Upload checkpoints_vlm_gym_counting_mark_all_one_image_lr2e_5_ce_ins/checkpoints_vlm_gym_counting_mark_all_one_image_lr2e_5_ce_ins

checkpoints_vlm_gym_counting_mark_all_one_image_lr2e_5_ce_ins/checkpoints_vlm_gym_counting_mark_all_one_image_lr2e_5_ce_ins/wandb/offline-run-20260125_205640-checkpoints_vlm_gym_counting_mark_all_one_image_lr2e_5_ce_ins-run0/files/config.yaml

@@ -0,0 +1,185 @@
+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: 1769374601
+    t:
+      1:
+      - 1
+      - 5
+      - 11
+      - 41
+      - 49
+      - 53
+      - 71
+      - 105
+      2:
+      - 1
+      - 5
+      - 11
+      - 41
+      - 49
+      - 53
+      - 71
+      - 105
+      3:
+      - 4
+      - 13
+      - 14
+      - 37
+      - 42
+      4: 3.11.10
+      5: 0.23.1
+      6: 4.49.0
+      13: linux-x86_64
+    e:
+      qivwap4wkbv1rrpyy3tc8ok85jcfa4gf:
+        os: Linux-6.6.93+-x86_64-with-glibc2.35
+        python: CPython 3.11.10
+        started_at: '2026-01-25T20:56:40.641636Z'
+        args:
+        - --dataset_config_file
+        - ./data/configs/vlm_gym_counting_mark_all_train_celoss.yaml
+        - --eval_dataset_config_file
+        - ./data/configs/vlm_gym_counting_mark_all_eval_celoss.yaml
+        - --viz_dataset_config_file
+        - ./data/configs/vlm_gym_counting_mark_all_eval_celoss.yaml
+        - --inference_hash_file
+        - /home/clouduser/Code/Github/launch_new/hashes_test_set_v10.json
+        - --task_name
+        - counting-mark_all_v5
+        - --instructions_dir
+        - ./data/instructions
+        - --train_data_dir
+        - /home/clouduser/Code/data/gym/counting-mark_all_v5/train/
+        - --train_jsonl_path
+        - /home/clouduser/Code/data/gym/counting-mark_all_v5/train/
+        - --eval_data_dir
+        - /home/clouduser/Code/data/gym/counting-mark_all_v5/val/
+        - --eval_jsonl_path
+        - /home/clouduser/Code/data/gym/counting-mark_all_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_counting_mark_all_one_image_lr2e_5_ce_ins
+        - --wandb_project
+        - bagel
+        - --wandb_name
+        - checkpoints_vlm_gym_counting_mark_all_one_image_lr2e_5_ce_ins
+        - --wandb_dir
+        - /dev/shm/models/checkpoints_vlm_gym_counting_mark_all_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: 45495bf06d28509bc54cbbda532f4b97404a7d66
+        root: /dev/shm/models/checkpoints_vlm_gym_counting_mark_all_one_image_lr2e_5_ce_ins
+        host: junyizhang-launch-new-225900672-1-0
+        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: '354507616256'
+        memory:
+          total: '1437332611072'
+        gpu_nvidia:
+        - name: NVIDIA A100-SXM4-80GB
+          memory_total: '85899345920'
+          cuda_cores: 6912
+          architecture: Ampere
+          uuid: GPU-27013fed-9784-d445-a1eb-01629cf403cc
+        - name: NVIDIA A100-SXM4-80GB
+          memory_total: '85899345920'
+          cuda_cores: 6912
+          architecture: Ampere
+          uuid: GPU-c4922cf6-bc87-9458-c12f-23210cb43686
+        - name: NVIDIA A100-SXM4-80GB
+          memory_total: '85899345920'
+          cuda_cores: 6912
+          architecture: Ampere
+          uuid: GPU-1af9405a-c062-486e-383f-7ea6c6ef5158
+        - name: NVIDIA A100-SXM4-80GB
+          memory_total: '85899345920'
+          cuda_cores: 6912
+          architecture: Ampere
+          uuid: GPU-793b7211-7436-7429-8bd7-cc05be70cc75
+        - name: NVIDIA A100-SXM4-80GB
+          memory_total: '85899345920'
+          cuda_cores: 6912
+          architecture: Ampere
+          uuid: GPU-5eb44009-8d7d-911d-0730-f219cb50498c
+        - name: NVIDIA A100-SXM4-80GB
+          memory_total: '85899345920'
+          cuda_cores: 6912
+          architecture: Ampere
+          uuid: GPU-62c85054-47c8-b915-18e9-e4433fc0f9bb
