset vlm_gym_mental_rotation_3d_pad3_by_axis_celoss_no_mse_evalonce/vlm_gym_mental_rotation_3d_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_pad3_by_axis_celoss_no_mse_evalonce'}]
+  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_3d_pad3_by_axis_celoss_no_mse_evalonce'}]
+  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_3d_pad3_by_axis_celoss_no_mse_evalonce'}]
+ce_avg: 0.04602212458848953, mse_avg: 0.0
+base_dir is /dev/shm/models/checkpoints_vlm_gym_mental_rotation_3d_pad3_by_axis_one_image_lr2e_5_ce_no_mse_ins/eval_used_rows, step_tag is checkpoints_vlm_gym_mental_rotation_3d_pad3_by_axis_one_image_lr2e_5_ce_no_mse_ins_step4500
+Preparing Dataset vlm_gym_mental_rotation_3d_pad3_by_axis_celoss_no_mse_evalonce/vlm_gym_mental_rotation_3d_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_pad3_by_axis_celoss_no_mse_evalonce'}]
+  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_3d_pad3_by_axis_celoss_no_mse_evalonce'}]
+  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_3d_pad3_by_axis_celoss_no_mse_evalonce'}]
+ce_avg: 0.04616258293390274, mse_avg: 0.0
+base_dir is /dev/shm/models/checkpoints_vlm_gym_mental_rotation_3d_pad3_by_axis_one_image_lr2e_5_ce_no_mse_ins/eval_used_rows, step_tag is checkpoints_vlm_gym_mental_rotation_3d_pad3_by_axis_one_image_lr2e_5_ce_no_mse_ins_step5000
+Preparing Dataset vlm_gym_mental_rotation_3d_pad3_by_axis_celoss_no_mse_evalonce/vlm_gym_mental_rotation_3d_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_pad3_by_axis_celoss_no_mse_evalonce'}]
+  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_3d_pad3_by_axis_celoss_no_mse_evalonce'}]
+  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_3d_pad3_by_axis_celoss_no_mse_evalonce'}]
+ce_avg: 0.04619377478957176, mse_avg: 0.0
 [2026-01-30 00:27:58] (step=0003687) Train Loss mse: 0.0000, Train Loss ce: 0.0334, Train Steps/Sec: 0.47,
 [2026-01-30 00:28:01] (step=0003688) Train Loss mse: 0.0000, Train Loss ce: 0.0278, Train Steps/Sec: 0.46,
 [2026-01-30 00:28:03] (step=0003689) Train Loss mse: 0.0000, Train Loss ce: 0.0315, Train Steps/Sec: 0.47,
@@ -4092,20 +4120,105 @@ Preparing Dataset vlm_gym_mental_rotation_3d_pad3_by_axis_celoss_no_mse_evalonce
 [2026-01-30 00:34:40] (step=0003876) Train Loss mse: 0.0000, Train Loss ce: 0.0315, Train Steps/Sec: 0.52,
 [2026-01-30 00:34:42] (step=0003877) Train Loss mse: 0.0000, Train Loss ce: 0.0304, Train Steps/Sec: 0.43,
 [2026-01-30 00:34:44] (step=0003878) Train Loss mse: 0.0000, Train Loss ce: 0.0308, Train Steps/Sec: 0.51,
-[2026-01-29 23:48:13] (step=0002567) Train Loss mse: 0.0000, Train Loss ce: 0.0333, Train Steps/Sec: 0.46,
-base_dir is /dev/shm/models/checkpoints_vlm_gym_mental_rotation_3d_pad3_by_axis_one_image_lr2e_5_ce_no_mse_ins/eval_used_rows, step_tag is checkpoints_vlm_gym_mental_rotation_3d_pad3_by_axis_one_image_lr2e_5_ce_no_mse_ins_step4000
