Upload checkpoints_vlm_gym_patch_reassembly_alt_one_image_lr2e_5_ce_ins/checkpoints_vlm_gym_patch_reassembly_alt_one_image_lr2e_5_ce_ins

checkpoints_vlm_gym_patch_reassembly_alt_one_image_lr2e_5_ce_ins/checkpoints_vlm_gym_patch_reassembly_alt_one_image_lr2e_5_ce_ins/0005000/ema.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:73b567c0952d33306f37941298efd623616a326df8cd00072efe55db64f1592d
+size 58429204680

checkpoints_vlm_gym_patch_reassembly_alt_one_image_lr2e_5_ce_ins/checkpoints_vlm_gym_patch_reassembly_alt_one_image_lr2e_5_ce_ins/wandb/offline-run-20260129_223919-checkpoints_vlm_gym_patch_reassembly_alt_one_image_lr2e_5_ce_ins-run0/files/config.yaml

@@ -0,0 +1,457 @@
+wandb_version: 1
+
+_wandb:
+  desc: null
+  value:
+    python_version: 3.11.10
+    cli_version: 0.23.1
+    framework: huggingface
+    huggingface_version: 4.49.0
+    is_jupyter_run: false
+    is_kaggle_kernel: false
+    start_time: 1769726360
+    t:
+      1:
+      - 1
+      - 5
+      - 11
+      - 41
+      - 49
+      - 53
+      - 71
+      - 105
+      2:
+      - 1
+      - 5
+      - 11
+      - 41
+      - 49
+      - 53
+      - 71
+      - 105
+      3:
+      - 2
+      - 4
+      - 13
+      - 14
+      - 37
+      - 42
+      - 61
+      4: 3.11.10
+      5: 0.23.1
+      6: 4.49.0
+      13: linux-x86_64
+    e:
+      pfbo5dtoypojh5ejq637djdym70xh66v:
+        os: Linux-6.6.93+-x86_64-with-glibc2.35
+        python: CPython 3.11.10
+        started_at: '2026-01-29T22:39:19.822640Z'
+        args:
+        - --dataset_config_file
+        - ./data/configs/vlm_gym_patch_reassembly_alt_train_celoss.yaml
+        - --eval_dataset_config_file
+        - ./data/configs/vlm_gym_patch_reassembly_alt_eval_celoss.yaml
+        - --viz_dataset_config_file
+        - ./data/configs/vlm_gym_patch_reassembly_alt_eval_celoss.yaml
+        - --inference_hash_file
+        - /home/clouduser/Code/Github/launch_new/hashes_test_set_v10.json
+        - --task_name
+        - patch_reassembly_v5
+        - --instructions_dir
+        - ./data/instructions
+        - --train_data_dir
+        - /home/clouduser/Code/data/gym/patch_reassembly_alt_v5/train/
+        - --train_jsonl_path
+        - /home/clouduser/Code/data/gym/patch_reassembly_alt_v5/train/
+        - --eval_data_dir
+        - /home/clouduser/Code/data/gym/patch_reassembly_alt_v5/val/
+        - --eval_jsonl_path
+        - /home/clouduser/Code/data/gym/patch_reassembly_alt_v5/val/
+        - --model_path
+        - /home/clouduser/Code/Models/BAGEL-7B-MoT
+        - --layer_module
+        - Qwen2MoTDecoderLayer
+        - --max_latent_size
+        - '64'
+        - --resume-from
+        - /home/clouduser/Code/Models/BAGEL-7B-MoT
+        - --finetune_from_hf
+        - 'True'
+        - --auto_resume
+        - 'False'
+        - --resume-model-only
+        - 'True'
+        - --finetune-from-ema
+        - 'True'
+        - --log_every
+        - '1'
+        - --lr
+        - 2e-5
+        - --warmup_steps
+        - '300'
+        - --lr_scheduler
+        - cosine
+        - --num_worker
+        - '1'
+        - --expected_num_tokens
+        - '30000'
+        - --max_num_tokens
+        - '30000'
+        - --max_num_tokens_per_sample
+        - '30000'
+        - --visual_und
+        - 'True'
+        - --save_every
+        - '2500'
+        - --total_steps
+        - '5000'
+        - --text_cond_dropout_prob
+        - '0.0'
+        - --vae_cond_dropout_prob
+        - '0.3'
+        - --vit_cond_dropout_prob
+        - '0.0'
+        - --ema
+        - '0.993'
+        - --checkpoint_dir
+        - /dev/shm/models/checkpoints_vlm_gym_patch_reassembly_alt_one_image_lr2e_5_ce_ins
+        - --wandb_project
+        - bagel
+        - --wandb_name
+        - checkpoints_vlm_gym_patch_reassembly_alt_one_image_lr2e_5_ce_ins
+        - --wandb_dir
+        - /dev/shm/models/checkpoints_vlm_gym_patch_reassembly_alt_one_image_lr2e_5_ce_ins
+        - --wandb_offline
+        - 'True'
+        program: /home/clouduser/Code/Github/unified_world_model/train/pretrain_unified_navit.py
+        code_path: train/pretrain_unified_navit.py
+        code_path_local: train/pretrain_unified_navit.py
+        git:
+          remote_url: https://github.com/para-lost/unified_world_model
+          commit: 8d7b26b7e552fc87b592cf3be94d93be7aeca2a9
+        root: /dev/shm/models/checkpoints_vlm_gym_patch_reassembly_alt_one_image_lr2e_5_ce_ins
+        host: junyizhang-launch-new-226785731-1-0
+        executable: /opt/conda/bin/python3.11
+        cpu_count: 48
+        cpu_count_logical: 96
+        gpu_type: NVIDIA A100-SXM4-80GB
+        gpu_count: 8
+        disk:
+          /:
+            total: '1052461830144'
+            used: '179558117376'
+        memory:
+          total: '1437332606976'
+        gpu_nvidia:
+        - name: NVIDIA A100-SXM4-80GB
+          memory_total: '85899345920'
+          cuda_cores: 6912
+          architecture: Ampere
+          uuid: GPU-871da72d-3f3a-ff37-fc78-177139a0eeab
+        - name: NVIDIA A100-SXM4-80GB
+          memory_total: '85899345920'
