Hugging Face's logo

fabono
/
xvla_so101_black_get

Robotics
LeRobot
Safetensors
xvla
Model card Files
xet
 Community
fabono commited on
Commit
e7c077e
·
verified ·
1 Parent(s): d88f881

Upload policy weights, train config and readme

Browse files
Files changed (4) hide show
  1. README.md +172 -0
  2. config.json +223 -0
  3. model.safetensors +3 -0
  4. train_config.json +381 -0
README.md ADDED
@@ -0,0 +1,172 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ base_model: lerobot/xvla-base
3
+ datasets: fabono/get_black_cube_20260807_190819
4
+ library_name: lerobot
5
+ license: apache-2.0
6
+ model_name: xvla
7
+ pipeline_tag: robotics
8
+ tags:
9
+ - robotics
10
+ - lerobot
11
+ - xvla
12
+ ---
13
+
14
+ # Model Card for xvla
15
+
16
+ <!-- Provide a quick summary of what the model is/does. -->
17
+
18
+
19
+ [X-VLA](https://huggingface.co/papers/2510.10274) is a soft-prompted, flow-matching Vision-Language-Action framework that treats each robot or hardware setup as a "task" encoded with a small set of learnable Soft Prompt embeddings, letting a single model reconcile diverse robot morphologies, sensors, and action spaces.
20
+
21
+
22
+
23
+ <p align="center">
24
+ <img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/lerobot/xvla-architecture.png" alt="xvla architecture" width="85%"/>
25
+ </p>
26
+
27
+
28
+ <!-- A short demo is worth more than any description! Record a GIF/video of the policy
29
+ running on your robot, upload it to this repo, and embed it here:
30
+ <p align="center">
31
+ <img src="https://huggingface.co/<hf_user>/<policy_repo_id>/resolve/main/demo.gif" width="60%"/>
32
+ </p>
33
+ -->
34
+
35
+ This policy has been trained and pushed to the Hub using [LeRobot](https://github.com/huggingface/lerobot).
36
+
37
+ Learn how to train and run it in the [LeRobot xvla guide](https://huggingface.co/docs/lerobot/main/en/xvla), or browse the [full documentation](https://huggingface.co/docs/lerobot/index).
38
+
39
+
40
+ ---
41
+
42
+ ## Model Details
43
+
44
+ - **License:** apache-2.0
45
+ - **Fine-tuned from:** [lerobot/xvla-base](https://huggingface.co/lerobot/xvla-base)
46
+ - **Robot type:** `so_follower`
47
+ - **Cameras:** `wrist`, `top`
48
+
49
+
50
+ ## Inputs & Outputs
51
+
52
+ The policy consumes these observation features and produces these action features.
53
+
54
+ **Inputs**
55
+
56
+ | Feature | Type | Shape |
57
+ | --- | --- | --- |
58
+ | `observation.images.image` | VISUAL | `(3, 256, 256)` |
59
+ | `observation.images.image2` | VISUAL | `(3, 256, 256)` |
60
+ | `observation.state` | STATE | `(8,)` |
61
+ | `observation.images.image3` | VISUAL | `(3, 224, 224)` |
62
+
63
+ **Outputs**
64
+
65
+ | Feature | Type | Shape |
66
+ | --- | --- | --- |
67
+ | `action` | ACTION | `(6,)` |
68
+
69
+
70
+ ## Training Dataset
71
+
72
+ - **Repository:** [fabono/get_black_cube_20260807_190819](https://huggingface.co/datasets/fabono/get_black_cube_20260807_190819)
73
+ - **Episodes:** 30
74
+ - **Frames:** 26755
75
+ - **Frame rate:** 30 FPS
76
+ - **Task(s):** "Take the black cube out of the blue drawer with the yellow handle and put it on the ground."
