Upload checkpoints_vlm_gym_colorization_one_image_lr2e_5_mse_only_ins/checkpoints_vlm_gym_colorization_one_image_lr2e_5_mse_only_ins

checkpoints_vlm_gym_colorization_one_image_lr2e_5_mse_only_ins/checkpoints_vlm_gym_colorization_one_image_lr2e_5_mse_only_ins/wandb/offline-run-20260125_170309-vlm_gym_colorization_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_colorization_mse_loss_only/vlm_gym_colorization_train
-base_dir is /dev/shm/models/checkpoints_vlm_gym_colorization_one_image_lr2e_5_mse_only_ins/eval_used_rows, step_tag is vlm_gym_colorization_one_img_lr2e_5_mse_only_ins_step0
-Preparing Dataset vlm_gym_colorization_mse_loss_only_evalonce/vlm_gym_colorization_val
-[eval debug] first 3 batch fingerprints:
-  fp[0]: [{'data_indexes': [0], 'worker_id': 0, 'dataset_name': 'vlm_gym_colorization_mse_loss_only_evalonce'}]
-  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_colorization_mse_loss_only_evalonce'}]
-  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_colorization_mse_loss_only_evalonce'}]
-ce_avg: 0.0, mse_avg: 0.05326032266020775
-base_dir is /dev/shm/models/checkpoints_vlm_gym_colorization_one_image_lr2e_5_mse_only_ins/eval_used_rows, step_tag is vlm_gym_colorization_one_img_lr2e_5_mse_only_ins_step500
-Preparing Dataset vlm_gym_colorization_mse_loss_only_evalonce/vlm_gym_colorization_val
-[eval debug] first 3 batch fingerprints:
-  fp[0]: [{'data_indexes': [0], 'worker_id': 0, 'dataset_name': 'vlm_gym_colorization_mse_loss_only_evalonce'}]
-  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_colorization_mse_loss_only_evalonce'}]
-  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_colorization_mse_loss_only_evalonce'}]
-ce_avg: 0.0, mse_avg: 0.007997258566319942
 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/
@@ -920,20 +734,6 @@ wandb: For more information, check out the docs at: https://weave-docs.wandb.ai/
 [2026-01-25 21:32:29] (step=0000723) Train Loss mse: 0.0069, Train Loss ce: 0.0000, Train Steps/Sec: 0.04,
 [2026-01-25 21:32:54] (step=0000724) Train Loss mse: 0.0069, Train Loss ce: 0.0000, Train Steps/Sec: 0.04,
 [2026-01-25 21:33:17] (step=0000725) Train Loss mse: 0.0077, Train Loss ce: 0.0000, Train Steps/Sec: 0.04,
-base_dir is /dev/shm/models/checkpoints_vlm_gym_colorization_one_image_lr2e_5_mse_only_ins/eval_used_rows, step_tag is vlm_gym_colorization_one_img_lr2e_5_mse_only_ins_step1000
-Preparing Dataset vlm_gym_colorization_mse_loss_only_evalonce/vlm_gym_colorization_val
-[eval debug] first 3 batch fingerprints:
-  fp[0]: [{'data_indexes': [0], 'worker_id': 0, 'dataset_name': 'vlm_gym_colorization_mse_loss_only_evalonce'}]
-  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_colorization_mse_loss_only_evalonce'}]
-  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_colorization_mse_loss_only_evalonce'}]
-ce_avg: 0.0, mse_avg: 0.007652191445231438
-base_dir is /dev/shm/models/checkpoints_vlm_gym_colorization_one_image_lr2e_5_mse_only_ins/eval_used_rows, step_tag is vlm_gym_colorization_one_img_lr2e_5_mse_only_ins_step1500
-Preparing Dataset vlm_gym_colorization_mse_loss_only_evalonce/vlm_gym_colorization_val
-[eval debug] first 3 batch fingerprints:
