Upload checkpoints_vlm_gym_reference_dot_one_image_lr2e_5_mse_only_ins/checkpoints_vlm_gym_reference_dot_one_image_lr2e_5_mse_only_ins

checkpoints_vlm_gym_reference_dot_one_image_lr2e_5_mse_only_ins/checkpoints_vlm_gym_reference_dot_one_image_lr2e_5_mse_only_ins/wandb/offline-run-20260129_223634-vlm_gym_reference_dot_one_img_lr2e_5_mse_only_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: 1769726194
+    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:
+      452u6eq6qj9cpdr80e5nue3sk0cejbit:
+        os: Linux-6.6.93+-x86_64-with-glibc2.35
+        python: CPython 3.11.10
+        started_at: '2026-01-29T22:36:34.516284Z'
+        args:
+        - --dataset_config_file
+        - ./data/configs/vlm_gym_reference_dot_train_mseloss_only.yaml
+        - --eval_dataset_config_file
+        - ./data/configs/vlm_gym_reference_dot_eval_mseloss_only.yaml
+        - --viz_dataset_config_file
+        - ./data/configs/vlm_gym_reference_dot_eval_mseloss_only.yaml
+        - --inference_hash_file
+        - /home/clouduser/Code/Github/launch_new/hashes_test_set_v10.json
+        - --task_name
+        - reference_dot_v5
+        - --instructions_dir
+        - ./data/instructions
+        - --train_data_dir
+        - /home/clouduser/Code/data/gym/reference_dot_v5/train/
+        - --train_jsonl_path
+        - /home/clouduser/Code/data/gym/reference_dot_v5/train/
+        - --eval_data_dir
+        - /home/clouduser/Code/data/gym/reference_dot_v5/val/
+        - --eval_jsonl_path
+        - /home/clouduser/Code/data/gym/reference_dot_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
+        - '5000'
+        - --total_steps
+        - '5000'
+        - --text_cond_dropout_prob
+        - '0.0'
+        - --vae_cond_dropout_prob
+        - '0.0'
+        - --vit_cond_dropout_prob
+        - '0.0'
+        - --ema
+        - '0.993'
+        - --checkpoint_dir
+        - /dev/shm/models/checkpoints_vlm_gym_reference_dot_one_image_lr2e_5_mse_only_ins
+        - --wandb_project
+        - bagel
+        - --wandb_name
+        - vlm_gym_reference_dot_one_img_lr2e_5_mse_only_ins
+        - --wandb_dir
+        - /dev/shm/models/checkpoints_vlm_gym_reference_dot_one_image_lr2e_5_mse_only_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_reference_dot_one_image_lr2e_5_mse_only_ins
+        host: junyizhang-launch-new-226786220-1-0
+        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: '164465111040'
+        memory:
+          total: '1437332606976'
+        gpu_nvidia:
+        - name: NVIDIA A100-SXM4-80GB
+          memory_total: '85899345920'
+          cuda_cores: 6912
+          architecture: Ampere
+          uuid: GPU-6743ee30-8d44-3700-5053-982b7634dc72
+        - name: NVIDIA A100-SXM4-80GB
+          memory_total: '85899345920'
+          cuda_cores: 6912
+          architecture: Ampere
+          uuid: GPU-efdba308-be1c-fb5c-98af-5161d08502b8
+        - name: NVIDIA A100-SXM4-80GB
+          memory_total: '85899345920'
+          cuda_cores: 6912
+          architecture: Ampere
+          uuid: GPU-7a081892-5d05-b817-0e72-25e6d0f10d7a
+        - name: NVIDIA A100-SXM4-80GB
+          memory_total: '85899345920'
+          cuda_cores: 6912
+          architecture: Ampere
+          uuid: GPU-9be64e16-04d7-3a0f-eac3-d2b19251109b
+        - name: NVIDIA A100-SXM4-80GB
+          memory_total: '85899345920'
+          cuda_cores: 6912
+          architecture: Ampere
+          uuid: GPU-119e1d42-0ee4-815a-6fbf-d61349b517ef
+        - name: NVIDIA A100-SXM4-80GB
+          memory_total: '85899345920'
+          cuda_cores: 6912
+          architecture: Ampere
+          uuid: GPU-54053246-0cf6-b199-d31d-12946eb5de68
+        - name: NVIDIA A100-SXM4-80GB
+          memory_total: '85899345920'
+          cuda_cores: 6912
+          architecture: Ampere
+          uuid: GPU-3e4b216a-3bc7-064f-197e-5d85c243b0d3
+        - name: NVIDIA A100-SXM4-80GB
+          memory_total: '85899345920'
+          cuda_cores: 6912
+          architecture: Ampere
+          uuid: GPU-58fb92a9-aa05-fe07-638f-be3c40ba2436
+        cuda_version: '12.2'
+        writer_id: 452u6eq6qj9cpdr80e5nue3sk0cejbit
+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_reference_dot_one_image_lr2e_5_mse_only_ins
+wandb_project:
+  desc: null
+  value: bagel
+wandb_name:
+  desc: null
+  value: vlm_gym_reference_dot_one_img_lr2e_5_mse_only_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_reference_dot_one_image_lr2e_5_mse_only_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: 5000
+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_reference_dot_eval_mseloss_only.yaml
+viz_dataset_config_file:
+  desc: null
+  value: ./data/configs/vlm_gym_reference_dot_eval_mseloss_only.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: 0.0
+vit_cond_dropout_prob:
+  desc: null
+  value: 0.0
+dataset_config_file:
+  desc: null
+  value: ./data/configs/vlm_gym_reference_dot_train_mseloss_only.yaml
+train_data_dir:
+  desc: null
+  value: /home/clouduser/Code/data/gym/reference_dot_v5/train/
+train_jsonl_path:
+  desc: null
+  value: /home/clouduser/Code/data/gym/reference_dot_v5/train/
+eval_data_dir:
+  desc: null
+  value: /home/clouduser/Code/data/gym/reference_dot_v5/val/
+eval_jsonl_path:
+  desc: null
+  value: /home/clouduser/Code/data/gym/reference_dot_v5/val/
+inference_hash_file:
+  desc: null
+  value: /home/clouduser/Code/Github/launch_new/hashes_test_set_v10.json
+task_name:
+  desc: null
+  value: reference_dot_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_reference_dot_one_image_lr2e_5_mse_only_ins/checkpoints_vlm_gym_reference_dot_one_image_lr2e_5_mse_only_ins/wandb/offline-run-20260129_223634-vlm_gym_reference_dot_one_img_lr2e_5_mse_only_ins-run0/files/output.log

@@ -928,126 +928,6 @@ wandb: For more information, check out the docs at: https://weave-docs.wandb.ai/
 [2026-01-30 00:05:37] (step=0000917) Train Loss mse: 0.0048, Train Loss ce: 0.0000, Train Steps/Sec: 0.21,
 [2026-01-30 00:05:41] (step=0000918) Train Loss mse: 0.0047, Train Loss ce: 0.0000, Train Steps/Sec: 0.22,
 [2026-01-30 00:05:46] (step=0000919) Train Loss mse: 0.0048, Train Loss ce: 0.0000, Train Steps/Sec: 0.22,
-[2026-01-30 00:05:52] (step=0000920) Train Loss mse: 0.0035, Train Loss ce: 0.0000, Train Steps/Sec: 0.16,
-[2026-01-30 00:05:59] (step=0000921) Train Loss mse: 0.0069, Train Loss ce: 0.0000, Train Steps/Sec: 0.16,
-[2026-01-30 00:06:05] (step=0000922) Train Loss mse: 0.0039, Train Loss ce: 0.0000, Train Steps/Sec: 0.16,
-[2026-01-30 00:06:10] (step=0000923) Train Loss mse: 0.0113, Train Loss ce: 0.0000, Train Steps/Sec: 0.20,