+        - name: NVIDIA A100-SXM4-80GB
+          memory_total: '85899345920'
+          cuda_cores: 6912
+          architecture: Ampere
+          uuid: GPU-c3b59f2c-b6b6-7730-54ff-8cf5fee4ea9c
+        - name: NVIDIA A100-SXM4-80GB
+          memory_total: '85899345920'
+          cuda_cores: 6912
+          architecture: Ampere
+          uuid: GPU-e988baaf-6bc5-3bb9-91fb-ab2cb214233d
+        cuda_version: '12.2'
+        writer_id: qivwap4wkbv1rrpyy3tc8ok85jcfa4gf

checkpoints_vlm_gym_counting_mark_all_one_image_lr2e_5_ce_ins/checkpoints_vlm_gym_counting_mark_all_one_image_lr2e_5_ce_ins/wandb/offline-run-20260125_205640-checkpoints_vlm_gym_counting_mark_all_one_image_lr2e_5_ce_ins-run0/files/output.log

@@ -0,0 +1,184 @@
+wandb: Detected [huggingface_hub.inference] in use.
+wandb: Use W&B Weave for improved LLM call tracing. Install Weave with `pip install weave` then add `import weave` to the top of your script.
+wandb: For more information, check out the docs at: https://weave-docs.wandb.ai/
+[2026-01-25 20:56:47] Training arguments TrainingArguments(visual_gen=True, visual_und=True, results_dir='results', checkpoint_dir='/dev/shm/models/checkpoints_vlm_gym_counting_mark_all_one_image_lr2e_5_ce_ins', wandb_project='bagel', wandb_name='checkpoints_vlm_gym_counting_mark_all_one_image_lr2e_5_ce_ins', wandb_runid='0', wandb_resume='allow', wandb_offline=True, wandb_dir='/dev/shm/models/checkpoints_vlm_gym_counting_mark_all_one_image_lr2e_5_ce_ins', global_seed=4396, auto_resume=False, resume_from='/home/clouduser/Code/Models/BAGEL-7B-MoT', resume_model_only=True, finetune_from_ema=True, finetune_from_hf=True, log_every=1, save_every=2500, total_steps=5000, warmup_steps=300, lr_scheduler='cosine', lr=2e-05, min_lr=1e-07, beta1=0.9, beta2=0.95, eps=1e-15, ema=0.993, max_grad_norm=1.0, timestep_shift=1.0, mse_weight=1.0, ce_weight=1.0, ce_loss_reweighting=False, expected_num_tokens=30000, num_replicate=1, num_shard=8, sharding_strategy='HYBRID_SHARD', backward_prefetch='BACKWARD_PRE', cpu_offload=False, freeze_llm=False, freeze_vit=False, freeze_vae=True, freeze_und=False, copy_init_moe=True, use_flex=False, eval_every=500, num_eval_batches=20, use_ema_for_eval=True, eval_log_dir=None, eval_run_tag='', viz_every=500, viz_n=8, viz_outdir='results/viz', eval_dataset_config_file='./data/configs/vlm_gym_counting_mark_all_eval_celoss.yaml', viz_dataset_config_file='./data/configs/vlm_gym_counting_mark_all_eval_celoss.yaml', eval_print_n=3, save_ema_only=True, save_optimizer=False)
+[2026-01-25 20:56:47] Model arguments ModelArguments(model_path='/home/clouduser/Code/Models/BAGEL-7B-MoT', llm_path='hf/Qwen2.5-0.5B-Instruct/', llm_qk_norm=True, tie_word_embeddings=False, layer_module='Qwen2MoTDecoderLayer', vae_path='flux/vae/ae.safetensors', vit_path='hf/siglip-so400m-14-980-flash-attn2-navit/', max_latent_size=64, latent_patch_size=2, vit_patch_size=14, vit_max_num_patch_per_side=70, connector_act='gelu_pytorch_tanh', interpolate_pos=False, vit_select_layer=-2, vit_rope=False, text_cond_dropout_prob=0.0, vae_cond_dropout_prob=0.3, vit_cond_dropout_prob=0.0)
+[2026-01-25 20:56:47] Data arguments DataArguments(dataset_config_file='./data/configs/vlm_gym_counting_mark_all_train_celoss.yaml', train_data_dir='/home/clouduser/Code/data/gym/counting-mark_all_v5/train/', train_jsonl_path='/home/clouduser/Code/data/gym/counting-mark_all_v5/train/', eval_data_dir='/home/clouduser/Code/data/gym/counting-mark_all_v5/val/', eval_jsonl_path='/home/clouduser/Code/data/gym/counting-mark_all_v5/val/', inference_hash_file='/home/clouduser/Code/Github/launch_new/hashes_test_set_v10.json', task_name='counting-mark_all_v5', instructions_dir='./data/instructions', prefetch_factor=2, num_workers=1, max_num_tokens_per_sample=30000, max_num_tokens=30000, prefer_buffer_before=16384, max_buffer_size=50, data_seed=42)