+          cuda_cores: 6912
+          architecture: Ampere
+          uuid: GPU-68b93783-6768-a263-778d-fc8d1d8c5cec
+        - name: NVIDIA A100-SXM4-80GB
+          memory_total: '85899345920'
+          cuda_cores: 6912
+          architecture: Ampere
+          uuid: GPU-fb1c8d7d-0ecc-acab-6ba2-def6b47110f7
+        - name: NVIDIA A100-SXM4-80GB
+          memory_total: '85899345920'
+          cuda_cores: 6912
+          architecture: Ampere
+          uuid: GPU-0bac335f-a2a0-f393-7ac3-edbd8ca00890
+        - name: NVIDIA A100-SXM4-80GB
+          memory_total: '85899345920'
+          cuda_cores: 6912
+          architecture: Ampere
+          uuid: GPU-235c5877-2112-1bf2-5697-9a6bb2c1bdc6
+        - name: NVIDIA A100-SXM4-80GB
+          memory_total: '85899345920'
+          cuda_cores: 6912
+          architecture: Ampere
+          uuid: GPU-53b3bf28-8aed-e7f3-3dd3-92d1557f1088
+        - name: NVIDIA A100-SXM4-80GB
+          memory_total: '85899345920'
+          cuda_cores: 6912
+          architecture: Ampere
+          uuid: GPU-1b085487-7123-2309-c8da-1e209c198264
+        - name: NVIDIA A100-SXM4-80GB
+          memory_total: '85899345920'
+          cuda_cores: 6912
+          architecture: Ampere
+          uuid: GPU-e2dcec15-6e8e-e6a6-5845-bbf312b788d7
+        cuda_version: '12.2'
+        writer_id: pfbo5dtoypojh5ejq637djdym70xh66v
+visual_gen:
+  desc: null
+  value: true
+visual_und:
+  desc: null
+  value: true
+results_dir:
+  desc: null
+  value: results
+checkpoint_dir:
+  desc: null
+  value: /dev/shm/models/checkpoints_vlm_gym_patch_reassembly_alt_one_image_lr2e_5_ce_ins
+wandb_project:
+  desc: null
+  value: bagel
+wandb_name:
+  desc: null
+  value: checkpoints_vlm_gym_patch_reassembly_alt_one_image_lr2e_5_ce_ins
+wandb_runid:
+  desc: null
+  value: '0'
+wandb_resume:
+  desc: null
+  value: allow
+wandb_offline:
+  desc: null
+  value: true
+wandb_dir:
+  desc: null
+  value: /dev/shm/models/checkpoints_vlm_gym_patch_reassembly_alt_one_image_lr2e_5_ce_ins
+global_seed:
+  desc: null
+  value: 4396
+auto_resume:
+  desc: null
+  value: false
+resume_from:
+  desc: null
+  value: /home/clouduser/Code/Models/BAGEL-7B-MoT
+resume_model_only:
+  desc: null
+  value: true
+finetune_from_ema:
+  desc: null
+  value: true
+finetune_from_hf:
+  desc: null
+  value: true
+log_every:
+  desc: null
+  value: 1
+save_every:
+  desc: null
+  value: 2500
+total_steps:
+  desc: null
+  value: 5000
+warmup_steps:
+  desc: null
+  value: 300
+lr_scheduler:
+  desc: null
+  value: cosine
+lr:
+  desc: null
+  value: 2.0e-05
+min_lr:
+  desc: null
+  value: 1.0e-07
+beta1:
+  desc: null
+  value: 0.9
+beta2:
+  desc: null
+  value: 0.95
+eps:
+  desc: null
+  value: 1.0e-15
+ema:
+  desc: null
+  value: 0.993
+max_grad_norm:
+  desc: null
+  value: 1.0
+timestep_shift:
+  desc: null
+  value: 1.0
+mse_weight:
+  desc: null
+  value: 1.0
+ce_weight:
+  desc: null
+  value: 1.0
+ce_loss_reweighting:
+  desc: null
+  value: false
+expected_num_tokens:
+  desc: null
+  value: 30000
+num_replicate:
+  desc: null
+  value: 1
+num_shard:
+  desc: null
+  value: 8
+sharding_strategy:
+  desc: null
+  value: HYBRID_SHARD
+backward_prefetch:
+  desc: null
+  value: BACKWARD_PRE
+cpu_offload:
+  desc: null
+  value: false
+freeze_llm:
+  desc: null
+  value: false
+freeze_vit:
+  desc: null
+  value: false
+freeze_vae:
+  desc: null
+  value: true
+freeze_und:
+  desc: null
+  value: false
+copy_init_moe:
+  desc: null
+  value: true
+use_flex:
+  desc: null
+  value: false
+eval_every:
+  desc: null
+  value: 500
+num_eval_batches:
+  desc: null
+  value: 20
+use_ema_for_eval:
+  desc: null
+  value: true
+eval_log_dir:
+  desc: null
+  value: null
+eval_run_tag:
+  desc: null
+  value: ''
+viz_every:
+  desc: null
+  value: 500
+viz_n:
+  desc: null
+  value: 8
+viz_outdir:
+  desc: null
+  value: results/viz
+eval_dataset_config_file:
+  desc: null
+  value: ./data/configs/vlm_gym_patch_reassembly_alt_eval_celoss.yaml
+viz_dataset_config_file:
+  desc: null
+  value: ./data/configs/vlm_gym_patch_reassembly_alt_eval_celoss.yaml
+eval_print_n:
+  desc: null
+  value: 3
+save_ema_only:
+  desc: null
+  value: true
+save_optimizer:
+  desc: null
+  value: false
+model_path:
+  desc: null
+  value: /home/clouduser/Code/Models/BAGEL-7B-MoT
+llm_path:
+  desc: null
+  value: hf/Qwen2.5-0.5B-Instruct/
+llm_qk_norm:
+  desc: null
+  value: true
+tie_word_embeddings:
+  desc: null
+  value: false
+layer_module:
+  desc: null
+  value: Qwen2MoTDecoderLayer
+vae_path:
+  desc: null
+  value: flux/vae/ae.safetensors
+vit_path:
+  desc: null
+  value: hf/siglip-so400m-14-980-flash-attn2-navit/
+max_latent_size:
+  desc: null
+  value: 64
+latent_patch_size:
+  desc: null
+  value: 2
+vit_patch_size:
+  desc: null
+  value: 14
+vit_max_num_patch_per_side:
+  desc: null
+  value: 70
+connector_act:
+  desc: null
+  value: gelu_pytorch_tanh
+interpolate_pos:
+  desc: null
+  value: false
+vit_select_layer:
+  desc: null
+  value: -2
+vit_rope:
+  desc: null
+  value: false
+text_cond_dropout_prob:
+  desc: null
