Upload checkpoints_vlm_gym_colorization_one_image_lr2e_5_ce_ins/checkpoints_vlm_gym_colorization_one_image_lr2e_5_ce_ins

checkpoints_vlm_gym_colorization_one_image_lr2e_5_ce_ins/checkpoints_vlm_gym_colorization_one_image_lr2e_5_ce_ins/wandb/offline-run-20260125_170425-checkpoints_vlm_gym_colorization_one_image_lr2e_5_ce_ins-run0/files/output.log

@@ -1,3 +1,189 @@
+[2026-01-25 21:33:41] (step=0000814) Train Loss mse: 0.0219, Train Loss ce: 0.2812, Train Steps/Sec: 0.05,
+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_celoss/vlm_gym_colorization_train
+base_dir is /dev/shm/models/checkpoints_vlm_gym_colorization_one_image_lr2e_5_ce_ins/eval_used_rows, step_tag is checkpoints_vlm_gym_colorization_one_image_lr2e_5_ce_ins_step0
+Preparing Dataset vlm_gym_colorization_celoss_evalonce/vlm_gym_colorization_val
+[eval debug] first 3 batch fingerprints:
+  fp[0]: [{'data_indexes': [0], 'worker_id': 0, 'dataset_name': 'vlm_gym_colorization_celoss_evalonce'}]
+  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_colorization_celoss_evalonce'}]
+  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_colorization_celoss_evalonce'}]
+ce_avg: 1.3898617029190063, mse_avg: 0.05326032266020775
+base_dir is /dev/shm/models/checkpoints_vlm_gym_colorization_one_image_lr2e_5_ce_ins/eval_used_rows, step_tag is checkpoints_vlm_gym_colorization_one_image_lr2e_5_ce_ins_step500
+Preparing Dataset vlm_gym_colorization_celoss_evalonce/vlm_gym_colorization_val
+[eval debug] first 3 batch fingerprints:
+  fp[0]: [{'data_indexes': [0], 'worker_id': 0, 'dataset_name': 'vlm_gym_colorization_celoss_evalonce'}]
+  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_colorization_celoss_evalonce'}]
+  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_colorization_celoss_evalonce'}]
 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/
@@ -729,192 +915,6 @@ wandb: For more information, check out the docs at: https://weave-docs.wandb.ai/
 [2026-01-25 21:03:00] (step=0000718) Train Loss mse: 0.0111, Train Loss ce: 0.2815, Train Steps/Sec: 0.05,
 [2026-01-25 21:03:14] (step=0000719) Train Loss mse: 0.0212, Train Loss ce: 0.3005, Train Steps/Sec: 0.07,
 [2026-01-25 21:03:34] (step=0000720) Train Loss mse: 0.0175, Train Loss ce: 0.2827, Train Steps/Sec: 0.05,
-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_celoss/vlm_gym_colorization_train
-base_dir is /dev/shm/models/checkpoints_vlm_gym_colorization_one_image_lr2e_5_ce_ins/eval_used_rows, step_tag is checkpoints_vlm_gym_colorization_one_image_lr2e_5_ce_ins_step0
-Preparing Dataset vlm_gym_colorization_celoss_evalonce/vlm_gym_colorization_val
-[eval debug] first 3 batch fingerprints:
-  fp[0]: [{'data_indexes': [0], 'worker_id': 0, 'dataset_name': 'vlm_gym_colorization_celoss_evalonce'}]
-  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_colorization_celoss_evalonce'}]
-  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_colorization_celoss_evalonce'}]
-ce_avg: 1.3898617029190063, mse_avg: 0.05326032266020775
-base_dir is /dev/shm/models/checkpoints_vlm_gym_colorization_one_image_lr2e_5_ce_ins/eval_used_rows, step_tag is checkpoints_vlm_gym_colorization_one_image_lr2e_5_ce_ins_step500
-Preparing Dataset vlm_gym_colorization_celoss_evalonce/vlm_gym_colorization_val