-[2026-01-30 00:06:15] (step=0000924) Train Loss mse: 0.0051, Train Loss ce: 0.0000, Train Steps/Sec: 0.21,
-[2026-01-30 00:06:21] (step=0000925) Train Loss mse: 0.0054, Train Loss ce: 0.0000, Train Steps/Sec: 0.16,
-[2026-01-30 00:06:27] (step=0000926) Train Loss mse: 0.0053, Train Loss ce: 0.0000, Train Steps/Sec: 0.16,
-[2026-01-30 00:06:32] (step=0000927) Train Loss mse: 0.0053, Train Loss ce: 0.0000, Train Steps/Sec: 0.20,
-[2026-01-30 00:06:37] (step=0000928) Train Loss mse: 0.0049, Train Loss ce: 0.0000, Train Steps/Sec: 0.21,
-[2026-01-30 00:06:42] (step=0000929) Train Loss mse: 0.0062, Train Loss ce: 0.0000, Train Steps/Sec: 0.21,
-[2026-01-30 00:06:46] (step=0000930) Train Loss mse: 0.0122, Train Loss ce: 0.0000, Train Steps/Sec: 0.22,
-[2026-01-30 00:06:51] (step=0000931) Train Loss mse: 0.0056, Train Loss ce: 0.0000, Train Steps/Sec: 0.21,
-[2026-01-30 00:06:57] (step=0000932) Train Loss mse: 0.0069, Train Loss ce: 0.0000, Train Steps/Sec: 0.16,
-[2026-01-30 00:07:02] (step=0000933) Train Loss mse: 0.0061, Train Loss ce: 0.0000, Train Steps/Sec: 0.21,
-[2026-01-30 00:07:07] (step=0000934) Train Loss mse: 0.0071, Train Loss ce: 0.0000, Train Steps/Sec: 0.21,
-[2026-01-30 00:07:13] (step=0000935) Train Loss mse: 0.0064, Train Loss ce: 0.0000, Train Steps/Sec: 0.16,
-[2026-01-30 00:07:18] (step=0000936) Train Loss mse: 0.0062, Train Loss ce: 0.0000, Train Steps/Sec: 0.20,
-[2026-01-30 00:07:24] (step=0000937) Train Loss mse: 0.0049, Train Loss ce: 0.0000, Train Steps/Sec: 0.16,
-[2026-01-30 00:07:31] (step=0000938) Train Loss mse: 0.0051, Train Loss ce: 0.0000, Train Steps/Sec: 0.16,
-[2026-01-30 00:07:35] (step=0000939) Train Loss mse: 0.0066, Train Loss ce: 0.0000, Train Steps/Sec: 0.21,
-[2026-01-30 00:07:40] (step=0000940) Train Loss mse: 0.0065, Train Loss ce: 0.0000, Train Steps/Sec: 0.21,
-[2026-01-30 00:07:45] (step=0000941) Train Loss mse: 0.0077, Train Loss ce: 0.0000, Train Steps/Sec: 0.21,
-[2026-01-30 00:07:51] (step=0000942) Train Loss mse: 0.0053, Train Loss ce: 0.0000, Train Steps/Sec: 0.16,
-[2026-01-30 00:07:56] (step=0000943) Train Loss mse: 0.0047, Train Loss ce: 0.0000, Train Steps/Sec: 0.21,
-[2026-01-30 00:08:02] (step=0000944) Train Loss mse: 0.0054, Train Loss ce: 0.0000, Train Steps/Sec: 0.18,
-[2026-01-30 00:08:08] (step=0000945) Train Loss mse: 0.0053, Train Loss ce: 0.0000, Train Steps/Sec: 0.16,
-[2026-01-30 00:08:13] (step=0000946) Train Loss mse: 0.0049, Train Loss ce: 0.0000, Train Steps/Sec: 0.21,
-[2026-01-30 00:08:19] (step=0000947) Train Loss mse: 0.0058, Train Loss ce: 0.0000, Train Steps/Sec: 0.16,
-[2026-01-30 00:08:25] (step=0000948) Train Loss mse: 0.0055, Train Loss ce: 0.0000, Train Steps/Sec: 0.18,
-[2026-01-30 00:08:30] (step=0000949) Train Loss mse: 0.0057, Train Loss ce: 0.0000, Train Steps/Sec: 0.21,
-[2026-01-30 00:08:35] (step=0000950) Train Loss mse: 0.0057, Train Loss ce: 0.0000, Train Steps/Sec: 0.19,
-[2026-01-30 00:08:41] (step=0000951) Train Loss mse: 0.0051, Train Loss ce: 0.0000, Train Steps/Sec: 0.16,
-[2026-01-30 00:08:46] (step=0000952) Train Loss mse: 0.0063, Train Loss ce: 0.0000, Train Steps/Sec: 0.21,
-[2026-01-30 00:08:52] (step=0000953) Train Loss mse: 0.0065, Train Loss ce: 0.0000, Train Steps/Sec: 0.16,
-[2026-01-30 00:08:58] (step=0000954) Train Loss mse: 0.0048, Train Loss ce: 0.0000, Train Steps/Sec: 0.17,
-[2026-01-30 00:09:04] (step=0000955) Train Loss mse: 0.0047, Train Loss ce: 0.0000, Train Steps/Sec: 0.18,
-[2026-01-30 00:09:09] (step=0000956) Train Loss mse: 0.0056, Train Loss ce: 0.0000, Train Steps/Sec: 0.22,
-[2026-01-30 00:09:13] (step=0000957) Train Loss mse: 0.0067, Train Loss ce: 0.0000, Train Steps/Sec: 0.21,
-[2026-01-30 00:09:20] (step=0000958) Train Loss mse: 0.0065, Train Loss ce: 0.0000, Train Steps/Sec: 0.16,
-[2026-01-30 00:09:26] (step=0000959) Train Loss mse: 0.0050, Train Loss ce: 0.0000, Train Steps/Sec: 0.16,
-[2026-01-30 00:09:31] (step=0000960) Train Loss mse: 0.0068, Train Loss ce: 0.0000, Train Steps/Sec: 0.19,
-[2026-01-30 00:09:36] (step=0000961) Train Loss mse: 0.0043, Train Loss ce: 0.0000, Train Steps/Sec: 0.21,
-[2026-01-30 00:09:42] (step=0000962) Train Loss mse: 0.0062, Train Loss ce: 0.0000, Train Steps/Sec: 0.16,
-[2026-01-30 00:09:48] (step=0000963) Train Loss mse: 0.0068, Train Loss ce: 0.0000, Train Steps/Sec: 0.16,
-[2026-01-30 00:09:53] (step=0000964) Train Loss mse: 0.0070, Train Loss ce: 0.0000, Train Steps/Sec: 0.21,
-[2026-01-30 00:09:58] (step=0000965) Train Loss mse: 0.0052, Train Loss ce: 0.0000, Train Steps/Sec: 0.20,
-[2026-01-30 00:10:04] (step=0000966) Train Loss mse: 0.0059, Train Loss ce: 0.0000, Train Steps/Sec: 0.16,
-[2026-01-30 00:10:09] (step=0000967) Train Loss mse: 0.0079, Train Loss ce: 0.0000, Train Steps/Sec: 0.20,
-[2026-01-30 00:10:14] (step=0000968) Train Loss mse: 0.0074, Train Loss ce: 0.0000, Train Steps/Sec: 0.22,
-[2026-01-30 00:10:19] (step=0000969) Train Loss mse: 0.0052, Train Loss ce: 0.0000, Train Steps/Sec: 0.21,
-[2026-01-30 00:10:25] (step=0000970) Train Loss mse: 0.0053, Train Loss ce: 0.0000, Train Steps/Sec: 0.16,
-[2026-01-30 00:10:30] (step=0000971) Train Loss mse: 0.0046, Train Loss ce: 0.0000, Train Steps/Sec: 0.21,
-[2026-01-30 00:10:35] (step=0000972) Train Loss mse: 0.0047, Train Loss ce: 0.0000, Train Steps/Sec: 0.19,
-[2026-01-30 00:10:41] (step=0000973) Train Loss mse: 0.0042, Train Loss ce: 0.0000, Train Steps/Sec: 0.16,
-[2026-01-30 00:10:48] (step=0000974) Train Loss mse: 0.0057, Train Loss ce: 0.0000, Train Steps/Sec: 0.15,
-[2026-01-30 00:10:54] (step=0000975) Train Loss mse: 0.0084, Train Loss ce: 0.0000, Train Steps/Sec: 0.16,
-[2026-01-30 00:11:00] (step=0000976) Train Loss mse: 0.0054, Train Loss ce: 0.0000, Train Steps/Sec: 0.16,
-[2026-01-30 00:11:05] (step=0000977) Train Loss mse: 0.0048, Train Loss ce: 0.0000, Train Steps/Sec: 0.22,
-[2026-01-30 00:11:09] (step=0000978) Train Loss mse: 0.0051, Train Loss ce: 0.0000, Train Steps/Sec: 0.22,
-[2026-01-30 00:11:14] (step=0000979) Train Loss mse: 0.0046, Train Loss ce: 0.0000, Train Steps/Sec: 0.21,
-[2026-01-30 00:11:20] (step=0000980) Train Loss mse: 0.0051, Train Loss ce: 0.0000, Train Steps/Sec: 0.16,
-[2026-01-30 00:11:26] (step=0000981) Train Loss mse: 0.0054, Train Loss ce: 0.0000, Train Steps/Sec: 0.19,
-[2026-01-30 00:11:32] (step=0000982) Train Loss mse: 0.0040, Train Loss ce: 0.0000, Train Steps/Sec: 0.17,
-[2026-01-30 00:11:38] (step=0000983) Train Loss mse: 0.0047, Train Loss ce: 0.0000, Train Steps/Sec: 0.16,
-[2026-01-30 00:11:43] (step=0000984) Train Loss mse: 0.0044, Train Loss ce: 0.0000, Train Steps/Sec: 0.21,