77
+
78
+ <a class="flex" href="https://huggingface.co/spaces/lerobot/visualize_dataset?path=fabono/get_black_cube_20260807_190819">
79
+ <img class="block dark:hidden" src="https://huggingface.co/datasets/huggingface/badges/resolve/main/visualize-this-dataset-xl.svg"/>
80
+ <img class="hidden dark:block" src="https://huggingface.co/datasets/huggingface/badges/resolve/main/visualize-this-dataset-xl-dark.svg"/>
81
+ </a>
82
+
83
+
84
+ ## Training Configuration
85
+
86
+ | Setting | Value |
87
+ | --- | --- |
88
+ | Training steps | 10000 |
89
+ | Batch size | 16 |
90
+ | Optimizer | xvla-adamw |
91
+ | Learning rate | 0.0001 |
92
+ | Seed | 1000 |
93
+ | LeRobot version | 0.6.1 |
94
+
95
+ ---
96
+
97
+ ## How to Get Started with the Model
98
+
99
+ New to LeRobot? These guides cover the full workflow:
100
+
101
+ - **[Install LeRobot](https://huggingface.co/docs/lerobot/main/en/installation)** — set up the `lerobot` package.
102
+ - **[Hardware setup](https://huggingface.co/docs/lerobot/main/en/hardware_guide)** — assemble, wire, and calibrate your robot and cameras.
103
+ - **[Record data & train a policy](https://huggingface.co/docs/lerobot/en/il_robots)** — the end-to-end imitation-learning walkthrough.
104
+ - **[CLI cheat-sheet](https://huggingface.co/docs/lerobot/main/en/cheat-sheet)** — quick reference for the `lerobot-*` commands.
105
+
106
+ The short version to run and train this policy:
107
+
108
+ ### Run the policy on your robot
109
+
110
+ ```bash
111
+ lerobot-rollout \
112
+ --strategy.type=base \
113
+ --robot.type=so_follower \
114
+ --robot.port=<your_robot_port> \
115
+ --robot.cameras="{ <camera_1>: {type: opencv, index_or_path: <index_or_path>, width: 640, height: 480, fps: 30}, <camera_2>: {type: opencv, index_or_path: <index_or_path>, width: 640, height: 480, fps: 30}}" \
116
+ --policy.path=fabono/xvla_so101_black_get \
117
+ --task="Take the black cube out of the blue drawer with the yellow handle and put it on the ground." \
118
+ --duration=60
119
+ ```
120
+
121
+ Replace the remaining `<...>` placeholders with your own values: `--robot.port` and the camera names/indices are specific to your machine, and the camera names must match the observation keys this policy was trained on.
122
+
123
+ When `--strategy.type=base` is used the script doesn't record the episodes. Skipping duration will make the policy run indefinitely. For more information look at [rollout documentation](https://huggingface.co/docs/lerobot/main/en/inference).
124
+
125
+ ### Train your own policy
126
+
127
+ This policy type is usually fine-tuned from the pretrained base model [lerobot/xvla-base](https://huggingface.co/lerobot/xvla-base):
128
+
129
+ ```bash
130
+ lerobot-train \
131
+ --dataset.repo_id=${HF_USER}/<dataset> \
132
+ --policy.path=lerobot/xvla-base \
133
+ --output_dir=outputs/train/<policy_repo_id> \
134
+ --job_name=lerobot_training \
135
+ --policy.device=cuda \
136
+ --policy.repo_id=${HF_USER}/<policy_repo_id> \
137
+ --wandb.enable=true
138
+ ```
139
+
140
+ _Writes checkpoints to `outputs/train/<policy_repo_id>/checkpoints/`._
141
+
142
+ ---
143
+
144
+ ## Evaluation
145
+
146
+ <!-- Report real-robot results here: run the policy several times per task and count the
147
+ successes. Delete the "No evaluation results" line and fill in this table instead:
148
+
149
+ | Task | Trials | Successes | Success rate |
150
+ | ---- | ------ | --------- | ------------ |
151
+ | pick the lego brick | 10 | 8 | 80% |
152
+
153
+ Also worth noting: anything that affects difficulty (new object positions, lighting,
154
+ distractors, a different robot of the same type, ...).