+  value: 0.0
+vae_cond_dropout_prob:
+  desc: null
+  value: 0.3
+vit_cond_dropout_prob:
+  desc: null
+  value: 0.0
+dataset_config_file:
+  desc: null
+  value: ./data/configs/vlm_gym_patch_reassembly_alt_train_celoss.yaml
+train_data_dir:
+  desc: null
+  value: /home/clouduser/Code/data/gym/patch_reassembly_alt_v5/train/
+train_jsonl_path:
+  desc: null
+  value: /home/clouduser/Code/data/gym/patch_reassembly_alt_v5/train/
+eval_data_dir:
+  desc: null
+  value: /home/clouduser/Code/data/gym/patch_reassembly_alt_v5/val/
+eval_jsonl_path:
+  desc: null
+  value: /home/clouduser/Code/data/gym/patch_reassembly_alt_v5/val/
+inference_hash_file:
+  desc: null
+  value: /home/clouduser/Code/Github/launch_new/hashes_test_set_v10.json
+task_name:
+  desc: null
+  value: patch_reassembly_v5
+instructions_dir:
+  desc: null
+  value: ./data/instructions
+prefetch_factor:
+  desc: null
+  value: 2
+num_workers:
+  desc: null
+  value: 1
+max_num_tokens_per_sample:
+  desc: null
+  value: 30000
+max_num_tokens:
+  desc: null
+  value: 30000
+prefer_buffer_before:
+  desc: null
+  value: 16384
+max_buffer_size:
+  desc: null
+  value: 50
+data_seed:
+  desc: null
+  value: 42

checkpoints_vlm_gym_patch_reassembly_alt_one_image_lr2e_5_ce_ins/checkpoints_vlm_gym_patch_reassembly_alt_one_image_lr2e_5_ce_ins/wandb/offline-run-20260129_223919-checkpoints_vlm_gym_patch_reassembly_alt_one_image_lr2e_5_ce_ins-run0/files/output.log

@@ -829,6 +829,22 @@ wandb: For more information, check out the docs at: https://weave-docs.wandb.ai/
 [2026-01-30 03:09:56] (step=0000818) Train Loss mse: 0.0246, Train Loss ce: 0.1153, Train Steps/Sec: 0.06,
 [2026-01-30 03:10:18] (step=0000819) Train Loss mse: 0.0270, Train Loss ce: 0.0949, Train Steps/Sec: 0.05,
 [2026-01-30 03:10:38] (step=0000820) Train Loss mse: 0.0203, Train Loss ce: 0.0989, Train Steps/Sec: 0.05,
+[2026-01-30 03:10:51] (step=0000821) Train Loss mse: 0.0239, Train Loss ce: 0.0780, Train Steps/Sec: 0.07,
+[2026-01-30 03:11:08] (step=0000822) Train Loss mse: 0.0224, Train Loss ce: 0.0904, Train Steps/Sec: 0.06,
+[2026-01-30 03:11:25] (step=0000823) Train Loss mse: 0.0222, Train Loss ce: 0.1022, Train Steps/Sec: 0.06,
+[2026-01-30 03:11:44] (step=0000824) Train Loss mse: 0.0243, Train Loss ce: 0.1039, Train Steps/Sec: 0.05,
+[2026-01-30 03:12:02] (step=0000825) Train Loss mse: 0.0204, Train Loss ce: 0.0916, Train Steps/Sec: 0.06,
+[2026-01-30 03:12:16] (step=0000826) Train Loss mse: 0.0201, Train Loss ce: 0.0970, Train Steps/Sec: 0.07,
+[2026-01-30 03:12:37] (step=0000827) Train Loss mse: 0.0192, Train Loss ce: 0.1047, Train Steps/Sec: 0.05,
+[2026-01-30 03:12:57] (step=0000828) Train Loss mse: 0.0214, Train Loss ce: 0.1020, Train Steps/Sec: 0.05,
+[2026-01-30 03:13:18] (step=0000829) Train Loss mse: 0.0250, Train Loss ce: 0.0784, Train Steps/Sec: 0.05,
+[2026-01-30 03:13:36] (step=0000830) Train Loss mse: 0.0287, Train Loss ce: 0.0897, Train Steps/Sec: 0.06,
+[2026-01-30 03:13:57] (step=0000831) Train Loss mse: 0.0190, Train Loss ce: 0.0986, Train Steps/Sec: 0.05,
+[2026-01-30 03:14:13] (step=0000832) Train Loss mse: 0.0256, Train Loss ce: 0.0924, Train Steps/Sec: 0.06,
+[2026-01-30 03:14:25] (step=0000833) Train Loss mse: 0.0232, Train Loss ce: 0.0824, Train Steps/Sec: 0.08,
+[2026-01-30 03:14:42] (step=0000834) Train Loss mse: 0.0249, Train Loss ce: 0.0951, Train Steps/Sec: 0.06,
+[2026-01-30 03:14:58] (step=0000835) Train Loss mse: 0.0275, Train Loss ce: 0.0966, Train Steps/Sec: 0.06,
+[2026-01-30 03:15:19] (step=0000836) Train Loss mse: 0.0208, Train Loss ce: 0.1061, Train Steps/Sec: 0.05,
 FullyShardedDataParallel(
   (_fsdp_wrapped_module): Bagel(
     (language_model): Qwen2ForCausalLM(
@@ -1029,29 +1045,6 @@ Preparing Dataset vlm_gym_patch_reassembly_alt_celoss_evalonce/vlm_gym_patch_rea
   fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_patch_reassembly_alt_celoss_evalonce'}]
   fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_patch_reassembly_alt_celoss_evalonce'}]
 ce_avg: 0.08397598564624786, mse_avg: 0.015734810382127762
-base_dir is /dev/shm/models/checkpoints_vlm_gym_patch_reassembly_alt_one_image_lr2e_5_ce_ins/eval_used_rows, step_tag is checkpoints_vlm_gym_patch_reassembly_alt_one_image_lr2e_5_ce_ins_step2000
-Preparing Dataset vlm_gym_patch_reassembly_alt_celoss_evalonce/vlm_gym_patch_reassembly_alt_val
-[eval debug] first 3 batch fingerprints:
-  fp[0]: [{'data_indexes': [0], 'worker_id': 0, 'dataset_name': 'vlm_gym_patch_reassembly_alt_celoss_evalonce'}]
-  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_patch_reassembly_alt_celoss_evalonce'}]
-  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_patch_reassembly_alt_celoss_evalonce'}]
-ce_avg: 0.08462081849575043, mse_avg: 0.015557804144918919
-[2026-01-30 03:10:51] (step=0000821) Train Loss mse: 0.0239, Train Loss ce: 0.0780, Train Steps/Sec: 0.07,
-[2026-01-30 03:11:08] (step=0000822) Train Loss mse: 0.0224, Train Loss ce: 0.0904, Train Steps/Sec: 0.06,
-[2026-01-30 03:11:25] (step=0000823) Train Loss mse: 0.0222, Train Loss ce: 0.1022, Train Steps/Sec: 0.06,