+[2026-01-25 21:01:13] Loading checkpoint from /home/clouduser/Code/Models/BAGEL-7B-MoT.
+[2026-01-25 21:01:25] _IncompatibleKeys(missing_keys=['latent_pos_embed.pos_embed'], unexpected_keys=[])
+[2026-01-25 21:01:38] _IncompatibleKeys(missing_keys=['latent_pos_embed.pos_embed'], unexpected_keys=[])
+[2026-01-25 21:02:29] Training for 5000 steps, starting at 0...
+FullyShardedDataParallel(
+  (_fsdp_wrapped_module): Bagel(
+    (language_model): Qwen2ForCausalLM(
+      (model): Qwen2Model(
+        (embed_tokens): Embedding(152064, 3584)
+        (layers): ModuleList(
+          (0-27): 28 x FullyShardedDataParallel(
+            (_fsdp_wrapped_module): CheckpointWrapper(
+              (_checkpoint_wrapped_module): Qwen2MoTDecoderLayer(
+                (self_attn): PackedAttentionMoT(
+                  (q_proj): Linear(in_features=3584, out_features=3584, bias=True)
+                  (k_proj): Linear(in_features=3584, out_features=512, bias=True)
+                  (v_proj): Linear(in_features=3584, out_features=512, bias=True)
+                  (o_proj): Linear(in_features=3584, out_features=3584, bias=False)
+                  (q_norm): Qwen2RMSNorm((128,), eps=1e-06)
+                  (k_norm): Qwen2RMSNorm((128,), eps=1e-06)
+                  (q_norm_moe_gen): Qwen2RMSNorm((128,), eps=1e-06)
+                  (k_norm_moe_gen): Qwen2RMSNorm((128,), eps=1e-06)
+                  (q_proj_moe_gen): Linear(in_features=3584, out_features=3584, bias=True)
+                  (k_proj_moe_gen): Linear(in_features=3584, out_features=512, bias=True)
+                  (v_proj_moe_gen): Linear(in_features=3584, out_features=512, bias=True)
+                  (o_proj_moe_gen): Linear(in_features=3584, out_features=3584, bias=False)
+                )
+                (mlp): Qwen2MLP(
+                  (gate_proj): Linear(in_features=3584, out_features=18944, bias=False)
+                  (up_proj): Linear(in_features=3584, out_features=18944, bias=False)
+                  (down_proj): Linear(in_features=18944, out_features=3584, bias=False)
+                  (act_fn): SiLU()
+                )
+                (mlp_moe_gen): Qwen2MLP(
+                  (gate_proj): Linear(in_features=3584, out_features=18944, bias=False)
+                  (up_proj): Linear(in_features=3584, out_features=18944, bias=False)
+                  (down_proj): Linear(in_features=18944, out_features=3584, bias=False)
+                  (act_fn): SiLU()
+                )
+                (input_layernorm): Qwen2RMSNorm((3584,), eps=1e-06)
+                (input_layernorm_moe_gen): Qwen2RMSNorm((3584,), eps=1e-06)
+                (post_attention_layernorm): Qwen2RMSNorm((3584,), eps=1e-06)
+                (post_attention_layernorm_moe_gen): Qwen2RMSNorm((3584,), eps=1e-06)
+              )
+            )
+          )
+        )
+        (norm): Qwen2RMSNorm((3584,), eps=1e-06)
+        (norm_moe_gen): Qwen2RMSNorm((3584,), eps=1e-06)
+        (rotary_emb): Qwen2RotaryEmbedding()
+      )
+      (lm_head): Linear(in_features=3584, out_features=152064, bias=False)
+    )
+    (time_embedder): FullyShardedDataParallel(
+      (_fsdp_wrapped_module): TimestepEmbedder(
+        (mlp): Sequential(
+          (0): Linear(in_features=256, out_features=3584, bias=True)
+          (1): SiLU()
+          (2): Linear(in_features=3584, out_features=3584, bias=True)
+        )
+      )
+    )
+    (vae2llm): Linear(in_features=64, out_features=3584, bias=True)
+    (llm2vae): Linear(in_features=3584, out_features=64, bias=True)
+    (latent_pos_embed): FullyShardedDataParallel(
+      (_fsdp_wrapped_module): PositionEmbedding()
+    )
+    (vit_model): SiglipVisionModel(
