Upload checkpoints_vlm_gym_jigsaw_one_image_lr2e_5_mse_only_ema9999_hashed/checkpoints_vlm_gym_jigsaw_one_image_lr2e_5_mse_only_ema9999_hashed
Browse files
checkpoints_vlm_gym_jigsaw_one_image_lr2e_5_mse_only_ema9999_hashed/checkpoints_vlm_gym_jigsaw_one_image_lr2e_5_mse_only_ema9999_hashed/wandb/offline-run-20260111_235518-vlm_gym_jigsaw_one_img_lr2e_5_mse_only_ema9999_hashed-run0/files/config.yaml
CHANGED
|
@@ -0,0 +1,181 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
wandb_version: 1
|
| 2 |
+
|
| 3 |
+
_wandb:
|
| 4 |
+
desc: null
|
| 5 |
+
value:
|
| 6 |
+
python_version: 3.11.10
|
| 7 |
+
cli_version: 0.23.1
|
| 8 |
+
framework: huggingface
|
| 9 |
+
huggingface_version: 4.49.0
|
| 10 |
+
is_jupyter_run: false
|
| 11 |
+
is_kaggle_kernel: false
|
| 12 |
+
start_time: 1768175718
|
| 13 |
+
t:
|
| 14 |
+
1:
|
| 15 |
+
- 1
|
| 16 |
+
- 5
|
| 17 |
+
- 11
|
| 18 |
+
- 41
|
| 19 |
+
- 49
|
| 20 |
+
- 53
|
| 21 |
+
- 71
|
| 22 |
+
- 105
|
| 23 |
+
2:
|
| 24 |
+
- 1
|
| 25 |
+
- 5
|
| 26 |
+
- 11
|
| 27 |
+
- 41
|
| 28 |
+
- 49
|
| 29 |
+
- 53
|
| 30 |
+
- 71
|
| 31 |
+
- 105
|
| 32 |
+
3:
|
| 33 |
+
- 4
|
| 34 |
+
- 13
|
| 35 |
+
- 14
|
| 36 |
+
- 37
|
| 37 |
+
- 42
|
| 38 |
+
4: 3.11.10
|
| 39 |
+
5: 0.23.1
|
| 40 |
+
6: 4.49.0
|
| 41 |
+
13: linux-x86_64
|
| 42 |
+
e:
|
| 43 |
+
p78nkupquzvg4sp7iakb8i8e0ffje8id:
|
| 44 |
+
os: Linux-6.6.93+-x86_64-with-glibc2.35
|
| 45 |
+
python: CPython 3.11.10
|
| 46 |
+
started_at: '2026-01-11T23:55:18.222054Z'
|
| 47 |
+
args:
|
| 48 |
+
- --dataset_config_file
|
| 49 |
+
- ./data/configs/vlm_gym_jigsaw_train_mseloss_only.yaml
|
| 50 |
+
- --eval_dataset_config_file
|
| 51 |
+
- ./data/configs/vlm_gym_jigsaw_train_mseloss_only.yaml
|
| 52 |
+
- --viz_dataset_config_file
|
| 53 |
+
- ./data/configs/vlm_gym_jigsaw_train_mseloss_only.yaml
|
| 54 |
+
- --inference_hash_file
|
| 55 |
+
- /home/clouduser/Code/Github/launch_new/hashes_test_set_v10.json
|
| 56 |
+
- --train_data_dir
|
| 57 |
+
- /home/clouduser/Code/data/gym/jigsaw-swap_v5/train/
|
| 58 |
+
- --train_jsonl_path
|
| 59 |
+
- /home/clouduser/Code/data/gym/jigsaw-swap_v5/train/
|
| 60 |
+
- --eval_data_dir
|
| 61 |
+
- /home/clouduser/Code/data/gym/jigsaw-swap_v5/val/
|
| 62 |
+
- --eval_jsonl_path
|
| 63 |
+
- /home/clouduser/Code/data/gym/jigsaw-swap_v5/val/
|
| 64 |
+
- --model_path
|
| 65 |
+
- /home/clouduser/Code/Models/BAGEL-7B-MoT
|
| 66 |
+
- --layer_module
|
| 67 |
+
- Qwen2MoTDecoderLayer
|
| 68 |
+
- --max_latent_size
|
| 69 |
+
- '64'
|
| 70 |
+
- --resume-from
|
| 71 |
+
- /home/clouduser/Code/Models/BAGEL-7B-MoT
|
| 72 |
+
- --finetune_from_hf
|
| 73 |
+
- 'True'
|
| 74 |
+
- --auto_resume
|
| 75 |
+
- 'False'
|
| 76 |
+
- --resume-model-only
|
| 77 |
+
- 'True'
|
| 78 |
+
- --finetune-from-ema
|
| 79 |
+
- 'True'
|
| 80 |
+
- --log_every
|
| 81 |
+
- '1'
|
| 82 |
+
- --lr
|
| 83 |
+
- 2e-5
|
| 84 |
+
- --warmup_steps
|
| 85 |
+
- '300'
|
| 86 |
+
- --lr_scheduler
|
| 87 |
+
- cosine
|
| 88 |
+
- --num_worker
|
| 89 |
+
- '1'
|
| 90 |
+
- --expected_num_tokens
|
| 91 |
+
- '20000'
|
| 92 |
+
- --max_num_tokens
|
| 93 |
+
- '20000'
|
| 94 |
+
- --max_num_tokens_per_sample
|
| 95 |
+
- '20000'
|
| 96 |
+
- --visual_und
|
| 97 |
+
- 'True'
|
| 98 |
+
- --save_every
|
| 99 |
+
- '2500'
|
| 100 |
+
- --total_steps
