Upload checkpoints_vlm_gym_mental_rotation_2d_one_image_lr2e_5_mse_only_ins/checkpoints_vlm_gym_mental_rotation_2d_one_image_lr2e_5_mse_only_ins

checkpoints_vlm_gym_mental_rotation_2d_one_image_lr2e_5_mse_only_ins/checkpoints_vlm_gym_mental_rotation_2d_one_image_lr2e_5_mse_only_ins/wandb/offline-run-20260129_220034-vlm_gym_mental_rotation_2d_one_img_lr2e_5_mse_only_ins-run0/files/output.log

@@ -1,189 +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)
-    )
-    (time_embedder): FullyShardedDataParallel(
-      (_fsdp_wrapped_module): TimestepEmbedder(
-        (mlp): Sequential(
-          (0): Linear(in_features=256, out_features=3584, bias=True)
-          (1): SiLU()
-          (2): Linear(in_features=3584, out_features=3584, bias=True)
-        )
-      )
-    )
-    (vae2llm): Linear(in_features=64, out_features=3584, bias=True)
-    (llm2vae): Linear(in_features=3584, out_features=64, bias=True)
-    (latent_pos_embed): FullyShardedDataParallel(
-      (_fsdp_wrapped_module): PositionEmbedding()
-    )
-    (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
-time_embedder._fsdp_wrapped_module._flat_param True
-latent_pos_embed._fsdp_wrapped_module._flat_param False
-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_2d_mse_loss_only/vlm_gym_mental_rotation_2d_train
-base_dir is /dev/shm/models/checkpoints_vlm_gym_mental_rotation_2d_one_image_lr2e_5_mse_only_ins/eval_used_rows, step_tag is vlm_gym_mental_rotation_2d_one_img_lr2e_5_mse_only_ins_step0
-Preparing Dataset vlm_gym_mental_rotation_2d_mse_loss_only_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_mse_loss_only_evalonce'}]
-  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_2d_mse_loss_only_evalonce'}]
-  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_2d_mse_loss_only_evalonce'}]
-ce_avg: 0.0, mse_avg: 0.33117932081222534
-base_dir is /dev/shm/models/checkpoints_vlm_gym_mental_rotation_2d_one_image_lr2e_5_mse_only_ins/eval_used_rows, step_tag is vlm_gym_mental_rotation_2d_one_img_lr2e_5_mse_only_ins_step500
-Preparing Dataset vlm_gym_mental_rotation_2d_mse_loss_only_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_mse_loss_only_evalonce'}]
-  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_2d_mse_loss_only_evalonce'}]
-  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_2d_mse_loss_only_evalonce'}]
-ce_avg: 0.0, mse_avg: 0.09970466047525406
 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/
@@ -1182,20 +996,192 @@ wandb: For more information, check out the docs at: https://weave-docs.wandb.ai/
 [2026-01-30 00:05:52] (step=0000985) Train Loss mse: 0.0868, Train Loss ce: 0.0000, Train Steps/Sec: 0.16,
 [2026-01-30 00:05:59] (step=0000986) Train Loss mse: 0.0982, Train Loss ce: 0.0000, Train Steps/Sec: 0.13,
 [2026-01-30 00:06:06] (step=0000987) Train Loss mse: 0.1057, Train Loss ce: 0.0000, Train Steps/Sec: 0.14,
-[2026-01-30 00:06:14] (step=0000988) Train Loss mse: 0.0923, Train Loss ce: 0.0000, Train Steps/Sec: 0.13,
-[2026-01-30 00:06:21] (step=0000989) Train Loss mse: 0.0922, Train Loss ce: 0.0000, Train Steps/Sec: 0.13,
