Upload folder using huggingface_hub
Browse files- checkpoints/step-018000-epoch-87-loss=0.0409.pt +3 -0
- config.json +59 -0
- config.yaml +55 -0
- dataset_statistics.json +133 -0
- eval/EVAL-libero_spatial-instruct_vla-2025_09_08-20_42_54-step-018000-epoch-87-loss=0.0409.txt +0 -0
- eval/EVAL-libero_spatial-instruct_vla-2025_09_08-20_43_18-step-018000-epoch-87-loss=0.0409.txt +0 -0
- eval/EVAL-libero_spatial-instruct_vla-2025_09_08-20_43_52-step-018000-epoch-87-loss=0.0409.txt +0 -0
- run-metrics.jsonl +1 -0
- sys12_meta_query_action_only_sync_pretraining_v2_query_64_mlp_lora_libero_spatial_wrist--image_augstage1.jsonl +0 -0
checkpoints/step-018000-epoch-87-loss=0.0409.pt
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:97a101a64cdaa23b9291f25d363f45272213df23b69a008a1e0bc201cf248d90
|
3 |
+
size 11322825962
|
config.json
ADDED
@@ -0,0 +1,59 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"action_dim": 7,
|
3 |
+
"action_model_type": "DiT-B",
|
4 |
+
"data_root_dir": "/mnt/inspurfs/efm_t/robot_data/cache/LIBERO/dataset",
|
5 |
+
"debug": false,
|
6 |
+
"disable_instruction": false,
|
7 |
+
"fix_system1": false,
|
8 |
+
"future_action_window_size": 7,
|
9 |
+
"hf_token": "hf_token",
|
10 |
+
"image_aug": true,
|
11 |
+
"is_resume": false,
|
12 |
+
"load_all_data_for_training": true,
|
13 |
+
"num_of_meta_query": 64,
|
14 |
+
"past_action_window_size": 0,
|
15 |
+
"pretrained_checkpoint": null,
|
16 |
+
"repeated_diffusion_steps": 4,
|
17 |
+
"resume_epoch": null,
|
18 |
+
"resume_step": null,
|
19 |
+
"run_id": "sys12_meta_query_action_only_sync_pretraining_v2_query_64_mlp_lora_libero_spatial_wrist--image_augstage1",
|
20 |
+
"run_id_note": null,
|
21 |
+
"run_root_dir": "outputs/libero_wrist",
|
22 |
+
"save_interval": 1500,
|
23 |
+
"seed": 42,
|
24 |
+
"stage": "stage1",
|
25 |
+
"trackers": [
|
26 |
+
"jsonl",
|
27 |
+
"wandb"
|
28 |
+
],
|
29 |
+
"use_ema": false,
|
30 |
+
"use_mm": false,
|
31 |
+
"vla": {
|
32 |
+
"action_tokenizer": "extra_action_tokenizer",
|
33 |
+
"base_vlm": "/mnt/petrelfs/yangshuai1/yangshuai1/share_mllm/Eagle2-2B",
|
34 |
+
"data_mix": "libero_spatial_no_noops",
|
35 |
+
"enable_gradient_checkpointing": true,
|
36 |
+
"enable_mixed_precision_training": true,
|
37 |
+
"epochs": 100,
|
38 |
+
"expected_world_size": 8,
|
39 |
+
"freeze_llm_backbone": false,
|
40 |
+
"freeze_vision_backbone": false,
|
41 |
+
"global_batch_size": 256,
|
42 |
+
"learning_rate": 5e-05,
|
43 |
+
"lr_scheduler_type": "constant",
|
44 |
+
"max_grad_norm": 1.0,
|
45 |
+
"max_steps": null,
|
46 |
+
"per_device_batch_size": 32,
|
47 |
+
"reduce_in_full_precision": true,
|
48 |
+
"shuffle_buffer_size": 250000,
|
49 |
+
"train_strategy": "fsdp-full-shard",
|
50 |
+
"type": "prism-qwen25-dinosiglip-224px+0_5b",
|
51 |
+
"unfreeze_last_llm_layer": false,
|
52 |
+
"vla_id": "prism-qwen25-dinosiglip-224px+0_5b",