-  fp[0]: [{'data_indexes': [0], 'worker_id': 0, 'dataset_name': 'vlm_gym_colorization_mse_loss_only_evalonce'}]
-  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_colorization_mse_loss_only_evalonce'}]
-  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_colorization_mse_loss_only_evalonce'}]
-ce_avg: 0.0, mse_avg: 0.00800316222012043
 [2026-01-25 21:33:38] (step=0000726) Train Loss mse: 0.0078, Train Loss ce: 0.0000, Train Steps/Sec: 0.05,
 [2026-01-25 21:33:57] (step=0000727) Train Loss mse: 0.0079, Train Loss ce: 0.0000, Train Steps/Sec: 0.06,
 [2026-01-25 21:34:18] (step=0000728) Train Loss mse: 0.0087, Train Loss ce: 0.0000, Train Steps/Sec: 0.05,
@@ -967,6 +767,192 @@ ce_avg: 0.0, mse_avg: 0.00800316222012043
 [2026-01-25 21:44:25] (step=0000756) Train Loss mse: 0.0091, Train Loss ce: 0.0000, Train Steps/Sec: 0.05,
 [2026-01-25 21:44:45] (step=0000757) Train Loss mse: 0.0084, Train Loss ce: 0.0000, Train Steps/Sec: 0.05,
 [2026-01-25 21:45:11] (step=0000758) Train Loss mse: 0.0070, Train Loss ce: 0.0000, Train Steps/Sec: 0.04,
+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_colorization_mse_loss_only/vlm_gym_colorization_train
+base_dir is /dev/shm/models/checkpoints_vlm_gym_colorization_one_image_lr2e_5_mse_only_ins/eval_used_rows, step_tag is vlm_gym_colorization_one_img_lr2e_5_mse_only_ins_step0
+Preparing Dataset vlm_gym_colorization_mse_loss_only_evalonce/vlm_gym_colorization_val
+[eval debug] first 3 batch fingerprints:
+  fp[0]: [{'data_indexes': [0], 'worker_id': 0, 'dataset_name': 'vlm_gym_colorization_mse_loss_only_evalonce'}]
+  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_colorization_mse_loss_only_evalonce'}]
+  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_colorization_mse_loss_only_evalonce'}]
+ce_avg: 0.0, mse_avg: 0.05326032266020775
+base_dir is /dev/shm/models/checkpoints_vlm_gym_colorization_one_image_lr2e_5_mse_only_ins/eval_used_rows, step_tag is vlm_gym_colorization_one_img_lr2e_5_mse_only_ins_step500
+Preparing Dataset vlm_gym_colorization_mse_loss_only_evalonce/vlm_gym_colorization_val
+[eval debug] first 3 batch fingerprints:
+  fp[0]: [{'data_indexes': [0], 'worker_id': 0, 'dataset_name': 'vlm_gym_colorization_mse_loss_only_evalonce'}]
+  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_colorization_mse_loss_only_evalonce'}]
+  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_colorization_mse_loss_only_evalonce'}]
+ce_avg: 0.0, mse_avg: 0.007997258566319942
 [2026-01-25 21:45:29] (step=0000759) Train Loss mse: 0.0082, Train Loss ce: 0.0000, Train Steps/Sec: 0.06,
 [2026-01-25 21:45:51] (step=0000760) Train Loss mse: 0.0091, Train Loss ce: 0.0000, Train Steps/Sec: 0.04,
 [2026-01-25 21:46:13] (step=0000761) Train Loss mse: 0.0069, Train Loss ce: 0.0000, Train Steps/Sec: 0.05,
@@ -1963,20 +1949,6 @@ ce_avg: 0.0, mse_avg: 0.00800316222012043
 [2026-01-26 03:41:23] (step=0001752) Train Loss mse: 0.0081, Train Loss ce: 0.0000, Train Steps/Sec: 0.05,
 [2026-01-26 03:41:43] (step=0001753) Train Loss mse: 0.0081, Train Loss ce: 0.0000, Train Steps/Sec: 0.05,
 [2026-01-26 03:42:05] (step=0001754) Train Loss mse: 0.0073, Train Loss ce: 0.0000, Train Steps/Sec: 0.04,