155
+ -->
156
+
157
+ _No evaluation results have been provided for this policy yet._
158
+
159
+ ---
160
+
161
+ ## Citation
162
+
163
+ If you use this policy, please cite the method linked in the description above, along with LeRobot:
164
+
165
+ ```bibtex
166
+ @misc{cadene2024lerobot,
167
+ author = {Cadene, Remi and Alibert, Simon and Soare, Alexander and Gallouedec, Quentin and Zouitine, Adil and Palma, Steven and Kooijmans, Pepijn and Aractingi, Michel and Shukor, Mustafa and Aubakirova, Dana and Russi, Martino and Capuano, Francesco and Pascal, Caroline and Choghari, Jade and Moss, Jess and Wolf, Thomas},
168
+ title = {LeRobot: State-of-the-art Machine Learning for Real-World Robotics in Pytorch},
169
+ howpublished = "\url{https://github.com/huggingface/lerobot}",
170
+ year = {2024}
171
+ }
172
+ ```
config.json ADDED
@@ -0,0 +1,223 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "type": "xvla",
3
+ "n_obs_steps": 1,
4
+ "input_features": {
5
+ "observation.images.image": {
6
+ "type": "VISUAL",
7
+ "shape": [
8
+ 3,
9
+ 256,
10
+ 256
11
+ ]
12
+ },
13
+ "observation.images.image2": {
14
+ "type": "VISUAL",
15
+ "shape": [
16
+ 3,
17
+ 256,
18
+ 256
19
+ ]
20
+ },
21
+ "observation.state": {
22
+ "type": "STATE",
23
+ "shape": [
24
+ 8
25
+ ]
26
+ },
27
+ "observation.images.image3": {
28
+ "type": "VISUAL",
29
+ "shape": [
30
+ 3,
31
+ 224,
32
+ 224
33
+ ]
34
+ }
35
+ },
36
+ "output_features": {
37
+ "action": {
38
+ "type": "ACTION",
39
+ "shape": [
40
+ 6
41
+ ]
42
+ }
43
+ },
44
+ "device": "cuda",
45
+ "use_amp": false,
46
+ "use_peft": false,
47
+ "push_to_hub": true,
48
+ "repo_id": "fabono/xvla_so101_black_get",
49
+ "private": null,
50
+ "tags": null,
51
+ "license": null,
52
+ "pretrained_path": "lerobot/xvla-base",
53
+ "pretrained_revision": null,
54
+ "chunk_size": 30,
55
+ "n_action_steps": 30,
56
+ "dtype": "bfloat16",
57
+ "normalization_mapping": {
58
+ "STATE": "IDENTITY",
59
+ "ACTION": "MEAN_STD",
60
+ "VISUAL": "IDENTITY"
61
+ },
62
+ "florence_config": {
63
+ "model_type": "florence2",
64
+ "bos_token_id": 0,
65
+ "eos_token_id": 2,
66
+ "ignore_index": -100,
67
+ "pad_token_id": 1,
68
+ "projection_dim": 1024,
69
+ "text_config": {
70
+ "vocab_size": 51289,
71
+ "activation_dropout": 0.1,
72
+ "activation_function": "gelu",
73
+ "add_bias_logits": false,
74
+ "add_final_layer_norm": false,
75
+ "attention_dropout": 0.1,
76
+ "bos_token_id": 0,
77
+ "classif_dropout": 0.1,
78
+ "classifier_dropout": 0.0,
79
+ "d_model": 1024,
80
+ "decoder_attention_heads": 16,
81
+ "decoder_ffn_dim": 4096,
82
+ "decoder_layerdrop": 0.0,
83
+ "decoder_layers": 12,
84
+ "decoder_start_token_id": 2,
85
+ "dropout": 0.1,
86
+ "early_stopping": true,
87
+ "encoder_attention_heads": 16,
88
+ "encoder_ffn_dim": 4096,
89
+ "encoder_layerdrop": 0.0,
90
+ "encoder_layers": 12,
