Upload checkpoints_vlm_gym_match_equation_sos_one_image_lr2e_5_ce_no_mse_ins/checkpoints_vlm_gym_match_equation_sos_one_image_lr2e_5_ce_no_mse_ins

checkpoints_vlm_gym_match_equation_sos_one_image_lr2e_5_ce_no_mse_ins/checkpoints_vlm_gym_match_equation_sos_one_image_lr2e_5_ce_no_mse_ins/wandb/offline-run-20260126_192812-checkpoints_vlm_gym_match_equation_sos_one_image_lr2e_5_ce_no_mse_ins-run0/files/output.log

@@ -1,173 +1,3 @@
-FullyShardedDataParallel(
-  (_fsdp_wrapped_module): Bagel(
-    (language_model): Qwen2ForCausalLM(
-      (model): Qwen2Model(
-        (embed_tokens): Embedding(152064, 3584)
-        (layers): ModuleList(
-          (0-27): 28 x FullyShardedDataParallel(
-            (_fsdp_wrapped_module): CheckpointWrapper(
-              (_checkpoint_wrapped_module): Qwen2MoTDecoderLayer(
-                (self_attn): PackedAttentionMoT(
-                  (q_proj): Linear(in_features=3584, out_features=3584, bias=True)
-                  (k_proj): Linear(in_features=3584, out_features=512, bias=True)
-                  (v_proj): Linear(in_features=3584, out_features=512, bias=True)
-                  (o_proj): Linear(in_features=3584, out_features=3584, bias=False)
-                  (q_norm): Qwen2RMSNorm((128,), eps=1e-06)
-                  (k_norm): Qwen2RMSNorm((128,), eps=1e-06)
-                  (q_norm_moe_gen): Qwen2RMSNorm((128,), eps=1e-06)
-                  (k_norm_moe_gen): Qwen2RMSNorm((128,), eps=1e-06)
-                  (q_proj_moe_gen): Linear(in_features=3584, out_features=3584, bias=True)
-                  (k_proj_moe_gen): Linear(in_features=3584, out_features=512, bias=True)
-                  (v_proj_moe_gen): Linear(in_features=3584, out_features=512, bias=True)
-                  (o_proj_moe_gen): Linear(in_features=3584, out_features=3584, bias=False)
-                )
-                (mlp): Qwen2MLP(
-                  (gate_proj): Linear(in_features=3584, out_features=18944, bias=False)
-                  (up_proj): Linear(in_features=3584, out_features=18944, bias=False)
-                  (down_proj): Linear(in_features=18944, out_features=3584, bias=False)
-                  (act_fn): SiLU()
-                )
-                (mlp_moe_gen): Qwen2MLP(
-                  (gate_proj): Linear(in_features=3584, out_features=18944, bias=False)
-                  (up_proj): Linear(in_features=3584, out_features=18944, bias=False)
-                  (down_proj): Linear(in_features=18944, out_features=3584, bias=False)
-                  (act_fn): SiLU()
-                )
-                (input_layernorm): Qwen2RMSNorm((3584,), eps=1e-06)
-                (input_layernorm_moe_gen): Qwen2RMSNorm((3584,), eps=1e-06)
-                (post_attention_layernorm): Qwen2RMSNorm((3584,), eps=1e-06)
-                (post_attention_layernorm_moe_gen): Qwen2RMSNorm((3584,), eps=1e-06)
-              )
-            )
-          )
-        )
-        (norm): Qwen2RMSNorm((3584,), eps=1e-06)
-        (norm_moe_gen): Qwen2RMSNorm((3584,), eps=1e-06)
-        (rotary_emb): Qwen2RotaryEmbedding()
-      )
-      (lm_head): Linear(in_features=3584, out_features=152064, bias=False)
-    )
-    (vit_model): SiglipVisionModel(
-      (vision_model): FullyShardedDataParallel(
-        (_fsdp_wrapped_module): SiglipVisionTransformer(
-          (embeddings): SiglipVisionEmbeddings(
-            (position_embedding): Embedding(4900, 1152)
-            (patch_embedding): Linear(in_features=588, out_features=1152, bias=True)
-          )
-          (encoder): SiglipEncoder(
-            (layers): ModuleList(
-              (0-25): 26 x FullyShardedDataParallel(
-                (_fsdp_wrapped_module): CheckpointWrapper(
-                  (_checkpoint_wrapped_module): SiglipEncoderLayer(
-                    (self_attn): SiglipFlashAttention2(
-                      (k_proj): Linear(in_features=1152, out_features=1152, bias=True)
-                      (v_proj): Linear(in_features=1152, out_features=1152, bias=True)
-                      (q_proj): Linear(in_features=1152, out_features=1152, bias=True)
-                      (out_proj): Linear(in_features=1152, out_features=1152, bias=True)
-                    )
-                    (layer_norm1): LayerNorm((1152,), eps=1e-06, elementwise_affine=True)
-                    (mlp): SiglipMLP(
-                      (activation_fn): PytorchGELUTanh()
-                      (fc1): Linear(in_features=1152, out_features=4304, bias=True)
-                      (fc2): Linear(in_features=4304, out_features=1152, bias=True)
-                    )
-                    (layer_norm2): LayerNorm((1152,), eps=1e-06, elementwise_affine=True)
-                  )
-                )
-              )
-            )
-          )
-          (post_layernorm): LayerNorm((1152,), eps=1e-06, elementwise_affine=True)
-        )
-      )
-    )
-    (connector): FullyShardedDataParallel(
-      (_fsdp_wrapped_module): CheckpointWrapper(
-        (_checkpoint_wrapped_module): MLPconnector(
-          (activation_fn): PytorchGELUTanh()
-          (fc1): Linear(in_features=1152, out_features=3584, bias=True)
-          (fc2): Linear(in_features=3584, out_features=3584, bias=True)
-        )
-      )
-    )
-    (vit_pos_embed): FullyShardedDataParallel(
-      (_fsdp_wrapped_module): PositionEmbedding()
-    )
-  )
-)
-_flat_param True
-language_model.model.layers.0._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
-language_model.model.layers.1._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
-language_model.model.layers.2._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
-language_model.model.layers.3._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
-language_model.model.layers.4._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
-language_model.model.layers.5._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
-language_model.model.layers.6._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
-language_model.model.layers.7._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
-language_model.model.layers.8._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
-language_model.model.layers.9._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
-language_model.model.layers.10._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
-language_model.model.layers.11._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