-[eval debug] first 3 batch fingerprints:
-  fp[0]: [{'data_indexes': [0], 'worker_id': 0, 'dataset_name': 'vlm_gym_colorization_celoss_evalonce'}]
-  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_colorization_celoss_evalonce'}]
-  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_colorization_celoss_evalonce'}]
-ce_avg: 0.2838270962238312, mse_avg: 0.00850633904337883
 [2026-01-25 21:03:53] (step=0000721) Train Loss mse: 0.0203, Train Loss ce: 0.2965, Train Steps/Sec: 0.05,
 [2026-01-25 21:04:11] (step=0000722) Train Loss mse: 0.0150, Train Loss ce: 0.2638, Train Steps/Sec: 0.05,
 [2026-01-25 21:04:33] (step=0000723) Train Loss mse: 0.0155, Train Loss ce: 0.2844, Train Steps/Sec: 0.05,
@@ -1008,7 +1008,28 @@ ce_avg: 0.2838270962238312, mse_avg: 0.00850633904337883
 [2026-01-25 21:32:44] (step=0000811) Train Loss mse: 0.0192, Train Loss ce: 0.2767, Train Steps/Sec: 0.06,
 [2026-01-25 21:33:04] (step=0000812) Train Loss mse: 0.0235, Train Loss ce: 0.2813, Train Steps/Sec: 0.05,
 [2026-01-25 21:33:19] (step=0000813) Train Loss mse: 0.0263, Train Loss ce: 0.2622, Train Steps/Sec: 0.06,
-[2026-01-25 21:33:41] (step=0000814) Train Loss mse: 0.0219, Train Loss ce: 0.2812, Train Steps/Sec: 0.05,
+ce_avg: 0.2838270962238312, mse_avg: 0.00850633904337883
+base_dir is /dev/shm/models/checkpoints_vlm_gym_colorization_one_image_lr2e_5_ce_ins/eval_used_rows, step_tag is checkpoints_vlm_gym_colorization_one_image_lr2e_5_ce_ins_step1000
+Preparing Dataset vlm_gym_colorization_celoss_evalonce/vlm_gym_colorization_val
+[eval debug] first 3 batch fingerprints:
+  fp[0]: [{'data_indexes': [0], 'worker_id': 0, 'dataset_name': 'vlm_gym_colorization_celoss_evalonce'}]
+  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_colorization_celoss_evalonce'}]
+  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_colorization_celoss_evalonce'}]
+ce_avg: 0.4483165144920349, mse_avg: 0.008000156842172146
+base_dir is /dev/shm/models/checkpoints_vlm_gym_colorization_one_image_lr2e_5_ce_ins/eval_used_rows, step_tag is checkpoints_vlm_gym_colorization_one_image_lr2e_5_ce_ins_step1500
+Preparing Dataset vlm_gym_colorization_celoss_evalonce/vlm_gym_colorization_val
+[eval debug] first 3 batch fingerprints:
+  fp[0]: [{'data_indexes': [0], 'worker_id': 0, 'dataset_name': 'vlm_gym_colorization_celoss_evalonce'}]
+  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_colorization_celoss_evalonce'}]
+  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_colorization_celoss_evalonce'}]
+ce_avg: 0.5879027843475342, mse_avg: 0.008525446057319641
+base_dir is /dev/shm/models/checkpoints_vlm_gym_colorization_one_image_lr2e_5_ce_ins/eval_used_rows, step_tag is checkpoints_vlm_gym_colorization_one_image_lr2e_5_ce_ins_step2000
+Preparing Dataset vlm_gym_colorization_celoss_evalonce/vlm_gym_colorization_val
+[eval debug] first 3 batch fingerprints:
+  fp[0]: [{'data_indexes': [0], 'worker_id': 0, 'dataset_name': 'vlm_gym_colorization_celoss_evalonce'}]
+  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_colorization_celoss_evalonce'}]
+  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_colorization_celoss_evalonce'}]
+ce_avg: 1.2331268787384033, mse_avg: 0.008829578757286072
