Upload folder using huggingface_hub
Browse files
qwen-0130v/checkpoints/steps_56000_pytorch_model.pt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:3a0ccfec364edc35e0971ca9d55ebbdf5c3526ead6c88cc2147c7af90f8621da
|
| 3 |
+
size 6870057558
|
qwen-0130v/config.json
CHANGED
|
@@ -34,7 +34,7 @@
|
|
| 34 |
"data_root_dir": "/data/LumosVLA/LumosVLA/data/starvla_intern",
|
| 35 |
"data_mix": "train_internAdapter",
|
| 36 |
"per_device_batch_size": 32,
|
| 37 |
-
"num_workers":
|
| 38 |
"action_type": "abs_ee",
|
| 39 |
"image_size": 256,
|
| 40 |
"video_backend": "pyav"
|
|
@@ -74,7 +74,9 @@
|
|
| 74 |
],
|
| 75 |
"eps": 1e-08,
|
| 76 |
"weight_decay": 0
|
| 77 |
-
}
|
|
|
|
|
|
|
| 78 |
},
|
| 79 |
"is_resume": false,
|
| 80 |
"resume_epoch": null,
|
|
|
|
| 34 |
"data_root_dir": "/data/LumosVLA/LumosVLA/data/starvla_intern",
|
| 35 |
"data_mix": "train_internAdapter",
|
| 36 |
"per_device_batch_size": 32,
|
| 37 |
+
"num_workers": 8,
|
| 38 |
"action_type": "abs_ee",
|
| 39 |
"image_size": 256,
|
| 40 |
"video_backend": "pyav"
|
|
|
|
| 74 |
],
|
| 75 |
"eps": 1e-08,
|
| 76 |
"weight_decay": 0
|
| 77 |
+
},
|
| 78 |
+
"is_resume": true,
|
| 79 |
+
"resume_step": 56000
|
| 80 |
},
|
| 81 |
"is_resume": false,
|
| 82 |
"resume_epoch": null,
|
qwen-0130v/config.yaml
CHANGED
|
@@ -1,49 +1,75 @@
|
|
| 1 |
-
|
| 2 |
-
|
| 3 |
-
|
| 4 |
-
|
| 5 |
-
|
| 6 |
-
|
| 7 |
-
|
|
|
|
|
|
|
| 8 |
framework:
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 9 |
action_model:
|
| 10 |
-
|
| 11 |
-
action_query_num: 64
|
| 12 |
hidden_dim: 2048
|
| 13 |
-
|
| 14 |
-
phase: Training
|
| 15 |
state_dim: 20
|
|
|
|
| 16 |
use_pro_version: true
|
| 17 |
use_proprio: true
|
| 18 |
-
|
| 19 |
-
|
| 20 |
-
|
| 21 |
-
|
| 22 |
-
|
| 23 |
-
|
| 24 |
-
|
| 25 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
| 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 |
learning_rate:
|
| 32 |
-
action_model: 1.0e-05
|
| 33 |
base: 4.0e-06
|
| 34 |
qwen_vl_interface: 4.0e-06
|
| 35 |
-
|
| 36 |
lr_scheduler_type: cosine_with_min_lr
|
| 37 |
-
|
| 38 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 39 |
optimizer:
|
|
|
|
| 40 |
betas:
|
| 41 |
- 0.9
|
| 42 |
- 0.95
|
| 43 |
eps: 1.0e-08
|
| 44 |
weight_decay: 0
|
| 45 |
-
|
| 46 |
-
|
| 47 |
-
|
| 48 |
-
|
| 49 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
run_id: QwenAdapter-0130v-eep
|
| 2 |
+
run_root_dir: /data/LumosVLA/LumosVLA/results
|
| 3 |
+
seed: 42
|
| 4 |
+
trackers:
|
| 5 |
+
- jsonl
|
| 6 |
+
- wandb
|
| 7 |
+