-Preparing Dataset vlm_gym_mental_rotation_3d_pad3_by_axis_celoss_no_mse_evalonce/vlm_gym_mental_rotation_3d_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_pad3_by_axis_celoss_no_mse_evalonce'}]
-  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_3d_pad3_by_axis_celoss_no_mse_evalonce'}]
-  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_3d_pad3_by_axis_celoss_no_mse_evalonce'}]
-ce_avg: 0.04602212458848953, mse_avg: 0.0
-base_dir is /dev/shm/models/checkpoints_vlm_gym_mental_rotation_3d_pad3_by_axis_one_image_lr2e_5_ce_no_mse_ins/eval_used_rows, step_tag is checkpoints_vlm_gym_mental_rotation_3d_pad3_by_axis_one_image_lr2e_5_ce_no_mse_ins_step4500
-Preparing Dataset vlm_gym_mental_rotation_3d_pad3_by_axis_celoss_no_mse_evalonce/vlm_gym_mental_rotation_3d_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_pad3_by_axis_celoss_no_mse_evalonce'}]
-  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_3d_pad3_by_axis_celoss_no_mse_evalonce'}]
-  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_3d_pad3_by_axis_celoss_no_mse_evalonce'}]
+[2026-01-30 00:34:46] (step=0003879) Train Loss mse: 0.0000, Train Loss ce: 0.0313, Train Steps/Sec: 0.47,
+[2026-01-30 00:34:48] (step=0003880) Train Loss mse: 0.0000, Train Loss ce: 0.0305, Train Steps/Sec: 0.47,
+[2026-01-30 00:34:50] (step=0003881) Train Loss mse: 0.0000, Train Loss ce: 0.0298, Train Steps/Sec: 0.47,
+[2026-01-30 00:34:52] (step=0003882) Train Loss mse: 0.0000, Train Loss ce: 0.0307, Train Steps/Sec: 0.46,
+[2026-01-30 00:34:54] (step=0003883) Train Loss mse: 0.0000, Train Loss ce: 0.0307, Train Steps/Sec: 0.52,
+[2026-01-30 00:34:57] (step=0003884) Train Loss mse: 0.0000, Train Loss ce: 0.0319, Train Steps/Sec: 0.42,
+[2026-01-30 00:34:59] (step=0003885) Train Loss mse: 0.0000, Train Loss ce: 0.0299, Train Steps/Sec: 0.47,
+[2026-01-30 00:35:01] (step=0003886) Train Loss mse: 0.0000, Train Loss ce: 0.0336, Train Steps/Sec: 0.47,
+[2026-01-30 00:35:03] (step=0003887) Train Loss mse: 0.0000, Train Loss ce: 0.0345, Train Steps/Sec: 0.52,
+[2026-01-30 00:35:05] (step=0003888) Train Loss mse: 0.0000, Train Loss ce: 0.0271, Train Steps/Sec: 0.47,
+[2026-01-30 00:35:07] (step=0003889) Train Loss mse: 0.0000, Train Loss ce: 0.0273, Train Steps/Sec: 0.46,
+[2026-01-30 00:35:09] (step=0003890) Train Loss mse: 0.0000, Train Loss ce: 0.0301, Train Steps/Sec: 0.46,
+[2026-01-30 00:35:11] (step=0003891) Train Loss mse: 0.0000, Train Loss ce: 0.0294, Train Steps/Sec: 0.52,
+[2026-01-30 00:35:14] (step=0003892) Train Loss mse: 0.0000, Train Loss ce: 0.0313, Train Steps/Sec: 0.46,
+[2026-01-30 00:35:16] (step=0003893) Train Loss mse: 0.0000, Train Loss ce: 0.0308, Train Steps/Sec: 0.47,