-[2026-01-30 03:11:44] (step=0000824) Train Loss mse: 0.0243, Train Loss ce: 0.1039, Train Steps/Sec: 0.05,
-[2026-01-30 03:12:02] (step=0000825) Train Loss mse: 0.0204, Train Loss ce: 0.0916, Train Steps/Sec: 0.06,
-[2026-01-30 03:12:16] (step=0000826) Train Loss mse: 0.0201, Train Loss ce: 0.0970, Train Steps/Sec: 0.07,
-[2026-01-30 03:12:37] (step=0000827) Train Loss mse: 0.0192, Train Loss ce: 0.1047, Train Steps/Sec: 0.05,
-[2026-01-30 03:12:57] (step=0000828) Train Loss mse: 0.0214, Train Loss ce: 0.1020, Train Steps/Sec: 0.05,
-[2026-01-30 03:13:18] (step=0000829) Train Loss mse: 0.0250, Train Loss ce: 0.0784, Train Steps/Sec: 0.05,
-[2026-01-30 03:13:36] (step=0000830) Train Loss mse: 0.0287, Train Loss ce: 0.0897, Train Steps/Sec: 0.06,
-[2026-01-30 03:13:57] (step=0000831) Train Loss mse: 0.0190, Train Loss ce: 0.0986, Train Steps/Sec: 0.05,
-[2026-01-30 03:14:13] (step=0000832) Train Loss mse: 0.0256, Train Loss ce: 0.0924, Train Steps/Sec: 0.06,
-[2026-01-30 03:14:25] (step=0000833) Train Loss mse: 0.0232, Train Loss ce: 0.0824, Train Steps/Sec: 0.08,
-[2026-01-30 03:14:42] (step=0000834) Train Loss mse: 0.0249, Train Loss ce: 0.0951, Train Steps/Sec: 0.06,
-[2026-01-30 03:14:58] (step=0000835) Train Loss mse: 0.0275, Train Loss ce: 0.0966, Train Steps/Sec: 0.06,
-[2026-01-30 03:15:19] (step=0000836) Train Loss mse: 0.0208, Train Loss ce: 0.1061, Train Steps/Sec: 0.05,
 [2026-01-30 03:15:35] (step=0000837) Train Loss mse: 0.0218, Train Loss ce: 0.0916, Train Steps/Sec: 0.06,
 [2026-01-30 03:15:53] (step=0000838) Train Loss mse: 0.0238, Train Loss ce: 0.1028, Train Steps/Sec: 0.05,
 [2026-01-30 03:16:12] (step=0000839) Train Loss mse: 0.0211, Train Loss ce: 0.1001, Train Steps/Sec: 0.05,
@@ -2162,6 +2155,20 @@ ce_avg: 0.08462081849575043, mse_avg: 0.015557804144918919
 [2026-01-30 09:06:52] (step=0001944) Train Loss mse: 0.0175, Train Loss ce: 0.0748, Train Steps/Sec: 0.06,
 [2026-01-30 09:07:12] (step=0001945) Train Loss mse: 0.0267, Train Loss ce: 0.0836, Train Steps/Sec: 0.05,
 [2026-01-30 09:07:26] (step=0001946) Train Loss mse: 0.0200, Train Loss ce: 0.0730, Train Steps/Sec: 0.07,
+base_dir is /dev/shm/models/checkpoints_vlm_gym_patch_reassembly_alt_one_image_lr2e_5_ce_ins/eval_used_rows, step_tag is checkpoints_vlm_gym_patch_reassembly_alt_one_image_lr2e_5_ce_ins_step2000
+Preparing Dataset vlm_gym_patch_reassembly_alt_celoss_evalonce/vlm_gym_patch_reassembly_alt_val
+[eval debug] first 3 batch fingerprints:
+  fp[0]: [{'data_indexes': [0], 'worker_id': 0, 'dataset_name': 'vlm_gym_patch_reassembly_alt_celoss_evalonce'}]
+  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_patch_reassembly_alt_celoss_evalonce'}]
+  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_patch_reassembly_alt_celoss_evalonce'}]
+ce_avg: 0.08462081849575043, mse_avg: 0.015557804144918919
+base_dir is /dev/shm/models/checkpoints_vlm_gym_patch_reassembly_alt_one_image_lr2e_5_ce_ins/eval_used_rows, step_tag is checkpoints_vlm_gym_patch_reassembly_alt_one_image_lr2e_5_ce_ins_step2500
+Preparing Dataset vlm_gym_patch_reassembly_alt_celoss_evalonce/vlm_gym_patch_reassembly_alt_val
+[eval debug] first 3 batch fingerprints:
+  fp[0]: [{'data_indexes': [0], 'worker_id': 0, 'dataset_name': 'vlm_gym_patch_reassembly_alt_celoss_evalonce'}]
+  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_patch_reassembly_alt_celoss_evalonce'}]
+  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_patch_reassembly_alt_celoss_evalonce'}]
+ce_avg: 0.3942395746707916, mse_avg: 0.015481945127248764
 [2026-01-30 09:07:46] (step=0001947) Train Loss mse: 0.0181, Train Loss ce: 0.0834, Train Steps/Sec: 0.05,
 [2026-01-30 09:08:06] (step=0001948) Train Loss mse: 0.0197, Train Loss ce: 0.0996, Train Steps/Sec: 0.05,
 [2026-01-30 09:08:26] (step=0001949) Train Loss mse: 0.0215, Train Loss ce: 0.0908, Train Steps/Sec: 0.05,
@@ -2199,13 +2206,6 @@ ce_avg: 0.08462081849575043, mse_avg: 0.015557804144918919
 [2026-01-30 09:18:42] (step=0001981) Train Loss mse: 0.0167, Train Loss ce: 0.0742, Train Steps/Sec: 0.06,
 [2026-01-30 09:19:04] (step=0001982) Train Loss mse: 0.0220, Train Loss ce: 0.0905, Train Steps/Sec: 0.05,
 [2026-01-30 09:19:20] (step=0001983) Train Loss mse: 0.0224, Train Loss ce: 0.0735, Train Steps/Sec: 0.06,
-base_dir is /dev/shm/models/checkpoints_vlm_gym_patch_reassembly_alt_one_image_lr2e_5_ce_ins/eval_used_rows, step_tag is checkpoints_vlm_gym_patch_reassembly_alt_one_image_lr2e_5_ce_ins_step2500
-Preparing Dataset vlm_gym_patch_reassembly_alt_celoss_evalonce/vlm_gym_patch_reassembly_alt_val
-[eval debug] first 3 batch fingerprints:
-  fp[0]: [{'data_indexes': [0], 'worker_id': 0, 'dataset_name': 'vlm_gym_patch_reassembly_alt_celoss_evalonce'}]
-  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_patch_reassembly_alt_celoss_evalonce'}]
-  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_patch_reassembly_alt_celoss_evalonce'}]
-ce_avg: 0.3942395746707916, mse_avg: 0.015481945127248764
 [2026-01-30 09:19:42] (step=0001984) Train Loss mse: 0.0192, Train Loss ce: 0.0788, Train Steps/Sec: 0.05,
 [2026-01-30 09:20:04] (step=0001985) Train Loss mse: 0.0141, Train Loss ce: 0.0955, Train Steps/Sec: 0.05,