wandb_entity: xspadex-nanyang-technological-university-singapore
|
| 8 |
+
wandb_project: qwenvl-0130v-eep
|
| 9 |
+
is_debug: false
|
| 10 |
framework:
|
| 11 |
+
name: QwenAdapter
|
| 12 |
+
qwenvl:
|
| 13 |
+
base_vlm: /data/LumosVLA/LumosVLA/playground/Pretrained_models/hub/models--Qwen--Qwen3-VL-2B-Instruct/snapshots/89644892e4d85e24eaac8bacfd4f463576704203
|
| 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 |
+
phase: Training
|
| 25 |
+
num_actions_chunk: 30
|
| 26 |
+
datasets:
|
| 27 |
+
vla_data:
|
| 28 |
+
dataset_py: lerobot_datasets
|
| 29 |
+
data_root_dir: /data/LumosVLA/LumosVLA/data/starvla_intern
|
| 30 |
+
data_mix: train_internAdapter
|
| 31 |
+
per_device_batch_size: 32
|
| 32 |
+
num_workers: 8
|
| 33 |
+
action_type: abs_ee
|
| 34 |
+
image_size: 256
|
| 35 |
+
video_backend: pyav
|
| 36 |
trainer:
|
| 37 |
+
epochs: 100
|
| 38 |
+
max_train_steps: 125000
|
| 39 |
+
num_warmup_steps: 1000
|
| 40 |
+
save_interval: 1000
|
| 41 |
eval_interval: 100000
|
|
|
|
|
|
|
|
|
|
| 42 |
learning_rate:
|
|
|
|
| 43 |
base: 4.0e-06
|
| 44 |
qwen_vl_interface: 4.0e-06
|
| 45 |
+
action_model: 1.0e-05
|
| 46 |
lr_scheduler_type: cosine_with_min_lr
|
| 47 |
+
scheduler_specific_kwargs:
|
| 48 |
+
min_lr: 5.0e-07
|
| 49 |
+
freeze_modules: qwen_vl_interface.model.model.language_model
|
| 50 |
+
loss_scale:
|
| 51 |
+
vla: 1.0
|
| 52 |
+
vlm: 0.05
|
| 53 |
+
max_grad_norm: 1.0
|
| 54 |
+
warmup_ratio: 0.1
|
| 55 |
+
weight_decay: 0.0
|
| 56 |
+
logging_frequency: 10
|
| 57 |
+
gradient_clipping: 1.0
|
| 58 |
+
gradient_accumulation_steps: 1
|
| 59 |
optimizer:
|
| 60 |
+
name: AdamW
|
| 61 |
betas:
|
| 62 |
- 0.9
|
| 63 |
- 0.95
|
| 64 |
eps: 1.0e-08
|
| 65 |
weight_decay: 0
|
| 66 |
+
is_resume: true
|
| 67 |
+
resume_step: 56000
|
| 68 |
+
is_resume: false
|
| 69 |
+
resume_epoch: null
|
| 70 |
+
resume_step: null
|
| 71 |
+
enable_gradient_checkpointing: true
|
| 72 |
+
enable_mixed_precision_training: true
|
| 73 |
+
report_to: wandb
|
| 74 |
+
wandb_run_name: v1
|
| 75 |
+
output_dir: /data/LumosVLA/LumosVLA/results/QwenAdapter-0130v-eep
|
qwen-0130v/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,
|
|
@@ -40,7 +40,7 @@
|
|
| 40 |
3.1415927410125732,
|
| 41 |
1.5707963705062866,
|
| 42 |
3.1415927410125732,
|
| 43 |
-
0.
|
| 44 |
0.5489550232887268,
|
| 45 |
0.45939499139785767,
|
| 46 |
0.6456500291824341,
|
|
@@ -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,
|
|
@@ -154,7 +154,7 @@
|
|
| 154 |
3.1415927410125732,
|
| 155 |
1.5707963705062866,
|
| 156 |
3.1415927410125732,
|
| 157 |
-
0.
