Upload folder using huggingface_hub
Browse files
QwenAdapter-0130-pre-fintune-v/checkpoints/steps_51000_pytorch_model.pt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:2401ef0016ae0a91893523307d0a495b8eda57ca65d71a4cd2019b74650b09e4
|
| 3 |
+
size 6870058775
|
QwenAdapter-0130-pre-fintune-v/config.json
CHANGED
|
@@ -35,7 +35,7 @@
|
|
| 35 |
"data_root_dir": "/lumos-vePFS/shenzhen/data/data_warehouse/data_warehouse_output",
|
| 36 |
"data_mix": "train_internAdapter",
|
| 37 |
"per_device_batch_size": 40,
|
| 38 |
-
"num_workers":
|
| 39 |
"action_type": "abs_ee",
|
| 40 |
"image_size": 256,
|
| 41 |
"video_backend": "pyav"
|
|
@@ -77,7 +77,7 @@
|
|
| 77 |
"weight_decay": 0
|
| 78 |
},
|
| 79 |
"is_resume": true,
|
| 80 |
-
"resume_step":
|
| 81 |
},
|
| 82 |
"is_resume": false,
|
| 83 |
"resume_epoch": null,
|
|
|
|
| 35 |
"data_root_dir": "/lumos-vePFS/shenzhen/data/data_warehouse/data_warehouse_output",
|
| 36 |
"data_mix": "train_internAdapter",
|
| 37 |
"per_device_batch_size": 40,
|
| 38 |
+
"num_workers": 8,
|
| 39 |
"action_type": "abs_ee",
|
| 40 |
"image_size": 256,
|
| 41 |
"video_backend": "pyav"
|
|
|
|
| 77 |
"weight_decay": 0
|
| 78 |
},
|
| 79 |
"is_resume": true,
|
| 80 |
+
"resume_step": 40000
|
| 81 |
},
|
| 82 |
"is_resume": false,
|
| 83 |
"resume_epoch": null,
|
QwenAdapter-0130-pre-fintune-v/config.yaml
CHANGED
|
@@ -1,76 +1,48 @@
|
|
| 1 |
-
|
| 2 |
-
|
| 3 |
-
|
| 4 |
-
|
| 5 |
-
|
| 6 |
-
|
| 7 |
-
|
| 8 |
-
wandb_project: qwenvl-0130-pre-fintune+vit-eep
|
| 9 |
-
is_debug: false
|
| 10 |
framework:
|
| 11 |
-
name: QwenAdapter
|
| 12 |
-
qwenvl:
|
| 13 |
-
base_vlm: /root/jinchengp/LumosVLA/base_vl/Qwen3-VL-2B-Instruct
|
| 14 |
-
attn_implementation: flash_attention_2
|
| 15 |
-
vl_hidden_dim: 2048
|
| 16 |
action_model:
|
| 17 |
-
action_model_type: VLA-Adapter
|
| 18 |
-
hidden_dim: 2048
|
| 19 |
action_dim: 20
|
| 20 |
-
state_dim: 20
|
| 21 |
action_query_num: 64
|
|
|
|
|
|
|
|
|
|
|
|
|
| 22 |
use_pro_version: true
|
| 23 |
use_proprio: true
|
| 24 |
-
|
| 25 |
-
|
| 26 |
-
|
| 27 |
-
|
| 28 |
-
|
| 29 |
-
|
| 30 |
-
|
| 31 |
-
|
| 32 |
-
per_device_batch_size: 40
|
| 33 |
-
num_workers: 2
|
| 34 |
-
action_type: abs_ee
|
| 35 |
-
image_size: 256
|
| 36 |
-
video_backend: pyav
|
| 37 |
trainer:
|
| 38 |
-
epochs: 100
|
| 39 |
-
max_train_steps: 85000
|
| 40 |
-
num_warmup_steps: 1000
|
| 41 |
-
save_interval: 1000
|
| 42 |
eval_interval: 100000
|
|
|
|
|
|
|
|
|
|
|
|
|
| 43 |
learning_rate:
|
|
|
|
| 44 |
base: 4.0e-06
|
| 45 |
qwen_vl_interface: 4.0e-06
|
| 46 |
-
action_model: 1.0e-05
|
| 47 |
-
lr_scheduler_type: cosine_with_min_lr
|
| 48 |
-
scheduler_specific_kwargs:
|
| 49 |
-
min_lr: 5.0e-07
|
| 50 |
-
freeze_modules: qwen_vl_interface.model.model.language_model