 [2026-01-30 09:20:25] (step=0001986) Train Loss mse: 0.0198, Train Loss ce: 0.0789, Train Steps/Sec: 0.05,
@@ -3072,27 +3072,6 @@ ce_avg: 0.3942395746707916, mse_avg: 0.015481945127248764
 [2026-01-30 13:55:59] (step=0002844) Train Loss mse: 0.0198, Train Loss ce: 0.0788, Train Steps/Sec: 0.05,
 [2026-01-30 13:56:17] (step=0002845) Train Loss mse: 0.0168, Train Loss ce: 0.0757, Train Steps/Sec: 0.05,
 [2026-01-30 13:56:37] (step=0002846) Train Loss mse: 0.0185, Train Loss ce: 0.0793, Train Steps/Sec: 0.05,
-[2026-01-30 13:56:59
-base_dir is /dev/shm/models/checkpoints_vlm_gym_patch_reassembly_alt_one_image_lr2e_5_ce_ins/eval_used_rows, step_tag is checkpoints_vlm_gym_patch_reassembly_alt_one_image_lr2e_5_ce_ins_step3000
-Preparing Dataset vlm_gym_patch_reassembly_alt_celoss_evalonce/vlm_gym_patch_reassembly_alt_val
-[eval debug] first 3 batch fingerprints:
-  fp[0]: [{'data_indexes': [0], 'worker_id': 0, 'dataset_name': 'vlm_gym_patch_reassembly_alt_celoss_evalonce'}]
-  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_patch_reassembly_alt_celoss_evalonce'}]
-  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_patch_reassembly_alt_celoss_evalonce'}]
-ce_avg: 0.07592044025659561, mse_avg: 0.013209872879087925
-base_dir is /dev/shm/models/checkpoints_vlm_gym_patch_reassembly_alt_one_image_lr2e_5_ce_ins/eval_used_rows, step_tag is checkpoints_vlm_gym_patch_reassembly_alt_one_image_lr2e_5_ce_ins_step3500
-Preparing Dataset vlm_gym_patch_reassembly_alt_celoss_evalonce/vlm_gym_patch_reassembly_alt_val
-[eval debug] first 3 batch fingerprints:
-  fp[0]: [{'data_indexes': [0], 'worker_id': 0, 'dataset_name': 'vlm_gym_patch_reassembly_alt_celoss_evalonce'}]
-  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_patch_reassembly_alt_celoss_evalonce'}]
-  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_patch_reassembly_alt_celoss_evalonce'}]
-ce_avg: 0.07461568713188171, mse_avg: 0.013009372167289257
-base_dir is /dev/shm/models/checkpoints_vlm_gym_patch_reassembly_alt_one_image_lr2e_5_ce_ins/eval_used_rows, step_tag is checkpoints_vlm_gym_patch_reassembly_alt_one_image_lr2e_5_ce_ins_step4000
-Preparing Dataset vlm_gym_patch_reassembly_alt_celoss_evalonce/vlm_gym_patch_reassembly_alt_val
-[eval debug] first 3 batch fingerprints:
-  fp[0]: [{'data_indexes': [0], 'worker_id': 0, 'dataset_name': 'vlm_gym_patch_reassembly_alt_celoss_evalonce'}]
-  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_patch_reassembly_alt_celoss_evalonce'}]
-  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_patch_reassembly_alt_celoss_evalonce'}]
 [2026-01-30 13:56:59] (step=0002847) Train Loss mse: 0.0204, Train Loss ce: 0.0870, Train Steps/Sec: 0.05,
 [2026-01-30 13:57:16] (step=0002848) Train Loss mse: 0.0180, Train Loss ce: 0.0854, Train Steps/Sec: 0.06,
 [2026-01-30 13:57:38] (step=0002849) Train Loss mse: 0.0181, Train Loss ce: 0.0848, Train Steps/Sec: 0.05,
@@ -3332,17 +3311,27 @@ Preparing Dataset vlm_gym_patch_reassembly_alt_celoss_evalonce/vlm_gym_patch_rea
 [2026-01-30 15:14:49] (step=0003083) Train Loss mse: 0.0170, Train Loss ce: 0.0682, Train Steps/Sec: 0.05,
 [2026-01-30 15:15:11] (step=0003084) Train Loss mse: 0.0136, Train Loss ce: 0.0838, Train Steps/Sec: 0.04,
 [2026-01-30 15:15:34] (step=0003085) Train Loss mse: 0.0167, Train Loss ce: 0.0730, Train Steps/Sec: 0.04,
-[2026-01-30 15:15:58] (step=0003086) Train Loss mse: 0.0185, Train Loss ce: 0.0791, Train Steps/Sec: 0.04,
-[2026-01-30 15:16:17] (step=0003087) Train Loss mse: 0.0131, Train Loss ce: 0.0739, Train Steps/Sec: 0.05,
-[2026-01-30 15:16:38] (step=0003088) Train Loss mse: 0.0154, Train Loss ce: 0.0845, Train Steps/Sec: 0.05,
-[2026-01-30 15:17:00] (step=0003089) Train Loss mse: 0.0206, Train Loss ce: 0.0767, Train Steps/Sec: 0.05,
-[2026-01-30 15:17:16] (step=0003090) Train Loss mse: 0.0185, Train Loss ce: 0.0730, Train Steps/Sec: 0.06,
-[2026-01-30 15:17:34] (step=0003091) Train Loss mse: 0.0181, Train Loss ce: 0.0746, Train Steps/Sec: 0.05,
-[2026-01-30 15:17:55] (step=0003092) Train Loss mse: 0.0198, Train Loss ce: 0.0845, Train Steps/Sec: 0.05,
-[2026-01-30 15:18:16] (step=0003093) Train Loss mse: 0.0185, Train Loss ce: 0.0748, Train Steps/Sec: 0.05,
-[2026-01-30 15:18:33] (step=0003094) Train Loss mse: 0.0219, Train Loss ce: 0.0700, Train Steps/Sec: 0.06,
-[2026-01-30 15:18:52] (step=0003095) Train Loss mse: 0.0138, Train Loss ce: 0.0704, Train Steps/Sec: 0.05,
-[2026-01-30 15:19:12] (step=0003096) Train Loss mse: 0.0226, Train Loss ce: 0.0813, Train Steps/Sec: 0.05,
+base_dir is /dev/shm/models/checkpoints_vlm_gym_patch_reassembly_alt_one_image_lr2e_5_ce_ins/eval_used_rows, step_tag is checkpoints_vlm_gym_patch_reassembly_alt_one_image_lr2e_5_ce_ins_step3000
+Preparing Dataset vlm_gym_patch_reassembly_alt_celoss_evalonce/vlm_gym_patch_reassembly_alt_val
+[eval debug] first 3 batch fingerprints:
+  fp[0]: [{'data_indexes': [0], 'worker_id': 0, 'dataset_name': 'vlm_gym_patch_reassembly_alt_celoss_evalonce'}]
+  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_patch_reassembly_alt_celoss_evalonce'}]