 [2026-01-25 21:33:57] (step=0000815) Train Loss mse: 0.0234, Train Loss ce: 0.2941, Train Steps/Sec: 0.07,
 [2026-01-25 21:34:16] (step=0000816) Train Loss mse: 0.0220, Train Loss ce: 0.2732, Train Steps/Sec: 0.05,
 [2026-01-25 21:34:34] (step=0000817) Train Loss mse: 0.0237, Train Loss ce: 0.2544, Train Steps/Sec: 0.05,
@@ -2082,41 +2103,6 @@ ce_avg: 0.2838270962238312, mse_avg: 0.00850633904337883
 [2026-01-26 03:14:59] (step=0001885) Train Loss mse: 0.0153, Train Loss ce: 0.2766, Train Steps/Sec: 0.05,
 [2026-01-26 03:15:19] (step=0001886) Train Loss mse: 0.0292, Train Loss ce: 0.2450, Train Steps/Sec: 0.05,
 [2026-01-26 03:15:36] (step=0001887) Train Loss mse: 0.0145, Train Loss ce: 0.2653, Train Steps/Sec: 0.06,
-base_dir is /dev/shm/models/checkpoints_vlm_gym_colorization_one_image_lr2e_5_ce_ins/eval_used_rows, step_tag is checkpoints_vlm_gym_colorization_one_image_lr2e_5_ce_ins_step1000
-Preparing Dataset vlm_gym_colorization_celoss_evalonce/vlm_gym_colorization_val
-[eval debug] first 3 batch fingerprints:
-  fp[0]: [{'data_indexes': [0], 'worker_id': 0, 'dataset_name': 'vlm_gym_colorization_celoss_evalonce'}]
-  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_colorization_celoss_evalonce'}]
-  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_colorization_celoss_evalonce'}]
-ce_avg: 0.4483165144920349, mse_avg: 0.008000156842172146
-base_dir is /dev/shm/models/checkpoints_vlm_gym_colorization_one_image_lr2e_5_ce_ins/eval_used_rows, step_tag is checkpoints_vlm_gym_colorization_one_image_lr2e_5_ce_ins_step1500
-Preparing Dataset vlm_gym_colorization_celoss_evalonce/vlm_gym_colorization_val
-[eval debug] first 3 batch fingerprints:
-  fp[0]: [{'data_indexes': [0], 'worker_id': 0, 'dataset_name': 'vlm_gym_colorization_celoss_evalonce'}]
-  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_colorization_celoss_evalonce'}]
-  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_colorization_celoss_evalonce'}]
-ce_avg: 0.5879027843475342, mse_avg: 0.008525446057319641
-base_dir is /dev/shm/models/checkpoints_vlm_gym_colorization_one_image_lr2e_5_ce_ins/eval_used_rows, step_tag is checkpoints_vlm_gym_colorization_one_image_lr2e_5_ce_ins_step2000
-Preparing Dataset vlm_gym_colorization_celoss_evalonce/vlm_gym_colorization_val
-[eval debug] first 3 batch fingerprints:
-  fp[0]: [{'data_indexes': [0], 'worker_id': 0, 'dataset_name': 'vlm_gym_colorization_celoss_evalonce'}]
-  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_colorization_celoss_evalonce'}]
-  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_colorization_celoss_evalonce'}]
-ce_avg: 1.2331268787384033, mse_avg: 0.008829578757286072
-base_dir is /dev/shm/models/checkpoints_vlm_gym_colorization_one_image_lr2e_5_ce_ins/eval_used_rows, step_tag is checkpoints_vlm_gym_colorization_one_image_lr2e_5_ce_ins_step2500
-Preparing Dataset vlm_gym_colorization_celoss_evalonce/vlm_gym_colorization_val
-[eval debug] first 3 batch fingerprints:
-  fp[0]: [{'data_indexes': [0], 'worker_id': 0, 'dataset_name': 'vlm_gym_colorization_celoss_evalonce'}]
-  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_colorization_celoss_evalonce'}]
-  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_colorization_celoss_evalonce'}]
-ce_avg: 1.185328722000122, mse_avg: 0.008346919901669025