91
+ "eos_token_id": 2,
92
+ "forced_eos_token_id": 2,
93
+ "forced_bos_token_id": 0,
94
+ "gradient_checkpointing": false,
95
+ "init_std": 0.02,
96
+ "is_encoder_decoder": true,
97
+ "label2id": {
98
+ "LABEL_0": 0,
99
+ "LABEL_1": 1,
100
+ "LABEL_2": 2
101
+ },
102
+ "max_position_embeddings": 4096,
103
+ "no_repeat_ngram_size": 3,
104
+ "normalize_before": false,
105
+ "num_hidden_layers": 12,
106
+ "pad_token_id": 1,
107
+ "scale_embedding": false,
108
+ "num_beams": 3
109
+ },
110
+ "vision_config": {
111
+ "model_type": "davit",
112
+ "drop_path_rate": 0.1,
113
+ "patch_size": [
114
+ 7,
115
+ 3,
116
+ 3,
117
+ 3
118
+ ],
119
+ "patch_stride": [
120
+ 4,
121
+ 2,
122
+ 2,
123
+ 2
124
+ ],
125
+ "patch_padding": [
126
+ 3,
127
+ 1,
128
+ 1,
129
+ 1
130
+ ],
131
+ "patch_prenorm": [
132
+ false,
133
+ true,
134
+ true,
135
+ true
136
+ ],
137
+ "enable_checkpoint": false,
138
+ "dim_embed": [
139
+ 256,
140
+ 512,
141
+ 1024,
142
+ 2048
143
+ ],
144
+ "num_heads": [
145
+ 8,
146
+ 16,
147
+ 32,
148
+ 64
149
+ ],
150
+ "num_groups": [
151
+ 8,
152
+ 16,
153
+ 32,
154
+ 64
155
+ ],
156
+ "depths": [
157
+ 1,
158
+ 1,
159
+ 9,
160
+ 1
161
+ ],
162
+ "window_size": 12,
163
+ "projection_dim": 1024,
164
+ "visual_temporal_embedding": {
165
+ "type": "COSINE",
166
+ "max_temporal_embeddings": 100
167
+ },
168
+ "image_pos_embed": {
169
+ "type": "learned_abs_2d",
170
+ "max_pos_embeddings": 50
171
+ },
172
+ "image_feature_source": [
173
+ "spatial_avg_pool",
174
+ "temporal_avg_pool"
175
+ ]
176
+ },
177
+ "vocab_size": 51289,
178
+ "torch_dtype": "float32",
179
+ "is_encoder_decoder": true
180
+ },
181
+ "tokenizer_name": "facebook/bart-large",
182
+ "tokenizer_max_length": 1024,
183
+ "tokenizer_padding_side": "right",
184
+ "pad_language_to": "max_length",
185
+ "hidden_size": 1024,
186
+ "depth": 24,
187
+ "num_heads": 16,
188
+ "mlp_ratio": 4.0,
189
+ "num_domains": 30,
190
+ "len_soft_prompts": 32,
191
+ "dim_time": 32,
192
+ "max_len_seq": 512,
193
+ "use_hetero_proj": false,
194
+ "action_mode": "auto",
195
+ "num_denoising_steps": 10,
196
+ "use_proprio": true,
197
+ "max_state_dim": 20,
198
+ "max_action_dim": 20,
199
+ "domain_feature_key": null,
200
+ "resize_imgs_with_padding": [
201
+ 224,
202
+ 224
203
+ ],
204
+ "num_image_views": 3,
205
+ "empty_cameras": 0,
206
+ "freeze_vision_encoder": false,
207
+ "freeze_language_encoder": false,
208
+ "train_policy_transformer": true,
209
+ "train_soft_prompts": true,
210
+ "optimizer_lr": 0.0001,
211
+ "optimizer_betas": [
212
+ 0.9,
213
+ 0.95
214
+ ],
215
+ "optimizer_eps": 1e-08,
216
+ "optimizer_weight_decay": 0.0001,
217
+ "optimizer_grad_clip_norm": 10.0,
218
+ "optimizer_soft_prompt_lr_scale": 1.0,
219
+ "optimizer_soft_prompt_warmup_lr_scale": null,
220
+ "scheduler_warmup_steps": 1000,
221
