Hugging Face's logo

Junyi42
/
gym

Safetensors
Model card Files
xet
 Community
Junyi42 commited on
Commit
60f6d9b
·
verified ·
1 Parent(s): 0457f1d

Upload checkpoints_vlm_gym_toy_maze_3d_one_image_lr2e_5_ce_ins/checkpoints_vlm_gym_toy_maze_3d_one_image_lr2e_5_ce_ins

Browse files
Files changed (3) hide show
  1. checkpoints_vlm_gym_toy_maze_3d_one_image_lr2e_5_ce_ins/checkpoints_vlm_gym_toy_maze_3d_one_image_lr2e_5_ce_ins/wandb/offline-run-20260129_223621-checkpoints_vlm_gym_toy_maze_3d_one_image_lr2e_5_ce_ins-run0/files/config.yaml +185 -0
  2. checkpoints_vlm_gym_toy_maze_3d_one_image_lr2e_5_ce_ins/checkpoints_vlm_gym_toy_maze_3d_one_image_lr2e_5_ce_ins/wandb/offline-run-20260130_174942-checkpoints_vlm_gym_toy_maze_3d_one_image_lr2e_5_ce_ins-run0/files/config.yaml +185 -0
  3. checkpoints_vlm_gym_toy_maze_3d_one_image_lr2e_5_ce_ins/checkpoints_vlm_gym_toy_maze_3d_one_image_lr2e_5_ce_ins/wandb/offline-run-20260130_203229-checkpoints_vlm_gym_toy_maze_3d_one_image_lr2e_5_ce_ins-run0/files/config.yaml +456 -0
checkpoints_vlm_gym_toy_maze_3d_one_image_lr2e_5_ce_ins/checkpoints_vlm_gym_toy_maze_3d_one_image_lr2e_5_ce_ins/wandb/offline-run-20260129_223621-checkpoints_vlm_gym_toy_maze_3d_one_image_lr2e_5_ce_ins-run0/files/config.yaml CHANGED
@@ -0,0 +1,185 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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: 1769726181
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
+ zcxxh1ni9bhfnmriu0rizt0w5oidbkhb:
44
+ os: Linux-6.6.93+-x86_64-with-glibc2.35
45
+ python: CPython 3.11.10
46
+ started_at: '2026-01-29T22:36:21.153529Z'
47
+ args:
48
+ - --dataset_config_file
49
+ - ./data/configs/vlm_gym_toy_maze_3d_train_celoss.yaml
50
+ - --eval_dataset_config_file
51
+ - ./data/configs/vlm_gym_toy_maze_3d_eval_celoss.yaml
52
+ - --viz_dataset_config_file
53
+ - ./data/configs/vlm_gym_toy_maze_3d_eval_celoss.yaml
54
+ - --inference_hash_file
55
+ - /home/clouduser/Code/Github/launch_new/hashes_test_set_v10.json
56
+ - --task_name
57
+ - toy_maze_3d_v5
58
+ - --instructions_dir
59
+ - ./data/instructions
60
+ - --train_data_dir
61
+ - /home/clouduser/Code/data/gym/toy_maze_3d_v5/train/
62
+ - --train_jsonl_path
63
+ - /home/clouduser/Code/data/gym/toy_maze_3d_v5/train/
64
+ - --eval_data_dir
65
+ - /home/clouduser/Code/data/gym/toy_maze_3d_v5/val/
66
+ - --eval_jsonl_path
67
+ - /home/clouduser/Code/data/gym/toy_maze_3d_v5/val/
68
+ - --model_path
69
+ - /home/clouduser/Code/Models/BAGEL-7B-MoT
70
+ - --layer_module
71
+ - Qwen2MoTDecoderLayer
72
+ - --max_latent_size
73
+ - '64'
74
+ - --resume-from
75
+ - /home/clouduser/Code/Models/BAGEL-7B-MoT
76
+ - --finetune_from_hf
77
+ - 'True'
78
+ - --auto_resume
79
+ - 'False'