-base_dir is /dev/shm/models/checkpoints_vlm_gym_colorization_one_image_lr2e_5_mse_only_ins/eval_used_rows, step_tag is vlm_gym_colorization_one_img_lr2e_5_mse_only_ins_step2000
-Preparing Dataset vlm_gym_colorization_mse_loss_only_evalonce/vlm_gym_colorization_val
-[eval debug] first 3 batch fingerprints:
-  fp[0]: [{'data_indexes': [0], 'worker_id': 0, 'dataset_name': 'vlm_gym_colorization_mse_loss_only_evalonce'}]
-  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_colorization_mse_loss_only_evalonce'}]
-  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_colorization_mse_loss_only_evalonce'}]
-ce_avg: 0.0, mse_avg: 0.0081106498837471
-base_dir is /dev/shm/models/checkpoints_vlm_gym_colorization_one_image_lr2e_5_mse_only_ins/eval_used_rows, step_tag is vlm_gym_colorization_one_img_lr2e_5_mse_only_ins_step2500
-Preparing Dataset vlm_gym_colorization_mse_loss_only_evalonce/vlm_gym_colorization_val
-[eval debug] first 3 batch fingerprints:
-  fp[0]: [{'data_indexes': [0], 'worker_id': 0, 'dataset_name': 'vlm_gym_colorization_mse_loss_only_evalonce'}]
-  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_colorization_mse_loss_only_evalonce'}]
-  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_colorization_mse_loss_only_evalonce'}]
-ce_avg: 0.0, mse_avg: 0.007652428932487965
 [2026-01-26 03:42:23] (step=0001755) Train Loss mse: 0.0076, Train Loss ce: 0.0000, Train Steps/Sec: 0.05,
 [2026-01-26 03:42:41] (step=0001756) Train Loss mse: 0.0076, Train Loss ce: 0.0000, Train Steps/Sec: 0.06,
 [2026-01-26 03:43:02] (step=0001757) Train Loss mse: 0.0065, Train Loss ce: 0.0000, Train Steps/Sec: 0.05,
@@ -2043,6 +2015,34 @@ ce_avg: 0.0, mse_avg: 0.007652428932487965
 [2026-01-26 04:04:48] (step=0001818) Train Loss mse: 0.0081, Train Loss ce: 0.0000, Train Steps/Sec: 0.06,
 [2026-01-26 04:05:17] (step=0001819) Train Loss mse: 0.0070, Train Loss ce: 0.0000, Train Steps/Sec: 0.04,
 [2026-01-26 04:05:40] (step=0001820) Train Loss mse: 0.0080, Train Loss ce: 0.0000, Train Steps/Sec: 0.04,
+base_dir is /dev/shm/models/checkpoints_vlm_gym_colorization_one_image_lr2e_5_mse_only_ins/eval_used_rows, step_tag is vlm_gym_colorization_one_img_lr2e_5_mse_only_ins_step1000
+Preparing Dataset vlm_gym_colorization_mse_loss_only_evalonce/vlm_gym_colorization_val
+[eval debug] first 3 batch fingerprints:
+  fp[0]: [{'data_indexes': [0], 'worker_id': 0, 'dataset_name': 'vlm_gym_colorization_mse_loss_only_evalonce'}]
+  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_colorization_mse_loss_only_evalonce'}]
+  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_colorization_mse_loss_only_evalonce'}]
+ce_avg: 0.0, mse_avg: 0.007652191445231438
+base_dir is /dev/shm/models/checkpoints_vlm_gym_colorization_one_image_lr2e_5_mse_only_ins/eval_used_rows, step_tag is vlm_gym_colorization_one_img_lr2e_5_mse_only_ins_step1500
+Preparing Dataset vlm_gym_colorization_mse_loss_only_evalonce/vlm_gym_colorization_val
+[eval debug] first 3 batch fingerprints:
+  fp[0]: [{'data_indexes': [0], 'worker_id': 0, 'dataset_name': 'vlm_gym_colorization_mse_loss_only_evalonce'}]