-[2026-01-30 00:11:47] (step=0000985) Train Loss mse: 0.0048, Train Loss ce: 0.0000, Train Steps/Sec: 0.22,
-[2026-01-30 00:11:52] (step=0000986) Train Loss mse: 0.0063, Train Loss ce: 0.0000, Train Steps/Sec: 0.22,
-[2026-01-30 00:11:58] (step=0000987) Train Loss mse: 0.0035, Train Loss ce: 0.0000, Train Steps/Sec: 0.16,
-[2026-01-30 00:12:04] (step=0000988) Train Loss mse: 0.0053, Train Loss ce: 0.0000, Train Steps/Sec: 0.17,
-[2026-01-30 00:12:09] (step=0000989) Train Loss mse: 0.0052, Train Loss ce: 0.0000, Train Steps/Sec: 0.22,
-[2026-01-30 00:12:14] (step=0000990) Train Loss mse: 0.0067, Train Loss ce: 0.0000, Train Steps/Sec: 0.20,
-[2026-01-30 00:12:18] (step=0000991) Train Loss mse: 0.0045, Train Loss ce: 0.0000, Train Steps/Sec: 0.21,
-[2026-01-30 00:12:25] (step=0000992) Train Loss mse: 0.0044, Train Loss ce: 0.0000, Train Steps/Sec: 0.16,
-[2026-01-30 00:12:31] (step=0000993) Train Loss mse: 0.0047, Train Loss ce: 0.0000, Train Steps/Sec: 0.16,
-[2026-01-30 00:12:36] (step=0000994) Train Loss mse: 0.0039, Train Loss ce: 0.0000, Train Steps/Sec: 0.22,
-[2026-01-30 00:12:41] (step=0000995) Train Loss mse: 0.0051, Train Loss ce: 0.0000, Train Steps/Sec: 0.19,
-[2026-01-30 00:12:46] (step=0000996) Train Loss mse: 0.0039, Train Loss ce: 0.0000, Train Steps/Sec: 0.20,
-[2026-01-30 00:12:52] (step=0000997) Train Loss mse: 0.0042, Train Loss ce: 0.0000, Train Steps/Sec: 0.16,
-[2026-01-30 00:12:57] (step=0000998) Train Loss mse: 0.0043, Train Loss ce: 0.0000, Train Steps/Sec: 0.22,
-[2026-01-30 00:13:02] (step=0000999) Train Loss mse: 0.0057, Train Loss ce: 0.0000, Train Steps/Sec: 0.19,
-[2026-01-30 00:13:33] (step=0001000) Train Loss mse: 0.0065, Train Loss ce: 0.0000, Train Steps/Sec: 0.03,
-[2026-01-30 00:13:37] (step=0001001) Train Loss mse: 0.0048, Train Loss ce: 0.0000, Train Steps/Sec: 0.22,
-[2026-01-30 00:13:43] (step=0001002) Train Loss mse: 0.0054, Train Loss ce: 0.0000, Train Steps/Sec: 0.16,
-[2026-01-30 00:13:50] (step=0001003) Train Loss mse: 0.0046, Train Loss ce: 0.0000, Train Steps/Sec: 0.16,
-[2026-01-30 00:13:56] (step=0001004) Train Loss mse: 0.0052, Train Loss ce: 0.0000, Train Steps/Sec: 0.16,
-[2026-01-30 00:14:02] (step=0001005) Train Loss mse: 0.0053, Train Loss ce: 0.0000, Train Steps/Sec: 0.16,
-[2026-01-30 00:14:08] (step=0001006) Train Loss mse: 0.0042, Train Loss ce: 0.0000, Train Steps/Sec: 0.16,
-[2026-01-30 00:14:13] (step=0001007) Train Loss mse: 0.0050, Train Loss ce: 0.0000, Train Steps/Sec: 0.20,
-[2026-01-30 00:14:20] (step=0001008) Train Loss mse: 0.0055, Train Loss ce: 0.0000, Train Steps/Sec: 0.16,
-[2026-01-30 00:14:24] (step=0001009) Train Loss mse: 0.0051, Train Loss ce: 0.0000, Train Steps/Sec: 0.21,
-[2026-01-30 00:14:29] (step=0001010) Train Loss mse: 0.0051, Train Loss ce: 0.0000, Train Steps/Sec: 0.22,
-[2026-01-30 00:14:33] (step=0001011) Train Loss mse: 0.0060, Train Loss ce: 0.0000, Train Steps/Sec: 0.22,
-[2026-01-30 00:14:37] (step=0001012) Train Loss mse: 0.0056, Train Loss ce: 0.0000, Train Steps/Sec: 0.25,
-[2026-01-30 00:14:44] (step=0001013) Train Loss mse: 0.0049, Train Loss ce: 0.0000, Train Steps/Sec: 0.16,
-[2026-01-30 00:14:50] (step=0001014) Train Loss mse: 0.0048, Train Loss ce: 0.0000, Train Steps/Sec: 0.15,
-[2026-01-30 00:14:55] (step=0001015) Train Loss mse: 0.0050, Train Loss ce: 0.0000, Train Steps/Sec: 0.22,
-[2026-01-30 00:15:01] (step=0001016) Train Loss mse: 0.0049, Train Loss ce: 0.0000, Train Steps/Sec: 0.16,
-[2026-01-30 00:15:07] (step=0001017) Train Loss mse: 0.0054, Train Loss ce: 0.0000, Train Steps/Sec: 0.16,
-[2026-01-30 00:15:12] (step=0001018) Train Loss mse: 0.0051, Train Loss ce: 0.0000, Train Steps/Sec: 0.22,
-[2026-01-30 00:15:18] (step=0001019) Train Loss mse: 0.0055, Train Loss ce: 0.0000, Train Steps/Sec: 0.16,
-[2026-01-30 00:15:23] (step=0001020) Train Loss mse: 0.0050, Train Loss ce: 0.0000, Train Steps/Sec: 0.22,
-[2026-01-30 00:15:29] (step=0001021) Train Loss mse: 0.0046, Train Loss ce: 0.0000, Train Steps/Sec: 0.15,
-[2026-01-30 00:15:35] (step=0001022) Train Loss mse: 0.0056, Train Loss ce: 0.0000, Train Steps/Sec: 0.17,
-[2026-01-30 00:15:42] (step=0001023) Train Loss mse: 0.0043, Train Loss ce: 0.0000, Train Steps/Sec: 0.15,
-[2026-01-30 00:15:48] (step=0001024) Train Loss mse: 0.0045, Train Loss ce: 0.0000, Train Steps/Sec: 0.16,
-[2026-01-30 00:15:53] (step=0001025) Train Loss mse: 0.0046, Train Loss ce: 0.0000, Train Steps/Sec: 0.21,
-[2026-01-30 00:15:58] (step=0001026) Train Loss mse: 0.0053, Train Loss ce: 0.0000, Train Steps/Sec: 0.19,
-[2026-01-30 00:16:03] (step=0001027) Train Loss mse: 0.0079, Train Loss ce: 0.0000, Train Steps/Sec: 0.22,
-[2026-01-30 00:16:09] (step=0001028) Train Loss mse: 0.0054, Train Loss ce: 0.0000, Train Steps/Sec: 0.15,
-[2026-01-30 00:16:16] (step=0001029) Train Loss mse: 0.0049, Train Loss ce: 0.0000, Train Steps/Sec: 0.16,
-[2026-01-30 00:16:21] (step=0001030) Train Loss mse: 0.0066, Train Loss ce: 0.0000, Train Steps/Sec: 0.18,
-[2026-01-30 00:16:26] (step=0001031) Train Loss mse: 0.0048, Train Loss ce: 0.0000, Train Steps/Sec: 0.22,
-[2026-01-30 00:16:32] (step=0001032) Train Loss mse: 0.0049, Train Loss ce: 0.0000, Train Steps/Sec: 0.17,
-[2026-01-30 00:16:36] (step=0001033) Train Loss mse: 0.0049, Train Loss ce: 0.0000, Train Steps/Sec: 0.21,
-[2026-01-30 00:16:43] (step=0001034) Train Loss mse: 0.0054, Train Loss ce: 0.0000, Train Steps/Sec: 0.16,
-[2026-01-30 00:16:48] (step=0001035) Train Loss mse: 0.0061, Train Loss ce: 0.0000, Train Steps/Sec: 0.20,
-[2026-01-30 00:16:52] (step=0001036) Train Loss mse: 0.0040, Train Loss ce: 0.0000, Train Steps/Sec: 0.23,
-[2026-01-30 00:16:57] (step=0001037) Train Loss mse: 0.0040, Train Loss ce: 0.0000, Train Steps/Sec: 0.22,
-[2026-01-30 00:17:03] (step=0001038) Train Loss mse: 0.0043, Train Loss ce: 0.0000, Train Steps/Sec: 0.16,
-[2026-01-30 00:17:09] (step=0001039) Train Loss mse: 0.0046, Train Loss ce: 0.0000, Train Steps/Sec: 0.16,
 FullyShardedDataParallel(
   (_fsdp_wrapped_module): Bagel(
     (language_model): Qwen2ForCausalLM(
@@ -1234,13 +1114,6 @@ Preparing Dataset vlm_gym_reference_dot_mse_loss_only_evalonce/vlm_gym_reference
   fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_reference_dot_mse_loss_only_evalonce'}]
   fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_reference_dot_mse_loss_only_evalonce'}]
 ce_avg: 0.0, mse_avg: 0.0055910381488502026
-base_dir is /dev/shm/models/checkpoints_vlm_gym_reference_dot_one_image_lr2e_5_mse_only_ins/eval_used_rows, step_tag is vlm_gym_reference_dot_one_img_lr2e_5_mse_only_ins_step1000
-Preparing Dataset vlm_gym_reference_dot_mse_loss_only_evalonce/vlm_gym_reference_dot_val
