Timsty commited on
Commit
28dd5af
·
verified ·
1 Parent(s): f8880d9

Delete config.json

Browse files
Files changed (1) hide show
  1. config.json +0 -121
config.json DELETED
@@ -1,121 +0,0 @@
1
- {
2
- "run_id": "0418_QwenLatent_13tasks_actionstate_30k",
3
- "run_root_dir": "./runs",
4
- "seed": 42,
5
- "trackers": [
6
- "jsonl",
7
- "wandb"
8
- ],
9
- "wandb_entity": "timsty",
10
- "wandb_project": "vla_jepa",
11
- "is_debug": false,
12
- "framework": {
13
- "name": "QwenLatent",
14
- "qwenvl": {
15
- "base_vlm": "/mnt/data/fangyu/model/Qwen/Qwen3-VL-2B-Instruct",
16
- "attn_implementation": "flash_attention_2",
17
- "vl_hidden_dim": 2048,
18
- "num_data_tokens": 32
19
- },
20
- "action_model": {
21
- "ckpt_path": "/mnt/data/fangyu/code/reward_new/runs/0418_Action_13tasks_actionstate_fixchunk15/final_model/pytorch_model.pt",
22
- "action_size": 37,
23
- "state_size": 74,
24
- "use_state": "${datasets.vla_data.state_use_action_chunk}",
25
- "hidden_size": 1024,
26
- "intermediate_size": 3072,
27
- "dataset_vocab_size": 256,
28
- "num_data_tokens": 32,
29
- "num_t_samples": 4,
30
- "min_action_len": 5,
31
- "num_encoder_layers": 28,
32
- "num_decoder_layers": 28,
33
- "num_attention_heads": 16,
34
- "num_key_value_heads": 8,
35
- "head_dim": 128,
36
- "max_position_embeddings": 2048,
37
- "max_action_chunk_size": 50,
38
- "rms_norm_eps": 1e-06,
39
- "attention_dropout": 0.0,
40
- "use_vae_reparameterization": false,
41
- "use_ema": false,
42
- "chunk_size": "${datasets.vla_data.chunk_size}",
43
- "loss_mode": "full",
44
- "qwen3_pretrained_name_or_path": "/mnt/data/fangyu/model/Qwen/Qwen3-0.6B"
45
- }
46
- },
47
- "datasets": {
48
- "vla_data": {
49
- "dataset_py": "lerobot_datasets",
50
- "data_root_dir": "/mnt/data/fangyu/dataset/IPEC-COMMUNITY",
51
- "data_mix": "cross_embodiedment_13tasks",
52
- "CoT_prompt": "Task: {instruction}. What are the next 15 actions to take?",
53
- "default_image_resolution": [
54
- 3,
55
- 224,
56
- 224
57
- ],
58
- "per_device_batch_size": 32,
59
- "load_all_data_for_training": true,
60
- "obs": [
61
- "image_0"
62
- ],
63
- "image_size": [
64
- 224,
65
- 224
66
- ],
67
- "video_backend": "torchcodec",
68
- "load_video": true,
69
- "chunk_size": 15,
70
- "state_use_action_chunk": true,
71
- "num_history_steps": 0,
72
- "include_state": "${datasets.vla_data.state_use_action_chunk}"
73
- }
74
- },
75
- "trainer": {
76
- "epochs": 100,
77
- "max_train_steps": 50000,
78
- "num_warmup_steps": 5000,
79
- "num_stable_steps": 0,
80
- "mode": "decay_aux_loss",
81
- "loss_weights_decay_steps": 5000,
82
- "save_interval": 5000,
83
- "eval_interval": 50,
84
- "max_checkpoints_to_keep": 20,
85
- "learning_rate": {
86
- "base": 2.5e-05,
87
- "qwen_vl_interface": 2.5e-05,
88
- "action_model": 2.5e-05
89
- },
90
- "lr_scheduler_type": "warmup_stable_cosine",
91
- "scheduler_specific_kwargs": {
92
- "min_lr_ratio": 0.001
93
- },
94
- "freeze_modules": "",
95
- "loss_scale": {
96
- "align_loss": 1.0,
97
- "recon_loss": 1.0,
98
- "predict_loss": 1.0
99
- },
100
- "warmup_ratio": 0.1,
101
- "weight_decay": 0.0,
102
- "logging_frequency": 10,
103
- "gradient_clipping": 5.0,
104
- "gradient_accumulation_steps": 1,
105
- "optimizer": {
106
- "name": "AdamW",
107
- "betas": [
108
- 0.9,
109
- 0.95
110
- ],
111
- "eps": 1e-08,
112
- "weight_decay": 1e-08
113
- },
114
- "is_resume": false,
115
- "resume_epoch": null,
116
- "resume_step": null,
117
- "enable_gradient_checkpointing": true,
118
- "enable_mixed_precision_training": true
119
- },
120
- "output_dir": "./runs/0418_QwenLatent_13tasks_actionstate_30k"
121
- }