+  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_colorization_mse_loss_only_evalonce'}]
+  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_colorization_mse_loss_only_evalonce'}]
+ce_avg: 0.0, mse_avg: 0.00800316222012043
+base_dir is /dev/shm/models/checkpoints_vlm_gym_colorization_one_image_lr2e_5_mse_only_ins/eval_used_rows, step_tag is vlm_gym_colorization_one_img_lr2e_5_mse_only_ins_step2000
+Preparing Dataset vlm_gym_colorization_mse_loss_only_evalonce/vlm_gym_colorization_val
+[eval debug] first 3 batch fingerprints:
+  fp[0]: [{'data_indexes': [0], 'worker_id': 0, 'dataset_name': 'vlm_gym_colorization_mse_loss_only_evalonce'}]
+  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_colorization_mse_loss_only_evalonce'}]
+  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_colorization_mse_loss_only_evalonce'}]
+ce_avg: 0.0, mse_avg: 0.0081106498837471
+base_dir is /dev/shm/models/checkpoints_vlm_gym_colorization_one_image_lr2e_5_mse_only_ins/eval_used_rows, step_tag is vlm_gym_colorization_one_img_lr2e_5_mse_only_ins_step2500
+Preparing Dataset vlm_gym_colorization_mse_loss_only_evalonce/vlm_gym_colorization_val
+[eval debug] first 3 batch fingerprints:
+  fp[0]: [{'data_indexes': [0], 'worker_id': 0, 'dataset_name': 'vlm_gym_colorization_mse_loss_only_evalonce'}]
+  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_colorization_mse_loss_only_evalonce'}]
+  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_colorization_mse_loss_only_evalonce'}]
+ce_avg: 0.0, mse_avg: 0.007652428932487965
 [2026-01-26 04:05:59] (step=0001821) Train Loss mse: 0.0070, Train Loss ce: 0.0000, Train Steps/Sec: 0.05,
 [2026-01-26 04:06:23] (step=0001822) Train Loss mse: 0.0059, Train Loss ce: 0.0000, Train Steps/Sec: 0.04,
 [2026-01-26 04:06:47] (step=0001823) Train Loss mse: 0.0062, Train Loss ce: 0.0000, Train Steps/Sec: 0.04,
@@ -2988,20 +2988,6 @@ ce_avg: 0.0, mse_avg: 0.007652428932487965
 [2026-01-26 09:44:17] (step=0002763) Train Loss mse: 0.0071, Train Loss ce: 0.0000, Train Steps/Sec: 0.05,
 [2026-01-26 09:44:39] (step=0002764) Train Loss mse: 0.0068, Train Loss ce: 0.0000, Train Steps/Sec: 0.04,
 [2026-01-26 09:44:56] (step=0002765) Train Loss mse: 0.0072, Train Loss ce: 0.0000, Train Steps/Sec: 0.06,
-base_dir is /dev/shm/models/checkpoints_vlm_gym_colorization_one_image_lr2e_5_mse_only_ins/eval_used_rows, step_tag is vlm_gym_colorization_one_img_lr2e_5_mse_only_ins_step3000
-Preparing Dataset vlm_gym_colorization_mse_loss_only_evalonce/vlm_gym_colorization_val
-[eval debug] first 3 batch fingerprints:
-  fp[0]: [{'data_indexes': [0], 'worker_id': 0, 'dataset_name': 'vlm_gym_colorization_mse_loss_only_evalonce'}]
-  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_colorization_mse_loss_only_evalonce'}]
-  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_colorization_mse_loss_only_evalonce'}]
-ce_avg: 0.0, mse_avg: 0.007834003306925297
-base_dir is /dev/shm/models/checkpoints_vlm_gym_colorization_one_image_lr2e_5_mse_only_ins/eval_used_rows, step_tag is vlm_gym_colorization_one_img_lr2e_5_mse_only_ins_step3500
-Preparing Dataset vlm_gym_colorization_mse_loss_only_evalonce/vlm_gym_colorization_val
-[eval debug] first 3 batch fingerprints:
-  fp[0]: [{'data_indexes': [0], 'worker_id': 0, 'dataset_name': 'vlm_gym_colorization_mse_loss_only_evalonce'}]
-  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_colorization_mse_loss_only_evalonce'}]
-  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_colorization_mse_loss_only_evalonce'}]
-ce_avg: 0.0, mse_avg: 0.007766008842736483
 [2026-01-26 09:45:20] (step=0002766) Train Loss mse: 0.0084, Train Loss ce: 0.0000, Train Steps/Sec: 0.04,
 [2026-01-26 09:45:43] (step=0002767) Train Loss mse: 0.0068, Train Loss ce: 0.0000, Train Steps/Sec: 0.04,
 [2026-01-26 09:46:03] (step=0002768) Train Loss mse: 0.0074, Train Loss ce: 0.0000, Train Steps/Sec: 0.05,
@@ -3098,6 +3084,20 @@ ce_avg: 0.0, mse_avg: 0.007766008842736483
 [2026-01-26 10:18:28] (step=0002859) Train Loss mse: 0.0073, Train Loss ce: 0.0000, Train Steps/Sec: 0.05,
 [2026-01-26 10:18:49] (step=0002860) Train Loss mse: 0.0080, Train Loss ce: 0.0000, Train Steps/Sec: 0.05,
 [2026-01-26 10:19:12] (step=0002861) Train Loss mse: 0.0067, Train Loss ce: 0.0000, Train Steps/Sec: 0.04,
+base_dir is /dev/shm/models/checkpoints_vlm_gym_colorization_one_image_lr2e_5_mse_only_ins/eval_used_rows, step_tag is vlm_gym_colorization_one_img_lr2e_5_mse_only_ins_step3000
+Preparing Dataset vlm_gym_colorization_mse_loss_only_evalonce/vlm_gym_colorization_val
+[eval debug] first 3 batch fingerprints:
+  fp[0]: [{'data_indexes': [0], 'worker_id': 0, 'dataset_name': 'vlm_gym_colorization_mse_loss_only_evalonce'}]
+  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_colorization_mse_loss_only_evalonce'}]
+  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_colorization_mse_loss_only_evalonce'}]
+ce_avg: 0.0, mse_avg: 0.007834003306925297
+base_dir is /dev/shm/models/checkpoints_vlm_gym_colorization_one_image_lr2e_5_mse_only_ins/eval_used_rows, step_tag is vlm_gym_colorization_one_img_lr2e_5_mse_only_ins_step3500
+Preparing Dataset vlm_gym_colorization_mse_loss_only_evalonce/vlm_gym_colorization_val
+[eval debug] first 3 batch fingerprints:
+  fp[0]: [{'data_indexes': [0], 'worker_id': 0, 'dataset_name': 'vlm_gym_colorization_mse_loss_only_evalonce'}]
+  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_colorization_mse_loss_only_evalonce'}]
+  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_colorization_mse_loss_only_evalonce'}]
+ce_avg: 0.0, mse_avg: 0.007766008842736483
 [2026-01-26 10:19:31] (step=0002862) Train Loss mse: 0.0065, Train Loss ce: 0.0000, Train Steps/Sec: 0.05,
 [2026-01-26 10:19:53] (step=0002863) Train Loss mse: 0.0073, Train Loss ce: 0.0000, Train Steps/Sec: 0.04,
 [2026-01-26 10:20:13] (step=0002864) Train Loss mse: 0.0084, Train Loss ce: 0.0000, Train Steps/Sec: 0.05,
@@ -3740,20 +3740,6 @@ ce_avg: 0.0, mse_avg: 0.007766008842736483
 [2026-01-26 14:09:22] (step=0003501) Train Loss mse: 0.0066, Train Loss ce: 0.0000, Train Steps/Sec: 0.04,
 [2026-01-26 14:09:44] (step=0003502) Train Loss mse: 0.0062, Train Loss ce: 0.0000, Train Steps/Sec: 0.05,
 [2026-01-26 14:10:07] (step=0003503) Train Loss mse: 0.0075, Train Loss ce: 0.0000, Train Steps/Sec: 0.04,