+  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_patch_reassembly_alt_celoss_evalonce'}]
+ce_avg: 0.07592044025659561, mse_avg: 0.013209872879087925
+base_dir is /dev/shm/models/checkpoints_vlm_gym_patch_reassembly_alt_one_image_lr2e_5_ce_ins/eval_used_rows, step_tag is checkpoints_vlm_gym_patch_reassembly_alt_one_image_lr2e_5_ce_ins_step3500
+Preparing Dataset vlm_gym_patch_reassembly_alt_celoss_evalonce/vlm_gym_patch_reassembly_alt_val
+[eval debug] first 3 batch fingerprints:
+  fp[0]: [{'data_indexes': [0], 'worker_id': 0, 'dataset_name': 'vlm_gym_patch_reassembly_alt_celoss_evalonce'}]
+  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_patch_reassembly_alt_celoss_evalonce'}]
+  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_patch_reassembly_alt_celoss_evalonce'}]
+ce_avg: 0.07461568713188171, mse_avg: 0.013009372167289257
+base_dir is /dev/shm/models/checkpoints_vlm_gym_patch_reassembly_alt_one_image_lr2e_5_ce_ins/eval_used_rows, step_tag is checkpoints_vlm_gym_patch_reassembly_alt_one_image_lr2e_5_ce_ins_step4000
+Preparing Dataset vlm_gym_patch_reassembly_alt_celoss_evalonce/vlm_gym_patch_reassembly_alt_val
+[eval debug] first 3 batch fingerprints:
+  fp[0]: [{'data_indexes': [0], 'worker_id': 0, 'dataset_name': 'vlm_gym_patch_reassembly_alt_celoss_evalonce'}]
+  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_patch_reassembly_alt_celoss_evalonce'}]
+  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_patch_reassembly_alt_celoss_evalonce'}]
+ce_avg: 0.4290580153465271, mse_avg: 0.012517130933701992
 [2026-01-30 15:19:30] (step=0003097) Train Loss mse: 0.0185, Train Loss ce: 0.0726, Train Steps/Sec: 0.05,
 [2026-01-30 15:19:48] (step=0003098) Train Loss mse: 0.0155, Train Loss ce: 0.0671, Train Steps/Sec: 0.06,
 [2026-01-30 15:20:07] (step=0003099) Train Loss mse: 0.0187, Train Loss ce: 0.0632, Train Steps/Sec: 0.05,
@@ -4255,53 +4244,17 @@ Preparing Dataset vlm_gym_patch_reassembly_alt_celoss_evalonce/vlm_gym_patch_rea
 [2026-01-30 20:11:11] (step=0004006) Train Loss mse: 0.0208, Train Loss ce: 0.0885, Train Steps/Sec: 0.04,
 [2026-01-30 20:11:29] (step=0004007) Train Loss mse: 0.0143, Train Loss ce: 0.0732, Train Steps/Sec: 0.05,
 [2026-01-30 20:11:49] (step=0004008) Train Loss mse: 0.0147, Train Loss ce: 0.0707, Train Steps/Sec: 0.05,
-[2026-01-30 20:12:06] (step=0004009) Train Loss mse: 0.0179, Train Loss ce: 0.0831, Train Steps/Sec: 0.06,
-[2026-01-30 20:12:24] (step=0004010) Train Loss mse: 0.0206, Train Loss ce: 0.0700, Train Steps/Sec: 0.05,
-[2026-01-30 20:12:41] (step=0004011) Train Loss mse: 0.0176, Train Loss ce: 0.0685, Train Steps/Sec: 0.06,
-[2026-01-30 20:13:00] (step=0004012) Train Loss mse: 0.0187, Train Loss ce: 0.0644, Train Steps/Sec: 0.05,
-[2026-01-30 20:13:21] (step=0004013) Train Loss mse: 0.0137, Train Loss ce: 0.0782, Train Steps/Sec: 0.05,
-[2026-01-30 20:13:36] (step=0004014) Train Loss mse: 0.0164, Train Loss ce: 0.0727, Train Steps/Sec: 0.07,
-[2026-01-30 20:13:51] (step=0004015) Train Loss mse: 0.0166, Train Loss ce: 0.0657, Train Steps/Sec: 0.07,
-[2026-01-30 20:14:07] (step=0004016) Train Loss mse: 0.0183, Train Loss ce: 0.0631, Train Steps/Sec: 0.06,
-[2026-01-30 20:14:29] (step=0004017) Train Loss mse: 0.0177, Train Loss ce: 0.0605, Train Steps/Sec: 0.05,
-[2026-01-30 20:14:52] (step=0004018) Train Loss mse: 0.0166, Train Loss ce: 0.0792, Train Steps/Sec: 0.04,
-[2026-01-30 20:15:14] (step=0004019) Train Loss mse: 0.0147, Train Loss ce: 0.0865, Train Steps/Sec: 0.05,
-[2026-01-30 20:15:30] (step=0004020) Train Loss mse: 0.0176, Train Loss ce: 0.0630, Train Steps/Sec: 0.06,
-[2026-01-30 20:15:48] (step=0004021) Train Loss mse: 0.0191, Train Loss ce: 0.0688, Train Steps/Sec: 0.06,
-[2026-01-30 20:16:10] (step=0004022) Train Loss mse: 0.0132, Train Loss ce: 0.0891, Train Steps/Sec: 0.05,
-[2026-01-30 20:16:32] (step=0004023) Train Loss mse: 0.0154, Train Loss ce: 0.0771, Train Steps/Sec: 0.04,
-[2026-01-30 20:16:54] (step=0004024) Train Loss mse: 0.0163, Train Loss ce: 0.0733, Train Steps/Sec: 0.05,
-[2026-01-30 20:17:09] (step=0004025) Train Loss mse: 0.0140, Train Loss ce: 0.0760, Train Steps/Sec: 0.06,
-[2026-01-30 20:17:26] (step=0004026) Train Loss mse: 0.0225, Train Loss ce: 0.0853, Train Steps/Sec: 0.06,
-[2026-01-30 20:17:48] (step=0004027) Train Loss mse: 0.0173, Train Loss ce: 0.0815, Train Steps/Sec: 0.04,
-[2026-01-30 20:18:03] (step=0004028) Train Loss mse: 0.0156, Train Loss ce: 0.0705, Train Steps/Sec: 0.07,
-[2026-01-30 20:18:25] (step=0004029) Train Loss mse: 0.0211, Train Loss ce: 0.0749, Train Steps/Sec: 0.05,
-[2026-01-30 20:18:44] (step=0004030) Train Loss mse: 0.0201, Train Loss ce: 0.0623, Train Steps/Sec: 0.05,
-[2026-01-30 20:19:01] (step=0004031) Train Loss mse: 0.0188, Train Loss ce: 0.0554, Train Steps/Sec: 0.06,
-[2026-01-30 20:19:20] (step=0004032) Train Loss mse: 0.0122, Train Loss ce: 0.0831, Train Steps/Sec: 0.05,
-[2026-01-30 20:19:37] (step=0004033) Train Loss mse: 0.0150, Train Loss ce: 0.0704, Train Steps/Sec: 0.06,
-[2026-01-30 20:19:53] (step=0004034) Train Loss mse: 0.0151, Train Loss ce: 0.0717, Train Steps/Sec: 0.06,