80
+ - --resume-model-only
81
+ - 'True'
82
+ - --finetune-from-ema
83
+ - 'True'
84
+ - --log_every
85
+ - '1'
86
+ - --lr
87
+ - 2e-5
88
+ - --warmup_steps
89
+ - '300'
90
+ - --lr_scheduler
91
+ - cosine
92
+ - --num_worker
93
+ - '1'
94
+ - --expected_num_tokens
95
+ - '30000'
96
+ - --max_num_tokens
97
+ - '30000'
98
+ - --max_num_tokens_per_sample
99
+ - '30000'
100
+ - --visual_und
101
+ - 'True'
102
+ - --save_every
103
+ - '2500'
104
+ - --total_steps
105
+ - '5000'
106
+ - --text_cond_dropout_prob
107
+ - '0.0'
108
+ - --vae_cond_dropout_prob
109
+ - '0.3'
110
+ - --vit_cond_dropout_prob
111
+ - '0.0'
112
+ - --ema
113
+ - '0.993'
114
+ - --checkpoint_dir
115
+ - /dev/shm/models/checkpoints_vlm_gym_toy_maze_3d_one_image_lr2e_5_ce_ins
116
+ - --wandb_project
117
+ - bagel
118
+ - --wandb_name
119
+ - checkpoints_vlm_gym_toy_maze_3d_one_image_lr2e_5_ce_ins
120
+ - --wandb_dir
121
+ - /dev/shm/models/checkpoints_vlm_gym_toy_maze_3d_one_image_lr2e_5_ce_ins
122
+ - --wandb_offline
123
+ - 'True'
124
+ program: /home/clouduser/Code/Github/unified_world_model/train/pretrain_unified_navit.py
125
+ code_path: train/pretrain_unified_navit.py
126
+ code_path_local: train/pretrain_unified_navit.py
127
+ git:
128
+ remote_url: https://github.com/para-lost/unified_world_model
129
+ commit: 8d7b26b7e552fc87b592cf3be94d93be7aeca2a9
130
+ root: /dev/shm/models/checkpoints_vlm_gym_toy_maze_3d_one_image_lr2e_5_ce_ins
131
+ host: junyizhang-launch-new-226786748-1-0
132
+ executable: /opt/conda/bin/python3.11
133
+ cpu_count: 48
134
+ cpu_count_logical: 96
135
+ gpu_type: NVIDIA A100-SXM4-80GB
136
+ gpu_count: 8
137
+ disk:
138
+ /:
139
+ total: '1052461830144'
140
+ used: '217227292672'
141
+ memory:
142
+ total: '1437332606976'
143
+ gpu_nvidia:
144
+ - name: NVIDIA A100-SXM4-80GB
145
+ memory_total: '85899345920'
146
+ cuda_cores: 6912
147
+ architecture: Ampere
148
+ uuid: GPU-999aba75-4d4c-b194-b420-5ef33dafe906
149
+ - name: NVIDIA A100-SXM4-80GB
150
+ memory_total: '85899345920'
151
+ cuda_cores: 6912
152
+ architecture: Ampere
153
+ uuid: GPU-3ce11b96-3e25-3bfb-10e3-02c0a26edefe
154
+ - name: NVIDIA A100-SXM4-80GB
155
+ memory_total: '85899345920'
156
+ cuda_cores: 6912
157
+ architecture: Ampere
158
+ uuid: GPU-5b7e1303-96ed-21e0-2d39-fce64aa2bbe0
159
+ - name: NVIDIA A100-SXM4-80GB
160
+ memory_total: '85899345920'
161
+ cuda_cores: 6912
162
+ architecture: Ampere
163
+ uuid: GPU-73198aab-be13-b25a-9acc-0a1132c8c4c5
164
+ - name: NVIDIA A100-SXM4-80GB
165
+ memory_total: '85899345920'
166
+ cuda_cores: 6912
167
+ architecture: Ampere
168
+ uuid: GPU-769c424a-528f-8900-dc0f-4f7ee15abd8a
169
+ - name: NVIDIA A100-SXM4-80GB
170
+ memory_total: '85899345920'