+      (vision_model): FullyShardedDataParallel(
+        (_fsdp_wrapped_module): SiglipVisionTransformer(
+          (embeddings): SiglipVisionEmbeddings(
+            (position_embedding): Embedding(4900, 1152)
+            (patch_embedding): Linear(in_features=588, out_features=1152, bias=True)
+          )
+          (encoder): SiglipEncoder(
+            (layers): ModuleList(
+              (0-25): 26 x FullyShardedDataParallel(
+                (_fsdp_wrapped_module): CheckpointWrapper(
+                  (_checkpoint_wrapped_module): SiglipEncoderLayer(
+                    (self_attn): SiglipFlashAttention2(
+                      (k_proj): Linear(in_features=1152, out_features=1152, bias=True)
+                      (v_proj): Linear(in_features=1152, out_features=1152, bias=True)
+                      (q_proj): Linear(in_features=1152, out_features=1152, bias=True)
+                      (out_proj): Linear(in_features=1152, out_features=1152, bias=True)
+                    )
+                    (layer_norm1): LayerNorm((1152,), eps=1e-06, elementwise_affine=True)
+                    (mlp): SiglipMLP(
+                      (activation_fn): PytorchGELUTanh()
+                      (fc1): Linear(in_features=1152, out_features=4304, bias=True)
+                      (fc2): Linear(in_features=4304, out_features=1152, bias=True)
+                    )
+                    (layer_norm2): LayerNorm((1152,), eps=1e-06, elementwise_affine=True)
+                  )
+                )
+              )
+            )
+          )
+          (post_layernorm): LayerNorm((1152,), eps=1e-06, elementwise_affine=True)
+        )
+      )
+    )
+    (connector): FullyShardedDataParallel(
+      (_fsdp_wrapped_module): CheckpointWrapper(
+        (_checkpoint_wrapped_module): MLPconnector(
+          (activation_fn): PytorchGELUTanh()
+          (fc1): Linear(in_features=1152, out_features=3584, bias=True)
+          (fc2): Linear(in_features=3584, out_features=3584, bias=True)
+        )
+      )
+    )
+    (vit_pos_embed): FullyShardedDataParallel(
+      (_fsdp_wrapped_module): PositionEmbedding()
+    )
+  )
+)
+_flat_param True
+language_model.model.layers.0._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+language_model.model.layers.1._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+language_model.model.layers.2._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+language_model.model.layers.3._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+language_model.model.layers.4._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+language_model.model.layers.5._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+language_model.model.layers.6._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+language_model.model.layers.7._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+language_model.model.layers.8._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+language_model.model.layers.9._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+language_model.model.layers.10._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+language_model.model.layers.11._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+language_model.model.layers.12._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+language_model.model.layers.13._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+language_model.model.layers.14._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+language_model.model.layers.15._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+language_model.model.layers.16._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+language_model.model.layers.17._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+language_model.model.layers.18._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+language_model.model.layers.19._