-[eval debug] first 3 batch fingerprints:
-  fp[0]: [{'data_indexes': [0], 'worker_id': 0, 'dataset_name': 'vlm_gym_reference_dot_mse_loss_only_evalonce'}]
-  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_reference_dot_mse_loss_only_evalonce'}]
-  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_reference_dot_mse_loss_only_evalonce'}]
-ce_avg: 0.0, mse_avg: 0.005657645873725414
 base_dir is /dev/shm/models/checkpoints_vlm_gym_reference_dot_one_image_lr2e_5_mse_only_ins/eval_used_rows, step_tag is vlm_gym_reference_dot_one_img_lr2e_5_mse_only_ins_step1500
 Preparing Dataset vlm_gym_reference_dot_mse_loss_only_evalonce/vlm_gym_reference_dot_val
 [eval debug] first 3 batch fingerprints:
@@ -1248,13 +1121,126 @@ Preparing Dataset vlm_gym_reference_dot_mse_loss_only_evalonce/vlm_gym_reference
   fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_reference_dot_mse_loss_only_evalonce'}]
   fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_reference_dot_mse_loss_only_evalonce'}]
 ce_avg: 0.0, mse_avg: 0.0055648270063102245
-base_dir is /dev/shm/models/checkpoints_vlm_gym_reference_dot_one_image_lr2e_5_mse_only_ins/eval_used_rows, step_tag is vlm_gym_reference_dot_one_img_lr2e_5_mse_only_ins_step2000
-Preparing Dataset vlm_gym_reference_dot_mse_loss_only_evalonce/vlm_gym_reference_dot_val
-[eval debug] first 3 batch fingerprints:
-  fp[0]: [{'data_indexes': [0], 'worker_id': 0, 'dataset_name': 'vlm_gym_reference_dot_mse_loss_only_evalonce'}]
-  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_reference_dot_mse_loss_only_evalonce'}]
-  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_reference_dot_mse_loss_only_evalonce'}]
-ce_avg: 0.0, mse_avg: 0.0056802802719175816
+[2026-01-30 00:05:52] (step=0000920) Train Loss mse: 0.0035, Train Loss ce: 0.0000, Train Steps/Sec: 0.16,
+[2026-01-30 00:05:59] (step=0000921) Train Loss mse: 0.0069, Train Loss ce: 0.0000, Train Steps/Sec: 0.16,
+[2026-01-30 00:06:05] (step=0000922) Train Loss mse: 0.0039, Train Loss ce: 0.0000, Train Steps/Sec: 0.16,
+[2026-01-30 00:06:10] (step=0000923) Train Loss mse: 0.0113, Train Loss ce: 0.0000, Train Steps/Sec: 0.20,
+[2026-01-30 00:06:15] (step=0000924) Train Loss mse: 0.0051, Train Loss ce: 0.0000, Train Steps/Sec: 0.21,
+[2026-01-30 00:06:21] (step=0000925) Train Loss mse: 0.0054, Train Loss ce: 0.0000, Train Steps/Sec: 0.16,
+[2026-01-30 00:06:27] (step=0000926) Train Loss mse: 0.0053, Train Loss ce: 0.0000, Train Steps/Sec: 0.16,
+[2026-01-30 00:06:32] (step=0000927) Train Loss mse: 0.0053, Train Loss ce: 0.0000, Train Steps/Sec: 0.20,
+[2026-01-30 00:06:37] (step=0000928) Train Loss mse: 0.0049, Train Loss ce: 0.0000, Train Steps/Sec: 0.21,
+[2026-01-30 00:06:42] (step=0000929) Train Loss mse: 0.0062, Train Loss ce: 0.0000, Train Steps/Sec: 0.21,
+[2026-01-30 00:06:46] (step=0000930) Train Loss mse: 0.0122, Train Loss ce: 0.0000, Train Steps/Sec: 0.22,
+[2026-01-30 00:06:51] (step=0000931) Train Loss mse: 0.0056, Train Loss ce: 0.0000, Train Steps/Sec: 0.21,
+[2026-01-30 00:06:57] (step=0000932) Train Loss mse: 0.0069, Train Loss ce: 0.0000, Train Steps/Sec: 0.16,
+[2026-01-30 00:07:02] (step=0000933) Train Loss mse: 0.0061, Train Loss ce: 0.0000, Train Steps/Sec: 0.21,
+[2026-01-30 00:07:07] (step=0000934) Train Loss mse: 0.0071, Train Loss ce: 0.0000, Train Steps/Sec: 0.21,
+[2026-01-30 00:07:13] (step=0000935) Train Loss mse: 0.0064, Train Loss ce: 0.0000, Train Steps/Sec: 0.16,
+[2026-01-30 00:07:18] (step=0000936) Train Loss mse: 0.0062, Train Loss ce: 0.0000, Train Steps/Sec: 0.20,
+[2026-01-30 00:07:24] (step=0000937) Train Loss mse: 0.0049, Train Loss ce: 0.0000, Train Steps/Sec: 0.16,
+[2026-01-30 00:07:31] (step=0000938) Train Loss mse: 0.0051, Train Loss ce: 0.0000, Train Steps/Sec: 0.16,
+[2026-01-30 00:07:35] (step=0000939) Train Loss mse: 0.0066, Train Loss ce: 0.0000, Train Steps/Sec: 0.21,
+[2026-01-30 00:07:40] (step=0000940) Train Loss mse: 0.0065, Train Loss ce: 0.0000, Train Steps/Sec: 0.21,
+[2026-01-30 00:07:45] (step=0000941) Train Loss mse: 0.0077, Train Loss ce: 0.0000, Train Steps/Sec: 0.21,
+[2026-01-30 00:07:51] (step=0000942) Train Loss mse: 0.0053, Train Loss ce: 0.0000, Train Steps/Sec: 0.16,
+[2026-01-30 00:07:56] (step=0000943) Train Loss mse: 0.0047, Train Loss ce: 0.0000, Train Steps/Sec: 0.21,
+[2026-01-30 00:08:02] (step=0000944) Train Loss mse: 0.0054, Train Loss ce: 0.0000, Train Steps/Sec: 0.18,
+[2026-01-30 00:08:08] (step=0000945) Train Loss mse: 0.0053, Train Loss ce: 0.0000, Train Steps/Sec: 0.16,
+[2026-01-30 00:08:13] (step=0000946) Train Loss mse: 0.0049, Train Loss ce: 0.0000, Train Steps/Sec: 0.21,
+[2026-01-30 00:08:19] (step=0000947) Train Loss mse: 0.0058, Train Loss ce: 0.0000, Train Steps/Sec: 0.16,
+[2026-01-30 00:08:25] (step=0000948) Train Loss mse: 0.0055, Train Loss ce: 0.0000, Train Steps/Sec: 0.18,
+[2026-01-30 00:08:30] (step=0000949) Train Loss mse: 0.0057, Train Loss ce: 0.0000, Train Steps/Sec: 0.21,
+[2026-01-30 00:08:35] (step=0000950) Train Loss mse: 0.0057, Train Loss ce: 0.0000, Train Steps/Sec: 0.19,
+[2026-01-30 00:08:41] (step=0000951) Train Loss mse: 0.0051, Train Loss ce: 0.0000, Train Steps/Sec: 0.16,
+[2026-01-30 00:08:46] (step=0000952) Train Loss mse: 0.0063, Train Loss ce: 0.0000, Train Steps/Sec: 0.21,
+[2026-01-30 00:08:52] (step=0000953) Train Loss mse: 0.0065, Train Loss ce: 0.0000, Train Steps/Sec: 0.16,
+[2026-01-30 00:08:58] (step=0000954) Train Loss mse: 0.0048, Train Loss ce: 0.0000, Train Steps/Sec: 0.17,
+[2026-01-30 00:09:04] (step=0000955) Train Loss mse: 0.0047, Train Loss ce: 0.0000, Train Steps/Sec: 0.18,
+[2026-01-30 00:09:09] (step=0000956) Train Loss mse: 0.0056, Train Loss ce: 0.0000, Train Steps/Sec: 0.22,
+[2026-01-30 00:09:13] (step=0000957) Train Loss mse: 0.0067, Train Loss ce: 0.0000, Train Steps/Sec: 0.21,
+[2026-01-30 00:09:20] (step=0000958) Train Loss mse: 0.0065, Train Loss ce: 0.0000, Train Steps/Sec: 0.16,
+[2026-01-30 00:09:26] (step=0000959) Train Loss mse: 0.0050, Train Loss ce: 0.0000, Train Steps/Sec: 0.16,
+[2026-01-30 00:09:31] (step=0000960) Train Loss mse: 0.0068, Train Loss ce: 0.0000, Train Steps/Sec: 0.19,
+[2026-01-30 00:09:36] (step=0000961) Train Loss mse: 0.0043, Train Loss ce: 0.0000, Train Steps/Sec: 0.21,
+[2026-01-30 00:09:42] (step=0000962) Train Loss mse: 0.0062, Train Loss ce: 0.0000, Train Steps/Sec: 0.16,
+[2026-01-30 00:09:48] (step=0000963) Train Loss mse: 0.0068, Train Loss ce: 0.0000, Train Steps/Sec: 0.16,
+[2026-01-30 00:09:53] (step=0000964) Train Loss mse: 0.0070, Train Loss ce: 0.0000, Train Steps/Sec: 0.21,