171
+ cuda_cores: 6912
172
+ architecture: Ampere
173
+ uuid: GPU-ae997134-ba6c-c3a6-fa85-aca15004a0b2
174
+ - name: NVIDIA A100-SXM4-80GB
175
+ memory_total: '85899345920'
176
+ cuda_cores: 6912
177
+ architecture: Ampere
178
+ uuid: GPU-6ee941f0-ec7a-486c-ba97-8837582f3e51
179
+ - name: NVIDIA A100-SXM4-80GB
180
+ memory_total: '85899345920'
181
+ cuda_cores: 6912
182
+ architecture: Ampere
183
+ uuid: GPU-f81bd588-16c5-3039-5613-cbadbf7adfb2
184
+ cuda_version: '12.2'
185
+ writer_id: zcxxh1ni9bhfnmriu0rizt0w5oidbkhb
checkpoints_vlm_gym_toy_maze_3d_one_image_lr2e_5_ce_ins/checkpoints_vlm_gym_toy_maze_3d_one_image_lr2e_5_ce_ins/wandb/offline-run-20260130_174942-checkpoints_vlm_gym_toy_maze_3d_one_image_lr2e_5_ce_ins-run0/files/config.yaml CHANGED
@@ -0,0 +1,185 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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: 1769795383
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
+ dcj5d66lnp5q4nhs12j4rc50hc9ofe7y:
44
+ os: Linux-6.6.93+-x86_64-with-glibc2.35
45
+ python: CPython 3.11.10
46
+ started_at: '2026-01-30T17:49:42.622929Z'
47
+ args:
48
+ - --dataset_config_file
49
+ - ./data/configs/vlm_gym_toy_maze_3d_train_celoss.yaml
50
+ - --eval_dataset_config_file
51
+ - ./data/configs/vlm_gym_toy_maze_3d_eval_celoss.yaml
52
+ - --viz_dataset_config_file
53
+ - ./data/configs/vlm_gym_toy_maze_3d_eval_celoss.yaml
54
+ - --inference_hash_file
55
+ - /home/clouduser/Code/Github/launch_new/hashes_test_set_v10.json
56
+ - --task_name
57
+ - toy_maze_3d_v5
58
+ - --instructions_dir
59
+ - ./data/instructions
60
+ - --train_data_dir
61
+ - /home/clouduser/Code/data/gym/toy_maze_3d_v5/train/
62
+ - --train_jsonl_path
63
+ - /home/clouduser/Code/data/gym/toy_maze_3d_v5/train/
64
+ - --eval_data_dir
65
+ - /home/clouduser/Code/data/gym/toy_maze_3d_v5/val/
66
+ - --eval_jsonl_path
67
+ - /home/clouduser/Code/data/gym/toy_maze_3d_v5/val/
68
+ - --model_path
69
+ - /home/clouduser/Code/Models/BAGEL-7B-MoT
70
+ - --layer_module
71
+ - Qwen2MoTDecoderLayer
72
+ - --max_latent_size
73
+ - '64'
74
+ - --resume-from
75
+ - /home/clouduser/Code/Models/BAGEL-7B-MoT
76
+ - --finetune_from_hf
77
+ - 'True'
78
+ - --auto_resume
79
+ - 'False'
80
+ - --resume-model-only
81
+ - 'True'
82
+ - --finetune-from-ema
83
+ - 'True'
84
+ - --log_every
85
+ - '1'
86
+ - --lr
87
+ - 2e-5
88
+ - --warmup_steps
89
+ - '300'
90
+ - --lr_scheduler
91
+ - cosine
92
+ - --num_worker
93
+ - '1'
94
+ - --expected_num_tokens
95
+ - '40000'
96
+ - --max_num_tokens
97
+ - '40000'
98
+ - --max_num_tokens_per_sample
99
+ - '40000'
100
+ - --visual_und
101
+ - 'True'
102
+ - --save_every
103
+ - '2500'
104
+ - --total_steps
105
+ - '5000'
106