+[2026-01-30 00:35:18] (step=0003894) Train Loss mse: 0.0000, Train Loss ce: 0.0290, Train Steps/Sec: 0.47,
+[2026-01-30 00:35:20] (step=0003895) Train Loss mse: 0.0000, Train Loss ce: 0.0299, Train Steps/Sec: 0.47,
+[2026-01-30 00:35:22] (step=0003896) Train Loss mse: 0.0000, Train Loss ce: 0.0287, Train Steps/Sec: 0.47,
+[2026-01-30 00:35:24] (step=0003897) Train Loss mse: 0.0000, Train Loss ce: 0.0318, Train Steps/Sec: 0.52,
+[2026-01-30 00:35:26] (step=0003898) Train Loss mse: 0.0000, Train Loss ce: 0.0280, Train Steps/Sec: 0.47,
+[2026-01-30 00:35:28] (step=0003899) Train Loss mse: 0.0000, Train Loss ce: 0.0302, Train Steps/Sec: 0.52,
+[2026-01-30 00:35:31] (step=0003900) Train Loss mse: 0.0000, Train Loss ce: 0.0312, Train Steps/Sec: 0.42,
+[2026-01-30 00:35:32] (step=0003901) Train Loss mse: 0.0000, Train Loss ce: 0.0344, Train Steps/Sec: 0.51,
+[2026-01-30 00:35:35] (step=0003902) Train Loss mse: 0.0000, Train Loss ce: 0.0311, Train Steps/Sec: 0.42,
+[2026-01-30 00:35:37] (step=0003903) Train Loss mse: 0.0000, Train Loss ce: 0.0297, Train Steps/Sec: 0.43,
+[2026-01-30 00:35:39] (step=0003904) Train Loss mse: 0.0000, Train Loss ce: 0.0293, Train Steps/Sec: 0.52,
+[2026-01-30 00:35:41] (step=0003905) Train Loss mse: 0.0000, Train Loss ce: 0.0291, Train Steps/Sec: 0.52,
+[2026-01-30 00:35:43] (step=0003906) Train Loss mse: 0.0000, Train Loss ce: 0.0315, Train Steps/Sec: 0.46,
+[2026-01-30 00:35:45] (step=0003907) Train Loss mse: 0.0000, Train Loss ce: 0.0293, Train Steps/Sec: 0.47,
+[2026-01-30 00:35:48] (step=0003908) Train Loss mse: 0.0000, Train Loss ce: 0.0331, Train Steps/Sec: 0.45,
+[2026-01-30 00:35:50] (step=0003909) Train Loss mse: 0.0000, Train Loss ce: 0.0302, Train Steps/Sec: 0.46,
+[2026-01-30 00:35:52] (step=0003910) Train Loss mse: 0.0000, Train Loss ce: 0.0321, Train Steps/Sec: 0.43,
+[2026-01-30 00:35:54] (step=0003911) Train Loss mse: 0.0000, Train Loss ce: 0.0285, Train Steps/Sec: 0.51,
+[2026-01-30 00:35:56] (step=0003912) Train Loss mse: 0.0000, Train Loss ce: 0.0301, Train Steps/Sec: 0.51,
+[2026-01-30 00:35:58] (step=0003913) Train Loss mse: 0.0000, Train Loss ce: 0.0304, Train Steps/Sec: 0.46,
+[2026-01-30 00:36:00] (step=0003914) Train Loss mse: 0.0000, Train Loss ce: 0.0280, Train Steps/Sec: 0.52,
+[2026-01-30 00:36:02] (step=0003915) Train Loss mse: 0.0000, Train Loss ce: 0.0283, Train Steps/Sec: 0.46,
+[2026-01-30 00:36:04] (step=0003916) Train Loss mse: 0.0000, Train Loss ce: 0.0293, Train Steps/Sec: 0.47,
+[2026-01-30 00:36:07] (step=0003917) Train Loss mse: 0.0000, Train Loss ce: 0.0319, Train Steps/Sec: 0.43,
+[2026-01-30 00:36:09] (step=0003918) Train Loss mse: 0.0000, Train Loss ce: 0.0295, Train Steps/Sec: 0.46,