+[2026-01-30 00:09:58] (step=0000965) Train Loss mse: 0.0052, Train Loss ce: 0.0000, Train Steps/Sec: 0.20,
+[2026-01-30 00:10:04] (step=0000966) Train Loss mse: 0.0059, Train Loss ce: 0.0000, Train Steps/Sec: 0.16,
+[2026-01-30 00:10:09] (step=0000967) Train Loss mse: 0.0079, Train Loss ce: 0.0000, Train Steps/Sec: 0.20,
+[2026-01-30 00:10:14] (step=0000968) Train Loss mse: 0.0074, Train Loss ce: 0.0000, Train Steps/Sec: 0.22,
+[2026-01-30 00:10:19] (step=0000969) Train Loss mse: 0.0052, Train Loss ce: 0.0000, Train Steps/Sec: 0.21,
+[2026-01-30 00:10:25] (step=0000970) Train Loss mse: 0.0053, Train Loss ce: 0.0000, Train Steps/Sec: 0.16,
+[2026-01-30 00:10:30] (step=0000971) Train Loss mse: 0.0046, Train Loss ce: 0.0000, Train Steps/Sec: 0.21,
+[2026-01-30 00:10:35] (step=0000972) Train Loss mse: 0.0047, Train Loss ce: 0.0000, Train Steps/Sec: 0.19,
+[2026-01-30 00:10:41] (step=0000973) Train Loss mse: 0.0042, Train Loss ce: 0.0000, Train Steps/Sec: 0.16,
+[2026-01-30 00:10:48] (step=0000974) Train Loss mse: 0.0057, Train Loss ce: 0.0000, Train Steps/Sec: 0.15,
+[2026-01-30 00:10:54] (step=0000975) Train Loss mse: 0.0084, Train Loss ce: 0.0000, Train Steps/Sec: 0.16,
+[2026-01-30 00:11:00] (step=0000976) Train Loss mse: 0.0054, Train Loss ce: 0.0000, Train Steps/Sec: 0.16,
+[2026-01-30 00:11:05] (step=0000977) Train Loss mse: 0.0048, Train Loss ce: 0.0000, Train Steps/Sec: 0.22,
+[2026-01-30 00:11:09] (step=0000978) Train Loss mse: 0.0051, Train Loss ce: 0.0000, Train Steps/Sec: 0.22,
+[2026-01-30 00:11:14] (step=0000979) Train Loss mse: 0.0046, Train Loss ce: 0.0000, Train Steps/Sec: 0.21,
+[2026-01-30 00:11:20] (step=0000980) Train Loss mse: 0.0051, Train Loss ce: 0.0000, Train Steps/Sec: 0.16,
+[2026-01-30 00:11:26] (step=0000981) Train Loss mse: 0.0054, Train Loss ce: 0.0000, Train Steps/Sec: 0.19,
+[2026-01-30 00:11:32] (step=0000982) Train Loss mse: 0.0040, Train Loss ce: 0.0000, Train Steps/Sec: 0.17,
+[2026-01-30 00:11:38] (step=0000983) Train Loss mse: 0.0047, Train Loss ce: 0.0000, Train Steps/Sec: 0.16,
+[2026-01-30 00:11:43] (step=0000984) Train Loss mse: 0.0044, Train Loss ce: 0.0000, Train Steps/Sec: 0.21,
+[2026-01-30 00:11:47] (step=0000985) Train Loss mse: 0.0048, Train Loss ce: 0.0000, Train Steps/Sec: 0.22,
+[2026-01-30 00:11:52] (step=0000986) Train Loss mse: 0.0063, Train Loss ce: 0.0000, Train Steps/Sec: 0.22,
+[2026-01-30 00:11:58] (step=0000987) Train Loss mse: 0.0035, Train Loss ce: 0.0000, Train Steps/Sec: 0.16,
+[2026-01-30 00:12:04] (step=0000988) Train Loss mse: 0.0053, Train Loss ce: 0.0000, Train Steps/Sec: 0.17,
+[2026-01-30 00:12:09] (step=0000989) Train Loss mse: 0.0052, Train Loss ce: 0.0000, Train Steps/Sec: 0.22,
+[2026-01-30 00:12:14] (step=0000990) Train Loss mse: 0.0067, Train Loss ce: 0.0000, Train Steps/Sec: 0.20,
+[2026-01-30 00:12:18] (step=0000991) Train Loss mse: 0.0045, Train Loss ce: 0.0000, Train Steps/Sec: 0.21,
+[2026-01-30 00:12:25] (step=0000992) Train Loss mse: 0.0044, Train Loss ce: 0.0000, Train Steps/Sec: 0.16,
+[2026-01-30 00:12:31] (step=0000993) Train Loss mse: 0.0047, Train Loss ce: 0.0000, Train Steps/Sec: 0.16,
+[2026-01-30 00:12:36] (step=0000994) Train Loss mse: 0.0039, Train Loss ce: 0.0000, Train Steps/Sec: 0.22,
+[2026-01-30 00:12:41] (step=0000995) Train Loss mse: 0.0051, Train Loss ce: 0.0000, Train Steps/Sec: 0.19,
+[2026-01-30 00:12:46] (step=0000996) Train Loss mse: 0.0039, Train Loss ce: 0.0000, Train Steps/Sec: 0.20,
+[2026-01-30 00:12:52] (step=0000997) Train Loss mse: 0.0042, Train Loss ce: 0.0000, Train Steps/Sec: 0.16,
+[2026-01-30 00:12:57] (step=0000998) Train Loss mse: 0.0043, Train Loss ce: 0.0000, Train Steps/Sec: 0.22,
+[2026-01-30 00:13:02] (step=0000999) Train Loss mse: 0.0057, Train Loss ce: 0.0000, Train Steps/Sec: 0.19,
+[2026-01-30 00:13:33] (step=0001000) Train Loss mse: 0.0065, Train Loss ce: 0.0000, Train Steps/Sec: 0.03,
+[2026-01-30 00:13:37] (step=0001001) Train Loss mse: 0.0048, Train Loss ce: 0.0000, Train Steps/Sec: 0.22,
+[2026-01-30 00:13:43] (step=0001002) Train Loss mse: 0.0054, Train Loss ce: 0.0000, Train Steps/Sec: 0.16,
+[2026-01-30 00:13:50] (step=0001003) Train Loss mse: 0.0046, Train Loss ce: 0.0000, Train Steps/Sec: 0.16,
+[2026-01-30 00:13:56] (step=0001004) Train Loss mse: 0.0052, Train Loss ce: 0.0000, Train Steps/Sec: 0.16,
+[2026-01-30 00:14:02] (step=0001005) Train Loss mse: 0.0053, Train Loss ce: 0.0000, Train Steps/Sec: 0.16,
+[2026-01-30 00:14:08] (step=0001006) Train Loss mse: 0.0042, Train Loss ce: 0.0000, Train Steps/Sec: 0.16,
+[2026-01-30 00:14:13] (step=0001007) Train Loss mse: 0.0050, Train Loss ce: 0.0000, Train Steps/Sec: 0.20,
+[2026-01-30 00:14:20] (step=0001008) Train Loss mse: 0.0055, Train Loss ce: 0.0000, Train Steps/Sec: 0.16,
+[2026-01-30 00:14:24] (step=0001009) Train Loss mse: 0.0051, Train Loss ce: 0.0000, Train Steps/Sec: 0.21,
+[2026-01-30 00:14:29] (step=0001010) Train Loss mse: 0.0051, Train Loss ce: 0.0000, Train Steps/Sec: 0.22,
+[2026-01-30 00:14:33] (step=0001011) Train Loss mse: 0.0060, Train Loss ce: 0.0000, Train Steps/Sec: 0.22,
+[2026-01-30 00:14:37] (step=0001012) Train Loss mse: 0.0056, Train Loss ce: 0.0000, Train Steps/Sec: 0.25,
+[2026-01-30 00:14:44] (step=0001013) Train Loss mse: 0.0049, Train Loss ce: 0.0000, Train Steps/Sec: 0.16,
+[2026-01-30 00:14:50] (step=0001014) Train Loss mse: 0.0048, Train Loss ce: 0.0000, Train Steps/Sec: 0.15,
+[2026-01-30 00:14:55] (step=0001015) Train Loss mse: 0.0050, Train Loss ce: 0.0000, Train Steps/Sec: 0.22,
+[2026-01-30 00:15:01] (step=0001016) Train Loss mse: 0.0049, Train Loss ce: 0.0000, Train Steps/Sec: 0.16,
+[2026-01-30 00:15:07] (step=0001017) Train Loss mse: 0.0054, Train Loss ce: 0.0000, Train Steps/Sec: 0.16,
+[2026-01-30 00:15:12] (step=0001018) Train Loss mse: 0.0051, Train Loss ce: 0.0000, Train Steps/Sec: 0.22,
+[2026-01-30 00:15:18] (step=0001019) Train Loss mse: 0.0055, Train Loss ce: 0.0000, Train Steps/Sec: 0.16,
+[2026-01-30 00:15:23] (step=0001020) Train Loss mse: 0.0050, Train Loss ce: 0.0000, Train Steps/Sec: 0.22,
+[2026-01-30 00:15:29] (step=0001021) Train Loss mse: 0.0046, Train Loss ce: 0.0000, Train Steps/Sec: 0.15,
+[2026-01-30 00:15:35] (step=0001022) Train Loss mse: 0.0056, Train Loss ce: 0.0000, Train Steps/Sec: 0.17,
+[2026-01-30 00:15:42] (step=0001023) Train Loss mse: 0.0043, Train Loss ce: 0.0000, Train Steps/Sec: 0.15,
+[2026-01-30 00:15:48] (step=0001024) Train Loss mse: 0.0045, Train Loss ce: 0.0000, Train Steps/Sec: 0.16,
+[2026-01-30 00:15:53] (step=0001025) Train Loss mse: 0.0046, Train Loss ce: 0.0000, Train Steps/Sec: 0.21,