-[2026-01-30 00:06:29] (step=0000990) Train Loss mse: 0.0976, Train Loss ce: 0.0000, Train Steps/Sec: 0.13,
-[2026-01-30 00:06:34] (step=0000991) Train Loss mse: 0.0941, Train Loss ce: 0.0000, Train Steps/Sec: 0.18,
-[2026-01-30 00:06:40] (step=0000992) Train Loss mse: 0.0977, Train Loss ce: 0.0000, Train Steps/Sec: 0.18,
-[2026-01-30 00:06:48] (step=0000993) Train Loss mse: 0.0935, Train Loss ce: 0.0000, Train Steps/Sec: 0.13,
-[2026-01-30 00:06:54] (step=0000994) Train Loss mse: 0.0901, Train Loss ce: 0.0000, Train Steps/Sec: 0.15,
-base_dir is /dev/shm/models/checkpoints_vlm_gym_mental_rotation_2d_one_image_lr2e_5_mse_only_ins/eval_used_rows, step_tag is vlm_gym_mental_rotation_2d_one_img_lr2e_5_mse_only_ins_step1000
+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)
+    )
+    (time_embedder): FullyShardedDataParallel(
+      (_fsdp_wrapped_module): TimestepEmbedder(
+        (mlp): Sequential(
+          (0): Linear(in_features=256, out_features=3584, bias=True)
+          (1): SiLU()
+          (2): Linear(in_features=3584, out_features=3584, bias=True)
+        )
+      )
+    )
+    (vae2llm): Linear(in_features=64, out_features=3584, bias=True)
+    (llm2vae): Linear(in_features=3584, out_features=64, bias=True)
+    (latent_pos_embed): FullyShardedDataParallel(
+      (_fsdp_wrapped_module): PositionEmbedding()
+    )
+    (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
+time_embedder._fsdp_wrapped_module._flat_param True
+latent_pos_embed._fsdp_wrapped_module._flat_param False
+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_2d_mse_loss_only/vlm_gym_mental_rotation_2d_train
+base_dir is /dev/shm/models/checkpoints_vlm_gym_mental_rotation_2d_one_image_lr2e_5_mse_only_ins/eval_used_rows, step_tag is vlm_gym_mental_rotation_2d_one_img_lr2e_5_mse_only_ins_step0
+Preparing Dataset vlm_gym_mental_rotation_2d_mse_loss_only_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_mse_loss_only_evalonce'}]
+  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_2d_mse_loss_only_evalonce'}]
+  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_2d_mse_loss_only_evalonce'}]
+ce_avg: 0.0, mse_avg: 0.33117932081222534
+base_dir is /dev/shm/models/checkpoints_vlm_gym_mental_rotation_2d_one_image_lr2e_5_mse_only_ins/eval_used_rows, step_tag is vlm_gym_mental_rotation_2d_one_img_lr2e_5_mse_only_ins_step500
 Preparing Dataset vlm_gym_mental_rotation_2d_mse_loss_only_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_mse_loss_only_evalonce'}]
   fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_2d_mse_loss_only_evalonce'}]
   fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_2d_mse_loss_only_evalonce'}]
-ce_avg: 0.0, mse_avg: 0.09500830620527267
+ce_avg: 0.0, mse_avg: 0.09970466047525406
 base_dir is /dev/shm/models/checkpoints_vlm_gym_mental_rotation_2d_one_image_lr2e_5_mse_only_ins/eval_used_rows, step_tag is vlm_gym_mental_rotation_2d_one_img_lr2e_5_mse_only_ins_step1500
 Preparing Dataset vlm_gym_mental_rotation_2d_mse_loss_only_evalonce/vlm_gym_mental_rotation_2d_val