-language_model.model.layers.12._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
-language_model.model.layers.13._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
-language_model.model.layers.14._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
-language_model.model.layers.15._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
-language_model.model.layers.16._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
-language_model.model.layers.17._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
-language_model.model.layers.18._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
-language_model.model.layers.19._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
-language_model.model.layers.20._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
-language_model.model.layers.21._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
-language_model.model.layers.22._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
-language_model.model.layers.23._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
-language_model.model.layers.24._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
-language_model.model.layers.25._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
-language_model.model.layers.26._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
-language_model.model.layers.27._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
-vit_model.vision_model._fsdp_wrapped_module._flat_param True
-vit_model.vision_model._fsdp_wrapped_module.encoder.layers.0._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
-vit_model.vision_model._fsdp_wrapped_module.encoder.layers.1._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
-vit_model.vision_model._fsdp_wrapped_module.encoder.layers.2._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
-vit_model.vision_model._fsdp_wrapped_module.encoder.layers.3._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
-vit_model.vision_model._fsdp_wrapped_module.encoder.layers.4._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
-vit_model.vision_model._fsdp_wrapped_module.encoder.layers.5._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
-vit_model.vision_model._fsdp_wrapped_module.encoder.layers.6._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
-vit_model.vision_model._fsdp_wrapped_module.encoder.layers.7._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
-vit_model.vision_model._fsdp_wrapped_module.encoder.layers.8._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
-vit_model.vision_model._fsdp_wrapped_module.encoder.layers.9._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
-vit_model.vision_model._fsdp_wrapped_module.encoder.layers.10._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
-vit_model.vision_model._fsdp_wrapped_module.encoder.layers.11._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
-vit_model.vision_model._fsdp_wrapped_module.encoder.layers.12._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
-vit_model.vision_model._fsdp_wrapped_module.encoder.layers.13._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
-vit_model.vision_model._fsdp_wrapped_module.encoder.layers.14._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
-vit_model.vision_model._fsdp_wrapped_module.encoder.layers.15._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
-vit_model.vision_model._fsdp_wrapped_module.encoder.layers.16._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
-vit_model.vision_model._fsdp_wrapped_module.encoder.layers.17._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
-vit_model.vision_model._fsdp_wrapped_module.encoder.layers.18._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
-vit_model.vision_model._fsdp_wrapped_module.encoder.layers.19._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
-vit_model.vision_model._fsdp_wrapped_module.encoder.layers.20._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
-vit_model.vision_model._fsdp_wrapped_module.encoder.layers.21._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
-vit_model.vision_model._fsdp_wrapped_module.encoder.layers.22._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
-vit_model.vision_model._fsdp_wrapped_module.encoder.layers.23._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
-vit_model.vision_model._fsdp_wrapped_module.encoder.layers.24._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
-vit_model.vision_model._fsdp_wrapped_module.encoder.layers.25._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
-connector._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
-vit_pos_embed._fsdp_wrapped_module._flat_param False
-Preparing Dataset vlm_gym_match_equation_sos_celoss_no_mse/vlm_gym_match_equation_sos_train
-base_dir is /dev/shm/models/checkpoints_vlm_gym_match_equation_sos_one_image_lr2e_5_ce_no_mse_ins/eval_used_rows, step_tag is checkpoints_vlm_gym_match_equation_sos_one_image_lr2e_5_ce_no_mse_ins_step0
-Preparing Dataset vlm_gym_match_equation_sos_celoss_no_mse_evalonce/vlm_gym_match_equation_sos_val
-[eval debug] first 3 batch fingerprints:
-  fp[0]: [{'data_indexes': [0], 'worker_id': 0, 'dataset_name': 'vlm_gym_match_equation_sos_celoss_no_mse_evalonce'}]
-  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_match_equation_sos_celoss_no_mse_evalonce'}]
-  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_match_equation_sos_celoss_no_mse_evalonce'}]
-ce_avg: 1.5381660461425781, mse_avg: 0.0
-base_dir is /dev/shm/models/checkpoints_vlm_gym_match_equation_sos_one_image_lr2e_5_ce_no_mse_ins/eval_used_rows, step_tag is checkpoints_vlm_gym_match_equation_sos_one_image_lr2e_5_ce_no_mse_ins_step500
-Preparing Dataset vlm_gym_match_equation_sos_celoss_no_mse_evalonce/vlm_gym_match_equation_sos_val
-[eval debug] first 3 batch fingerprints:
-  fp[0]: [{'data_indexes': [0], 'worker_id': 0, 'dataset_name': 'vlm_gym_match_equation_sos_celoss_no_mse_evalonce'}]
-  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_match_equation_sos_celoss_no_mse_evalonce'}]
-  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_match_equation_sos_celoss_no_mse_evalonce'}]
-ce_avg: 0.06237730756402016, mse_avg: 0.0
 wandb: Detected [huggingface_hub.inference] in use.
 wandb: Use W&B Weave for improved LLM call tracing. Install Weave with `pip install weave` then add `import weave` to the top of your script.