+[2026-01-30 00:15:58] (step=0001026) Train Loss mse: 0.0053, Train Loss ce: 0.0000, Train Steps/Sec: 0.19,
+[2026-01-30 00:16:03] (step=0001027) Train Loss mse: 0.0079, Train Loss ce: 0.0000, Train Steps/Sec: 0.22,
+[2026-01-30 00:16:09] (step=0001028) Train Loss mse: 0.0054, Train Loss ce: 0.0000, Train Steps/Sec: 0.15,
+[2026-01-30 00:16:16] (step=0001029) Train Loss mse: 0.0049, Train Loss ce: 0.0000, Train Steps/Sec: 0.16,
+[2026-01-30 00:16:21] (step=0001030) Train Loss mse: 0.0066, Train Loss ce: 0.0000, Train Steps/Sec: 0.18,
+[2026-01-30 00:16:26] (step=0001031) Train Loss mse: 0.0048, Train Loss ce: 0.0000, Train Steps/Sec: 0.22,
+[2026-01-30 00:16:32] (step=0001032) Train Loss mse: 0.0049, Train Loss ce: 0.0000, Train Steps/Sec: 0.17,
+[2026-01-30 00:16:36] (step=0001033) Train Loss mse: 0.0049, Train Loss ce: 0.0000, Train Steps/Sec: 0.21,
+[2026-01-30 00:16:43] (step=0001034) Train Loss mse: 0.0054, Train Loss ce: 0.0000, Train Steps/Sec: 0.16,
+[2026-01-30 00:16:48] (step=0001035) Train Loss mse: 0.0061, Train Loss ce: 0.0000, Train Steps/Sec: 0.20,
+[2026-01-30 00:16:52] (step=0001036) Train Loss mse: 0.0040, Train Loss ce: 0.0000, Train Steps/Sec: 0.23,
+[2026-01-30 00:16:57] (step=0001037) Train Loss mse: 0.0040, Train Loss ce: 0.0000, Train Steps/Sec: 0.22,
+[2026-01-30 00:17:03] (step=0001038) Train Loss mse: 0.0043, Train Loss ce: 0.0000, Train Steps/Sec: 0.16,
+[2026-01-30 00:17:09] (step=0001039) Train Loss mse: 0.0046, Train Loss ce: 0.0000, Train Steps/Sec: 0.16,
 [2026-01-30 00:17:14] (step=0001040) Train Loss mse: 0.0034, Train Loss ce: 0.0000, Train Steps/Sec: 0.21,
 [2026-01-30 00:17:19] (step=0001041) Train Loss mse: 0.0067, Train Loss ce: 0.0000, Train Steps/Sec: 0.21,
 [2026-01-30 00:17:24] (step=0001042) Train Loss mse: 0.0063, Train Loss ce: 0.0000, Train Steps/Sec: 0.20,
@@ -2518,6 +2504,27 @@ ce_avg: 0.0, mse_avg: 0.0056802802719175816
 [2026-01-30 02:13:09] (step=0002300) Train Loss mse: 0.0045, Train Loss ce: 0.0000, Train Steps/Sec: 0.16,
 [2026-01-30 02:13:14] (step=0002301) Train Loss mse: 0.0044, Train Loss ce: 0.0000, Train Steps/Sec: 0.21,
 [2026-01-30 02:13:20] (step=0002302) Train Loss mse: 0.0046, Train Loss ce: 0.0000, Train Steps/Sec: 0.17,
+base_dir is /dev/shm/models/checkpoints_vlm_gym_reference_dot_one_image_lr2e_5_mse_only_ins/eval_used_rows, step_tag is vlm_gym_reference_dot_one_img_lr2e_5_mse_only_ins_step2000
+Preparing Dataset vlm_gym_reference_dot_mse_loss_only_evalonce/vlm_gym_reference_dot_val
+[eval debug] first 3 batch fingerprints:
+  fp[0]: [{'data_indexes': [0], 'worker_id': 0, 'dataset_name': 'vlm_gym_reference_dot_mse_loss_only_evalonce'}]
+  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_reference_dot_mse_loss_only_evalonce'}]
+  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_reference_dot_mse_loss_only_evalonce'}]
+ce_avg: 0.0, mse_avg: 0.0056802802719175816
+base_dir is /dev/shm/models/checkpoints_vlm_gym_reference_dot_one_image_lr2e_5_mse_only_ins/eval_used_rows, step_tag is vlm_gym_reference_dot_one_img_lr2e_5_mse_only_ins_step2500
+Preparing Dataset vlm_gym_reference_dot_mse_loss_only_evalonce/vlm_gym_reference_dot_val
+[eval debug] first 3 batch fingerprints:
+  fp[0]: [{'data_indexes': [0], 'worker_id': 0, 'dataset_name': 'vlm_gym_reference_dot_mse_loss_only_evalonce'}]
+  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_reference_dot_mse_loss_only_evalonce'}]
+  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_reference_dot_mse_loss_only_evalonce'}]
+ce_avg: 0.0, mse_avg: 0.005605650134384632
+base_dir is /dev/shm/models/checkpoints_vlm_gym_reference_dot_one_image_lr2e_5_mse_only_ins/eval_used_rows, step_tag is vlm_gym_reference_dot_one_img_lr2e_5_mse_only_ins_step3000
+Preparing Dataset vlm_gym_reference_dot_mse_loss_only_evalonce/vlm_gym_reference_dot_val
+[eval debug] first 3 batch fingerprints:
+  fp[0]: [{'data_indexes': [0], 'worker_id': 0, 'dataset_name': 'vlm_gym_reference_dot_mse_loss_only_evalonce'}]
+  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_reference_dot_mse_loss_only_evalonce'}]
+  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_reference_dot_mse_loss_only_evalonce'}]
+ce_avg: 0.0, mse_avg: 0.005620267707854509
 [2026-01-30 02:13:25] (step=0002303) Train Loss mse: 0.0041, Train Loss ce: 0.0000, Train Steps/Sec: 0.17,
 [2026-01-30 02:13:30] (step=0002304) Train Loss mse: 0.0048, Train Loss ce: 0.0000, Train Steps/Sec: 0.22,
 [2026-01-30 02:13:36] (step=0002305) Train Loss mse: 0.0039, Train Loss ce: 0.0000, Train Steps/Sec: 0.15,
@@ -2621,20 +2628,6 @@ ce_avg: 0.0, mse_avg: 0.0056802802719175816
 [2026-01-30 02:22:40] (step=0002403) Train Loss mse: 0.0045, Train Loss ce: 0.0000, Train Steps/Sec: 0.16,
 [2026-01-30 02:22:45] (step=0002404) Train Loss mse: 0.0047, Train Loss ce: 0.0000, Train Steps/Sec: 0.20,
 [2026-01-30 02:22:50] (step=0002405) Train Loss mse: 0.0048, Train Loss ce: 0.0000, Train Steps/Sec: 0.22,
-base_dir is /dev/shm/models/checkpoints_vlm_gym_reference_dot_one_image_lr2e_5_mse_only_ins/eval_used_rows, step_tag is vlm_gym_reference_dot_one_img_lr2e_5_mse_only_ins_step2500
-Preparing Dataset vlm_gym_reference_dot_mse_loss_only_evalonce/vlm_gym_reference_dot_val
-[eval debug] first 3 batch fingerprints:
-  fp[0]: [{'data_indexes': [0], 'worker_id': 0, 'dataset_name': 'vlm_gym_reference_dot_mse_loss_only_evalonce'}]
-  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_reference_dot_mse_loss_only_evalonce'}]
-  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_reference_dot_mse_loss_only_evalonce'}]
-ce_avg: 0.0, mse_avg: 0.005605650134384632
-base_dir is /dev/shm/models/checkpoints_vlm_gym_reference_dot_one_image_lr2e_5_mse_only_ins/eval_used_rows, step_tag is vlm_gym_reference_dot_one_img_lr2e_5_mse_only_ins_step3000
-Preparing Dataset vlm_gym_reference_dot_mse_loss_only_evalonce/vlm_gym_reference_dot_val
-[eval debug] first 3 batch fingerprints:
-  fp[0]: [{'data_indexes': [0], 'worker_id': 0, 'dataset_name': 'vlm_gym_reference_dot_mse_loss_only_evalonce'}]
-  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_reference_dot_mse_loss_only_evalonce'}]
-  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_reference_dot_mse_loss_only_evalonce'}]
-ce_avg: 0.0, mse_avg: 0.005620267707854509
 [2026-01-30 02:22:56] (step=0002406) Train Loss mse: 0.0039, Train Loss ce: 0.0000, Train Steps/Sec: 0.16,
 [2026-01-30 02:23:01] (step=0002407) Train Loss mse: 0.0048, Train Loss ce: 0.0000, Train Steps/Sec: 0.19,
 [2026-01-30 02:23:06] (step=0002408) Train Loss mse: 0.0050, Train Loss ce: 0.0000, Train Steps/Sec: 0.21,
@@ -3549,6 +3542,27 @@ ce_avg: 0.0, mse_avg: 0.005620267707854509