-base_dir is /dev/shm/models/checkpoints_vlm_gym_colorization_one_image_lr2e_5_ce_ins/eval_used_rows, step_tag is checkpoints_vlm_gym_colorization_one_image_lr2e_5_ce_ins_step3000
-Preparing Dataset vlm_gym_colorization_celoss_evalonce/vlm_gym_colorization_val
-[eval debug] first 3 batch fingerprints:
-  fp[0]: [{'data_indexes': [0], 'worker_id': 0, 'dataset_name': 'vlm_gym_colorization_celoss_evalonce'}]
-  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_colorization_celoss_evalonce'}]
-  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_colorization_celoss_evalonce'}]
-ce_avg: 0.24374419450759888, mse_avg: 0.007726446725428104
 [2026-01-26 03:15:56] (step=0001888) Train Loss mse: 0.0274, Train Loss ce: 0.2608, Train Steps/Sec: 0.05,
 [2026-01-26 03:16:09] (step=0001889) Train Loss mse: 0.0220, Train Loss ce: 0.2631, Train Steps/Sec: 0.08,
 [2026-01-26 03:16:27] (step=0001890) Train Loss mse: 0.0193, Train Loss ce: 0.2902, Train Steps/Sec: 0.06,
@@ -2201,6 +2187,20 @@ ce_avg: 0.24374419450759888, mse_avg: 0.007726446725428104
 [2026-01-26 03:41:34] (step=0001969) Train Loss mse: 0.0250, Train Loss ce: 0.2561, Train Steps/Sec: 0.06,
 [2026-01-26 03:41:53] (step=0001970) Train Loss mse: 0.0162, Train Loss ce: 0.2314, Train Steps/Sec: 0.05,
 [2026-01-26 03:42:15] (step=0001971) Train Loss mse: 0.0232, Train Loss ce: 0.2660, Train Steps/Sec: 0.04,
+base_dir is /dev/shm/models/checkpoints_vlm_gym_colorization_one_image_lr2e_5_ce_ins/eval_used_rows, step_tag is checkpoints_vlm_gym_colorization_one_image_lr2e_5_ce_ins_step2500
+Preparing Dataset vlm_gym_colorization_celoss_evalonce/vlm_gym_colorization_val
+[eval debug] first 3 batch fingerprints:
+  fp[0]: [{'data_indexes': [0], 'worker_id': 0, 'dataset_name': 'vlm_gym_colorization_celoss_evalonce'}]
+  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_colorization_celoss_evalonce'}]
+  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_colorization_celoss_evalonce'}]
+ce_avg: 1.185328722000122, mse_avg: 0.008346919901669025
+base_dir is /dev/shm/models/checkpoints_vlm_gym_colorization_one_image_lr2e_5_ce_ins/eval_used_rows, step_tag is checkpoints_vlm_gym_colorization_one_image_lr2e_5_ce_ins_step3000
+Preparing Dataset vlm_gym_colorization_celoss_evalonce/vlm_gym_colorization_val
+[eval debug] first 3 batch fingerprints:
+  fp[0]: [{'data_indexes': [0], 'worker_id': 0, 'dataset_name': 'vlm_gym_colorization_celoss_evalonce'}]
+  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_colorization_celoss_evalonce'}]
+  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_colorization_celoss_evalonce'}]
+ce_avg: 0.24374419450759888, mse_avg: 0.007726446725428104
 [2026-01-26 03:42:35] (step=0001972) Train Loss mse: 0.0179, Train Loss ce: 0.2603, Train Steps/Sec: 0.05,
 [2026-01-26 03:42:54] (step=0001973) Train Loss mse: 0.0221, Train Loss ce: 0.2504, Train Steps/Sec: 0.05,
 [2026-01-26 03:43:14] (step=0001974) Train Loss mse: 0.0300, Train Loss ce: 0.2581, Train Steps/Sec: 0.05,
@@ -3231,6 +3231,13 @@ ce_avg: 0.24374419450759888, mse_avg: 0.007726446725428104
 [2026-01-26 09:12:03] (step=0002996) Train Loss mse: 0.0169, Train Loss ce: 0.2552, Train Steps/Sec: 0.06,
 [2026-01-26 09:12:21] (step=0002997) Train Loss mse: 0.0187, Train Loss ce: 0.2464, Train Steps/Sec: 0.06,