-base_dir is /dev/shm/models/checkpoints_vlm_gym_colorization_one_image_lr2e_5_mse_only_ins/eval_used_rows, step_tag is vlm_gym_colorization_one_img_lr2e_5_mse_only_ins_step4000
-Preparing Dataset vlm_gym_colorization_mse_loss_only_evalonce/vlm_gym_colorization_val
-[eval debug] first 3 batch fingerprints:
-  fp[0]: [{'data_indexes': [0], 'worker_id': 0, 'dataset_name': 'vlm_gym_colorization_mse_loss_only_evalonce'}]
-  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_colorization_mse_loss_only_evalonce'}]
-  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_colorization_mse_loss_only_evalonce'}]
-ce_avg: 0.0, mse_avg: 0.007558991201221943
-base_dir is /dev/shm/models/checkpoints_vlm_gym_colorization_one_image_lr2e_5_mse_only_ins/eval_used_rows, step_tag is vlm_gym_colorization_one_img_lr2e_5_mse_only_ins_step4500
-Preparing Dataset vlm_gym_colorization_mse_loss_only_evalonce/vlm_gym_colorization_val
-[eval debug] first 3 batch fingerprints:
-  fp[0]: [{'data_indexes': [0], 'worker_id': 0, 'dataset_name': 'vlm_gym_colorization_mse_loss_only_evalonce'}]
-  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_colorization_mse_loss_only_evalonce'}]
-  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_colorization_mse_loss_only_evalonce'}]
-ce_avg: 0.0, mse_avg: 0.007897508330643177
 [2026-01-26 14:10:28] (step=0003504) Train Loss mse: 0.0076, Train Loss ce: 0.0000, Train Steps/Sec: 0.05,
 [2026-01-26 14:10:53] (step=0003505) Train Loss mse: 0.0080, Train Loss ce: 0.0000, Train Steps/Sec: 0.04,
 [2026-01-26 14:11:12] (step=0003506) Train Loss mse: 0.0073, Train Loss ce: 0.0000, Train Steps/Sec: 0.05,
@@ -3870,6 +3856,20 @@ ce_avg: 0.0, mse_avg: 0.007897508330643177
 [2026-01-26 14:51:12] (step=0003617) Train Loss mse: 0.0059, Train Loss ce: 0.0000, Train Steps/Sec: 0.05,
 [2026-01-26 14:51:33] (step=0003618) Train Loss mse: 0.0073, Train Loss ce: 0.0000, Train Steps/Sec: 0.05,
 [2026-01-26 14:51:54] (step=0003619) Train Loss mse: 0.0076, Train Loss ce: 0.0000, Train Steps/Sec: 0.05,
+base_dir is /dev/shm/models/checkpoints_vlm_gym_colorization_one_image_lr2e_5_mse_only_ins/eval_used_rows, step_tag is vlm_gym_colorization_one_img_lr2e_5_mse_only_ins_step4000
+Preparing Dataset vlm_gym_colorization_mse_loss_only_evalonce/vlm_gym_colorization_val
+[eval debug] first 3 batch fingerprints:
+  fp[0]: [{'data_indexes': [0], 'worker_id': 0, 'dataset_name': 'vlm_gym_colorization_mse_loss_only_evalonce'}]
+  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_colorization_mse_loss_only_evalonce'}]
+  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_colorization_mse_loss_only_evalonce'}]
+ce_avg: 0.0, mse_avg: 0.007558991201221943
+base_dir is /dev/shm/models/checkpoints_vlm_gym_colorization_one_image_lr2e_5_mse_only_ins/eval_used_rows, step_tag is vlm_gym_colorization_one_img_lr2e_5_mse_only_ins_step4500
+Preparing Dataset vlm_gym_colorization_mse_loss_only_evalonce/vlm_gym_colorization_val
+[eval debug] first 3 batch fingerprints:
+  fp[0]: [{'data_indexes': [0], 'worker_id': 0, 'dataset_name': 'vlm_gym_colorization_mse_loss_only_evalonce'}]
+  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_colorization_mse_loss_only_evalonce'}]