|
| 51 |
-
loss_scale:
|
| 52 |
-
vla: 1.0
|
| 53 |
-
vlm: 0.05
|
| 54 |
-
max_grad_norm: 1.0
|
| 55 |
-
warmup_ratio: 0.1
|
| 56 |
-
weight_decay: 0.0
|
| 57 |
logging_frequency: 10
|
| 58 |
-
|
| 59 |
-
|
|
|
|
| 60 |
optimizer:
|
| 61 |
-
name: AdamW
|
| 62 |
betas:
|
| 63 |
- 0.9
|
| 64 |
- 0.95
|
| 65 |
eps: 1.0e-08
|
| 66 |
weight_decay: 0
|
| 67 |
-
|
| 68 |
-
|
| 69 |
-
|
| 70 |
-
resume_epoch: null
|
| 71 |
-
resume_step: null
|
| 72 |
-
enable_gradient_checkpointing: true
|
| 73 |
-
enable_mixed_precision_training: true
|
| 74 |
-
report_to: wandb
|
| 75 |
-
wandb_run_name: v1
|
| 76 |
-
output_dir: /root/jinchengp/LumosVLA/results/QwenAdapter-0130-pre-fintune+vit-eep
|
|
|
|
| 1 |
+
datasets:
|
| 2 |
+
vla_data:
|
| 3 |
+
data_mix: train_internAdapter
|
| 4 |
+
data_root_dir: /lumos-vePFS/shenzhen/data/data_warehouse/data_warehouse_output
|
| 5 |
+
dataset_py: lerobot_datasets
|
| 6 |
+
num_workers: 8
|
| 7 |
+
per_device_batch_size: 40
|
|
|
|
|
|
|
| 8 |
framework:
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 9 |
action_model:
|
|
|
|
|
|
|
| 10 |
action_dim: 20
|
|
|
|
| 11 |
action_query_num: 64
|
| 12 |
+
hidden_dim: 2048
|
| 13 |
+
num_actions_chunk: 30
|
| 14 |
+
phase: Training
|
| 15 |
+
state_dim: 20
|
| 16 |
use_pro_version: true
|
| 17 |
use_proprio: true
|
| 18 |
+
name: QwenAdapter
|
| 19 |
+
qwenvl:
|
| 20 |
+
base_vlm: /root/jinchengp/LumosVLA/base_vl/Qwen3-VL-2B-Instruct
|
| 21 |
+
vl_hidden_dim: 2048
|
| 22 |
+
output_dir: /root/jinchengp/LumosVLA/results/QwenAdapter-0130-pre-fintune+vit-eep
|
| 23 |
+
run_id: QwenAdapter-0130-pre-fintune+vit-eep
|
| 24 |
+
run_root_dir: /root/jinchengp/LumosVLA/results
|
| 25 |
+
seed: 42
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 26 |
trainer:
|
|
|
|
|
|
|
|
|
|
|
|
|
| 27 |
eval_interval: 100000
|
| 28 |
+
freeze_modules: qwen_vl_interface.model.model.language_model
|
| 29 |
+
gradient_accumulation_steps: 1
|
| 30 |
+
gradient_clipping: 1.0
|
| 31 |
+
is_resume: true
|
| 32 |
learning_rate:
|
| 33 |
+
action_model: 1.0e-05
|
| 34 |
base: 4.0e-06
|
| 35 |
qwen_vl_interface: 4.0e-06
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 36 |
logging_frequency: 10
|
| 37 |
+
lr_scheduler_type: cosine_with_min_lr
|
| 38 |
+
max_train_steps: 85000
|
| 39 |
+
num_warmup_steps: 1000
|
| 40 |
optimizer:
|
|
|
|
| 41 |
betas:
|
| 42 |
- 0.9
|
| 43 |
- 0.95
|
| 44 |
eps: 1.0e-08
|
| 45 |
weight_decay: 0
|
| 46 |
+
save_interval: 1000
|
| 47 |
+
scheduler_specific_kwargs:
|
| 48 |
+
min_lr: 5.0e-07
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
QwenAdapter-0130-pre-fintune-v/dataset_statistics.json
CHANGED
|
@@ -2,36 +2,36 @@
|
|
| 2 |
"new_embodiment": {
|
| 3 |
"action": {
|
| 4 |
"mean": [
|
| 5 |
-
0.