 [2026-01-26 09:12:42] (step=0002998) Train Loss mse: 0.0267, Train Loss ce: 0.2342, Train Steps/Sec: 0.05,
+base_dir is /dev/shm/models/checkpoints_vlm_gym_colorization_one_image_lr2e_5_ce_ins/eval_used_rows, step_tag is checkpoints_vlm_gym_colorization_one_image_lr2e_5_ce_ins_step3500
+Preparing Dataset vlm_gym_colorization_celoss_evalonce/vlm_gym_colorization_val
+[eval debug] first 3 batch fingerprints:
+  fp[0]: [{'data_indexes': [0], 'worker_id': 0, 'dataset_name': 'vlm_gym_colorization_celoss_evalonce'}]
+  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_colorization_celoss_evalonce'}]
+  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_colorization_celoss_evalonce'}]
+ce_avg: 0.2417636662721634, mse_avg: 0.007533130701631308
 [2026-01-26 09:13:01] (step=0002999) Train Loss mse: 0.0298, Train Loss ce: 0.2517, Train Steps/Sec: 0.05,
 [2026-01-26 09:14:50] (step=0003000) Train Loss mse: 0.0192, Train Loss ce: 0.2577, Train Steps/Sec: 0.01,
 [2026-01-26 09:15:09] (step=0003001) Train Loss mse: 0.0150, Train Loss ce: 0.2598, Train Steps/Sec: 0.05,
@@ -3256,13 +3263,20 @@ ce_avg: 0.24374419450759888, mse_avg: 0.007726446725428104
 [2026-01-26 09:21:26] (step=0003021) Train Loss mse: 0.0134, Train Loss ce: 0.2563, Train Steps/Sec: 0.05,
 [2026-01-26 09:21:45] (step=0003022) Train Loss mse: 0.0198, Train Loss ce: 0.2363, Train Steps/Sec: 0.05,
 [2026-01-26 09:22:06] (step=0003023) Train Loss mse: 0.0328, Train Loss ce: 0.2532, Train Steps/Sec: 0.05,
-base_dir is /dev/shm/models/checkpoints_vlm_gym_colorization_one_image_lr2e_5_ce_ins/eval_used_rows, step_tag is checkpoints_vlm_gym_colorization_one_image_lr2e_5_ce_ins_step3500
-Preparing Dataset vlm_gym_colorization_celoss_evalonce/vlm_gym_colorization_val
-[eval debug] first 3 batch fingerprints:
-  fp[0]: [{'data_indexes': [0], 'worker_id': 0, 'dataset_name': 'vlm_gym_colorization_celoss_evalonce'}]
-  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_colorization_celoss_evalonce'}]
-  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_colorization_celoss_evalonce'}]
-ce_avg: 0.2417636662721634, mse_avg: 0.007533130701631308
+[2026-01-26 09:22:24] (step=0003024) Train Loss mse: 0.0296, Train Loss ce: 0.2491, Train Steps/Sec: 0.06,
+[2026-01-26 09:22:40] (step=0003025) Train Loss mse: 0.0156, Train Loss ce: 0.2473, Train Steps/Sec: 0.06,
+[2026-01-26 09:22:57] (step=0003026) Train Loss mse: 0.0173, Train Loss ce: 0.2442, Train Steps/Sec: 0.06,
+[2026-01-26 09:23:17] (step=0003027) Train Loss mse: 0.0204, Train Loss ce: 0.2472, Train Steps/Sec: 0.05,
+[2026-01-26 09:23:36] (step=0003028) Train Loss mse: 0.0190, Train Loss ce: 0.2462, Train Steps/Sec: 0.05,
+[2026-01-26 09:23:57] (step=0003029) Train Loss mse: 0.0203, Train Loss ce: 0.2455, Train Steps/Sec: 0.05,
+[2026-01-26 09:24:19] (step=0003030) Train Loss mse: 0.0311, Train Loss ce: 0.2568, Train Steps/Sec: 0.05,
+[2026-01-26 09:24:38] (step=0003031) Train Loss mse: 0.0205, Train Loss ce: 0.2451, Train Steps/Sec: 0.05,
+[2026-01-26 09:24:55] (step=0003032) Train Loss mse: 0.0161, Train Loss ce: 0.2535, Train Steps/Sec: 0.06,
+[2026-01-26 09:25:10] (step=0003033) Train Loss mse: 0.0199, Train Loss ce: 0.2518, Train Steps/Sec: 0.07,