+ - --text_cond_dropout_prob
107
+ - '0.0'
108
+ - --vae_cond_dropout_prob
109
+ - '0.3'
110
+ - --vit_cond_dropout_prob
111
+ - '0.0'
112
+ - --ema
113
+ - '0.993'
114
+ - --checkpoint_dir
115
+ - /dev/shm/models/checkpoints_vlm_gym_toy_maze_3d_one_image_lr2e_5_ce_ins
116
+ - --wandb_project
117
+ - bagel
118
+ - --wandb_name
119
+ - checkpoints_vlm_gym_toy_maze_3d_one_image_lr2e_5_ce_ins
120
+ - --wandb_dir
121
+ - /dev/shm/models/checkpoints_vlm_gym_toy_maze_3d_one_image_lr2e_5_ce_ins
122
+ - --wandb_offline
123
+ - 'True'
124
+ program: /home/clouduser/Code/Github/unified_world_model/train/pretrain_unified_navit.py
125
+ code_path: train/pretrain_unified_navit.py
126
+ code_path_local: train/pretrain_unified_navit.py
127
+ git:
128
+ remote_url: https://github.com/para-lost/unified_world_model
129
+ commit: 20694bc0d8bf3c48fe817e31af63a14d6e1e3619
130
+ root: /dev/shm/models/checkpoints_vlm_gym_toy_maze_3d_one_image_lr2e_5_ce_ins
131
+ host: junyizhang-launch-new-226992214-1-0
132
+ executable: /opt/conda/bin/python3.11
133
+ cpu_count: 48
134
+ cpu_count_logical: 96
135
+ gpu_type: NVIDIA A100-SXM4-80GB
136
+ gpu_count: 8
137
+ disk:
138
+ /:
139
+ total: '1052461830144'
140
+ used: '217227833344'
141
+ memory:
142
+ total: '1437332606976'
143
+ gpu_nvidia:
144
+ - name: NVIDIA A100-SXM4-80GB
145
+ memory_total: '85899345920'
146
+ cuda_cores: 6912
147
+ architecture: Ampere
148
+ uuid: GPU-3a472f7c-aa92-cede-f401-880e3113dd3c
149
+ - name: NVIDIA A100-SXM4-80GB
150
+ memory_total: '85899345920'
151
+ cuda_cores: 6912
152
+ architecture: Ampere
153
+ uuid: GPU-bb20d5a7-d024-e881-c41c-5d603567e046
154
+ - name: NVIDIA A100-SXM4-80GB
155
+ memory_total: '85899345920'
156
+ cuda_cores: 6912
157
+ architecture: Ampere
158
+ uuid: GPU-70ec8ac9-ee3b-2275-64e3-2972e735b268
159
+ - name: NVIDIA A100-SXM4-80GB
160
+ memory_total: '85899345920'
161
+ cuda_cores: 6912
162
+ architecture: Ampere
163
+ uuid: GPU-c5e7d2f5-85f5-97a5-4852-925f2562dc4d
164
+ - name: NVIDIA A100-SXM4-80GB
165
+ memory_total: '85899345920'
166
+ cuda_cores: 6912
167
+ architecture: Ampere
168
+ uuid: GPU-2a401b4b-58e4-50e6-2c9d-7ddf1e845958
169
+ - name: NVIDIA A100-SXM4-80GB
170
+ memory_total: '85899345920'
171
+ cuda_cores: 6912
172
+ architecture: Ampere
173
+ uuid: GPU-2d6385c2-e99d-bf16-60b2-966bedcccdca
174
+ - name: NVIDIA A100-SXM4-80GB
175
+ memory_total: '85899345920'
176
+ cuda_cores: 6912
177
+ architecture: Ampere
178
+ uuid: GPU-16a524b0-3995-0081-61f9-eb8e462b1b91
179
+ - name: NVIDIA A100-SXM4-80GB
180
+ memory_total: '85899345920'
181
+ cuda_cores: 6912
182
+ architecture: Ampere
183
+ uuid: GPU-b2ac53f1-4b38-71f9-5f20-ff03aec335dc
184
+ cuda_version: '12.2'
185