|
| 6 |
-
-0.
|
| 7 |
-
0.
|
| 8 |
-
-1.
|
| 9 |
-
0.
|
| 10 |
-
1.
|
| 11 |
-
0.
|
| 12 |
-
0.
|
| 13 |
-
0.
|
| 14 |
-
0.
|
| 15 |
-
1.
|
| 16 |
-
0.
|
| 17 |
-
-1.
|
| 18 |
-
0.
|
| 19 |
],
|
| 20 |
"std": [
|
| 21 |
-
0.
|
| 22 |
-
0.
|
| 23 |
-
0.
|
| 24 |
-
2.
|
| 25 |
-
0.
|
| 26 |
-
2.
|
| 27 |
-
0.
|
| 28 |
-
0.
|
| 29 |
-
0.
|
| 30 |
-
0.
|
| 31 |
-
2.
|
| 32 |
-
0.
|
| 33 |
-
2.
|
| 34 |
-
0.
|
| 35 |
],
|
| 36 |
"max": [
|
| 37 |
0.5931400060653687,
|
|
@@ -58,7 +58,7 @@
|
|
| 58 |
-3.1410515308380127,
|
| 59 |
-0.0017500000540167093,
|
| 60 |
-0.03119499981403351,
|
| 61 |
-
-0.
|
| 62 |
0.06903599947690964,
|
| 63 |
-3.1410515308380127,
|
| 64 |
-0.7706850171089172,
|
|
@@ -116,36 +116,36 @@
|
|
| 116 |
},
|
| 117 |
"state": {
|
| 118 |
"mean": [
|
| 119 |
-
0.
|
| 120 |
-
-0.
|
| 121 |
-
0.
|
| 122 |
-
-1.
|
| 123 |
-
0.
|
| 124 |
-
1.
|
| 125 |
-
0.
|
| 126 |
-
0.
|
| 127 |
-
0.
|
| 128 |
-
0.
|
| 129 |
-
1.
|
| 130 |
-
0.
|
| 131 |
-
-1.
|
| 132 |
-
0.
|
| 133 |
],
|
| 134 |
"std": [
|
| 135 |
-
0.
|
| 136 |
-
0.
|
| 137 |
-
0.
|
| 138 |
-
2.
|
| 139 |
-
0.
|
| 140 |
-
2.
|
| 141 |
-
0.
|
| 142 |
-
0.
|
| 143 |
-
0.
|
| 144 |
-
0.
|
| 145 |
-
2.
|
| 146 |
-
0.
|
| 147 |
-
2.
|
| 148 |
-
0.
|
| 149 |
],
|
| 150 |
"max": [
|
| 151 |
0.5931400060653687,
|
|
@@ -172,7 +172,7 @@
|
|
| 172 |
-3.1410515308380127,
|
| 173 |
-0.0017500000540167093,
|
| 174 |
-0.03119499981403351,
|
| 175 |
-
-0.