+[2026-01-26 09:25:30] (step=0003034) Train Loss mse: 0.0306, Train Loss ce: 0.2664, Train Steps/Sec: 0.05,
+[2026-01-26 09:25:49] (step=0003035) Train Loss mse: 0.0209, Train Loss ce: 0.2695, Train Steps/Sec: 0.05,
+[2026-01-26 09:26:09] (step=0003036) Train Loss mse: 0.0244, Train Loss ce: 0.2414, Train Steps/Sec: 0.05,
+[2026-01-26 09:26:25] (step=0003037) Train Loss mse: 0.0370, Train Loss ce: 0.2761, Train Steps/Sec: 0.06,
 [2026-01-26 09:26:47] (step=0003038) Train Loss mse: 0.0324, Train Loss ce: 0.2545, Train Steps/Sec: 0.04,
 [2026-01-26 09:27:08] (step=0003039) Train Loss mse: 0.0328, Train Loss ce: 0.2373, Train Steps/Sec: 0.05,
 [2026-01-26 09:27:29] (step=0003040) Train Loss mse: 0.0207, Train Loss ce: 0.2709, Train Steps/Sec: 0.05,
@@ -4151,20 +4165,6 @@ ce_avg: 0.2417636662721634, mse_avg: 0.007533130701631308
 [2026-01-26 14:10:34] (step=0003923) Train Loss mse: 0.0186, Train Loss ce: 0.2431, Train Steps/Sec: 0.06,
 [2026-01-26 14:10:49] (step=0003924) Train Loss mse: 0.0123, Train Loss ce: 0.2697, Train Steps/Sec: 0.06,
 [2026-01-26 14:11:05] (step=0003925) Train Loss mse: 0.0176, Train Loss ce: 0.2535, Train Steps/Sec: 0.06,
-base_dir is /dev/shm/models/checkpoints_vlm_gym_colorization_one_image_lr2e_5_ce_ins/eval_used_rows, step_tag is checkpoints_vlm_gym_colorization_one_image_lr2e_5_ce_ins_step4000
-Preparing Dataset vlm_gym_colorization_celoss_evalonce/vlm_gym_colorization_val
-[eval debug] first 3 batch fingerprints:
-  fp[0]: [{'data_indexes': [0], 'worker_id': 0, 'dataset_name': 'vlm_gym_colorization_celoss_evalonce'}]
-  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_colorization_celoss_evalonce'}]
-  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_colorization_celoss_evalonce'}]
-ce_avg: 0.24075058102607727, mse_avg: 0.007322242017835379
-base_dir is /dev/shm/models/checkpoints_vlm_gym_colorization_one_image_lr2e_5_ce_ins/eval_used_rows, step_tag is checkpoints_vlm_gym_colorization_one_image_lr2e_5_ce_ins_step4500
-Preparing Dataset vlm_gym_colorization_celoss_evalonce/vlm_gym_colorization_val
-[eval debug] first 3 batch fingerprints:
-  fp[0]: [{'data_indexes': [0], 'worker_id': 0, 'dataset_name': 'vlm_gym_colorization_celoss_evalonce'}]
-  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_colorization_celoss_evalonce'}]
-  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_colorization_celoss_evalonce'}]
-ce_avg: 0.23980583250522614, mse_avg: 0.007650755811482668
 [2026-01-26 14:11:24] (step=0003926) Train Loss mse: 0.0236, Train Loss ce: 0.2457, Train Steps/Sec: 0.05,
 [2026-01-26 14:11:47] (step=0003927) Train Loss mse: 0.0151, Train Loss ce: 0.2381, Train Steps/Sec: 0.04,
 [2026-01-26 14:12:08] (step=0003928) Train Loss mse: 0.0155, Train Loss ce: 0.2358, Train Steps/Sec: 0.05,
@@ -4196,6 +4196,27 @@ ce_avg: 0.23980583250522614, mse_avg: 0.007650755811482668
 [2026-01-26 14:20:02] (step=0003954) Train Loss mse: 0.0229, Train Loss ce: 0.2695, Train Steps/Sec: 0.06,
 [2026-01-26 14:20:22] (step=0003955) Train Loss mse: 0.0199, Train Loss ce: 0.2492, Train Steps/Sec: 0.05,
 [2026-01-26 14:20:43] (step=0003956) Train Loss mse: 0.0114, Train Loss ce: 0.2503, Train Steps/Sec: 0.05,
+base_dir is /dev/shm/models/checkpoints_vlm_gym_colorization_one_image_lr2e_5_ce_ins/eval_used_rows, step_tag is checkpoints_vlm_gym_colorization_one_image_lr2e_5_ce_ins_step4000