 wandb: For more information, check out the docs at: https://weave-docs.wandb.ai/
@@ -1252,6 +1082,176 @@ wandb: For more information, check out the docs at: https://weave-docs.wandb.ai/
 [2026-01-26 19:53:23] (step=0001071) Train Loss mse: 0.0000, Train Loss ce: 0.0635, Train Steps/Sec: 1.02,
 [2026-01-26 19:53:24] (step=0001072) Train Loss mse: 0.0000, Train Loss ce: 0.0730, Train Steps/Sec: 1.01,
 [2026-01-26 19:53:25] (step=0001073) Train Loss mse: 0.0000, Train Loss ce: 0.0489, Train Steps/Sec: 1.01,
+FullyShardedDataParallel(
+  (_fsdp_wrapped_module): Bagel(
+    (language_model): Qwen2ForCausalLM(
+      (model): Qwen2Model(
+        (embed_tokens): Embedding(152064, 3584)
+        (layers): ModuleList(
+          (0-27): 28 x FullyShardedDataParallel(
+            (_fsdp_wrapped_module): CheckpointWrapper(
+              (_checkpoint_wrapped_module): Qwen2MoTDecoderLayer(
+                (self_attn): PackedAttentionMoT(
+                  (q_proj): Linear(in_features=3584, out_features=3584, bias=True)
+                  (k_proj): Linear(in_features=3584, out_features=512, bias=True)
+                  (v_proj): Linear(in_features=3584, out_features=512, bias=True)
+                  (o_proj): Linear(in_features=3584, out_features=3584, bias=False)
+                  (q_norm): Qwen2RMSNorm((128,), eps=1e-06)
+                  (k_norm): Qwen2RMSNorm((128,), eps=1e-06)
+                  (q_norm_moe_gen): Qwen2RMSNorm((128,), eps=1e-06)
+                  (k_norm_moe_gen): Qwen2RMSNorm((128,), eps=1e-06)
+                  (q_proj_moe_gen): Linear(in_features=3584, out_features=3584, bias=True)
+                  (k_proj_moe_gen): Linear(in_features=3584, out_features=512, bias=True)
+                  (v_proj_moe_gen): Linear(in_features=3584, out_features=512, bias=True)
+                  (o_proj_moe_gen): Linear(in_features=3584, out_features=3584, bias=False)
+                )
+                (mlp): Qwen2MLP(
+                  (gate_proj): Linear(in_features=3584, out_features=18944, bias=False)
+                  (up_proj): Linear(in_features=3584, out_features=18944, bias=False)
+                  (down_proj): Linear(in_features=18944, out_features=3584, bias=False)
+                  (act_fn): SiLU()
+                )
+                (mlp_moe_gen): Qwen2MLP(
+                  (gate_proj): Linear(in_features=3584, out_features=18944, bias=False)
+                  (up_proj): Linear(in_features=3584, out_features=18944, bias=False)
+                  (down_proj): Linear(in_features=18944, out_features=3584, bias=False)
+                  (act_fn): SiLU()
+                )
+                (input_layernorm): Qwen2RMSNorm((3584,), eps=1e-06)
+                (input_layernorm_moe_gen): Qwen2RMSNorm((3584,), eps=1e-06)
+                (post_attention_layernorm): Qwen2RMSNorm((3584,), eps=1e-06)
+                (post_attention_layernorm_moe_gen): Qwen2RMSNorm((3584,), eps=1e-06)
+              )
+            )
+          )
+        )
+        (norm): Qwen2RMSNorm((3584,), eps=1e-06)
+        (norm_moe_gen): Qwen2RMSNorm((3584,), eps=1e-06)
+        (rotary_emb): Qwen2RotaryEmbedding()
+      )
+      (lm_head): Linear(in_features=3584, out_features=152064, bias=False)
+    )
+    (vit_model): SiglipVisionModel(
+      (vision_model): FullyShardedDataParallel(
+        (_fsdp_wrapped_module): SiglipVisionTransformer(
+          (embeddings): SiglipVisionEmbeddings(
+            (position_embedding): Embedding(4900, 1152)
+            (patch_embedding): Linear(in_features=588, out_features=1152, bias=True)
+          )
+          (encoder): SiglipEncoder(
+            (layers): ModuleList(
+              (0-25): 26 x FullyShardedDataParallel(
+                (_fsdp_wrapped_module): CheckpointWrapper(
+                  (_checkpoint_wrapped_module): SiglipEncoderLayer(
+                    (self_attn): SiglipFlashAttention2(
+                      (k_proj): Linear(in_features=1152, out_features=1152, bias=True)
+                      (v_proj): Linear(in_features=1152, out_features=1152, bias=True)
+                      (q_proj): Linear(in_features=1152, out_features=1152, bias=True)
+                      (out_proj): Linear(in_features=1152, out_features=1152, bias=True)
+                    )
+                    (layer_norm1): LayerNorm((1152,), eps=1e-06, elementwise_affine=True)
+                    (mlp): SiglipMLP(
+                      (activation_fn): PytorchGELUTanh()
+                      (fc1): Linear(in_features=1152, out_features=4304, bias=True)
+                      (fc2): Linear(in_features=4304, out_features=1152, bias=True)
+                    )
+                    (layer_norm2): LayerNorm((1152,), eps=1e-06, elementwise_affine=True)
+                  )
+                )
+              )
+            )
+          )
+          (post_layernorm): LayerNorm((1152,), eps=1e-06, elementwise_affine=True)
+        )
+      )
+    )
+    (connector): FullyShardedDataParallel(
+      (_fsdp_wrapped_module): CheckpointWrapper(
+        (_checkpoint_wrapped_module): MLPconnector(
+          (activation_fn): PytorchGELUTanh()