+ writer_id: dcj5d66lnp5q4nhs12j4rc50hc9ofe7y
checkpoints_vlm_gym_toy_maze_3d_one_image_lr2e_5_ce_ins/checkpoints_vlm_gym_toy_maze_3d_one_image_lr2e_5_ce_ins/wandb/offline-run-20260130_203229-checkpoints_vlm_gym_toy_maze_3d_one_image_lr2e_5_ce_ins-run0/files/config.yaml CHANGED
@@ -0,0 +1,456 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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: 1769805149
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
+ - 61
39
+ 4: 3.11.10
40
+ 5: 0.23.1
41
+ 6: 4.49.0
42
+ 13: linux-x86_64
43
+ e:
44
+ 7mdpbdubtrx63j78hsn2tz4yj6qjvhql:
45
+ os: Linux-6.6.93+-x86_64-with-glibc2.35
46
+ python: CPython 3.11.10
47
+ started_at: '2026-01-30T20:32:29.312128Z'
48
+ args:
49
+ - --dataset_config_file
50
+ - ./data/configs/vlm_gym_toy_maze_3d_train_celoss.yaml
51
+ - --eval_dataset_config_file
52
+ - ./data/configs/vlm_gym_toy_maze_3d_eval_celoss.yaml
53
+ - --viz_dataset_config_file
54
+ - ./data/configs/vlm_gym_toy_maze_3d_eval_celoss.yaml
55
+ - --inference_hash_file
56
+ - /home/clouduser/Code/Github/launch_new/hashes_test_set_v10.json
57
+ - --task_name
58
+ - toy_maze_3d_v5
59
+ - --instructions_dir
60
+ - ./data/instructions
61
+ - --train_data_dir
62
+ - /home/clouduser/Code/data/gym/toy_maze_3d_v5/train/
63
+ - --train_jsonl_path
64
+ - /home/clouduser/Code/data/gym/toy_maze_3d_v5/train/
65
+ - --eval_data_dir
66
+ - /home/clouduser/Code/data/gym/toy_maze_3d_v5/val/
67
+ - --eval_jsonl_path
68
+ - /home/clouduser/Code/data/gym/toy_maze_3d_v5/val/
69
+ - --model_path
70
+ - /home/clouduser/Code/Models/BAGEL-7B-MoT
71
+ - --layer_module
72
+ - Qwen2MoTDecoderLayer
73
+ - --max_latent_size
74
+ - '64'
75
+ - --resume-from
76
+ - /home/clouduser/Code/Models/BAGEL-7B-MoT
77
+ - --finetune_from_hf
78
+ - 'True'
79
+ - --auto_resume
80
+ - 'False'
81
+ - --resume-model-only
82
+ - 'True'
83
+ - --finetune-from-ema
84
+ - 'True'
85
+ - --log_every
86
+ - '1'
87
+ - --lr
88
+ - 2e-5
89
+ - --warmup_steps
90
+ - '300'
91
+ - --lr_scheduler
92
+ - cosine
93
+ - --num_worker
94
+ - '1'
95
+ - --expected_num_tokens
96
+ - '50000'
97
+ - --max_num_tokens
98
+ - '50000'
99
+ - --max_num_tokens_per_sample
100
+ - '50000'
101
+ - --visual_und
102
+ - 'True'
103
+ - --save_every
104
+ - '2500'
105
+ - --total_steps
106
+ - '5000'
107
+ - --text_cond_dropout_prob
108
+ - '0.0'
109
+ - --vae_cond_dropout_prob
110
+ - '0.3'
111
+ - --vit_cond_dropout_prob
112
+ - '0.0'
113
+ - --ema
114
+ - '0.993'
115
+ - --checkpoint_dir
116
+ - /dev/shm/models/checkpoints_vlm_gym_toy_maze_3d_one_image_lr2e_5_ce_ins
117
+ - --wandb_project
118
+ - bagel
119
+ - --wandb_name
120
+ - checkpoints_vlm_gym_toy_maze_3d_one_image_lr2e_5_ce_ins
121
+ - --wandb_dir
122