-[2026-01-30 20:20:11] (step=0004035) Train Loss mse: 0.0177, Train Loss ce: 0.0796, Train Steps/Sec: 0.06,
-[2026-01-30 20:20:31] (step=0004036) Train Loss mse: 0.0131, Train Loss ce: 0.0839, Train Steps/Sec: 0.05,
-[2026-01-30 20:20:50] (step=0004037) Train Loss mse: 0.0204, Train Loss ce: 0.0773, Train Steps/Sec: 0.05,
-[2026-01-30 20:21:10] (step=0004038) Train Loss mse: 0.0142, Train Loss ce: 0.0768, Train Steps/Sec: 0.05,
-[2026-01-30 20:21:32] (step=0004039) Train Loss mse: 0.0164, Train Loss ce: 0.0707, Train Steps/Sec: 0.05,
-[2026-01-30 20:21:49] (step=0004040) Train Loss mse: 0.0159, Train Loss ce: 0.0670, Train Steps/Sec: 0.06,
-[2026-01-30 20:22:11] (step=0004041) Train Loss mse: 0.0131, Train Loss ce: 0.0709, Train Steps/Sec: 0.05,
-[2026-01-30 20:22:31] (step=0004042) Train Loss mse: 0.0176, Train Loss ce: 0.0648, Train Steps/Sec: 0.05,
-[2026-01-30 20:22:50] (step=0004043) Train Loss mse: 0.0205, Train Loss ce: 0.0746, Train Steps/Sec: 0.05,
-[2026-01-30 20:23:10] (step=0004044) Train Loss mse: 0.0144, Train Loss ce: 0.0847, Train Steps/Sec: 0.05,
-[2026-01-30 20:23:24] (step=0004045) Train Loss mse: 0.0179, Train Loss ce: 0.0575, Train Steps/Sec: 0.07,
-[2026-01-30 20:23:41] (step=0004046) Train Loss mse: 0.0134, Train Loss ce: 0.0550, Train Steps/Sec: 0.06,
-[2026-01-30 20:23:57] (step=0004047) Train Loss mse: 0.0185, Train Loss ce: 0.0675, Train Steps/Sec: 0.06,
-[2026-01-30 20:24:14] (step=0004048) Train Loss mse: 0.0152, Train Loss ce: 0.0608, Train Steps/Sec: 0.06,
-[2026-01-30 20:24:30] (step=0004049) Train Loss mse: 0.0157, Train Loss ce: 0.0600, Train Steps/Sec: 0.06,
-[2026-01-30 20:24:52] (step=0004050) Train Loss mse: 0.0159, Train Loss ce: 0.0711, Train Steps/Sec: 0.05,
-[2026-01-30 20:25:11] (step=0004051) Train Loss mse: 0.0189, Train Loss ce: 0.0743, Train Steps/Sec: 0.05,
-[2026-01-30 20:25:31] (step=0004052) Train Loss mse: 0.0175, Train Loss ce: 0.0711, Train Steps/Sec: 0.05,
-[2026-01-30 20:25:47] (step=0004053) Train Loss mse: 0.0159, Train Loss ce: 0.0579, Train Steps/Sec: 0.06,
-[2026-01-30 20:26:05] (step=0004054) Train Loss mse: 0.0185, Train Loss ce: 0.0781, Train Steps/Sec: 0.06,
-[2026-01-30 20:26:23] (step=0004055) Train Loss mse: 0.0159, Train Loss ce: 0.0670, Train Steps/Sec: 0.06,
+[2026-01-30 15:15:58] (step=0003086) Train Loss mse: 0.0185, Train Loss ce: 0.0791, Train Steps/Sec: 0.04,
+[2026-01-30 15:16:17] (step=0003087) Train Loss mse: 0.0131, Train Loss ce: 0.0739, Train Steps/Sec: 0.05,
+[2026-01-30 15:16:38] (step=0003088) Train Loss mse: 0.0154, Train Loss ce: 0.0845, Train Steps/Sec: 0.05,
+[2026-01-30 15:17:00] (step=0003089) Train Loss mse: 0.0206, Train Loss ce: 0.0767, Train Steps/Sec: 0.05,
+[2026-01-30 15:17:16] (step=0003090) Train Loss mse: 0.0185, Train Loss ce: 0.0730, Train Steps/Sec: 0.06,
+[2026-01-30 15:17:34] (step=0003091) Train Loss mse: 0.0181, Train Loss ce: 0.0746, Train Steps/Sec: 0.05,
+[2026-01-30 15:17:55] (step=0003092) Train Loss mse: 0.0198, Train Loss ce: 0.0845, Train Steps/Sec: 0.05,
+[2026-01-30 15:18:16] (step=0003093) Train Loss mse: 0.0185, Train Loss ce: 0.0748, Train Steps/Sec: 0.05,
+[2026-01-30 15:18:33] (step=0003094) Train Loss mse: 0.0219, Train Loss ce: 0.0700, Train Steps/Sec: 0.06,
+[2026-01-30 15:18:52] (step=0003095) Train Loss mse: 0.0138, Train Loss ce: 0.0704, Train Steps/Sec: 0.05,
+[2026-01-30 15:19:12] (step=0003096) Train Loss mse: 0.0226, Train Loss ce: 0.0813, Train Steps/Sec: 0.05,
 base_dir is /dev/shm/models/checkpoints_vlm_gym_patch_reassembly_alt_one_image_lr2e_5_ce_ins/eval_used_rows, step_tag is checkpoints_vlm_gym_patch_reassembly_alt_one_image_lr2e_5_ce_ins_step4500
 Preparing Dataset vlm_gym_patch_reassembly_alt_celoss_evalonce/vlm_gym_patch_reassembly_alt_val
 [eval debug] first 3 batch fingerprints:
@@ -4315,59 +4268,6 @@ Preparing Dataset vlm_gym_patch_reassembly_alt_celoss_evalonce/vlm_gym_patch_rea
   fp[0]: [{'data_indexes': [0], 'worker_id': 0, 'dataset_name': 'vlm_gym_patch_reassembly_alt_celoss_evalonce'}]
   fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_patch_reassembly_alt_celoss_evalonce'}]
   fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_patch_reassembly_alt_celoss_evalonce'}]
-ce_avg: 0.815917432308197, mse_avg: 0.012591413222253323
-[2026-01-30 20:26:41] (step=0004056) Train Loss mse: 0.0173, Train Loss ce: 0.0698, Train Steps/Sec: 0.06,
-[2026-01-30 20:26:59] (step=0004057) Train Loss mse: 0.0169, Train Loss ce: 0.0774, Train Steps/Sec: 0.05,
-[2026-01-30 20:27:18] (step=0004058) Train Loss mse: 0.0154, Train Loss ce: 0.0729, Train Steps/Sec: 0.06,
-[2026-01-30 20:27:35] (step=0004059) Train Loss mse: 0.0184, Train Loss ce: 0.0678, Train Steps/Sec: 0.06,
-[2026-01-30 20:27:52] (step=0004060) Train Loss mse: 0.0179, Train Loss ce: 0.0634, Train Steps/Sec: 0.06,
-[2026-01-30 20:28:11] (step=0004061) Train Loss mse: 0.0187, Train Loss ce: 0.0742, Train Steps/Sec: 0.05,
-[2026-01-30 20:28:28] (step=0004062) Train Loss mse: 0.0162, Train Loss ce: 0.0733, Train Steps/Sec: 0.06,
-[2026-01-30 20:28:44] (step=0004063) Train Loss mse: 0.0147, Train Loss ce: 0.0743, Train Steps/Sec: 0.06,