|
53 |
+
"warmup_ratio": 0.0,
|
54 |
+
"weight_decay": 0.0
|
55 |
+
},
|
56 |
+
"wandb_entity": "shuaiyang2003",
|
57 |
+
"wandb_project": "dual_sys_libero",
|
58 |
+
"with_pointing": true
|
59 |
+
}
|
config.yaml
ADDED
@@ -0,0 +1,55 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
action_dim: 7
|
2 |
+
action_model_type: DiT-B
|
3 |
+
data_root_dir: /mnt/inspurfs/efm_t/robot_data/cache/LIBERO/dataset
|
4 |
+
debug: false
|
5 |
+
disable_instruction: false
|
6 |
+
fix_system1: false
|
7 |
+
future_action_window_size: 7
|
8 |
+
hf_token: hf_token
|
9 |
+
image_aug: true
|
10 |
+
is_resume: false
|
11 |
+
load_all_data_for_training: true
|
12 |
+
num_of_meta_query: 64
|
13 |
+
past_action_window_size: 0
|
14 |
+
pretrained_checkpoint: null
|
15 |
+
repeated_diffusion_steps: 4
|
16 |
+
resume_epoch: null
|
17 |
+
resume_step: null
|
18 |
+
run_id: sys12_meta_query_action_only_sync_pretraining_v2_query_64_mlp_lora_libero_spatial_wrist--image_augstage1
|
19 |
+
run_id_note: null
|
20 |
+
run_root_dir: outputs/libero_wrist
|
21 |
+
save_interval: 1500
|
22 |
+
seed: 42
|
23 |
+
stage: stage1
|
24 |
+
trackers:
|
25 |
+
- jsonl
|
26 |
+
- wandb
|
27 |
+
use_ema: false
|
28 |
+
use_mm: false
|
29 |
+
vla:
|
30 |
+
action_tokenizer: extra_action_tokenizer
|
31 |
+
base_vlm: /mnt/petrelfs/yangshuai1/yangshuai1/share_mllm/Eagle2-2B
|
32 |
+
data_mix: libero_spatial_no_noops
|
33 |
+
enable_gradient_checkpointing: true
|
34 |
+
enable_mixed_precision_training: true
|
35 |
+
epochs: 100
|
36 |
+
expected_world_size: 8
|
37 |
+
freeze_llm_backbone: false
|
38 |
+
freeze_vision_backbone: false
|
39 |
+
global_batch_size: 256
|
40 |
+
learning_rate: 5.0e-05
|
41 |
+
lr_scheduler_type: constant
|
42 |
+
max_grad_norm: 1.0
|
43 |
+
max_steps: null
|
44 |
+
per_device_batch_size: 32
|
45 |
+
reduce_in_full_precision: true
|
46 |
+
shuffle_buffer_size: 250000
|
47 |
+
train_strategy: fsdp-full-shard
|
48 |
+
type: prism-qwen25-dinosiglip-224px+0_5b
|
49 |
+
unfreeze_last_llm_layer: false
|
50 |
+
vla_id: prism-qwen25-dinosiglip-224px+0_5b
|
51 |
+
warmup_ratio: 0.0
|
52 |
+
weight_decay: 0.0
|
53 |
+
wandb_entity: shuaiyang2003
|
54 |
+
wandb_project: dual_sys_libero
|
55 |
+
with_pointing: true
|
dataset_statistics.json
ADDED
@@ -0,0 +1,133 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"libero_spatial_no_noops": {
|
3 |
+
"action": {
|
4 |
+
"mean": [
|
5 |
+
0.15312479436397552,
|
6 |
+
0.13707277178764343,
|
7 |
+
-0.15526802837848663,
|
8 |
+
-0.005176450591534376,
|
9 |
+
-0.01120874285697937,
|
10 |
+
-0.020194264128804207,
|
11 |
+
0.4578818082809448
|
12 |
+
],
|
13 |
+
"std": [
|
14 |
+
0.41272708773612976,
|
15 |
+
0.34724321961402893,
|
16 |
+
0.50869220495224,
|
17 |
+
0.037266165018081665,
|
18 |
+
0.07244449853897095,
|
19 |
+
0.05762382969260216,
|
20 |
+
0.49827873706817627