+ - /dev/shm/models/checkpoints_vlm_gym_toy_maze_3d_one_image_lr2e_5_ce_ins
123
+ - --wandb_offline
124
+ - 'True'
125
+ program: /home/clouduser/Code/Github/unified_world_model/train/pretrain_unified_navit.py
126
+ code_path: train/pretrain_unified_navit.py
127
+ code_path_local: train/pretrain_unified_navit.py
128
+ git:
129
+ remote_url: https://github.com/para-lost/unified_world_model
130
+ commit: 3525b3ceed8df2fd8df555a87280a3194063b307
131
+ root: /dev/shm/models/checkpoints_vlm_gym_toy_maze_3d_one_image_lr2e_5_ce_ins
132
+ host: junyizhang-launch-new-227016951-1-0
133
+ executable: /opt/conda/bin/python3.11
134
+ cpu_count: 48
135
+ cpu_count_logical: 96
136
+ gpu_type: NVIDIA A100-SXM4-80GB
137
+ gpu_count: 8
138
+ disk:
139
+ /:
140
+ total: '1052461830144'
141
+ used: '217229398016'
142
+ memory:
143
+ total: '1437332606976'
144
+ gpu_nvidia:
145
+ - name: NVIDIA A100-SXM4-80GB
146
+ memory_total: '85899345920'
147
+ cuda_cores: 6912
148
+ architecture: Ampere
149
+ uuid: GPU-280260ef-7124-b301-3970-9c8561f32e71
150
+ - name: NVIDIA A100-SXM4-80GB
151
+ memory_total: '85899345920'
152
+ cuda_cores: 6912
153
+ architecture: Ampere
154
+ uuid: GPU-e42e5ead-f745-3375-eb3f-94287780189f
155
+ - name: NVIDIA A100-SXM4-80GB
156
+ memory_total: '85899345920'
157
+ cuda_cores: 6912
158
+ architecture: Ampere
159
+ uuid: GPU-33bf0174-b8ce-489c-215b-08b4b107eecc
160
+ - name: NVIDIA A100-SXM4-80GB
161
+ memory_total: '85899345920'
162
+ cuda_cores: 6912
163
+ architecture: Ampere
164
+ uuid: GPU-5973177d-a910-a5d6-45ed-b118a9e73a10
165
+ - name: NVIDIA A100-SXM4-80GB
166
+ memory_total: '85899345920'
167
+ cuda_cores: 6912
168
+ architecture: Ampere
169
+ uuid: GPU-0269d322-2fb5-7d6c-794a-b22ecc05762e
170
+ - name: NVIDIA A100-SXM4-80GB
171
+ memory_total: '85899345920'
172
+ cuda_cores: 6912
173
+ architecture: Ampere
174
+ uuid: GPU-99c90791-3cf8-cf47-0e56-2bf0308d3056
175
+ - name: NVIDIA A100-SXM4-80GB
176
+ memory_total: '85899345920'
177
+ cuda_cores: 6912
178
+ architecture: Ampere
179
+ uuid: GPU-2ac58fb0-f683-ee8e-b676-6e074a8141fa
180
+ - name: NVIDIA A100-SXM4-80GB
181
+ memory_total: '85899345920'
182
+ cuda_cores: 6912
183
+ architecture: Ampere
184
+ uuid: GPU-93d910ad-1813-c2cc-d788-949b06693659
185
+ cuda_version: '12.2'
186
+ writer_id: 7mdpbdubtrx63j78hsn2tz4yj6qjvhql
187
+ visual_gen:
188
+ desc: null
189
+ value: true
190
+ visual_und:
191
+ desc: null
192
+ value: true
193
+ results_dir:
194
+ desc: null
195
+ value: results
196
+ checkpoint_dir:
197
+ desc: null
198
+ value: /dev/shm/models/checkpoints_vlm_gym_toy_maze_3d_one_image_lr2e_5_ce_ins
199
+ wandb_project:
200
+ desc: null
201
+ value: bagel
202
+ wandb_name:
203
+ desc: null
204