+ "scheduler_decay_steps": 30000,
222
+ "scheduler_decay_lr": 2.5e-06
223
+ }
model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:a89fd441a8dd85e94fa56b38a8566bcf44b289fd06cd9a6e809589581766c5f6
3
+ size 1759596986
train_config.json ADDED
@@ -0,0 +1,381 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "dataset": {
3
+ "repo_id": "fabono/get_black_cube_20260807_190819",
4
+ "root": null,
5
+ "episodes": null,
6
+ "image_transforms": {
7
+ "enable": false,
8
+ "max_num_transforms": 3,
9
+ "random_order": false,
10
+ "tfs": {
11
+ "brightness": {
12
+ "weight": 1.0,
13
+ "type": "ColorJitter",
14
+ "kwargs": {
15
+ "brightness": [
16
+ 0.8,
17
+ 1.2
18
+ ]
19
+ }
20
+ },
21
+ "contrast": {
22
+ "weight": 1.0,
23
+ "type": "ColorJitter",
24
+ "kwargs": {
25
+ "contrast": [
26
+ 0.8,
27
+ 1.2
28
+ ]
29
+ }
30
+ },
31
+ "saturation": {
32
+ "weight": 1.0,
33
+ "type": "ColorJitter",
34
+ "kwargs": {
35
+ "saturation": [
36
+ 0.5,
37
+ 1.5
38
+ ]
39
+ }
40
+ },
41
+ "hue": {
42
+ "weight": 1.0,
43
+ "type": "ColorJitter",
44
+ "kwargs": {
45
+ "hue": [
46
+ -0.05,
47
+ 0.05
48
+ ]
49
+ }
50
+ },
51
+ "sharpness": {
52
+ "weight": 1.0,
53
+ "type": "SharpnessJitter",
54
+ "kwargs": {
55
+ "sharpness": [
56
+ 0.5,
57
+ 1.5
58
+ ]
59
+ }
60
+ },
61
+ "affine": {
62
+ "weight": 1.0,
63
+ "type": "RandomAffine",
64
+ "kwargs": {
65
+ "degrees": [
66
+ -5.0,
67
+ 5.0
68
+ ],
69
+ "translate": [
70
+ 0.05,
71
+ 0.05
72
+ ]
73
+ }
74
+ }
75
+ }
76
+ },
77
+ "revision": null,
78
+ "use_imagenet_stats": true,
79
+ "video_backend": "torchcodec",
80
+ "return_uint8": false,
81
+ "depth_output_unit": "mm",
82
+ "streaming": false,
83
+ "eval_split": 0.0
84
+ },
85
+ "env": null,
86
+ "policy": {
87
+ "type": "xvla",
88
+ "n_obs_steps": 1,
89
+ "input_features": {
90
+ "observation.images.image": {
91
+ "type": "VISUAL",
92
+ "shape": [
93
+ 3,
94
+ 256,
95
+ 256
96
+ ]
97
+ },
98
+ "observation.images.image2": {
99
+ "type": "VISUAL",
100
+ "shape": [
101
+ 3,
102
+ 256,
103
+ 256
104
+ ]
105
+ },
106
+ "observation.state": {
107
+ "type": "STATE",
108
+ "shape": [
109
+ 8
110
+ ]
111
+ },
112
+ "observation.images.image3": {
113
+ "type": "VISUAL",
114
+ "shape": [
115
+ 3,
116
+ 224,
117
+ 224
118
+ ]
119
+ }
120
+ },
121
+ "output_features": {
122
+ "action": {
123
+ "type": "ACTION",
124
+ "shape": [
125
+ 6
126
+ ]
127
+ }
128
+ },
129
+ "device": "cuda",
130
+ "use_amp": false,
131
+ "use_peft": false,
132
+ "push_to_hub": true,
133
+ "repo_id": "fabono/xvla_so101_black_get",
134
+ "private": null,
135
+ "tags": null,
136
+ "license": null,
137
+ "pretrained_path": "lerobot/xvla-base",
138
+ "pretrained_revision": null,
139
+ "chunk_size": 30,
140
+ "n_action_steps": 30,
141
+ "dtype": "bfloat16",
142
+ "normalization_mapping": {
143
+ "STATE": "IDENTITY",
144
+ "ACTION": "MEAN_STD",
145