+Preparing Dataset vlm_gym_colorization_celoss_evalonce/vlm_gym_colorization_val
+[eval debug] first 3 batch fingerprints:
+  fp[0]: [{'data_indexes': [0], 'worker_id': 0, 'dataset_name': 'vlm_gym_colorization_celoss_evalonce'}]
+  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_colorization_celoss_evalonce'}]
+  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_colorization_celoss_evalonce'}]
+ce_avg: 0.24075058102607727, mse_avg: 0.007322242017835379
+base_dir is /dev/shm/models/checkpoints_vlm_gym_colorization_one_image_lr2e_5_ce_ins/eval_used_rows, step_tag is checkpoints_vlm_gym_colorization_one_image_lr2e_5_ce_ins_step4500
+Preparing Dataset vlm_gym_colorization_celoss_evalonce/vlm_gym_colorization_val
+[eval debug] first 3 batch fingerprints:
+  fp[0]: [{'data_indexes': [0], 'worker_id': 0, 'dataset_name': 'vlm_gym_colorization_celoss_evalonce'}]
+  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_colorization_celoss_evalonce'}]
+  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_colorization_celoss_evalonce'}]
+ce_avg: 0.23980583250522614, mse_avg: 0.007650755811482668
+base_dir is /dev/shm/models/checkpoints_vlm_gym_colorization_one_image_lr2e_5_ce_ins/eval_used_rows, step_tag is checkpoints_vlm_gym_colorization_one_image_lr2e_5_ce_ins_step5000
+Preparing Dataset vlm_gym_colorization_celoss_evalonce/vlm_gym_colorization_val
+[eval debug] first 3 batch fingerprints:
+  fp[0]: [{'data_indexes': [0], 'worker_id': 0, 'dataset_name': 'vlm_gym_colorization_celoss_evalonce'}]
+  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_colorization_celoss_evalonce'}]
+  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_colorization_celoss_evalonce'}]
+ce_avg: 0.23948809504508972, mse_avg: 0.007586085703223944
 [2026-01-26 14:21:06] (step=0003957) Train Loss mse: 0.0343, Train Loss ce: 0.2657, Train Steps/Sec: 0.04,
 [2026-01-26 14:21:25] (step=0003958) Train Loss mse: 0.0271, Train Loss ce: 0.2690, Train Steps/Sec: 0.05,
 [2026-01-26 14:21:43] (step=0003959) Train Loss mse: 0.0122, Train Loss ce: 0.2418, Train Steps/Sec: 0.05,
@@ -5241,11 +5262,4 @@ ce_avg: 0.23980583250522614, mse_avg: 0.007650755811482668
 [2026-01-26 19:56:56] (step=0004999) Train Loss mse: 0.0202, Train Loss ce: 0.2401, Train Steps/Sec: 0.04,
 [2026-01-26 19:58:44] (step=0005000) Train Loss mse: 0.0209, Train Loss ce: 0.2473, Train Steps/Sec: 0.01,
 [2026-01-26 19:58:44] Saving checkpoint to /dev/shm/models/checkpoints_vlm_gym_colorization_one_image_lr2e_5_ce_ins/0005000.
-[2026-01-26 20:01:32] Done!
-base_dir is /dev/shm/models/checkpoints_vlm_gym_colorization_one_image_lr2e_5_ce_ins/eval_used_rows, step_tag is checkpoints_vlm_gym_colorization_one_image_lr2e_5_ce_ins_step5000
-Preparing Dataset vlm_gym_colorization_celoss_evalonce/vlm_gym_colorization_val
-[eval debug] first 3 batch fingerprints:
-  fp[0]: [{'data_indexes': [0], 'worker_id': 0, 'dataset_name': 'vlm_gym_colorization_celoss_evalonce'}]
-  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_colorization_celoss_evalonce'}]
-  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_colorization_celoss_evalonce'}]
-ce_avg: 0.23948809504508972, mse_avg: 0.007586085703223944
+[2026-01-26 20:01:32] Done!