+ value: checkpoints_vlm_gym_toy_maze_3d_one_image_lr2e_5_ce_ins
205
+ wandb_runid:
206
+ desc: null
207
+ value: '0'
208
+ wandb_resume:
209
+ desc: null
210
+ value: allow
211
+ wandb_offline:
212
+ desc: null
213
+ value: true
214
+ wandb_dir:
215
+ desc: null
216
+ value: /dev/shm/models/checkpoints_vlm_gym_toy_maze_3d_one_image_lr2e_5_ce_ins
217
+ global_seed:
218
+ desc: null
219
+ value: 4396
220
+ auto_resume:
221
+ desc: null
222
+ value: false
223
+ resume_from:
224
+ desc: null
225
+ value: /home/clouduser/Code/Models/BAGEL-7B-MoT
226
+ resume_model_only:
227
+ desc: null
228
+ value: true
229
+ finetune_from_ema:
230
+ desc: null
231
+ value: true
232
+ finetune_from_hf:
233
+ desc: null
234
+ value: true
235
+ log_every:
236
+ desc: null
237
+ value: 1
238
+ save_every:
239
+ desc: null
240
+ value: 2500
241
+ total_steps:
242
+ desc: null
243
+ value: 5000
244
+ warmup_steps:
245
+ desc: null
246
+ value: 300
247
+ lr_scheduler:
248
+ desc: null
249
+ value: cosine
250
+ lr:
251
+ desc: null
252
+ value: 2.0e-05
253
+ min_lr:
254
+ desc: null
255
+ value: 1.0e-07
256
+ beta1:
257
+ desc: null
258
+ value: 0.9
259
+ beta2:
260
+ desc: null
261
+ value: 0.95
262
+ eps:
263
+ desc: null
264
+ value: 1.0e-15
265
+ ema:
266
+ desc: null
267
+ value: 0.993
268
+ max_grad_norm:
269
+ desc: null
270
+ value: 1.0
271
+ timestep_shift:
272
+ desc: null
273
+ value: 1.0
274
+ mse_weight:
275
+ desc: null
276
+ value: 1.0
277
+ ce_weight:
278
+ desc: null
279
+ value: 1.0
280
+ ce_loss_reweighting:
281
+ desc: null
282
+ value: false
283
+ expected_num_tokens:
284
+ desc: null
285
+ value: 50000
286
+ num_replicate:
287
+ desc: null
288
+ value: 1
289
+ num_shard:
290
+ desc: null
291
+ value: 8
292
+ sharding_strategy:
293
+ desc: null
294
+ value: HYBRID_SHARD
295
+ backward_prefetch:
296
+ desc: null
297
+ value: BACKWARD_PRE
298
+ cpu_offload:
299
+ desc: null
300
+ value: false
301
+ freeze_llm:
302
+ desc: null
303
+ value: false
304
+ freeze_vit:
305
+ desc: null
306
+ value: false
307
+ freeze_vae:
308
+ desc: null
309
+ value: true
310
+ freeze_und:
311
+ desc: null
312
+ value: false
313
+ copy_init_moe:
314
+ desc: null
315
+ value: true
316
+ use_flex:
317
+ desc: null
318
+ value: false
319
+ eval_every:
320
+ desc: null
321
+ value: 500
322
+ num_eval_batches:
323
+ desc: null
324
+ value: 20
325
+ use_ema_for_eval:
326
+ desc: null
327
+ value: true
328
+ eval_log_dir:
329
+ desc: null
330
+ value: null
331
+ eval_run_tag:
332
+ desc: null
333
+ value: ''
334
+ viz_every:
335
+ desc: null
336
+ value: 500
337
+ viz_n:
338
+ desc: null
339
+ value: 8
340
+ viz_outdir:
341
+ desc: null
342
+ value: results/viz
343
+ eval_dataset_config_file:
344