 [2026-01-30 03:46:48] (step=0003317) Train Loss mse: 0.0037, Train Loss ce: 0.0000, Train Steps/Sec: 0.16,
 [2026-01-30 03:46:53] (step=0003318) Train Loss mse: 0.0042, Train Loss ce: 0.0000, Train Steps/Sec: 0.22,
 [2026-01-30 03:46:59] (step=0003319) Train Loss mse: 0.0049, Train Loss ce: 0.0000, Train Steps/Sec: 0.16,
+base_dir is /dev/shm/models/checkpoints_vlm_gym_reference_dot_one_image_lr2e_5_mse_only_ins/eval_used_rows, step_tag is vlm_gym_reference_dot_one_img_lr2e_5_mse_only_ins_step3500
+Preparing Dataset vlm_gym_reference_dot_mse_loss_only_evalonce/vlm_gym_reference_dot_val
+[eval debug] first 3 batch fingerprints:
+  fp[0]: [{'data_indexes': [0], 'worker_id': 0, 'dataset_name': 'vlm_gym_reference_dot_mse_loss_only_evalonce'}]
+  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_reference_dot_mse_loss_only_evalonce'}]
+  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_reference_dot_mse_loss_only_evalonce'}]
+ce_avg: 0.0, mse_avg: 0.005567264277487993
+base_dir is /dev/shm/models/checkpoints_vlm_gym_reference_dot_one_image_lr2e_5_mse_only_ins/eval_used_rows, step_tag is vlm_gym_reference_dot_one_img_lr2e_5_mse_only_ins_step4000
+Preparing Dataset vlm_gym_reference_dot_mse_loss_only_evalonce/vlm_gym_reference_dot_val
+[eval debug] first 3 batch fingerprints:
+  fp[0]: [{'data_indexes': [0], 'worker_id': 0, 'dataset_name': 'vlm_gym_reference_dot_mse_loss_only_evalonce'}]
+  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_reference_dot_mse_loss_only_evalonce'}]
+  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_reference_dot_mse_loss_only_evalonce'}]
+ce_avg: 0.0, mse_avg: 0.005860968492925167
+base_dir is /dev/shm/models/checkpoints_vlm_gym_reference_dot_one_image_lr2e_5_mse_only_ins/eval_used_rows, step_tag is vlm_gym_reference_dot_one_img_lr2e_5_mse_only_ins_step4500
+Preparing Dataset vlm_gym_reference_dot_mse_loss_only_evalonce/vlm_gym_reference_dot_val
+[eval debug] first 3 batch fingerprints:
+  fp[0]: [{'data_indexes': [0], 'worker_id': 0, 'dataset_name': 'vlm_gym_reference_dot_mse_loss_only_evalonce'}]
+  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_reference_dot_mse_loss_only_evalonce'}]
+  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_reference_dot_mse_loss_only_evalonce'}]
+ce_avg: 0.0, mse_avg: 0.0062998272478580475
 [2026-01-30 03:47:05] (step=0003320) Train Loss mse: 0.0041, Train Loss ce: 0.0000, Train Steps/Sec: 0.16,
 [2026-01-30 03:47:11] (step=0003321) Train Loss mse: 0.0060, Train Loss ce: 0.0000, Train Steps/Sec: 0.16,
 [2026-01-30 03:47:16] (step=0003322) Train Loss mse: 0.0057, Train Loss ce: 0.0000, Train Steps/Sec: 0.20,
@@ -3595,27 +3609,53 @@ ce_avg: 0.0, mse_avg: 0.005620267707854509
 [2026-01-30 03:51:02] (step=0003363) Train Loss mse: 0.0041, Train Loss ce: 0.0000, Train Steps/Sec: 0.19,
 [2026-01-30 03:51:07] (step=0003364) Train Loss mse: 0.0062, Train Loss ce: 0.0000, Train Steps/Sec: 0.20,
 [2026-01-30 03:51:13] (step=0003365) Train Loss mse: 0.0040, Train Loss ce: 0.0000, Train Steps/Sec: 0.19,
-base_dir is /dev/shm/models/checkpoints_vlm_gym_reference_dot_one_image_lr2e_5_mse_only_ins/eval_used_rows, step_tag is vlm_gym_reference_dot_one_img_lr2e_5_mse_only_ins_step3500
-Preparing Dataset vlm_gym_reference_dot_mse_loss_only_evalonce/vlm_gym_reference_dot_val
-[eval debug] first 3 batch fingerprints:
-  fp[0]: [{'data_indexes': [0], 'worker_id': 0, 'dataset_name': 'vlm_gym_reference_dot_mse_loss_only_evalonce'}]
-  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_reference_dot_mse_loss_only_evalonce'}]
-  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_reference_dot_mse_loss_only_evalonce'}]
-ce_avg: 0.0, mse_avg: 0.005567264277487993
-base_dir is /dev/shm/models/checkpoints_vlm_gym_reference_dot_one_image_lr2e_5_mse_only_ins/eval_used_rows, step_tag is vlm_gym_reference_dot_one_img_lr2e_5_mse_only_ins_step4000
-Preparing Dataset vlm_gym_reference_dot_mse_loss_only_evalonce/vlm_gym_reference_dot_val
-[eval debug] first 3 batch fingerprints:
-  fp[0]: [{'data_indexes': [0], 'worker_id': 0, 'dataset_name': 'vlm_gym_reference_dot_mse_loss_only_evalonce'}]
-  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_reference_dot_mse_loss_only_evalonce'}]
-  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_reference_dot_mse_loss_only_evalonce'}]
-ce_avg: 0.0, mse_avg: 0.005860968492925167
-base_dir is /dev/shm/models/checkpoints_vlm_gym_reference_dot_one_image_lr2e_5_mse_only_ins/eval_used_rows, step_tag is vlm_gym_reference_dot_one_img_lr2e_5_mse_only_ins_step4500
-Preparing Dataset vlm_gym_reference_dot_mse_loss_only_evalonce/vlm_gym_reference_dot_val
-[eval debug] first 3 batch fingerprints:
-  fp[0]: [{'data_indexes': [0], 'worker_id': 0, 'dataset_name': 'vlm_gym_reference_dot_mse_loss_only_evalonce'}]
-  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_reference_dot_mse_loss_only_evalonce'}]
-  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_reference_dot_mse_loss_only_evalonce'}]
-ce_avg: 0.0, mse_avg: 0.0062998272478580475
+[2026-01-30 03:51:17] (step=0003366) Train Loss mse: 0.0046, Train Loss ce: 0.0000, Train Steps/Sec: 0.21,
+[2026-01-30 03:51:22] (step=0003367) Train Loss mse: 0.0050, Train Loss ce: 0.0000, Train Steps/Sec: 0.21,
+[2026-01-30 03:51:28] (step=0003368) Train Loss mse: 0.0042, Train Loss ce: 0.0000, Train Steps/Sec: 0.16,
+[2026-01-30 03:51:35] (step=0003369) Train Loss mse: 0.0053, Train Loss ce: 0.0000, Train Steps/Sec: 0.15,
+[2026-01-30 03:51:40] (step=0003370) Train Loss mse: 0.0043, Train Loss ce: 0.0000, Train Steps/Sec: 0.20,
+[2026-01-30 03:51:45] (step=0003371) Train Loss mse: 0.0038, Train Loss ce: 0.0000, Train Steps/Sec: 0.20,
+[2026-01-30 03:51:50] (step=0003372) Train Loss mse: 0.0042, Train Loss ce: 0.0000, Train Steps/Sec: 0.20,
+[2026-01-30 03:51:55] (step=0003373) Train Loss mse: 0.0039, Train Loss ce: 0.0000, Train Steps/Sec: 0.19,
+[2026-01-30 03:52:00] (step=0003374) Train Loss mse: 0.0042, Train Loss ce: 0.0000, Train Steps/Sec: 0.22,
+[2026-01-30 03:52:05] (step=0003375) Train Loss mse: 0.0039, Train Loss ce: 0.0000, Train Steps/Sec: 0.20,
+[2026-01-30 03:52:10] (step=0003376) Train Loss mse: 0.0033, Train Loss ce: 0.0000, Train Steps/Sec: 0.18,
+[2026-01-30 03:52:15] (step=0003377) Train Loss mse: 0.0043, Train Loss ce: 0.0000, Train Steps/Sec: 0.21,
+[2026-01-30 03:52:20] (step=0003378) Train Loss mse: 0.0035, Train Loss ce: 0.0000, Train Steps/Sec: 0.22,
+[2026-01-30 03:52:26] (step=0003379) Train Loss mse: 0.0034, Train Loss ce: 0.0000, Train Steps/Sec: 0.17,
+[2026-01-30 03:52:32] (step=0003380) Train Loss mse: 0.0055, Train Loss ce: 0.0000, Train Steps/Sec: 0.16,