+          (fc1): Linear(in_features=1152, out_features=3584, bias=True)
+          (fc2): Linear(in_features=3584, out_features=3584, bias=True)
+        )
+      )
+    )
+    (vit_pos_embed): FullyShardedDataParallel(
+      (_fsdp_wrapped_module): PositionEmbedding()
+    )
+  )
+)
+_flat_param True
+language_model.model.layers.0._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+language_model.model.layers.1._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+language_model.model.layers.2._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+language_model.model.layers.3._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+language_model.model.layers.4._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+language_model.model.layers.5._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+language_model.model.layers.6._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+language_model.model.layers.7._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+language_model.model.layers.8._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+language_model.model.layers.9._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+language_model.model.layers.10._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+language_model.model.layers.11._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+language_model.model.layers.12._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+language_model.model.layers.13._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+language_model.model.layers.14._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+language_model.model.layers.15._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+language_model.model.layers.16._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+language_model.model.layers.17._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+language_model.model.layers.18._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+language_model.model.layers.19._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+language_model.model.layers.20._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+language_model.model.layers.21._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+language_model.model.layers.22._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+language_model.model.layers.23._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+language_model.model.layers.24._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+language_model.model.layers.25._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+language_model.model.layers.26._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+language_model.model.layers.27._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+vit_model.vision_model._fsdp_wrapped_module._flat_param True
+vit_model.vision_model._fsdp_wrapped_module.encoder.layers.0._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+vit_model.vision_model._fsdp_wrapped_module.encoder.layers.1._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+vit_model.vision_model._fsdp_wrapped_module.encoder.layers.2._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+vit_model.vision_model._fsdp_wrapped_module.encoder.layers.3._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+vit_model.vision_model._fsdp_wrapped_module.encoder.layers.4._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+vit_model.vision_model._fsdp_wrapped_module.encoder.layers.5._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+vit_model.vision_model._fsdp_wrapped_module.encoder.layers.6._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+vit_model.vision_model._fsdp_wrapped_module.encoder.layers.7._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+vit_model.vision_model._fsdp_wrapped_module.encoder.layers.8._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+vit_model.vision_model._fsdp_wrapped_module.encoder.layers.9._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+vit_model.vision_model._fsdp_wrapped_module.encoder.layers.10._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+vit_model.vision_model._fsdp_wrapped_module.encoder.layers.11._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+vit_model.vision_model._fsdp_wrapped_module.encoder.layers.12._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+vit_model.vision_model._fsdp_wrapped_module.encoder.layers.13._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+vit_model.vision_model._fsdp_wrapped_module.encoder.layers.14._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+vit_model.vision_model._fsdp_wrapped_module.encoder.layers.15._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+vit_model.vision_model._fsdp_wrapped_module.encoder.layers.16._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+vit_model.vision_model._fsdp_wrapped_module.encoder.layers.17._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+vit_model.vision_model._fsdp_wrapped_module.encoder.layers.18._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+vit_model.vision_model._fsdp_wrapped_module.encoder.layers.19._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+vit_model.vision_model._fsdp_wrapped_module.encoder.layers.20._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+vit_model.vision_model._fsdp_wrapped_module.encoder.layers.21._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+vit_model.vision_model._fsdp_wrapped_module.encoder.layers.22._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+vit_model.vision_model._fsdp_wrapped_module.encoder.layers.23._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+vit_model.vision_model._fsdp_wrapped_module.encoder.layers.24._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+vit_model.vision_model._fsdp_wrapped_module.encoder.layers.25._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+connector._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+vit_pos_embed._fsdp_wrapped_module._flat_param False