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+language_model.model.layers.20._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+language_model.model.layers.21._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+language_model.model.layers.22._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+language_model.model.layers.23._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+language_model.model.layers.24._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+language_model.model.layers.25._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+language_model.model.layers.26._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+language_model.model.layers.27._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+time_embedder._fsdp_wrapped_module._flat_param True
+latent_pos_embed._fsdp_wrapped_module._flat_param False
+vit_model.vision_model._fsdp_wrapped_module._flat_param True
+vit_model.vision_model._fsdp_wrapped_module.encoder.layers.0._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+vit_model.vision_model._fsdp_wrapped_module.encoder.layers.1._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+vit_model.vision_model._fsdp_wrapped_module.encoder.layers.2._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+vit_model.vision_model._fsdp_wrapped_module.encoder.layers.3._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+vit_model.vision_model._fsdp_wrapped_module.encoder.layers.4._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+vit_model.vision_model._fsdp_wrapped_module.encoder.layers.5._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+vit_model.vision_model._fsdp_wrapped_module.encoder.layers.6._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+vit_model.vision_model._fsdp_wrapped_module.encoder.layers.7._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+vit_model.vision_model._fsdp_wrapped_module.encoder.layers.8._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+vit_model.vision_model._fsdp_wrapped_module.encoder.layers.9._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+vit_model.vision_model._fsdp_wrapped_module.encoder.layers.10._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+vit_model.vision_model._fsdp_wrapped_module.encoder.layers.11._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+vit_model.vision_model._fsdp_wrapped_module.encoder.layers.12._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+vit_model.vision_model._fsdp_wrapped_module.encoder.layers.13._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+vit_model.vision_model._fsdp_wrapped_module.encoder.layers.14._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+vit_model.vision_model._fsdp_wrapped_module.encoder.layers.15._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+vit_model.vision_model._fsdp_wrapped_module.encoder.layers.16._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+vit_model.vision_model._fsdp_wrapped_module.encoder.layers.17._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+vit_model.vision_model._fsdp_wrapped_module.encoder.layers.18._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+vit_model.vision_model._fsdp_wrapped_module.encoder.layers.19._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+vit_model.vision_model._fsdp_wrapped_module.encoder.layers.20._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+vit_model.vision_model._fsdp_wrapped_module.encoder.layers.21._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+vit_model.vision_model._fsdp_wrapped_module.encoder.layers.22._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+vit_model.vision_model._fsdp_wrapped_module.encoder.layers.23._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+vit_model.vision_model._fsdp_wrapped_module.encoder.layers.24._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+vit_model.vision_model._fsdp_wrapped_module.encoder.layers.25._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+connector._fsdp_wrapped_module._checkpoint_wrapped_module._flat_param True