 [eval debug] first 3 batch fingerprints:
@@ -1210,6 +1196,13 @@ Preparing Dataset vlm_gym_mental_rotation_2d_mse_loss_only_evalonce/vlm_gym_ment
   fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_2d_mse_loss_only_evalonce'}]
   fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_2d_mse_loss_only_evalonce'}]
 ce_avg: 0.0, mse_avg: 0.09311425685882568
+[2026-01-30 00:06:14] (step=0000988) Train Loss mse: 0.0923, Train Loss ce: 0.0000, Train Steps/Sec: 0.13,
+[2026-01-30 00:06:21] (step=0000989) Train Loss mse: 0.0922, Train Loss ce: 0.0000, Train Steps/Sec: 0.13,
+[2026-01-30 00:06:29] (step=0000990) Train Loss mse: 0.0976, Train Loss ce: 0.0000, Train Steps/Sec: 0.13,
+[2026-01-30 00:06:34] (step=0000991) Train Loss mse: 0.0941, Train Loss ce: 0.0000, Train Steps/Sec: 0.18,
+[2026-01-30 00:06:40] (step=0000992) Train Loss mse: 0.0977, Train Loss ce: 0.0000, Train Steps/Sec: 0.18,
+[2026-01-30 00:06:48] (step=0000993) Train Loss mse: 0.0935, Train Loss ce: 0.0000, Train Steps/Sec: 0.13,
+[2026-01-30 00:06:54] (step=0000994) Train Loss mse: 0.0901, Train Loss ce: 0.0000, Train Steps/Sec: 0.15,
 [2026-01-30 00:07:02] (step=0000995) Train Loss mse: 0.1012, Train Loss ce: 0.0000, Train Steps/Sec: 0.14,
 [2026-01-30 00:07:09] (step=0000996) Train Loss mse: 0.1097, Train Loss ce: 0.0000, Train Steps/Sec: 0.13,
 [2026-01-30 00:07:16] (step=0000997) Train Loss mse: 0.1051, Train Loss ce: 0.0000, Train Steps/Sec: 0.15,
@@ -2600,20 +2593,6 @@ ce_avg: 0.0, mse_avg: 0.09311425685882568
 [2026-01-30 02:55:05] (step=0002382) Train Loss mse: 0.0867, Train Loss ce: 0.0000, Train Steps/Sec: 0.14,
 [2026-01-30 02:55:12] (step=0002383) Train Loss mse: 0.0865, Train Loss ce: 0.0000, Train Steps/Sec: 0.15,
 [2026-01-30 02:55:19] (step=0002384) Train Loss mse: 0.0934, Train Loss ce: 0.0000, Train Steps/Sec: 0.14,
-base_dir is /dev/shm/models/checkpoints_vlm_gym_mental_rotation_2d_one_image_lr2e_5_mse_only_ins/eval_used_rows, step_tag is vlm_gym_mental_rotation_2d_one_img_lr2e_5_mse_only_ins_step2500
-Preparing Dataset vlm_gym_mental_rotation_2d_mse_loss_only_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_mse_loss_only_evalonce'}]
-  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_2d_mse_loss_only_evalonce'}]
-  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_2d_mse_loss_only_evalonce'}]
-ce_avg: 0.0, mse_avg: 0.09291024506092072
-base_dir is /dev/shm/models/checkpoints_vlm_gym_mental_rotation_2d_one_image_lr2e_5_mse_only_ins/eval_used_rows, step_tag is vlm_gym_mental_rotation_2d_one_img_lr2e_5_mse_only_ins_step3000
-Preparing Dataset vlm_gym_mental_rotation_2d_mse_loss_only_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_mse_loss_only_evalonce'}]
-  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_2d_mse_loss_only_evalonce'}]
-  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_2d_mse_loss_only_evalonce'}]
-ce_avg: 0.0, mse_avg: 0.09515543282032013
 [2026-01-30 02:55:27] (step=0002385) Train Loss mse: 0.0960, Train Loss ce: 0.0000, Train Steps/Sec: 0.14,