-[2026-01-30 20:29:04] (step=0004064) Train Loss mse: 0.0178, Train Loss ce: 0.0793, Train Steps/Sec: 0.05,
-[2026-01-30 20:29:17] (step=0004065) Train Loss mse: 0.0224, Train Loss ce: 0.0725, Train Steps/Sec: 0.08,
-[2026-01-30 20:29:42] (step=0004066) Train Loss mse: 0.0126, Train Loss ce: 0.0976, Train Steps/Sec: 0.04,
-[2026-01-30 20:30:02] (step=0004067) Train Loss mse: 0.0172, Train Loss ce: 0.0725, Train Steps/Sec: 0.05,
-[2026-01-30 20:30:22] (step=0004068) Train Loss mse: 0.0154, Train Loss ce: 0.0704, Train Steps/Sec: 0.05,
-[2026-01-30 20:30:42] (step=0004069) Train Loss mse: 0.0148, Train Loss ce: 0.0818, Train Steps/Sec: 0.05,
-[2026-01-30 20:30:59] (step=0004070) Train Loss mse: 0.0164, Train Loss ce: 0.0676, Train Steps/Sec: 0.06,
-[2026-01-30 20:31:14] (step=0004071) Train Loss mse: 0.0175, Train Loss ce: 0.0627, Train Steps/Sec: 0.06,
-[2026-01-30 20:31:35] (step=0004072) Train Loss mse: 0.0143, Train Loss ce: 0.0685, Train Steps/Sec: 0.05,
-[2026-01-30 20:31:51] (step=0004073) Train Loss mse: 0.0224, Train Loss ce: 0.0662, Train Steps/Sec: 0.06,
-[2026-01-30 20:32:07] (step=0004074) Train Loss mse: 0.0171, Train Loss ce: 0.0672, Train Steps/Sec: 0.06,
-[2026-01-30 20:32:24] (step=0004075) Train Loss mse: 0.0154, Train Loss ce: 0.0681, Train Steps/Sec: 0.06,
-[2026-01-30 20:32:40] (step=0004076) Train Loss mse: 0.0194, Train Loss ce: 0.0653, Train Steps/Sec: 0.06,
-[2026-01-30 20:32:55] (step=0004077) Train Loss mse: 0.0226, Train Loss ce: 0.0734, Train Steps/Sec: 0.07,
-[2026-01-30 20:33:13] (step=0004078) Train Loss mse: 0.0161, Train Loss ce: 0.0689, Train Steps/Sec: 0.06,
-[2026-01-30 20:33:28] (step=0004079) Train Loss mse: 0.0197, Train Loss ce: 0.0657, Train Steps/Sec: 0.07,
-[2026-01-30 20:33:48] (step=0004080) Train Loss mse: 0.0163, Train Loss ce: 0.0757, Train Steps/Sec: 0.05,
-[2026-01-30 20:34:05] (step=0004081) Train Loss mse: 0.0170, Train Loss ce: 0.0834, Train Steps/Sec: 0.06,
-[2026-01-30 20:34:18] (step=0004082) Train Loss mse: 0.0207, Train Loss ce: 0.0604, Train Steps/Sec: 0.08,
-[2026-01-30 20:34:33] (step=0004083) Train Loss mse: 0.0176, Train Loss ce: 0.0622, Train Steps/Sec: 0.07,
-[2026-01-30 20:34:45] (step=0004084) Train Loss mse: 0.0172, Train Loss ce: 0.0625, Train Steps/Sec: 0.08,
-[2026-01-30 20:35:07] (step=0004085) Train Loss mse: 0.0128, Train Loss ce: 0.0653, Train Steps/Sec: 0.05,
-[2026-01-30 20:35:27] (step=0004086) Train Loss mse: 0.0206, Train Loss ce: 0.0769, Train Steps/Sec: 0.05,
-[2026-01-30 20:35:47] (step=0004087) Train Loss mse: 0.0148, Train Loss ce: 0.0588, Train Steps/Sec: 0.05,
-[2026-01-30 20:36:08] (step=0004088) Train Loss mse: 0.0203, Train Loss ce: 0.0695, Train Steps/Sec: 0.05,
-[2026-01-30 20:36:26] (step=0004089) Train Loss mse: 0.0194, Train Loss ce: 0.0745, Train Steps/Sec: 0.06,
-[2026-01-30 20:36:45] (step=0004090) Train Loss mse: 0.0174, Train Loss ce: 0.0675, Train Steps/Sec: 0.05,
-[2026-01-30 20:37:08] (step=0004091) Train Loss mse: 0.0142, Train Loss ce: 0.0756, Train Steps/Sec: 0.04,
-[2026-01-30 20:37:27] (step=0004092) Train Loss mse: 0.0145, Train Loss ce: 0.0764, Train Steps/Sec: 0.05,
-[2026-01-30 20:37:44] (step=0004093) Train Loss mse: 0.0168, Train Loss ce: 0.0729, Train Steps/Sec: 0.06,
-[2026-01-30 20:37:59] (step=0004094) Train Loss mse: 0.0173, Train Loss ce: 0.0667, Train Steps/Sec: 0.07,
-[2026-01-30 20:38:10] (step=0004095) Train Loss mse: 0.0189, Train Loss ce: 0.0711, Train Steps/Sec: 0.09,
-[2026-01-30 20:38:33] (step=0004096) Train Loss mse: 0.0143, Train Loss ce: 0.0711, Train Steps/Sec: 0.04,
-[2026-01-30 20:38:56] (step=0004097) Train Loss mse: 0.0149, Train Loss ce: 0.0693, Train Steps/Sec: 0.04,
-[2026-01-30 20:39:12] (step=0004098) Train Loss mse: 0.0169, Train Loss ce: 0.0657, Train Steps/Sec: 0.06,
-[2026-01-30 20:39:33] (step=0004099) Train Loss mse: 0.0142, Train Loss ce: 0.0780, Train Steps/Sec: 0.05,
-[2026-01-30 20:39:52] (step=0004100) Train Loss mse: 0.0160, Train Loss ce: 0.0752, Train Steps/Sec: 0.05,
-[2026-01-30 20:40:15] (step=0004101) Train Loss mse: 0.0160, Train Loss ce: 0.0739, Train Steps/Sec: 0.04,
-[2026-01-30 20:40:36] (step=0004102) Train Loss mse: 0.0144, Train Loss ce: 0.0622, Train Steps/Sec: 0.05,
-[2026-01-30 20:40:57] (step=0004103) Train Loss mse: 0.0170, Train Loss ce: 0.0804, Train Steps/Sec: 0.05,
-[2026-01-30 20:41:17] (step=0004104) Train Loss mse: 0.0181, Train Loss ce: 0.0598, Train Steps/Sec: 0.05,
-[2026-01-30 20:41:38] (step=0004105) Train Loss mse: 0.0121, Train Loss ce: 0.0770, Train Steps/Sec: 0.05,
-[2026-01-30 20:41:55] (step=0004106) Train Loss mse: 0.0152, Train Loss ce: 0.0821, Train Steps/Sec: 0.06,
-[2026-01-30 20:42:13] (step=0004107) Train Loss mse: 0.0151, Train Loss ce: 0.0657, Train Steps/Sec: 0.06,
 [2026-01-30 20:42:28] (step=0004108) Train Loss mse: 0.0194, Train Loss ce: 0.0764, Train Steps/Sec: 0.07,
 [2026-01-30 20:42:48] (step=0004109) Train Loss mse: 0.0146, Train Loss ce: 0.0788, Train Steps/Sec: 0.05,
 [2026-01-30 20:43:07] (step=0004110) Train Loss mse: 0.0180, Train Loss ce: 0.0578, Train Steps/Sec: 0.05,