|
| 101 |
+
- '5000'
|
| 102 |
+
- --text_cond_dropout_prob
|
| 103 |
+
- '0.0'
|
| 104 |
+
- --vae_cond_dropout_prob
|
| 105 |
+
- '0.0'
|
| 106 |
+
- --vit_cond_dropout_prob
|
| 107 |
+
- '0.0'
|
| 108 |
+
- --ema
|
| 109 |
+
- '0.9999'
|
| 110 |
+
- --checkpoint_dir
|
| 111 |
+
- /dev/shm/models/checkpoints_vlm_gym_jigsaw_one_image_lr2e_5_mse_only_ema9999_hashed
|
| 112 |
+
- --wandb_project
|
| 113 |
+
- bagel
|
| 114 |
+
- --wandb_name
|
| 115 |
+
- vlm_gym_jigsaw_one_img_lr2e_5_mse_only_ema9999_hashed
|
| 116 |
+
- --wandb_dir
|
| 117 |
+
- /dev/shm/models/checkpoints_vlm_gym_jigsaw_one_image_lr2e_5_mse_only_ema9999_hashed
|
| 118 |
+
- --wandb_offline
|
| 119 |
+
- 'True'
|
| 120 |
+
program: /home/clouduser/Code/Github/unified_world_model/train/pretrain_unified_navit.py
|
| 121 |
+
code_path: train/pretrain_unified_navit.py
|
| 122 |
+
code_path_local: train/pretrain_unified_navit.py
|
| 123 |
+
git:
|
| 124 |
+
remote_url: https://github.com/para-lost/unified_world_model
|
| 125 |
+
commit: b61f53a430c22fa4a17c69d2f903ce1fe266f48e
|
| 126 |
+
root: /dev/shm/models/checkpoints_vlm_gym_jigsaw_one_image_lr2e_5_mse_only_ema9999_hashed
|
| 127 |
+
host: junyizhang-launch-new-221746038-1-0
|
| 128 |
+
executable: /opt/conda/bin/python3.11
|
| 129 |
+
cpu_count: 48
|
| 130 |
+
cpu_count_logical: 96
|
| 131 |
+
gpu_type: NVIDIA A100-SXM4-80GB
|
| 132 |
+
gpu_count: 8
|
| 133 |
+
disk:
|
| 134 |
+
/:
|
| 135 |
+
total: '1052461830144'
|
| 136 |
+
used: '261625344000'
|
| 137 |
+
memory:
|
| 138 |
+
total: '1437332606976'
|
| 139 |
+
gpu_nvidia:
|
| 140 |
+
- name: NVIDIA A100-SXM4-80GB
|
| 141 |
+
memory_total: '85899345920'
|
| 142 |
+
cuda_cores: 6912
|
| 143 |
+
architecture: Ampere
|
| 144 |
+
uuid: GPU-5c8a9afd-07c9-5b67-4b7d-2d070676ab83
|
| 145 |
+
- name: NVIDIA A100-SXM4-80GB
|
| 146 |
+
memory_total: '85899345920'
|
| 147 |
+
cuda_cores: 6912
|
| 148 |
+
architecture: Ampere
|
| 149 |
+
uuid: GPU-f44fe102-70f0-4db8-5fd1-c30bfd0f9bb8
|
| 150 |
+
- name: NVIDIA A100-SXM4-80GB
|
| 151 |
+
memory_total: '85899345920'
|
| 152 |
+
cuda_cores: 6912
|
| 153 |
+
architecture: Ampere
|
| 154 |
+
uuid: GPU-97f9aee6-0f04-ce00-5732-7d6cab1f9170
|
| 155 |
+
- name: NVIDIA A100-SXM4-80GB
|
| 156 |
+
memory_total: '85899345920'
|
| 157 |
+
cuda_cores: 6912
|
| 158 |
+
architecture: Ampere
|
| 159 |
+
uuid: GPU-e645965f-7741-e8d5-63df-bf1e988d5549
|
| 160 |
+
- name: NVIDIA A100-SXM4-80GB
|
| 161 |
+
memory_total: '85899345920'
|
| 162 |
+
cuda_cores: 6912
|
| 163 |
+
architecture: Ampere
|
| 164 |
+
uuid: GPU-4123c973-b472-4c08-392e-9df433596e67
|
| 165 |
+
- name: NVIDIA A100-SXM4-80GB
|
| 166 |
+
memory_total: '85899345920'
|
| 167 |
+
cuda_cores: 6912
|
| 168 |
+
architecture: Ampere
|
| 169 |
+
uuid: GPU-8a672f68-5ec7-eb5e-8f00-748126c6ce99
|
| 170 |
+
- name: NVIDIA A100-SXM4-80GB
|
| 171 |
+
memory_total: '85899345920'
|
| 172 |
+
cuda_cores: 6912
|
| 173 |
+
architecture: Ampere
|
| 174 |
+
uuid: GPU-81b788fb-644b-899f-258e-b2ac8ffffb2b
|
| 175 |
+
- name: NVIDIA A100-SXM4-80GB
|
| 176 |
+
memory_total: '85899345920'
|
| 177 |
+
cuda_cores: 6912
|
| 178 |
+
architecture: Ampere
|
| 179 |
+
uuid: GPU-15e8faa7-0d4d-86aa-05db-cf5fb77e5111
|
| 180 |
+
cuda_version: '12.2'
|
| 181 |
+
writer_id: p78nkupquzvg4sp7iakb8i8e0ffje8id
|