|
21 |
+
],
|
22 |
+
"max": [
|
23 |
+
0.9375,
|
24 |
+
0.9375,
|
25 |
+
0.9375,
|
26 |
+
0.1971428543329239,
|
27 |
+
0.33642858266830444,
|
28 |
+
0.375,
|
29 |
+
1.0
|
30 |
+
],
|
31 |
+
"min": [
|
32 |
+
-0.9375,
|
33 |
+
-0.9375,
|
34 |
+
-0.9375,
|
35 |
+
-0.1875,
|
36 |
+
-0.3675000071525574,
|
37 |
+
-0.36000001430511475,
|
38 |
+
0.0
|
39 |
+
],
|
40 |
+
"q01": [
|
41 |
+
-0.7454732114076613,
|
42 |
+
-0.6616071462631226,
|
43 |
+
-0.9375,
|
44 |
+
-0.1071428582072258,
|
45 |
+
-0.20678570866584778,
|
46 |
+
-0.1842857152223587,
|
47 |
+
0.0
|
48 |
+
],
|
49 |
+
"q99": [
|
50 |
+
0.9375,
|
51 |
+
0.8758928775787354,
|
52 |
+
0.9321428537368774,
|
53 |
+
0.1039285734295845,
|
54 |
+
0.17678570747375488,
|
55 |
+
0.14571428298950195,
|
56 |
+
1.0
|
57 |
+
],
|
58 |
+
"mask": [
|
59 |
+
true,
|
60 |
+
true,
|
61 |
+
true,
|
62 |
+
true,
|
63 |
+
true,
|
64 |
+
true,
|
65 |
+
false
|
66 |
+
]
|
67 |
+
},
|
68 |
+
"proprio": {
|
69 |
+
"mean": [
|
70 |
+
-0.024462558329105377,
|
71 |
+
0.106529600918293,
|
72 |
+
1.0580483675003052,
|
73 |
+
3.0628468990325928,
|
74 |
+
-0.10464039444923401,
|
75 |
+
0.08307311683893204,
|
76 |
+
0.01995457336306572,
|
77 |
+
-0.020162804052233696
|
78 |
+
],
|
79 |
+
"std": [
|
80 |
+
0.1101478561758995,
|
81 |
+
0.13784688711166382,
|
82 |
+
0.1044282391667366,
|
83 |
+
0.10451053828001022,
|
84 |
+
0.4112098217010498,
|
85 |
+
0.2176690548658371,
|
86 |
+
0.017260896041989326,
|
87 |
+
0.0171116404235363
|
88 |
+
],
|
89 |
+
"max": [
|
90 |
+
0.1759040206670761,
|
91 |
+
0.3904820382595062,
|
92 |
+
1.3290715217590332,
|
93 |
+
3.4566118717193604,
|
94 |
+
1.2268599271774292,
|
95 |
+
1.0429412126541138,
|
96 |
+
0.041053611785173416,
|
97 |
+
0.000775813648942858
|
98 |
+
],
|
99 |
+
"min": [
|
100 |
+
-0.3095473051071167,
|
101 |
+
-0.29250794649124146,
|
102 |
+
0.9095591306686401,
|
103 |
+
2.497488260269165,
|
104 |
+
-1.8006486892700195,
|
105 |
+
-0.7207611203193665,
|
106 |
+
-0.0004703797458205372,
|
107 |
+
-0.041536275297403336
|
108 |
+
],
|
109 |
+
"q01": [
|
110 |
+
-0.2727657300233841,
|
111 |
+
-0.23721413239836692,
|
112 |
+
0.9160063165426254,
|
113 |
+
2.77949666261673,
|
114 |
+
-1.3187511622905732,
|
115 |
+
-0.41989982962608335,
|
116 |
+
0.001503719249740243,
|
117 |
+
-0.03989770736545324
|
118 |
+
],
|
119 |
+
"q99": [
|
120 |
+
0.13529365032911292,
|
121 |
+
0.3629165390133857,
|
122 |
+
1.2862326657772063,
|
123 |
+
3.2829698753356933,
|
124 |
+
0.9332760351896285,
|
125 |
+
0.6325724506378171,
|
126 |
+
0.039933966137468815,
|
127 |
+
-0.001671919699292631
|
128 |
+
]
|
129 |
+
},
|
130 |
+
"num_transitions": 52970,
|
131 |
+
"num_trajectories": 432
|
132 |
+
}
|
133 |
+
}
|
eval/EVAL-libero_spatial-instruct_vla-2025_09_08-20_42_54-step-018000-epoch-87-loss=0.0409.txt
ADDED
The diff for this file is too large to render.
