JianZhangAI commited on
Commit
77586d7
·
1 Parent(s): 4042307

add steps

Browse files
dobot_pour_water_lerobot_1760255391_MolmoE-7B-10131629-5000_openai_seq824_l1_regression_three_images_resize_proprio-8_ft_ah_fullyft_llm_bs224/{step7000-unsharded → step8000-unsharded}/config.yaml RENAMED
File without changes
dobot_pour_water_lerobot_1760255391_MolmoE-7B-10131629-5000_openai_seq824_l1_regression_three_images_resize_proprio-8_ft_ah_fullyft_llm_bs224/{step7000-unsharded → step8000-unsharded}/model.pt RENAMED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:1bb14af1bf2fcfff2033f1697824c82767f5f6028d6df0ddc1740f949835e3a0
3
  size 32960667415
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:57a8703f2ce524e11223e2920581d69e38fc35fa3e9a1a9e51fdf58442bad620
3
  size 32960667415
dobot_tidy_up_the_desk_lerobot_1760321094_MolmoE-7B-10131629-5000_openai_seq768_flow_matching_three_images_resize_proprio-8_ft_ah_fullyft_llm_bs288/step6500-unsharded/config.yaml DELETED
@@ -1,331 +0,0 @@
1
- run_name: dobot_tidy_up_the_desk_lerobot_1760321094_MolmoE-7B-10131629-5000_openai_seq768_flow_matching_three_images_resize_proprio-8_ft_ah_fullyft_llm_bs288_20251103_163413
2
- seed: 6198
3
- epoch: null
4
- dry_run: false
5
- model:
6
- d_model: 3584
7
- n_heads: 28
8
- n_kv_heads: 4
9
- qkv_bias: true
10
- clip_qkv: null
11
- n_layers: 28
12
- mlp_ratio: 4
13
- mlp_hidden_size: 37888
14
- activation_type: swiglu
15
- block_type: sequential
16
- block_group_size: 1
17
- rope: true
18
- rope_full_precision: true
19
- rope_theta: 1000000.0
20
- vision_backbone:
21
- image_model_type: openai
22
- image_default_input_size:
23
- - 336
24
- - 336
25
- image_patch_size: 14
26
- image_pos_patch_size: 14
27
- image_emb_dim: 1024
28
- image_num_heads: 16
29
- image_num_key_value_heads: 16
30
- image_num_layers: 23
31
- image_head_dim: 64
32
- image_mlp_dim: 4096
33
- image_mlp_activations: quick_gelu
34
- image_dropout_rate: 0.0
35
- image_num_pos: 577
36
- image_norm_eps: 1.0e-05
37
- attention_dropout: 0.0
38
- residual_dropout: 0.0
39
- initializer_range: 0.02
40
- fsdp_wrap: false
41
- resize_mode: default
42
- vit_load_path: /weka/oe-training-default/mm-olmo/pretrained_image_encoders/vit-l-14-336.pt
43
- llm_load_path: /weka/oe-training-default/mm-olmo/pretrained_llms/qwen2-7b.pt
44
- low_cpu_fsdp: true
45
- attention_type: sdpa
46
- float32_attention: true
47
- attention_dropout: 0.0
48
- attention_layer_norm: false
49
- residual_dropout: 0.1
50
- response_residual_dropout: 0.0
51
- embedding_dropout: 0.0
52
- layer_norm_type: rms
53
- layer_norm_with_affine: true
54
- layer_norm_eps: 1.0e-06
55
- attention_layer_norm_with_affine: true
56
- max_sequence_length: 4096
57
- max_position_embeddings: null
58
- include_bias: false
59
- bias_for_layer_norm: null
60
- scale_logits: false
61
- vocab_size: 152064
62
- embedding_size: 152064
63
- ff_out_size: 152192
64
- additional_vocab_size: 128
65
- new_embedding_init_range: 0.02
66
- weight_tying: false
67
- init_device: cpu
68
- init_fn: normal
69
- init_std: 0.02
70
- init_cutoff_factor: null
71
- norm_after: false
72
- precision: amp_bf16
73
- max_crops: 12
74
- crop_mode: resize
75
- use_col_tokens: true
76
- prompt_type: uber_model
77
- system_prompt_kind: demo_or_style
78
- message_formatting: role
79
- always_start_with_space: true
80
- multi_annotation_weighting: root_subsegments
81
- default_inference_len: 128
82
- overlap_margins:
83
- - 4
84
- - 4
85
- pad_value: 0.0
86
- image_padding_embed: pad_and_partial_pad
87
- fix_image_padding: true
88
- vit_layers:
89
- - -2
90
- - -9
91
- image_pooling_h: 2
92
- image_pooling_w: 2
93
- image_pooling_2d: attention_meanq
94
- image_projector: mlp
95
- image_feature_dropout: 0.0
96
- initializer_range: 0.02
97
- normalize_input_embeds: false
98
- use_position_ids: true
99
- head_dim: null
100
- action_tokenizer:
101
- identifier: physical-intelligence/fast
102
- tokenizer_dir: null