+Preparing Dataset vlm_gym_match_equation_sos_celoss_no_mse/vlm_gym_match_equation_sos_train
+base_dir is /dev/shm/models/checkpoints_vlm_gym_match_equation_sos_one_image_lr2e_5_ce_no_mse_ins/eval_used_rows, step_tag is checkpoints_vlm_gym_match_equation_sos_one_image_lr2e_5_ce_no_mse_ins_step0
+Preparing Dataset vlm_gym_match_equation_sos_celoss_no_mse_evalonce/vlm_gym_match_equation_sos_val
+[eval debug] first 3 batch fingerprints:
+  fp[0]: [{'data_indexes': [0], 'worker_id': 0, 'dataset_name': 'vlm_gym_match_equation_sos_celoss_no_mse_evalonce'}]
+  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_match_equation_sos_celoss_no_mse_evalonce'}]
+  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_match_equation_sos_celoss_no_mse_evalonce'}]
+ce_avg: 1.5381660461425781, mse_avg: 0.0
+base_dir is /dev/shm/models/checkpoints_vlm_gym_match_equation_sos_one_image_lr2e_5_ce_no_mse_ins/eval_used_rows, step_tag is checkpoints_vlm_gym_match_equation_sos_one_image_lr2e_5_ce_no_mse_ins_step500
+Preparing Dataset vlm_gym_match_equation_sos_celoss_no_mse_evalonce/vlm_gym_match_equation_sos_val
+[eval debug] first 3 batch fingerprints:
+  fp[0]: [{'data_indexes': [0], 'worker_id': 0, 'dataset_name': 'vlm_gym_match_equation_sos_celoss_no_mse_evalonce'}]
+  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_match_equation_sos_celoss_no_mse_evalonce'}]
+  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_match_equation_sos_celoss_no_mse_evalonce'}]
+ce_avg: 0.06237730756402016, mse_avg: 0.0
 [2026-01-26 19:53:26] (step=0001074) Train Loss mse: 0.0000, Train Loss ce: 0.0603, Train Steps/Sec: 1.02,
 [2026-01-26 19:53:27] (step=0001075) Train Loss mse: 0.0000, Train Loss ce: 0.0678, Train Steps/Sec: 0.76,
 [2026-01-26 19:53:28] (step=0001076) Train Loss mse: 0.0000, Train Loss ce: 0.0544, Train Steps/Sec: 0.82,
@@ -1289,27 +1289,6 @@ wandb: For more information, check out the docs at: https://weave-docs.wandb.ai/
 [2026-01-26 19:54:02] (step=0001108) Train Loss mse: 0.0000, Train Loss ce: 0.0604, Train Steps/Sec: 1.01,
 [2026-01-26 19:54:03] (step=0001109) Train Loss mse: 0.0000, Train Loss ce: 0.0570, Train Steps/Sec: 0.81,
 [2026-01-26 19:54:04] (step=0001110) Train Loss mse: 0.0000, Train Loss ce: 0.0559, Train Steps/Sec: 1.01,
-base_dir is /dev/shm/models/checkpoints_vlm_gym_match_equation_sos_one_image_lr2e_5_ce_no_mse_ins/eval_used_rows, step_tag is checkpoints_vlm_gym_match_equation_sos_one_image_lr2e_5_ce_no_mse_ins_step1500
-Preparing Dataset vlm_gym_match_equation_sos_celoss_no_mse_evalonce/vlm_gym_match_equation_sos_val
-[eval debug] first 3 batch fingerprints:
-  fp[0]: [{'data_indexes': [0], 'worker_id': 0, 'dataset_name': 'vlm_gym_match_equation_sos_celoss_no_mse_evalonce'}]
-  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_match_equation_sos_celoss_no_mse_evalonce'}]
-  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_match_equation_sos_celoss_no_mse_evalonce'}]
-ce_avg: 0.05330345034599304, mse_avg: 0.0
-base_dir is /dev/shm/models/checkpoints_vlm_gym_match_equation_sos_one_image_lr2e_5_ce_no_mse_ins/eval_used_rows, step_tag is checkpoints_vlm_gym_match_equation_sos_one_image_lr2e_5_ce_no_mse_ins_step2000
-Preparing Dataset vlm_gym_match_equation_sos_celoss_no_mse_evalonce/vlm_gym_match_equation_sos_val
-[eval debug] first 3 batch fingerprints:
-  fp[0]: [{'data_indexes': [0], 'worker_id': 0, 'dataset_name': 'vlm_gym_match_equation_sos_celoss_no_mse_evalonce'}]
-  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_match_equation_sos_celoss_no_mse_evalonce'}]
-  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_match_equation_sos_celoss_no_mse_evalonce'}]
-ce_avg: 0.05127852410078049, mse_avg: 0.0
-base_dir is /dev/shm/models/checkpoints_vlm_gym_match_equation_sos_one_image_lr2e_5_ce_no_mse_ins/eval_used_rows, step_tag is checkpoints_vlm_gym_match_equation_sos_one_image_lr2e_5_ce_no_mse_ins_step2500
-Preparing Dataset vlm_gym_match_equation_sos_celoss_no_mse_evalonce/vlm_gym_match_equation_sos_val
-[eval debug] first 3 batch fingerprints:
-  fp[0]: [{'data_indexes': [0], 'worker_id': 0, 'dataset_name': 'vlm_gym_match_equation_sos_celoss_no_mse_evalonce'}]
-  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_match_equation_sos_celoss_no_mse_evalonce'}]
-  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_match_equation_sos_celoss_no_mse_evalonce'}]
-ce_avg: 0.05304804816842079, mse_avg: 0.0