+ "VISUAL": "IDENTITY"
146
+ },
147
+ "florence_config": {
148
+ "model_type": "florence2",
149
+ "bos_token_id": 0,
150
+ "eos_token_id": 2,
151
+ "ignore_index": -100,
152
+ "pad_token_id": 1,
153
+ "projection_dim": 1024,
154
+ "text_config": {
155
+ "vocab_size": 51289,
156
+ "activation_dropout": 0.1,
157
+ "activation_function": "gelu",
158
+ "add_bias_logits": false,
159
+ "add_final_layer_norm": false,
160
+ "attention_dropout": 0.1,
161
+ "bos_token_id": 0,
162
+ "classif_dropout": 0.1,
163
+ "classifier_dropout": 0.0,
164
+ "d_model": 1024,
165
+ "decoder_attention_heads": 16,
166
+ "decoder_ffn_dim": 4096,
167
+ "decoder_layerdrop": 0.0,
168
+ "decoder_layers": 12,
169
+ "decoder_start_token_id": 2,
170
+ "dropout": 0.1,
171
+ "early_stopping": true,
172
+ "encoder_attention_heads": 16,
173
+ "encoder_ffn_dim": 4096,
174
+ "encoder_layerdrop": 0.0,
175
+ "encoder_layers": 12,
176
+ "eos_token_id": 2,
177
+ "forced_eos_token_id": 2,
178
+ "forced_bos_token_id": 0,
179
+ "gradient_checkpointing": false,
180
+ "init_std": 0.02,
181
+ "is_encoder_decoder": true,
182
+ "label2id": {
183
+ "LABEL_0": 0,
184
+ "LABEL_1": 1,
185
+ "LABEL_2": 2
186
+ },
187
+ "max_position_embeddings": 4096,
188
+ "no_repeat_ngram_size": 3,
189
+ "normalize_before": false,
190
+ "num_hidden_layers": 12,
191
+ "pad_token_id": 1,
192
+ "scale_embedding": false,
193
+ "num_beams": 3
194
+ },
195
+ "vision_config": {
196
+ "model_type": "davit",
197
+ "drop_path_rate": 0.1,
198
+ "patch_size": [
199
+ 7,
200
+ 3,
201
+ 3,
202
+ 3
203
+ ],
204
+ "patch_stride": [
205
+ 4,
206
+ 2,
207
+ 2,
208
+ 2
209
+ ],
210
+ "patch_padding": [
211
+ 3,
212
+ 1,
213
+ 1,
214
+ 1
215
+ ],
216
+ "patch_prenorm": [
217
+ false,
218
+ true,
219
+ true,
220
+ true
221
+ ],
222
+ "enable_checkpoint": false,
223
+ "dim_embed": [
224
+ 256,
225
+ 512,
226
+ 1024,
227
+ 2048
228
+ ],
229
+ "num_heads": [
230
+ 8,
231
+ 16,
232
+ 32,
233
+ 64
234
+ ],
235
+ "num_groups": [
236
+ 8,
237
+ 16,
238
+ 32,
239
+ 64
240
+ ],
241
+ "depths": [
242
+ 1,
243
+ 1,
244
+ 9,
245
+ 1
246
+ ],
247
+ "window_size": 12,
248
+ "projection_dim": 1024,
249
+ "visual_temporal_embedding": {
250
+ "type": "COSINE",
251
+ "max_temporal_embeddings": 100
252
+ },
253
+ "image_pos_embed": {
254
+ "type": "learned_abs_2d",
255
+ "max_pos_embeddings": 50
256
+ },
257
+ "image_feature_source": [
258
+ "spatial_avg_pool",
259
+ "temporal_avg_pool"
260
+ ]
261
+ },
262
+ "vocab_size": 51289,
263
+ "torch_dtype": "float32",
264
+ "is_encoder_decoder": true
265
+ },
266
+ "tokenizer_name": "facebook/bart-large",
267
+ "tokenizer_max_length": 1024,
268
+ "tokenizer_padding_side": "right",
269
+ "pad_language_to": "max_length",
270
+ "hidden_size": 1024,
271
+ "depth": 24,
272
+ "num_heads": 16,
273
+ "mlp_ratio": 4.0,
274
+ "num_domains": 30,
275
+ "len_soft_prompts": 32,