+[2026-01-30 00:36:11] (step=0003919) Train Loss mse: 0.0000, Train Loss ce: 0.0283, Train Steps/Sec: 0.47,
+[2026-01-30 00:36:13] (step=0003920) Train Loss mse: 0.0000, Train Loss ce: 0.0292, Train Steps/Sec: 0.52,
+[2026-01-30 00:36:15] (step=0003921) Train Loss mse: 0.0000, Train Loss ce: 0.0293, Train Steps/Sec: 0.51,
+[2026-01-30 00:36:17] (step=0003922) Train Loss mse: 0.0000, Train Loss ce: 0.0302, Train Steps/Sec: 0.42,
+[2026-01-30 00:36:20] (step=0003923) Train Loss mse: 0.0000, Train Loss ce: 0.0309, Train Steps/Sec: 0.45,
+[2026-01-30 00:36:22] (step=0003924) Train Loss mse: 0.0000, Train Loss ce: 0.0296, Train Steps/Sec: 0.46,
+[2026-01-30 00:36:24] (step=0003925) Train Loss mse: 0.0000, Train Loss ce: 0.0289, Train Steps/Sec: 0.42,
+[2026-01-30 00:36:26] (step=0003926) Train Loss mse: 0.0000, Train Loss ce: 0.0292, Train Steps/Sec: 0.52,
+[2026-01-30 00:36:28] (step=0003927) Train Loss mse: 0.0000, Train Loss ce: 0.0281, Train Steps/Sec: 0.47,
+[2026-01-30 00:36:30] (step=0003928) Train Loss mse: 0.0000, Train Loss ce: 0.0272, Train Steps/Sec: 0.47,
+[2026-01-30 00:36:32] (step=0003929) Train Loss mse: 0.0000, Train Loss ce: 0.0330, Train Steps/Sec: 0.51,
+[2026-01-30 00:36:34] (step=0003930) Train Loss mse: 0.0000, Train Loss ce: 0.0296, Train Steps/Sec: 0.47,
+[2026-01-30 00:36:36] (step=0003931) Train Loss mse: 0.0000, Train Loss ce: 0.0323, Train Steps/Sec: 0.51,
+[2026-01-30 00:36:39] (step=0003932) Train Loss mse: 0.0000, Train Loss ce: 0.0301, Train Steps/Sec: 0.41,
+[2026-01-30 00:36:41] (step=0003933) Train Loss mse: 0.0000, Train Loss ce: 0.0316, Train Steps/Sec: 0.47,
+[2026-01-30 00:36:43] (step=0003934) Train Loss mse: 0.0000, Train Loss ce: 0.0287, Train Steps/Sec: 0.47,
+[2026-01-30 00:36:45] (step=0003935) Train Loss mse: 0.0000, Train Loss ce: 0.0328, Train Steps/Sec: 0.51,
+[2026-01-30 00:36:47] (step=0003936) Train Loss mse: 0.0000, Train Loss ce: 0.0307, Train Steps/Sec: 0.46,
+[2026-01-30 00:36:49] (step=0003937) Train Loss mse: 0.0000, Train Loss ce: 0.0336, Train Steps/Sec: 0.47,
+[2026-01-30 00:36:51] (step=0003938) Train Loss mse: 0.0000, Train Loss ce: 0.0322, Train Steps/Sec: 0.47,
+[2026-01-30 00:36:54] (step=0003939) Train Loss mse: 0.0000, Train Loss ce: 0.0276, Train Steps/Sec: 0.46,
+[2026-01-30 00:36:56] (step=0003940) Train Loss mse: 0.0000, Train Loss ce: 0.0299, Train Steps/Sec: 0.51,
+[2026-01-30 00:36:58] (step=0003941) Train Loss mse: 0.0000, Train Loss ce: 0.0283, Train Steps/Sec: 0.42,
+[2026-01-30 00:37:00] (step=0003942) Train Loss mse: 0.0000, Train Loss ce: 0.0321, Train Steps/Sec: 0.52,
+[2026-01-30 00:37:02] (step=0003943) Train Loss mse: 0.0000, Train Loss ce: 0.0288, Train Steps/Sec: 0.47,
+[2026-01-30 00:37:04] (step=0003944) Train Loss mse: 0.0000, Train Loss ce: 0.0342, Train Steps/Sec: 0.47,