|
| 176 |
0.06903599947690964,
|
| 177 |
-3.1410515308380127,
|
| 178 |
-0.7706850171089172,
|
|
@@ -212,7 +212,7 @@
|
|
| 212 |
0.05992000177502632
|
| 213 |
]
|
| 214 |
},
|
| 215 |
-
"num_transitions":
|
| 216 |
-
"num_trajectories":
|
| 217 |
}
|
| 218 |
}
|
|
|
|
| 2 |
"new_embodiment": {
|
| 3 |
"action": {
|
| 4 |
"mean": [
|
| 5 |
+
0.25654253200308913,
|
| 6 |
+
-0.07550963895373071,
|
| 7 |
+
0.19405086379345146,
|
| 8 |
+
-1.2614229729730786,
|
| 9 |
+
0.41015166799499564,
|
| 10 |
+
1.6125561946058933,
|
| 11 |
+
0.033000507073043146,
|
| 12 |
+
0.2293869228395697,
|
| 13 |
+
0.08791677572139318,
|
| 14 |
+
0.20820339411905364,
|
| 15 |
+
1.5555991992762643,
|
| 16 |
+
0.5074784118835239,
|
| 17 |
+
-1.6657334892717126,
|
| 18 |
+
0.034744793676758465
|
| 19 |
],
|
| 20 |
"std": [
|
| 21 |
+
0.10175063885577025,
|
| 22 |
+
0.10670231743920426,
|
| 23 |
+
0.06647325286023913,
|
| 24 |
+
2.6082373231142837,
|
| 25 |
+
0.32410331954665494,
|
| 26 |
+
2.26503174399618,
|
| 27 |
+
0.028480157379054398,
|
| 28 |
+
0.08840424754002822,
|
| 29 |
+
0.07915035920705979,
|
| 30 |
+
0.054107928205173586,
|
| 31 |
+
2.467586778069468,
|
| 32 |
+
0.2909265959253485,
|
| 33 |
+
2.259582180770817,
|
| 34 |
+
0.028410108222688427
|
| 35 |
],
|
| 36 |
"max": [
|
| 37 |
0.5931400060653687,
|
|
|
|
| 58 |
-3.1410515308380127,
|
| 59 |
-0.0017500000540167093,
|
| 60 |
-0.03119499981403351,
|
| 61 |
+
-0.3384290039539337,
|
| 62 |
0.06903599947690964,
|
| 63 |
-3.1410515308380127,
|
| 64 |
-0.7706850171089172,
|
|
|
|
| 116 |
},
|
| 117 |
"state": {
|
| 118 |
"mean": [
|
| 119 |
+
0.25654253200308913,
|
| 120 |
+
-0.07550963895373071,
|
| 121 |
+
0.19405086379345146,
|
| 122 |
+
-1.2614229729730786,
|
| 123 |
+
0.41015166799499564,
|
| 124 |
+
1.6125561946058933,
|
| 125 |
+
0.033000507073043146,
|
| 126 |
+
0.2293869228395697,
|
| 127 |
+
0.08791677572139318,
|
| 128 |
+
0.20820339411905364,
|
| 129 |
+
1.5555991992762643,
|
| 130 |
+
0.5074784118835239,
|
| 131 |
+
-1.6657334892717126,
|
| 132 |
+
0.034744793676758465
|
| 133 |
],
|
| 134 |
"std": [
|
| 135 |
+
0.10175063885577025,
|
| 136 |
+
0.10670231743920426,
|
| 137 |
+
0.06647325286023913,
|
| 138 |
+
2.6082373231142837,
|
| 139 |
+
0.32410331954665494,
|
| 140 |
+
2.26503174399618,
|
| 141 |
+
0.028480157379054398,
|
| 142 |
+
0.08840424754002822,
|
| 143 |
+
0.07915035920705979,
|
| 144 |
+
0.054107928205173586,
|
| 145 |
+
2.467586778069468,
|
| 146 |
+
0.2909265959253485,
|
| 147 |
+
2.259582180770817,
|
| 148 |
+
0.028410108222688427
|
| 149 |
],
|
| 150 |
"max": [
|
| 151 |
0.5931400060653687,
|
|
|
|
| 172 |
-3.1410515308380127,
|
| 173 |
-0.0017500000540167093,
|
| 174 |
-0.03119499981403351,
|
| 175 |
+
-0.3384290039539337,
|
| 176 |
0.06903599947690964,
|
| 177 |
-3.1410515308380127,
|
| 178 |
-0.7706850171089172,
|
|
|
|
| 212 |
0.05992000177502632
|
| 213 |
]
|
| 214 |
},
|
| 215 |
+
"num_transitions": 2477942,
|
| 216 |
+
"num_trajectories": 1945
|
| 217 |
}
|
| 218 |
}
|