+vit_pos_embed._fsdp_wrapped_module._flat_param False
+Preparing Dataset vlm_gym_counting_mark_all_celoss/vlm_gym_counting_mark_all_train
+base_dir is /dev/shm/models/checkpoints_vlm_gym_counting_mark_all_one_image_lr2e_5_ce_ins/eval_used_rows, step_tag is checkpoints_vlm_gym_counting_mark_all_one_image_lr2e_5_ce_ins_step0
+Preparing Dataset vlm_gym_counting_mark_all_celoss_evalonce/vlm_gym_counting_mark_all_val

checkpoints_vlm_gym_counting_mark_all_one_image_lr2e_5_ce_ins/checkpoints_vlm_gym_counting_mark_all_one_image_lr2e_5_ce_ins/wandb/offline-run-20260125_205640-checkpoints_vlm_gym_counting_mark_all_one_image_lr2e_5_ce_ins-run0/files/wandb-metadata.json

@@ -0,0 +1 @@
+{"os": "Linux-6.6.93+-x86_64-with-glibc2.35", "python": "CPython 3.11.10", "started_at": "2026-01-25T20:56:40.641636Z", "args": ["--dataset_config_file", "./data/configs/vlm_gym_counting_mark_all_train_celoss.yaml", "--eval_dataset_config_file", "./data/configs/vlm_gym_counting_mark_all_eval_celoss.yaml", "--viz_dataset_config_file", "./data/configs/vlm_gym_counting_mark_all_eval_celoss.yaml", "--inference_hash_file", "/home/clouduser/Code/Github/launch_new/hashes_test_set_v10.json", "--task_name", "counting-mark_all_v5", "--instructions_dir", "./data/instructions", "--train_data_dir", "/home/clouduser/Code/data/gym/counting-mark_all_v5/train/", "--train_jsonl_path", "/home/clouduser/Code/data/gym/counting-mark_all_v5/train/", "--eval_data_dir", "/home/clouduser/Code/data/gym/counting-mark_all_v5/val/", "--eval_jsonl_path", "/home/clouduser/Code/data/gym/counting-mark_all_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_counting_mark_all_one_image_lr2e_5_ce_ins", "--wandb_project", "bagel", "--wandb_name", "checkpoints_vlm_gym_counting_mark_all_one_image_lr2e_5_ce_ins", "--wandb_dir", "/dev/shm/models/checkpoints_vlm_gym_counting_mark_all_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": "45495bf06d28509bc54cbbda532f4b97404a7d66"}, "root": "/dev/shm/models/checkpoints_vlm_gym_counting_mark_all_one_image_lr2e_5_ce_ins", "host": "junyizhang-launch-new-225900672-1-0", "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": "354507616256"}}, "memory": {"total": "1437332611072"}, "gpu_nvidia": [{"name": "NVIDIA A100-SXM4-80GB", "memory_total": "85899345920", "cuda_cores": 6912, "architecture": "Ampere", "uuid": "GPU-27013fed-9784-d445-a1eb-01629cf403cc"}, {"name": "NVIDIA A100-SXM4-80GB", "memory_total": "85899345920", "cuda_cores": 6912, "architecture": "Ampere", "uuid": "GPU-c4922cf6-bc87-9458-c12f-23210cb43686"}, {"name": "NVIDIA A100-SXM4-80GB", "memory_total": "85899345920", "cuda_cores": 6912, "architecture": "Ampere", "uuid": "GPU-1af9405a-c062-486e-383f-7ea6c6ef5158"}, {"name": "NVIDIA A100-SXM4-80GB", "memory_total": "85899345920", "cuda_cores": 6912, "architecture": "Ampere", "uuid": "GPU-793b7211-7436-7429-8bd7-cc05be70cc75"}, {"name": "NVIDIA A100-SXM4-80GB", "memory_total": "85899345920", "cuda_cores": 6912, "architecture": "Ampere", "uuid": "GPU-5eb44009-8d7d-911d-0730-f219cb50498c"}, {"name": "NVIDIA A100-SXM4-80GB", "memory_total": "85899345920", "cuda_cores": 6912, "architecture": "Ampere", "uuid": "GPU-62c85054-47c8-b915-18e9-e4433fc0f9bb"}, {"name": "NVIDIA A100-SXM4-80GB", "memory_total": "85899345920", "cuda_cores": 6912, "architecture": "Ampere", "uuid": "GPU-c3b59f2c-b6b6-7730-54ff-8cf5fee4ea9c"}, {"name": "NVIDIA A100-SXM4-80GB", "memory_total": "85899345920", "cuda_cores": 6912, "architecture": "Ampere", "uuid": "GPU-e988baaf-6bc5-3bb9-91fb-ab2cb214233d"}], "cuda_version": "12.2", "writer_id": "qivwap4wkbv1rrpyy3tc8ok85jcfa4gf"}