 [2026-01-26 19:54:05] (step=0001111) Train Loss mse: 0.0000, Train Loss ce: 0.0425, Train Steps/Sec: 1.01,
 [2026-01-26 19:54:06] (step=0001112) Train Loss mse: 0.0000, Train Loss ce: 0.0792, Train Steps/Sec: 1.01,
 [2026-01-26 19:54:08] (step=0001113) Train Loss mse: 0.0000, Train Loss ce: 0.0488, Train Steps/Sec: 0.76,
@@ -2764,6 +2743,27 @@ ce_avg: 0.05304804816842079, mse_avg: 0.0
 [2026-01-26 20:19:55] (step=0002562) Train Loss mse: 0.0000, Train Loss ce: 0.0472, Train Steps/Sec: 1.02,
 [2026-01-26 20:19:56] (step=0002563) Train Loss mse: 0.0000, Train Loss ce: 0.0490, Train Steps/Sec: 1.02,
 [2026-01-26 20:19:57] (step=0002564) Train Loss mse: 0.0000, Train Loss ce: 0.0634, Train Steps/Sec: 1.02,
+base_dir is /dev/shm/models/checkpoints_vlm_gym_match_equation_sos_one_image_lr2e_5_ce_no_mse_ins/eval_used_rows, step_tag is checkpoints_vlm_gym_match_equation_sos_one_image_lr2e_5_ce_no_mse_ins_step1000
+Preparing Dataset vlm_gym_match_equation_sos_celoss_no_mse_evalonce/vlm_gym_match_equation_sos_val
+[eval debug] first 3 batch fingerprints:
+  fp[0]: [{'data_indexes': [0], 'worker_id': 0, 'dataset_name': 'vlm_gym_match_equation_sos_celoss_no_mse_evalonce'}]
+  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_match_equation_sos_celoss_no_mse_evalonce'}]
+  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_match_equation_sos_celoss_no_mse_evalonce'}]
+ce_avg: 0.05690578743815422, mse_avg: 0.0
+base_dir is /dev/shm/models/checkpoints_vlm_gym_match_equation_sos_one_image_lr2e_5_ce_no_mse_ins/eval_used_rows, step_tag is checkpoints_vlm_gym_match_equation_sos_one_image_lr2e_5_ce_no_mse_ins_step1500
+Preparing Dataset vlm_gym_match_equation_sos_celoss_no_mse_evalonce/vlm_gym_match_equation_sos_val
+[eval debug] first 3 batch fingerprints:
+  fp[0]: [{'data_indexes': [0], 'worker_id': 0, 'dataset_name': 'vlm_gym_match_equation_sos_celoss_no_mse_evalonce'}]
+  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_match_equation_sos_celoss_no_mse_evalonce'}]
+  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_match_equation_sos_celoss_no_mse_evalonce'}]
+ce_avg: 0.05330345034599304, mse_avg: 0.0
+base_dir is /dev/shm/models/checkpoints_vlm_gym_match_equation_sos_one_image_lr2e_5_ce_no_mse_ins/eval_used_rows, step_tag is checkpoints_vlm_gym_match_equation_sos_one_image_lr2e_5_ce_no_mse_ins_step2000
+Preparing Dataset vlm_gym_match_equation_sos_celoss_no_mse_evalonce/vlm_gym_match_equation_sos_val
+[eval debug] first 3 batch fingerprints:
+  fp[0]: [{'data_indexes': [0], 'worker_id': 0, 'dataset_name': 'vlm_gym_match_equation_sos_celoss_no_mse_evalonce'}]
+  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_match_equation_sos_celoss_no_mse_evalonce'}]
+  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_match_equation_sos_celoss_no_mse_evalonce'}]
+ce_avg: 0.05127852410078049, mse_avg: 0.0
 [2026-01-26 20:19:58] (step=0002565) Train Loss mse: 0.0000, Train Loss ce: 0.0394, Train Steps/Sec: 1.02,
 [2026-01-26 20:19:59] (step=0002566) Train Loss mse: 0.0000, Train Loss ce: 0.0322, Train Steps/Sec: 1.01,
 [2026-01-26 20:20:00] (step=0002567) Train Loss mse: 0.0000, Train Loss ce: 0.0585, Train Steps/Sec: 0.76,
@@ -2813,20 +2813,6 @@ ce_avg: 0.05304804816842079, mse_avg: 0.0
 [2026-01-26 20:20:47] (step=0002611) Train Loss mse: 0.0000, Train Loss ce: 0.0525, Train Steps/Sec: 1.01,
 [2026-01-26 20:20:48] (step=0002612) Train Loss mse: 0.0000, Train Loss ce: 0.0433, Train Steps/Sec: 0.99,
 [2026-01-26 20:20:49] (step=0002613) Train Loss mse: 0.0000, Train Loss ce: 0.0471, Train Steps/Sec: 0.76,
-base_dir is /dev/shm/models/checkpoints_vlm_gym_match_equation_sos_one_image_lr2e_5_ce_no_mse_ins/eval_used_rows, step_tag is checkpoints_vlm_gym_match_equation_sos_one_image_lr2e_5_ce_no_mse_ins_step3000
-Preparing Dataset vlm_gym_match_equation_sos_celoss_no_mse_evalonce/vlm_gym_match_equation_sos_val
-[eval debug] first 3 batch fingerprints:
-  fp[0]: [{'data_indexes': [0], 'worker_id': 0, 'dataset_name': 'vlm_gym_match_equation_sos_celoss_no_mse_evalonce'}]
-  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_match_equation_sos_celoss_no_mse_evalonce'}]
-  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_match_equation_sos_celoss_no_mse_evalonce'}]
-ce_avg: 0.058707185089588165, mse_avg: 0.0
-base_dir is /dev/shm/models/checkpoints_vlm_gym_match_equation_sos_one_image_lr2e_5_ce_no_mse_ins/eval_used_rows, step_tag is checkpoints_vlm_gym_match_equation_sos_one_image_lr2e_5_ce_no_mse_ins_step3500
-Preparing Dataset vlm_gym_match_equation_sos_celoss_no_mse_evalonce/vlm_gym_match_equation_sos_val
-[eval debug] first 3 batch fingerprints:
-  fp[0]: [{'data_indexes': [0], 'worker_id': 0, 'dataset_name': 'vlm_gym_match_equation_sos_celoss_no_mse_evalonce'}]
-  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_match_equation_sos_celoss_no_mse_evalonce'}]
-  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_match_equation_sos_celoss_no_mse_evalonce'}]
-ce_avg: 0.10416685044765472, mse_avg: 0.0
 [2026-01-26 20:20:50] (step=0002614) Train Loss mse: 0.0000, Train Loss ce: 0.0586, Train Steps/Sec: 1.01,