|
| 158 |
0.5489550232887268,
|
| 159 |
0.45939499139785767,
|
| 160 |
0.6456500291824341,
|
|
@@ -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 |
+
0.2536350143583198,
|
| 6 |
+
-0.07743038404671672,
|
| 7 |
+
0.1966707251573864,
|
| 8 |
+
-1.3468584815661115,
|
| 9 |
+
0.40754710399267974,
|
| 10 |
+
1.6235627918912656,
|
| 11 |
+
0.034987629603659906,
|
| 12 |
+
0.23352106752102836,
|
| 13 |
+
0.09337638377359042,
|
| 14 |
+
0.20792568958642196,
|
| 15 |
+
1.5060326024367099,
|
| 16 |
+
0.4999388032837918,
|
| 17 |
+
-1.7593652281844825,
|
| 18 |
+
0.0355335969412536
|
| 19 |
],
|
| 20 |
"std": [
|
| 21 |
+
0.09796345941892504,
|
| 22 |
+
0.10853129549959974,
|
| 23 |
+
0.0681641731581649,
|
| 24 |
+
2.571989629620975,
|
| 25 |
+
0.3175618094699909,
|
| 26 |
+
2.272367583216257,
|
| 27 |
+
0.028098021946224527,
|
| 28 |
+
0.08823442171678504,
|
| 29 |
+
0.07837648996356361,
|
| 30 |
+
0.05453627898660852,
|
| 31 |
+
2.5246083674731827,
|
| 32 |
+
0.28385594171744144,
|
| 33 |
+
2.189727010118041,
|
| 34 |
+
0.02824469944283456
|
| 35 |
],
|
| 36 |
"max": [
|
| 37 |
0.5931400060653687,
|
|
|
|
| 40 |
3.1415927410125732,
|
| 41 |
1.5707963705062866,
|
| 42 |
3.1415927410125732,
|
| 43 |
+
0.0810599997639656,
|
| 44 |
0.5489550232887268,
|
| 45 |
0.45939499139785767,
|
| 46 |
0.6456500291824341,
|
|
|
|
| 116 |
},
|
| 117 |
"state": {
|
| 118 |
"mean": [
|
| 119 |
+
0.2536350143583198,
|
| 120 |
+
-0.07743038404671672,
|
| 121 |
+
0.1966707251573864,
|
| 122 |
+
-1.3468584815661115,
|
| 123 |
+
0.40754710399267974,
|
| 124 |
+
1.6235627918912656,
|
| 125 |
+
0.034987629603659906,
|
| 126 |
+
0.23352106752102836,
|
| 127 |
+
0.09337638377359042,
|
| 128 |
+
0.20792568958642196,
|
| 129 |
+
1.5060326024367099,
|
| 130 |
+
0.4999388032837918,
|
| 131 |
+
-1.7593652281844825,
|
| 132 |
+
0.0355335969412536
|
| 133 |
],
|
| 134 |
"std": [
|
| 135 |
+
0.09796345941892504,
|
| 136 |
+
0.10853129549959974,
|
| 137 |
+
0.0681641731581649,
|
| 138 |
+
2.571989629620975,
|
| 139 |
+
0.3175618094699909,
|
| 140 |
+
2.272367583216257,
|
| 141 |
+
0.028098021946224527,
|
| 142 |
+
0.08823442171678504,
|
| 143 |
+
0.07837648996356361,
|
| 144 |
+
0.05453627898660852,
|
| 145 |
+
2.5246083674731827,
|
| 146 |
+
0.28385594171744144,
|
| 147 |
+
2.189727010118041,
|
| 148 |
+
0.02824469944283456
|
| 149 |
],
|
| 150 |
"max": [
|
| 151 |
0.5931400060653687,
|
|
|
|
| 154 |
3.1415927410125732,
|
| 155 |
1.5707963705062866,
|
| 156 |
3.1415927410125732,
|
| 157 |
+
0.0810599997639656,
|
| 158 |
0.5489550232887268,
|
| 159 |
0.45939499139785767,
|
| 160 |
0.6456500291824341,
|
|
|
|
| 212 |
0.05992000177502632
|
| 213 |
]
|
| 214 |
},
|
| 215 |
+
"num_transitions": 2167483,
|
| 216 |
+
"num_trajectories": 1802
|
| 217 |
}
|
| 218 |
}
|