+[2026-01-30 00:37:06] (step=0003945) Train Loss mse: 0.0000, Train Loss ce: 0.0294, Train Steps/Sec: 0.51,
+[2026-01-30 00:37:09] (step=0003946) Train Loss mse: 0.0000, Train Loss ce: 0.0295, Train Steps/Sec: 0.39,
+[2026-01-30 00:37:11] (step=0003947) Train Loss mse: 0.0000, Train Loss ce: 0.0301, Train Steps/Sec: 0.52,
+[2026-01-30 00:37:13] (step=0003948) Train Loss mse: 0.0000, Train Loss ce: 0.0314, Train Steps/Sec: 0.52,
+[2026-01-30 00:37:15] (step=0003949) Train Loss mse: 0.0000, Train Loss ce: 0.0268, Train Steps/Sec: 0.46,
+[2026-01-30 00:37:17] (step=0003950) Train Loss mse: 0.0000, Train Loss ce: 0.0295, Train Steps/Sec: 0.45,
+[2026-01-30 00:37:19] (step=0003951) Train Loss mse: 0.0000, Train Loss ce: 0.0281, Train Steps/Sec: 0.51,
+[2026-01-30 00:37:21] (step=0003952) Train Loss mse: 0.0000, Train Loss ce: 0.0322, Train Steps/Sec: 0.46,
+[2026-01-30 00:37:23] (step=0003953) Train Loss mse: 0.0000, Train Loss ce: 0.0315, Train Steps/Sec: 0.46,
+[2026-01-30 00:37:26] (step=0003954) Train Loss mse: 0.0000, Train Loss ce: 0.0323, Train Steps/Sec: 0.43,
+[2026-01-30 00:37:28] (step=0003955) Train Loss mse: 0.0000, Train Loss ce: 0.0320, Train Steps/Sec: 0.51,
+[2026-01-30 00:37:29] (step=0003956) Train Loss mse: 0.0000, Train Loss ce: 0.0308, Train Steps/Sec: 0.51,
+[2026-01-30 00:37:32] (step=0003957) Train Loss mse: 0.0000, Train Loss ce: 0.0284, Train Steps/Sec: 0.47,
+[2026-01-30 00:37:34] (step=0003958) Train Loss mse: 0.0000, Train Loss ce: 0.0317, Train Steps/Sec: 0.41,
+[2026-01-30 00:37:36] (step=0003959) Train Loss mse: 0.0000, Train Loss ce: 0.0298, Train Steps/Sec: 0.51,
+[2026-01-30 00:37:38] (step=0003960) Train Loss mse: 0.0000, Train Loss ce: 0.0315, Train Steps/Sec: 0.42,
+[2026-01-30 00:37:41] (step=0003961) Train Loss mse: 0.0000, Train Loss ce: 0.0302, Train Steps/Sec: 0.43,
+[2026-01-30 00:37:43] (step=0003962) Train Loss mse: 0.0000, Train Loss ce: 0.0295, Train Steps/Sec: 0.47,
+[2026-01-30 00:37:45] (step=0003963) Train Loss mse: 0.0000, Train Loss ce: 0.0318, Train Steps/Sec: 0.52,
+[2026-01-30 00:37:47] (step=0003964) Train Loss mse: 0.0000, Train Loss ce: 0.0300, Train Steps/Sec: 0.47,
+[2026-01-30 00:37:49] (step=0003965) Train Loss mse: 0.0000, Train Loss ce: 0.0346, Train Steps/Sec: 0.51,
+[2026-01-30 00:37:51] (step=0003966) Train Loss mse: 0.0000, Train Loss ce: 0.0300, Train Steps/Sec: 0.41,
+[2026-01-30 00:37:54] (step=0003967) Train Loss mse: 0.0000, Train Loss ce: 0.0305, Train Steps/Sec: 0.42,
+[2026-01-30 00:37:56] (step=0003968) Train Loss mse: 0.0000, Train Loss ce: 0.0291, Train Steps/Sec: 0.51,
+[2026-01-30 00:37:58] (step=0003969) Train Loss mse: 0.0000, Train Loss ce: 0.0325, Train Steps/Sec: 0.43,
+[2026-01-30 00:38:00] (step=0003970) Train Loss mse: 0.0000, Train Loss ce: 0.0314, Train Steps/Sec: 0.47,