 [2026-01-30 02:55:32] (step=0002386) Train Loss mse: 0.0849, Train Loss ce: 0.0000, Train Steps/Sec: 0.18,
 [2026-01-30 02:55:40] (step=0002387) Train Loss mse: 0.0905, Train Loss ce: 0.0000, Train Steps/Sec: 0.13,
@@ -2661,6 +2640,27 @@ ce_avg: 0.0, mse_avg: 0.09515543282032013
 [2026-01-30 03:00:42] (step=0002429) Train Loss mse: 0.0940, Train Loss ce: 0.0000, Train Steps/Sec: 0.15,
 [2026-01-30 03:00:50] (step=0002430) Train Loss mse: 0.0971, Train Loss ce: 0.0000, Train Steps/Sec: 0.13,
 [2026-01-30 03:00:57] (step=0002431) Train Loss mse: 0.0891, Train Loss ce: 0.0000, Train Steps/Sec: 0.16,
+base_dir is /dev/shm/models/checkpoints_vlm_gym_mental_rotation_2d_one_image_lr2e_5_mse_only_ins/eval_used_rows, step_tag is vlm_gym_mental_rotation_2d_one_img_lr2e_5_mse_only_ins_step2500
+Preparing Dataset vlm_gym_mental_rotation_2d_mse_loss_only_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_mse_loss_only_evalonce'}]
+  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_2d_mse_loss_only_evalonce'}]
+  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_2d_mse_loss_only_evalonce'}]
+ce_avg: 0.0, mse_avg: 0.09291024506092072
+base_dir is /dev/shm/models/checkpoints_vlm_gym_mental_rotation_2d_one_image_lr2e_5_mse_only_ins/eval_used_rows, step_tag is vlm_gym_mental_rotation_2d_one_img_lr2e_5_mse_only_ins_step3000
+Preparing Dataset vlm_gym_mental_rotation_2d_mse_loss_only_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_mse_loss_only_evalonce'}]
+  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_2d_mse_loss_only_evalonce'}]
+  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_2d_mse_loss_only_evalonce'}]
+ce_avg: 0.0, mse_avg: 0.09515543282032013
+base_dir is /dev/shm/models/checkpoints_vlm_gym_mental_rotation_2d_one_image_lr2e_5_mse_only_ins/eval_used_rows, step_tag is vlm_gym_mental_rotation_2d_one_img_lr2e_5_mse_only_ins_step3500
+Preparing Dataset vlm_gym_mental_rotation_2d_mse_loss_only_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_mse_loss_only_evalonce'}]
+  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_2d_mse_loss_only_evalonce'}]
+  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_2d_mse_loss_only_evalonce'}]
+ce_avg: 0.0, mse_avg: 0.09188222140073776
 [2026-01-30 03:01:04] (step=0002432) Train Loss mse: 0.0877, Train Loss ce: 0.0000, Train Steps/Sec: 0.13,
 [2026-01-30 03:01:12] (step=0002433) Train Loss mse: 0.0990, Train Loss ce: 0.0000, Train Steps/Sec: 0.13,
 [2026-01-30 03:01:17] (step=0002434) Train Loss mse: 0.0945, Train Loss ce: 0.0000, Train Steps/Sec: 0.18,
@@ -3613,27 +3613,6 @@ ce_avg: 0.0, mse_avg: 0.09515543282032013
 [2026-01-30 04:56:32] (step=0003381) Train Loss mse: 0.0866, Train Loss ce: 0.0000, Train Steps/Sec: 0.13,
 [2026-01-30 04:56:39] (step=0003382) Train Loss mse: 0.0849, Train Loss ce: 0.0000, Train Steps/Sec: 0.14,
 [2026-01-30 04:56:47] (step=0003383) Train Loss mse: 0.0870, Train Loss ce: 0.0000, Train Steps/Sec: 0.13,
-base_dir is /dev/shm/models/checkpoints_vlm_gym_mental_rotation_2d_one_image_lr2e_5_mse_only_ins/eval_used_rows, step_tag is vlm_gym_mental_rotation_2d_one_img_lr2e_5_mse_only_ins_step3500