+  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_colorization_mse_loss_only_evalonce'}]
+ce_avg: 0.0, mse_avg: 0.007897508330643177
 [2026-01-26 14:52:15] (step=0003620) Train Loss mse: 0.0077, Train Loss ce: 0.0000, Train Steps/Sec: 0.05,
 [2026-01-26 14:52:35] (step=0003621) Train Loss mse: 0.0074, Train Loss ce: 0.0000, Train Steps/Sec: 0.05,
 [2026-01-26 14:52:53] (step=0003622) Train Loss mse: 0.0076, Train Loss ce: 0.0000, Train Steps/Sec: 0.05,
@@ -4848,13 +4848,6 @@ ce_avg: 0.0, mse_avg: 0.007897508330643177
 [2026-01-26 20:43:22] (step=0004595) Train Loss mse: 0.0062, Train Loss ce: 0.0000, Train Steps/Sec: 0.05,
 [2026-01-26 20:43:40] (step=0004596) Train Loss mse: 0.0086, Train Loss ce: 0.0000, Train Steps/Sec: 0.05,
 [2026-01-26 20:44:01] (step=0004597) Train Loss mse: 0.0064, Train Loss ce: 0.0000, Train Steps/Sec: 0.05,
-base_dir is /dev/shm/models/checkpoints_vlm_gym_colorization_one_image_lr2e_5_mse_only_ins/eval_used_rows, step_tag is vlm_gym_colorization_one_img_lr2e_5_mse_only_ins_step5000
-Preparing Dataset vlm_gym_colorization_mse_loss_only_evalonce/vlm_gym_colorization_val
-[eval debug] first 3 batch fingerprints:
-  fp[0]: [{'data_indexes': [0], 'worker_id': 0, 'dataset_name': 'vlm_gym_colorization_mse_loss_only_evalonce'}]
-  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_colorization_mse_loss_only_evalonce'}]
-  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_colorization_mse_loss_only_evalonce'}]
-ce_avg: 0.0, mse_avg: 0.007832281291484833
 [2026-01-26 20:44:23] (step=0004598) Train Loss mse: 0.0061, Train Loss ce: 0.0000, Train Steps/Sec: 0.04,
 [2026-01-26 20:44:48] (step=0004599) Train Loss mse: 0.0056, Train Loss ce: 0.0000, Train Steps/Sec: 0.04,
 [2026-01-26 20:45:10] (step=0004600) Train Loss mse: 0.0070, Train Loss ce: 0.0000, Train Steps/Sec: 0.05,
@@ -4951,6 +4944,13 @@ ce_avg: 0.0, mse_avg: 0.007832281291484833
 [2026-01-26 21:17:01] (step=0004691) Train Loss mse: 0.0067, Train Loss ce: 0.0000, Train Steps/Sec: 0.05,
 [2026-01-26 21:17:21] (step=0004692) Train Loss mse: 0.0072, Train Loss ce: 0.0000, Train Steps/Sec: 0.05,
 [2026-01-26 21:17:43] (step=0004693) Train Loss mse: 0.0069, Train Loss ce: 0.0000, Train Steps/Sec: 0.04,
+base_dir is /dev/shm/models/checkpoints_vlm_gym_colorization_one_image_lr2e_5_mse_only_ins/eval_used_rows, step_tag is vlm_gym_colorization_one_img_lr2e_5_mse_only_ins_step5000
+Preparing Dataset vlm_gym_colorization_mse_loss_only_evalonce/vlm_gym_colorization_val
+[eval debug] first 3 batch fingerprints:
+  fp[0]: [{'data_indexes': [0], 'worker_id': 0, 'dataset_name': 'vlm_gym_colorization_mse_loss_only_evalonce'}]
+  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_colorization_mse_loss_only_evalonce'}]
+  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_colorization_mse_loss_only_evalonce'}]
+ce_avg: 0.0, mse_avg: 0.007832281291484833
 [2026-01-26 21:18:07] (step=0004694) Train Loss mse: 0.0077, Train Loss ce: 0.0000, Train Steps/Sec: 0.04,
 [2026-01-26 21:18:31] (step=0004695) Train Loss mse: 0.0081, Train Loss ce: 0.0000, Train Steps/Sec: 0.04,
 [2026-01-26 21:18:52] (step=0004696) Train Loss mse: 0.0081, Train Loss ce: 0.0000, Train Steps/Sec: 0.05,