+[2026-01-30 00:38:02] (step=0003971) Train Loss mse: 0.0000, Train Loss ce: 0.0304, Train Steps/Sec: 0.51,
+[2026-01-30 00:38:04] (step=0003972) Train Loss mse: 0.0000, Train Loss ce: 0.0286, Train Steps/Sec: 0.46,
+[2026-01-30 00:38:06] (step=0003973) Train Loss mse: 0.0000, Train Loss ce: 0.0294, Train Steps/Sec: 0.46,
+[2026-01-30 00:38:09] (step=0003974) Train Loss mse: 0.0000, Train Loss ce: 0.0310, Train Steps/Sec: 0.46,
+[2026-01-30 00:38:11] (step=0003975) Train Loss mse: 0.0000, Train Loss ce: 0.0335, Train Steps/Sec: 0.42,
+[2026-01-30 00:38:13] (step=0003976) Train Loss mse: 0.0000, Train Loss ce: 0.0308, Train Steps/Sec: 0.47,
+[2026-01-30 00:38:15] (step=0003977) Train Loss mse: 0.0000, Train Loss ce: 0.0291, Train Steps/Sec: 0.47,
 [2026-01-30 00:38:17] (step=0003978) Train Loss mse: 0.0000, Train Loss ce: 0.0309, Train Steps/Sec: 0.47,
 [2026-01-30 00:38:20] (step=0003979) Train Loss mse: 0.0000, Train Loss ce: 0.0288, Train Steps/Sec: 0.46,
 [2026-01-30 00:38:22] (step=0003980) Train Loss mse: 0.0000, Train Loss ce: 0.0315, Train Steps/Sec: 0.52,
@@ -5132,12 +5245,4 @@ Preparing Dataset vlm_gym_mental_rotation_3d_pad3_by_axis_celoss_no_mse_evalonce
 [2026-01-30 01:14:45] Saving checkpoint to /dev/shm/models/checkpoints_vlm_gym_mental_rotation_3d_pad3_by_axis_one_image_lr2e_5_ce_no_mse_ins/0005000.
 /opt/conda/lib/python3.11/site-packages/torch/distributed/fsdp/fully_sharded_data_parallel.py:690: FutureWarning: FSDP.state_dict_type() and FSDP.set_state_dict_type() are being deprecated. Please use APIs, get_state_dict() and set_state_dict(), which can support different parallelisms, FSDP1, FSDP2, DDP. API doc: https://pytorch.org/docs/stable/distributed.checkpoint.html#torch.distributed.checkpoint.state_dict.get_state_dict .Tutorial: https://pytorch.org/tutorials/recipes/distributed_checkpoint_recipe.html .
   warnings.warn(
-[2026-01-30 01:17:18] Done!
-[2026-01-30 00:38:17] (step=0003978) Train Loss mse: 0.0000, Train Loss ce: 0.0309, Train Steps/Sec: 0.47,
-base_dir is /dev/shm/models/checkpoints_vlm_gym_mental_rotation_3d_pad3_by_axis_one_image_lr2e_5_ce_no_mse_ins/eval_used_rows, step_tag is checkpoints_vlm_gym_mental_rotation_3d_pad3_by_axis_one_image_lr2e_5_ce_no_mse_ins_step5000
-Preparing Dataset vlm_gym_mental_rotation_3d_pad3_by_axis_celoss_no_mse_evalonce/vlm_gym_mental_rotation_3d_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_pad3_by_axis_celoss_no_mse_evalonce'}]
-  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_3d_pad3_by_axis_celoss_no_mse_evalonce'}]
-  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_3d_pad3_by_axis_celoss_no_mse_evalonce'}]
-ce_avg: 0.04619377478957176, mse_avg: 0.0
+[2026-01-30 01:17:18] Done!