-Preparing Dataset vlm_gym_mental_rotation_2d_mse_loss_only_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_mse_loss_only_evalonce'}]
-  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_2d_mse_loss_only_evalonce'}]
-  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_2d_mse_loss_only_evalonce'}]
-ce_avg: 0.0, mse_avg: 0.09188222140073776
-base_dir is /dev/shm/models/checkpoints_vlm_gym_mental_rotation_2d_one_image_lr2e_5_mse_only_ins/eval_used_rows, step_tag is vlm_gym_mental_rotation_2d_one_img_lr2e_5_mse_only_ins_step4000
-Preparing Dataset vlm_gym_mental_rotation_2d_mse_loss_only_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_mse_loss_only_evalonce'}]
-  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_2d_mse_loss_only_evalonce'}]
-  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_2d_mse_loss_only_evalonce'}]
-ce_avg: 0.0, mse_avg: 0.09225528687238693
-base_dir is /dev/shm/models/checkpoints_vlm_gym_mental_rotation_2d_one_image_lr2e_5_mse_only_ins/eval_used_rows, step_tag is vlm_gym_mental_rotation_2d_one_img_lr2e_5_mse_only_ins_step4500
-Preparing Dataset vlm_gym_mental_rotation_2d_mse_loss_only_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_mse_loss_only_evalonce'}]
-  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_2d_mse_loss_only_evalonce'}]
-  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_2d_mse_loss_only_evalonce'}]
-ce_avg: 0.0, mse_avg: 0.0914882943034172
 [2026-01-30 04:56:54] (step=0003384) Train Loss mse: 0.0835, Train Loss ce: 0.0000, Train Steps/Sec: 0.14,
 [2026-01-30 04:57:02] (step=0003385) Train Loss mse: 0.0951, Train Loss ce: 0.0000, Train Steps/Sec: 0.13,
 [2026-01-30 04:57:09] (step=0003386) Train Loss mse: 0.0884, Train Loss ce: 0.0000, Train Steps/Sec: 0.13,
@@ -3664,6 +3643,20 @@ ce_avg: 0.0, mse_avg: 0.0914882943034172
 [2026-01-30 05:00:10] (step=0003411) Train Loss mse: 0.0851, Train Loss ce: 0.0000, Train Steps/Sec: 0.14,
 [2026-01-30 05:00:17] (step=0003412) Train Loss mse: 0.0867, Train Loss ce: 0.0000, Train Steps/Sec: 0.14,
 [2026-01-30 05:00:25] (step=0003413) Train Loss mse: 0.0948, Train Loss ce: 0.0000, Train Steps/Sec: 0.13,
+base_dir is /dev/shm/models/checkpoints_vlm_gym_mental_rotation_2d_one_image_lr2e_5_mse_only_ins/eval_used_rows, step_tag is vlm_gym_mental_rotation_2d_one_img_lr2e_5_mse_only_ins_step4000
+Preparing Dataset vlm_gym_mental_rotation_2d_mse_loss_only_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_mse_loss_only_evalonce'}]
+  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_2d_mse_loss_only_evalonce'}]
+  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_2d_mse_loss_only_evalonce'}]
+ce_avg: 0.0, mse_avg: 0.09225528687238693
+base_dir is /dev/shm/models/checkpoints_vlm_gym_mental_rotation_2d_one_image_lr2e_5_mse_only_ins/eval_used_rows, step_tag is vlm_gym_mental_rotation_2d_one_img_lr2e_5_mse_only_ins_step4500
+Preparing Dataset vlm_gym_mental_rotation_2d_mse_loss_only_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_mse_loss_only_evalonce'}]