See raw diff
|
|
eval/EVAL-libero_spatial-instruct_vla-2025_09_08-20_43_18-step-018000-epoch-87-loss=0.0409.txt
ADDED
The diff for this file is too large to render.
See raw diff
|
|
eval/EVAL-libero_spatial-instruct_vla-2025_09_08-20_43_52-step-018000-epoch-87-loss=0.0409.txt
ADDED
The diff for this file is too large to render.
See raw diff
|
|
run-metrics.jsonl
ADDED
@@ -0,0 +1 @@
|
|
|
|
|
1 |
+
{"hparams": {"action_dim": 7, "action_model_type": "DiT-B", "data_root_dir": "/mnt/inspurfs/efm_t/robot_data/cache/LIBERO/dataset", "debug": false, "disable_instruction": false, "fix_system1": false, "future_action_window_size": 7, "hf_token": "hf_token", "image_aug": true, "is_resume": false, "load_all_data_for_training": true, "num_of_meta_query": 64, "past_action_window_size": 0, "pretrained_checkpoint": null, "repeated_diffusion_steps": 4, "resume_epoch": null, "resume_step": null, "run_id": "sys12_meta_query_action_only_sync_pretraining_v2_query_64_mlp_lora_libero_spatial_wrist--image_augstage1", "run_id_note": null, "run_root_dir": "outputs/libero_wrist", "save_interval": 1500, "seed": 42, "stage": "stage1", "trackers": ["jsonl", "wandb"], "use_ema": false, "use_mm": false, "vla": {"action_tokenizer": "extra_action_tokenizer", "base_vlm": "/mnt/petrelfs/yangshuai1/yangshuai1/share_mllm/Eagle2-2B", "data_mix": "libero_spatial_no_noops", "enable_gradient_checkpointing": true, "enable_mixed_precision_training": true, "epochs": 100, "expected_world_size": 8, "freeze_llm_backbone": false, "freeze_vision_backbone": false, "global_batch_size": 256, "learning_rate": 5e-05, "lr_scheduler_type": "constant", "max_grad_norm": 1.0, "max_steps": null, "per_device_batch_size": 32, "reduce_in_full_precision": true, "shuffle_buffer_size": 250000, "train_strategy": "fsdp-full-shard", "type": "prism-qwen25-dinosiglip-224px+0_5b", "unfreeze_last_llm_layer": false, "vla_id": "prism-qwen25-dinosiglip-224px+0_5b", "warmup_ratio": 0.0, "weight_decay": 0.0}, "wandb_entity": "shuaiyang2003", "wandb_project": "dual_sys_libero", "with_pointing": true}, "run_id": "sys12_meta_query_action_only_sync_pretraining_v2_query_64_mlp_lora_libero_spatial_wrist--image_augstage1"}
|
sys12_meta_query_action_only_sync_pretraining_v2_query_64_mlp_lora_libero_spatial_wrist--image_augstage1.jsonl
ADDED
The diff for this file is too large to render.
See raw diff
|
|