 [2026-01-26 20:20:51] (step=0002615) Train Loss mse: 0.0000, Train Loss ce: 0.0357, Train Steps/Sec: 1.01,
 [2026-01-26 20:20:52] (step=0002616) Train Loss mse: 0.0000, Train Loss ce: 0.0313, Train Steps/Sec: 1.00,
@@ -3800,6 +3786,48 @@ ce_avg: 0.10416685044765472, mse_avg: 0.0
 [2026-01-26 20:38:10] (step=0003584) Train Loss mse: 0.0000, Train Loss ce: 0.0436, Train Steps/Sec: 1.01,
 [2026-01-26 20:38:11] (step=0003585) Train Loss mse: 0.0000, Train Loss ce: 0.0705, Train Steps/Sec: 1.01,
 [2026-01-26 20:38:12] (step=0003586) Train Loss mse: 0.0000, Train Loss ce: 0.0235, Train Steps/Sec: 0.81,
+base_dir is /dev/shm/models/checkpoints_vlm_gym_match_equation_sos_one_image_lr2e_5_ce_no_mse_ins/eval_used_rows, step_tag is checkpoints_vlm_gym_match_equation_sos_one_image_lr2e_5_ce_no_mse_ins_step2500
+Preparing Dataset vlm_gym_match_equation_sos_celoss_no_mse_evalonce/vlm_gym_match_equation_sos_val
+[eval debug] first 3 batch fingerprints:
+  fp[0]: [{'data_indexes': [0], 'worker_id': 0, 'dataset_name': 'vlm_gym_match_equation_sos_celoss_no_mse_evalonce'}]
+  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_match_equation_sos_celoss_no_mse_evalonce'}]
+  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_match_equation_sos_celoss_no_mse_evalonce'}]
+ce_avg: 0.05304804816842079, mse_avg: 0.0
+base_dir is /dev/shm/models/checkpoints_vlm_gym_match_equation_sos_one_image_lr2e_5_ce_no_mse_ins/eval_used_rows, step_tag is checkpoints_vlm_gym_match_equation_sos_one_image_lr2e_5_ce_no_mse_ins_step3000
+Preparing Dataset vlm_gym_match_equation_sos_celoss_no_mse_evalonce/vlm_gym_match_equation_sos_val
+[eval debug] first 3 batch fingerprints:
+  fp[0]: [{'data_indexes': [0], 'worker_id': 0, 'dataset_name': 'vlm_gym_match_equation_sos_celoss_no_mse_evalonce'}]
+  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_match_equation_sos_celoss_no_mse_evalonce'}]
+  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_match_equation_sos_celoss_no_mse_evalonce'}]
+ce_avg: 0.058707185089588165, mse_avg: 0.0
+base_dir is /dev/shm/models/checkpoints_vlm_gym_match_equation_sos_one_image_lr2e_5_ce_no_mse_ins/eval_used_rows, step_tag is checkpoints_vlm_gym_match_equation_sos_one_image_lr2e_5_ce_no_mse_ins_step3500
+Preparing Dataset vlm_gym_match_equation_sos_celoss_no_mse_evalonce/vlm_gym_match_equation_sos_val
+[eval debug] first 3 batch fingerprints:
+  fp[0]: [{'data_indexes': [0], 'worker_id': 0, 'dataset_name': 'vlm_gym_match_equation_sos_celoss_no_mse_evalonce'}]
+  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_match_equation_sos_celoss_no_mse_evalonce'}]
+  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_match_equation_sos_celoss_no_mse_evalonce'}]
+ce_avg: 0.10416685044765472, mse_avg: 0.0
+base_dir is /dev/shm/models/checkpoints_vlm_gym_match_equation_sos_one_image_lr2e_5_ce_no_mse_ins/eval_used_rows, step_tag is checkpoints_vlm_gym_match_equation_sos_one_image_lr2e_5_ce_no_mse_ins_step4000
+Preparing Dataset vlm_gym_match_equation_sos_celoss_no_mse_evalonce/vlm_gym_match_equation_sos_val
+[eval debug] first 3 batch fingerprints:
+  fp[0]: [{'data_indexes': [0], 'worker_id': 0, 'dataset_name': 'vlm_gym_match_equation_sos_celoss_no_mse_evalonce'}]
+  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_match_equation_sos_celoss_no_mse_evalonce'}]
+  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_match_equation_sos_celoss_no_mse_evalonce'}]
+ce_avg: 0.09295430034399033, mse_avg: 0.0
+base_dir is /dev/shm/models/checkpoints_vlm_gym_match_equation_sos_one_image_lr2e_5_ce_no_mse_ins/eval_used_rows, step_tag is checkpoints_vlm_gym_match_equation_sos_one_image_lr2e_5_ce_no_mse_ins_step4500
+Preparing Dataset vlm_gym_match_equation_sos_celoss_no_mse_evalonce/vlm_gym_match_equation_sos_val
+[eval debug] first 3 batch fingerprints:
+  fp[0]: [{'data_indexes': [0], 'worker_id': 0, 'dataset_name': 'vlm_gym_match_equation_sos_celoss_no_mse_evalonce'}]
+  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_match_equation_sos_celoss_no_mse_evalonce'}]
+  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_match_equation_sos_celoss_no_mse_evalonce'}]
+ce_avg: 0.09719827771186829, mse_avg: 0.0
+base_dir is /dev/shm/models/checkpoints_vlm_gym_match_equation_sos_one_image_lr2e_5_ce_no_mse_ins/eval_used_rows, step_tag is checkpoints_vlm_gym_match_equation_sos_one_image_lr2e_5_ce_no_mse_ins_step5000
+Preparing Dataset vlm_gym_match_equation_sos_celoss_no_mse_evalonce/vlm_gym_match_equation_sos_val
+[eval debug] first 3 batch fingerprints:
+  fp[0]: [{'data_indexes': [0], 'worker_id': 0, 'dataset_name': 'vlm_gym_match_equation_sos_celoss_no_mse_evalonce'}]
+  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_match_equation_sos_celoss_no_mse_evalonce'}]
+  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_match_equation_sos_celoss_no_mse_evalonce'}]
+ce_avg: 0.10010946542024612, mse_avg: 0.0
 [2026-01-26 20:38:13] (step=0003587) Train Loss mse: 0.0000, Train Loss ce: 0.0596, Train Steps/Sec: 1.01,
 [2026-01-26 20:38:14] (step=0003588) Train Loss mse: 0.0000, Train Loss ce: 0.0493, Train Steps/Sec: 1.01,
 [2026-01-26 20:38:15] (step=0003589) Train Loss mse: 0.0000, Train Loss ce: 0.0348, Train Steps/Sec: 1.01,