+  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_2d_mse_loss_only_evalonce'}]
+  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_2d_mse_loss_only_evalonce'}]
+ce_avg: 0.0, mse_avg: 0.0914882943034172
 [2026-01-30 05:00:32] (step=0003414) Train Loss mse: 0.0811, Train Loss ce: 0.0000, Train Steps/Sec: 0.13,
 [2026-01-30 05:00:40] (step=0003415) Train Loss mse: 0.0855, Train Loss ce: 0.0000, Train Steps/Sec: 0.13,
 [2026-01-30 05:00:47] (step=0003416) Train Loss mse: 0.0879, Train Loss ce: 0.0000, Train Steps/Sec: 0.13,
@@ -5094,13 +5087,6 @@ ce_avg: 0.0, mse_avg: 0.0914882943034172
 [2026-01-30 07:53:22] (step=0004841) Train Loss mse: 0.0809, Train Loss ce: 0.0000, Train Steps/Sec: 0.15,
 [2026-01-30 07:53:29] (step=0004842) Train Loss mse: 0.0935, Train Loss ce: 0.0000, Train Steps/Sec: 0.14,
 [2026-01-30 07:53:37] (step=0004843) Train Loss mse: 0.0973, Train Loss ce: 0.0000, Train Steps/Sec: 0.13,
-base_dir is /dev/shm/models/checkpoints_vlm_gym_mental_rotation_2d_one_image_lr2e_5_mse_only_ins/eval_used_rows, step_tag is vlm_gym_mental_rotation_2d_one_img_lr2e_5_mse_only_ins_step5000
-Preparing Dataset vlm_gym_mental_rotation_2d_mse_loss_only_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_mse_loss_only_evalonce'}]
-  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_2d_mse_loss_only_evalonce'}]
-  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_2d_mse_loss_only_evalonce'}]
-ce_avg: 0.0, mse_avg: 0.09255406260490417
 [2026-01-30 07:53:44] (step=0004844) Train Loss mse: 0.0932, Train Loss ce: 0.0000, Train Steps/Sec: 0.13,
 [2026-01-30 07:53:51] (step=0004845) Train Loss mse: 0.0846, Train Loss ce: 0.0000, Train Steps/Sec: 0.16,
 [2026-01-30 07:53:58] (step=0004846) Train Loss mse: 0.0876, Train Loss ce: 0.0000, Train Steps/Sec: 0.13,
@@ -5126,6 +5112,13 @@ ce_avg: 0.0, mse_avg: 0.09255406260490417
 [2026-01-30 07:56:13] (step=0004866) Train Loss mse: 0.0875, Train Loss ce: 0.0000, Train Steps/Sec: 0.13,
 [2026-01-30 07:56:20] (step=0004867) Train Loss mse: 0.0830, Train Loss ce: 0.0000, Train Steps/Sec: 0.14,
 [2026-01-30 07:56:27] (step=0004868) Train Loss mse: 0.0884, Train Loss ce: 0.0000, Train Steps/Sec: 0.13,
+base_dir is /dev/shm/models/checkpoints_vlm_gym_mental_rotation_2d_one_image_lr2e_5_mse_only_ins/eval_used_rows, step_tag is vlm_gym_mental_rotation_2d_one_img_lr2e_5_mse_only_ins_step5000
+Preparing Dataset vlm_gym_mental_rotation_2d_mse_loss_only_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_mse_loss_only_evalonce'}]
+  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_2d_mse_loss_only_evalonce'}]
+  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_mental_rotation_2d_mse_loss_only_evalonce'}]
+ce_avg: 0.0, mse_avg: 0.09255406260490417
 [2026-01-30 07:56:34] (step=0004869) Train Loss mse: 0.0773, Train Loss ce: 0.0000, Train Steps/Sec: 0.15,
 [2026-01-30 07:56:42] (step=0004870) Train Loss mse: 0.0833, Train Loss ce: 0.0000, Train Steps/Sec: 0.13,
 [2026-01-30 07:56:49] (step=0004871) Train Loss mse: 0.0900, Train Loss ce: 0.0000, Train Steps/Sec: 0.14,