+ desc: null
345
+ value: ./data/configs/vlm_gym_toy_maze_3d_eval_celoss.yaml
346
+ viz_dataset_config_file:
347
+ desc: null
348
+ value: ./data/configs/vlm_gym_toy_maze_3d_eval_celoss.yaml
349
+ eval_print_n:
350
+ desc: null
351
+ value: 3
352
+ save_ema_only:
353
+ desc: null
354
+ value: true
355
+ save_optimizer:
356
+ desc: null
357
+ value: false
358
+ model_path:
359
+ desc: null
360
+ value: /home/clouduser/Code/Models/BAGEL-7B-MoT
361
+ llm_path:
362
+ desc: null
363
+ value: hf/Qwen2.5-0.5B-Instruct/
364
+ llm_qk_norm:
365
+ desc: null
366
+ value: true
367
+ tie_word_embeddings:
368
+ desc: null
369
+ value: false
370
+ layer_module:
371
+ desc: null
372
+ value: Qwen2MoTDecoderLayer
373
+ vae_path:
374
+ desc: null
375
+ value: flux/vae/ae.safetensors
376
+ vit_path:
377
+ desc: null
378
+ value: hf/siglip-so400m-14-980-flash-attn2-navit/
379
+ max_latent_size:
380
+ desc: null
381
+ value: 64
382
+ latent_patch_size:
383
+ desc: null
384
+ value: 2
385
+ vit_patch_size:
386
+ desc: null
387
+ value: 14
388
+ vit_max_num_patch_per_side:
389
+ desc: null
390
+ value: 70
391
+ connector_act:
392
+ desc: null
393
+ value: gelu_pytorch_tanh
394
+ interpolate_pos:
395
+ desc: null
396
+ value: false
397
+ vit_select_layer:
398
+ desc: null
399
+ value: -2
400
+ vit_rope:
401
+ desc: null
402
+ value: false
403
+ text_cond_dropout_prob:
404
+ desc: null
405
+ value: 0.0
406
+ vae_cond_dropout_prob:
407
+ desc: null
408
+ value: 0.3
409
+ vit_cond_dropout_prob:
410
+ desc: null
411
+ value: 0.0
412
+ dataset_config_file:
413
+ desc: null
414
+ value: ./data/configs/vlm_gym_toy_maze_3d_train_celoss.yaml
415
+ train_data_dir:
416
+ desc: null
417
+ value: /home/clouduser/Code/data/gym/toy_maze_3d_v5/train/
418
+ train_jsonl_path:
419
+ desc: null
420
+ value: /home/clouduser/Code/data/gym/toy_maze_3d_v5/train/
421
+ eval_data_dir:
422
+ desc: null
423
+ value: /home/clouduser/Code/data/gym/toy_maze_3d_v5/val/
424
+ eval_jsonl_path:
425
+ desc: null
426
+ value: /home/clouduser/Code/data/gym/toy_maze_3d_v5/val/
427
+ inference_hash_file:
428
+ desc: null
429
+ value: /home/clouduser/Code/Github/launch_new/hashes_test_set_v10.json
430
+ task_name:
431
+ desc: null
432
+ value: toy_maze_3d_v5
433
+ instructions_dir:
434
+ desc: null
435
+ value: ./data/instructions
436
+ prefetch_factor:
437
+ desc: null
438
+ value: 2
439
+ num_workers:
440
+ desc: null
441
+ value: 1
442
+ max_num_tokens_per_sample:
443
+ desc: null
444
+ value: 50000
445
+ max_num_tokens:
446
+ desc: null
447
+ value: 50000
448
+ prefer_buffer_before:
449
+ desc: null
450
+ value: 16384
451
+ max_buffer_size:
452
+ desc: null
453
+ value: 50
454
+ data_seed:
455
+ desc: null
456
+ value: 42