@@ -3907,27 +3935,27 @@ ce_avg: 0.10416685044765472, mse_avg: 0.0
 [2026-01-26 20:40:04] (step=0003691) Train Loss mse: 0.0000, Train Loss ce: 0.0398, Train Steps/Sec: 1.00,
 [2026-01-26 20:40:05] (step=0003692) Train Loss mse: 0.0000, Train Loss ce: 0.0272, Train Steps/Sec: 1.02,
 [2026-01-26 20:40:06] (step=0003693) Train Loss mse: 0.0000, Train Loss ce: 0.0313, Train Steps/Sec: 1.02,
-base_dir is /dev/shm/models/checkpoints_vlm_gym_match_equation_sos_one_image_lr2e_5_ce_no_mse_ins/eval_used_rows, step_tag is checkpoints_vlm_gym_match_equation_sos_one_image_lr2e_5_ce_no_mse_ins_step4000
-Preparing Dataset vlm_gym_match_equation_sos_celoss_no_mse_evalonce/vlm_gym_match_equation_sos_val
-[eval debug] first 3 batch fingerprints:
-  fp[0]: [{'data_indexes': [0], 'worker_id': 0, 'dataset_name': 'vlm_gym_match_equation_sos_celoss_no_mse_evalonce'}]
-  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_match_equation_sos_celoss_no_mse_evalonce'}]
-  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_match_equation_sos_celoss_no_mse_evalonce'}]
-ce_avg: 0.09295430034399033, mse_avg: 0.0
-base_dir is /dev/shm/models/checkpoints_vlm_gym_match_equation_sos_one_image_lr2e_5_ce_no_mse_ins/eval_used_rows, step_tag is checkpoints_vlm_gym_match_equation_sos_one_image_lr2e_5_ce_no_mse_ins_step4500
-Preparing Dataset vlm_gym_match_equation_sos_celoss_no_mse_evalonce/vlm_gym_match_equation_sos_val
-[eval debug] first 3 batch fingerprints:
-  fp[0]: [{'data_indexes': [0], 'worker_id': 0, 'dataset_name': 'vlm_gym_match_equation_sos_celoss_no_mse_evalonce'}]
-  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_match_equation_sos_celoss_no_mse_evalonce'}]
-  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_match_equation_sos_celoss_no_mse_evalonce'}]
-ce_avg: 0.09719827771186829, mse_avg: 0.0
-base_dir is /dev/shm/models/checkpoints_vlm_gym_match_equation_sos_one_image_lr2e_5_ce_no_mse_ins/eval_used_rows, step_tag is checkpoints_vlm_gym_match_equation_sos_one_image_lr2e_5_ce_no_mse_ins_step5000
-Preparing Dataset vlm_gym_match_equation_sos_celoss_no_mse_evalonce/vlm_gym_match_equation_sos_val
-[eval debug] first 3 batch fingerprints:
-  fp[0]: [{'data_indexes': [0], 'worker_id': 0, 'dataset_name': 'vlm_gym_match_equation_sos_celoss_no_mse_evalonce'}]
-  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_match_equation_sos_celoss_no_mse_evalonce'}]
-  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_match_equation_sos_celoss_no_mse_evalonce'}]
-ce_avg: 0.10010946542024612, mse_avg: 0.0
+[2026-01-26 20:40:07] (step=0003694) Train Loss mse: 0.0000, Train Loss ce: 0.0528, Train Steps/Sec: 1.02,
+[2026-01-26 20:40:08] (step=0003695) Train Loss mse: 0.0000, Train Loss ce: 0.0489, Train Steps/Sec: 1.02,
+[2026-01-26 20:40:09] (step=0003696) Train Loss mse: 0.0000, Train Loss ce: 0.0267, Train Steps/Sec: 0.79,
+[2026-01-26 20:40:10] (step=0003697) Train Loss mse: 0.0000, Train Loss ce: 0.0365, Train Steps/Sec: 0.74,
+[2026-01-26 20:40:11] (step=0003698) Train Loss mse: 0.0000, Train Loss ce: 0.0334, Train Steps/Sec: 1.01,
+[2026-01-26 20:40:13] (step=0003699) Train Loss mse: 0.0000, Train Loss ce: 0.0528, Train Steps/Sec: 0.97,
+[2026-01-26 20:40:13] (step=0003700) Train Loss mse: 0.0000, Train Loss ce: 0.0260, Train Steps/Sec: 1.01,
+[2026-01-26 20:40:14] (step=0003701) Train Loss mse: 0.0000, Train Loss ce: 0.0514, Train Steps/Sec: 1.01,
+[2026-01-26 20:40:15] (step=0003702) Train Loss mse: 0.0000, Train Loss ce: 0.0590, Train Steps/Sec: 1.01,
+[2026-01-26 20:40:16] (step=0003703) Train Loss mse: 0.0000, Train Loss ce: 0.0275, Train Steps/Sec: 1.01,
+[2026-01-26 20:40:18] (step=0003704) Train Loss mse: 0.0000, Train Loss ce: 0.0255, Train Steps/Sec: 0.80,
+[2026-01-26 20:40:19] (step=0003705) Train Loss mse: 0.0000, Train Loss ce: 0.0308, Train Steps/Sec: 0.74,
+[2026-01-26 20:40:20] (step=0003706) Train Loss mse: 0.0000, Train Loss ce: 0.0407, Train Steps/Sec: 1.00,
+[2026-01-26 20:40:21] (step=0003707) Train Loss mse: 0.0000, Train Loss ce: 0.0359, Train Steps/Sec: 1.00,
+[2026-01-26 20:40:22] (step=0003708) Train Loss mse: 0.0000, Train Loss ce: 0.0362, Train Steps/Sec: 1.01,
+[2026-01-26 20:40:23] (step=0003709) Train Loss mse: 0.0000, Train Loss ce: 0.0446, Train Steps/Sec: 1.02,
+[2026-01-26 20:40:24] (step=0003710) Train Loss mse: 0.0000, Train Loss ce: 0.0333, Train Steps/Sec: 1.01,
+[2026-01-26 20:40:25] (step=0003711) Train Loss mse: 0.0000, Train Loss ce: 0.0461, Train Steps/Sec: 0.80,
+[2026-01-26 20:40:27] (step=0003712) Train Loss mse: 0.0000, Train Loss ce: 0.0473, Train Steps/Sec: 0.75,
+[2026-01-26 20:40:28] (step=0003713) Train Loss mse: 0.0000, Train Loss ce: 0.0338, Train Steps/Sec: 1.00,
+[2026-01-26 20:40:29] (step=0003714) Train Loss mse: 0.0000, Train Loss ce: 0.0298, Train Steps/Sec: 1.00,
 [2026-01-26 20:40:30] (step=0003715) Train Loss mse: 0.0000, Train Loss ce: 0.0235, Train Steps/Sec: 1.01,
 [2026-01-26 20:40:31] (step=0003716) Train Loss mse: 0.0000, Train Loss ce: 0.0562, Train Steps/Sec: 1.01,
 [2026-01-26 20:40:32] (step=0003717) Train Loss mse: 0.0000, Train Loss ce: 0.0530, Train Steps/Sec: 0.99,