+[2026-01-30 03:52:38] (step=0003381) Train Loss mse: 0.0040, Train Loss ce: 0.0000, Train Steps/Sec: 0.18,
+[2026-01-30 03:52:43] (step=0003382) Train Loss mse: 0.0044, Train Loss ce: 0.0000, Train Steps/Sec: 0.18,
+[2026-01-30 03:52:48] (step=0003383) Train Loss mse: 0.0048, Train Loss ce: 0.0000, Train Steps/Sec: 0.20,
+[2026-01-30 03:52:54] (step=0003384) Train Loss mse: 0.0043, Train Loss ce: 0.0000, Train Steps/Sec: 0.16,
+[2026-01-30 03:52:59] (step=0003385) Train Loss mse: 0.0053, Train Loss ce: 0.0000, Train Steps/Sec: 0.22,
+[2026-01-30 03:53:04] (step=0003386) Train Loss mse: 0.0034, Train Loss ce: 0.0000, Train Steps/Sec: 0.18,
+[2026-01-30 03:53:11] (step=0003387) Train Loss mse: 0.0042, Train Loss ce: 0.0000, Train Steps/Sec: 0.16,
+[2026-01-30 03:53:17] (step=0003388) Train Loss mse: 0.0044, Train Loss ce: 0.0000, Train Steps/Sec: 0.16,
+[2026-01-30 03:53:23] (step=0003389) Train Loss mse: 0.0045, Train Loss ce: 0.0000, Train Steps/Sec: 0.16,
+[2026-01-30 03:53:29] (step=0003390) Train Loss mse: 0.0042, Train Loss ce: 0.0000, Train Steps/Sec: 0.16,
+[2026-01-30 03:53:34] (step=0003391) Train Loss mse: 0.0038, Train Loss ce: 0.0000, Train Steps/Sec: 0.22,
+[2026-01-30 03:53:40] (step=0003392) Train Loss mse: 0.0039, Train Loss ce: 0.0000, Train Steps/Sec: 0.16,
+[2026-01-30 03:53:45] (step=0003393) Train Loss mse: 0.0042, Train Loss ce: 0.0000, Train Steps/Sec: 0.22,
+[2026-01-30 03:53:50] (step=0003394) Train Loss mse: 0.0054, Train Loss ce: 0.0000, Train Steps/Sec: 0.20,
+[2026-01-30 03:53:55] (step=0003395) Train Loss mse: 0.0042, Train Loss ce: 0.0000, Train Steps/Sec: 0.19,
+[2026-01-30 03:54:00] (step=0003396) Train Loss mse: 0.0050, Train Loss ce: 0.0000, Train Steps/Sec: 0.20,
+[2026-01-30 03:54:05] (step=0003397) Train Loss mse: 0.0046, Train Loss ce: 0.0000, Train Steps/Sec: 0.20,
+[2026-01-30 03:54:12] (step=0003398) Train Loss mse: 0.0043, Train Loss ce: 0.0000, Train Steps/Sec: 0.15,
+[2026-01-30 03:54:18] (step=0003399) Train Loss mse: 0.0037, Train Loss ce: 0.0000, Train Steps/Sec: 0.16,
+[2026-01-30 03:54:23] (step=0003400) Train Loss mse: 0.0046, Train Loss ce: 0.0000, Train Steps/Sec: 0.22,
+[2026-01-30 03:54:28] (step=0003401) Train Loss mse: 0.0067, Train Loss ce: 0.0000, Train Steps/Sec: 0.20,
+[2026-01-30 03:54:32] (step=0003402) Train Loss mse: 0.0051, Train Loss ce: 0.0000, Train Steps/Sec: 0.22,
+[2026-01-30 03:54:37] (step=0003403) Train Loss mse: 0.0043, Train Loss ce: 0.0000, Train Steps/Sec: 0.22,
+[2026-01-30 03:54:43] (step=0003404) Train Loss mse: 0.0034, Train Loss ce: 0.0000, Train Steps/Sec: 0.17,
+[2026-01-30 03:54:48] (step=0003405) Train Loss mse: 0.0046, Train Loss ce: 0.0000, Train Steps/Sec: 0.20,
+[2026-01-30 03:54:54] (step=0003406) Train Loss mse: 0.0042, Train Loss ce: 0.0000, Train Steps/Sec: 0.16,
+[2026-01-30 03:54:59] (step=0003407) Train Loss mse: 0.0044, Train Loss ce: 0.0000, Train Steps/Sec: 0.21,
+[2026-01-30 03:55:04] (step=0003408) Train Loss mse: 0.0041, Train Loss ce: 0.0000, Train Steps/Sec: 0.19,
+[2026-01-30 03:55:10] (step=0003409) Train Loss mse: 0.0048, Train Loss ce: 0.0000, Train Steps/Sec: 0.16,
+[2026-01-30 03:55:16] (step=0003410) Train Loss mse: 0.0041, Train Loss ce: 0.0000, Train Steps/Sec: 0.16,
+[2026-01-30 03:55:23] (step=0003411) Train Loss mse: 0.0039, Train Loss ce: 0.0000, Train Steps/Sec: 0.16,
+[2026-01-30 03:55:28] (step=0003412) Train Loss mse: 0.0054, Train Loss ce: 0.0000, Train Steps/Sec: 0.20,
 [2026-01-30 03:55:32] (step=0003413) Train Loss mse: 0.0039, Train Loss ce: 0.0000, Train Steps/Sec: 0.21,
 [2026-01-30 03:55:37] (step=0003414) Train Loss mse: 0.0050, Train Loss ce: 0.0000, Train Steps/Sec: 0.22,
 [2026-01-30 03:55:42] (step=0003415) Train Loss mse: 0.0047, Train Loss ce: 0.0000, Train Steps/Sec: 0.20,
@@ -4986,13 +5026,6 @@ ce_avg: 0.0, mse_avg: 0.0062998272478580475
 [2026-01-30 06:00:41] (step=0004780) Train Loss mse: 0.0041, Train Loss ce: 0.0000, Train Steps/Sec: 0.16,
 [2026-01-30 06:00:47] (step=0004781) Train Loss mse: 0.0059, Train Loss ce: 0.0000, Train Steps/Sec: 0.16,
 [2026-01-30 06:00:53] (step=0004782) Train Loss mse: 0.0037, Train Loss ce: 0.0000, Train Steps/Sec: 0.16,
-base_dir is /dev/shm/models/checkpoints_vlm_gym_reference_dot_one_image_lr2e_5_mse_only_ins/eval_used_rows, step_tag is vlm_gym_reference_dot_one_img_lr2e_5_mse_only_ins_step5000
-Preparing Dataset vlm_gym_reference_dot_mse_loss_only_evalonce/vlm_gym_reference_dot_val
-[eval debug] first 3 batch fingerprints:
-  fp[0]: [{'data_indexes': [0], 'worker_id': 0, 'dataset_name': 'vlm_gym_reference_dot_mse_loss_only_evalonce'}]
-  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_reference_dot_mse_loss_only_evalonce'}]
-  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_reference_dot_mse_loss_only_evalonce'}]
-ce_avg: 0.0, mse_avg: 0.005609693005681038
 [2026-01-30 06:01:00] (step=0004783) Train Loss mse: 0.0048, Train Loss ce: 0.0000, Train Steps/Sec: 0.16,
 [2026-01-30 06:01:04] (step=0004784) Train Loss mse: 0.0040, Train Loss ce: 0.0000, Train Steps/Sec: 0.21,
 [2026-01-30 06:01:09] (step=0004785) Train Loss mse: 0.0037, Train Loss ce: 0.0000, Train Steps/Sec: 0.21,
@@ -5019,6 +5052,13 @@ ce_avg: 0.0, mse_avg: 0.005609693005681038
 [2026-01-30 06:03:00] (step=0004806) Train Loss mse: 0.0042, Train Loss ce: 0.0000, Train Steps/Sec: 0.20,
 [2026-01-30 06:03:05] (step=0004807) Train Loss mse: 0.0041, Train Loss ce: 0.0000, Train Steps/Sec: 0.20,
 [2026-01-30 06:03:10] (step=0004808) Train Loss mse: 0.0053, Train Loss ce: 0.0000, Train Steps/Sec: 0.20,
+base_dir is /dev/shm/models/checkpoints_vlm_gym_reference_dot_one_image_lr2e_5_mse_only_ins/eval_used_rows, step_tag is vlm_gym_reference_dot_one_img_lr2e_5_mse_only_ins_step5000
+Preparing Dataset vlm_gym_reference_dot_mse_loss_only_evalonce/vlm_gym_reference_dot_val
+[eval debug] first 3 batch fingerprints:
+  fp[0]: [{'data_indexes': [0], 'worker_id': 0, 'dataset_name': 'vlm_gym_reference_dot_mse_loss_only_evalonce'}]
+  fp[1]: [{'data_indexes': [8], 'worker_id': 0, 'dataset_name': 'vlm_gym_reference_dot_mse_loss_only_evalonce'}]
+  fp[2]: [{'data_indexes': [16], 'worker_id': 0, 'dataset_name': 'vlm_gym_reference_dot_mse_loss_only_evalonce'}]
+ce_avg: 0.0, mse_avg: 0.005609693005681038
 [2026-01-30 06:03:15] (step=0004809) Train Loss mse: 0.0037, Train Loss ce: 0.0000, Train Steps/Sec: 0.20,
 [2026-01-30 06:03:21] (step=0004810) Train Loss mse: 0.0047, Train Loss ce: 0.0000, Train Steps/Sec: 0.18,
 [2026-01-30 06:03:27] (step=0004811) Train Loss mse: 0.0036, Train Loss ce: 0.0000, Train Steps/Sec: 0.17,