276
+ "dim_time": 32,
277
+ "max_len_seq": 512,
278
+ "use_hetero_proj": false,
279
+ "action_mode": "auto",
280
+ "num_denoising_steps": 10,
281
+ "use_proprio": true,
282
+ "max_state_dim": 20,
283
+ "max_action_dim": 20,
284
+ "domain_feature_key": null,
285
+ "resize_imgs_with_padding": [
286
+ 224,
287
+ 224
288
+ ],
289
+ "num_image_views": 3,
290
+ "empty_cameras": 0,
291
+ "freeze_vision_encoder": false,
292
+ "freeze_language_encoder": false,
293
+ "train_policy_transformer": true,
294
+ "train_soft_prompts": true,
295
+ "optimizer_lr": 0.0001,
296
+ "optimizer_betas": [
297
+ 0.9,
298
+ 0.95
299
+ ],
300
+ "optimizer_eps": 1e-08,
301
+ "optimizer_weight_decay": 0.0001,
302
+ "optimizer_grad_clip_norm": 10.0,
303
+ "optimizer_soft_prompt_lr_scale": 1.0,
304
+ "optimizer_soft_prompt_warmup_lr_scale": null,
305
+ "scheduler_warmup_steps": 1000,
306
+ "scheduler_decay_steps": 30000,
307
+ "scheduler_decay_lr": 2.5e-06
308
+ },
309
+ "reward_model": null,
310
+ "output_dir": "outputs/train/xvla_so101_black_get",
311
+ "job_name": "xvla",
312
+ "resume": false,
313
+ "seed": 1000,
314
+ "cudnn_deterministic": false,
315
+ "num_workers": 4,
316
+ "batch_size": 16,
317
+ "prefetch_factor": 4,
318
+ "persistent_workers": true,
319
+ "steps": 10000,
320
+ "env_eval_freq": 20000,
321
+ "log_freq": 200,
322
+ "eval_steps": 0,
323
+ "max_eval_samples": 0,
324
+ "tolerance_s": 0.0001,
325
+ "save_checkpoint": true,
326
+ "save_freq": 4000,
327
+ "use_policy_training_preset": true,
328
+ "optimizer": {
329
+ "type": "xvla-adamw",
330
+ "lr": 0.0001,
331
+ "weight_decay": 0.0001,
332
+ "grad_clip_norm": 10.0,
333
+ "betas": [
334
+ 0.9,
335
+ 0.95
336
+ ],
337
+ "eps": 1e-08,
338
+ "soft_prompt_lr_scale": 1.0,
339
+ "soft_prompt_warmup_lr_scale": null
340
+ },
341
+ "scheduler": {
342
+ "type": "cosine_decay_with_warmup",
343
+ "num_warmup_steps": 1000,
344
+ "num_decay_steps": 30000,
345
+ "peak_lr": 0.0001,
346
+ "decay_lr": 2.5e-06
347
+ },
348
+ "eval": {
349
+ "n_episodes": 50,
350
+ "batch_size": 9,
351
+ "use_async_envs": true,
352
+ "recording": false,
353
+ "recording_repo_id": null,
354
+ "recording_private": false
355
+ },
356
+ "wandb": {
357
+ "enable": false,
358
+ "disable_artifact": false,
359
+ "project": "lerobot",
360
+ "entity": null,
361
+ "notes": null,
362
+ "run_id": null,
363
+ "mode": null,
364
+ "add_tags": true
365
+ },
366
+ "peft": null,
367
+ "job": {
368
+ "target": null,
369
+ "image": "huggingface/lerobot-gpu:latest",
370
+ "timeout": "2d",
371
+ "detach": false,
372
+ "tags": []
373
+ },
374
+ "save_checkpoint_to_hub": false,
375
+ "sample_weighting": null,
376
+ "rename_map": {
377
+ "observation.images.top": "observation.images.image",
378
+ "observation.images.wrist": "observation.images.image2"
379
+ },
380
+ "checkpoint_path": null
381
+ }