103
- tokenizer:
104
- identifier: Qwen/Qwen2-7B
105
- tokenizer_dir: null
106
- pad_tokenizer: true
107
- moe_num_experts: 8
108
- moe_top_k: 2
109
- moe_mlp_impl: sparse
110
- moe_log_expert_assignment: false
111
- moe_shared_expert: false
112
- moe_lbl_in_fp32: false
113
- moe_interleave: false
114
- moe_loss_weight: 0.1
115
- moe_zloss_weight: null
116
- moe_dropless: true
117
- moe_capacity_factor: 1.25
118
- action_head: flow_matching
119
- action_dim: 14
120
- fixed_action_dim: 14
121
- right_end_effector_dim: 7
122
- left_end_effector_dim: 7
123
- mobile_base_dim: 3
124
- num_actions_chunk: 50
125
- proprio_dim: 14
126
- num_diffusion_steps: 1000
127
- num_diffusion_inference_steps: 30
128
- use_proprio: true
129
- action_head_dit_hidden_size: 1152
130
- action_head_dit_depth: 28
131
- action_head_dit_num_heads: 16
132
- action_head_flow_matching_dim: 1024
133
- action_head_flow_matching_layers: 28
134
- action_head_flow_matching_heads: 8
135
- action_head_flow_matching_intermediate_size: 2048
136
- llm_causal_attention: false
137
- action_use_left_eef: true
138
- action_use_mobile_base: false
139
- allow_resume: false
140
- ft_llm: true
141
- ft_vit: false
142
- ft_connector: false
143
- ft_embedding: lm_head
144
- lora: false
145
- use_lora: false
146
- lora_rank: 8
147
- lora_llm: false
148
- lora_vit: false
149
- lora_connector: false
150
- early_exit: false
151
- train_exit_random_layer: false
152
- optimizer:
153
- name: adamw
154
- learning_rate: 0.0001
155
- weight_decay: 0.01
156
- betas:
157
- - 0.9
158
- - 0.95
159
- eps: 1.0e-05
160
- connector_learning_rate: 0.0002
161
- vit_learning_rate: 6.0e-06
162
- llm_learning_rate: 5.0e-05
163
- connector_weight_decay: 0.0
164
- vit_weight_decay: 0.0
165
- llm_weight_decay: 0.0
166
- connector_betas:
167
- - 0.9
168
- - 0.95
169
- vit_betas:
170
- - 0.9
171
- - 0.95
172
- llm_betas:
173
- - 0.9
174
- - 0.95
175
- connector_eps: 1.0e-06
176
- vit_eps: 1.0e-06
177
- llm_eps: 1.0e-06
178
- metrics_log_interval: 20
179
- scheduler:
180
- name: multimodal
181
- units: steps
182
- t_warmup: 100
183
- t_max: null
184
- alpha_f: 0.1
185
- connector_t_warmup: 200
186
- vit_t_warmup: 2000
187
- llm_t_warmup: 2000
188
- grad_clip_warmup_steps: null
189
- grad_clip_warmup_factor: null
190
- warmup_min_lr: 0.0
191
- data:
192
- dataset: vla_dataset_realmachine
193
- mixture: null
194
- root_size_mixture: null
195
- split: train
196
- seed: 95818
197
- shuffle_messages: false
198
- pad: to_max
199
- sequence_length: 768
200
- shuffle: true
201
- for_inference: false
202
- multi_modal: torch
203
- num_workers: 0
204
- drop_last: true
205
- pin_memory: true
206
- prefetch_factor: null
207
- persistent_workers: false
208
- timeout: 0
209
- rlds_dataset_name: libero_4_task_suites_no_noops
210
- rlds_data_root_dir: /vast/users/xiaodan/zhangjian/HuggingFace/dataset/Dobot-Xtrainer/dobot_tidy_up_the_desk_lerobot_1760321094
211
- use_wrist_image: true
212
- use_proprio: true
213
- rlds_shuffle_buffer_size: 100000
214
- rlds_traj_threads: 8
215
- rlds_read_threads: 8
216
- lerobot_episode_index_start: null
217
- lerobot_episode_index_end: null
218
- restore_dataloader: true
219
- fast_forward_batches: null
220
- evaluators:
221
- - label: val
222
- data:
223
- dataset: vla_dataset_realmachine
224
- mixture: null
225
- root_size_mixture: null
226
- split: validation
227
- seed: null
228
- shuffle_messages: false
229
- pad: to_max
230
- sequence_length: 768
231
- shuffle: false
232
- for_inference: false
233
- multi_modal: torch
234
- num_workers: 0
235
- drop_last: true
236
- pin_memory: true
237
- prefetch_factor: null
238
- persistent_workers: true
239
- timeout: 0
240
- rlds_dataset_name: libero_4_task_suites_no_noops
241
- rlds_data_root_dir: /mnt/data/zhangjian/dataset/Simulation/datasets--openvla--modified_libero_rlds
242
- use_wrist_image: true
243
- use_proprio: true
244
- rlds_shuffle_buffer_size: 256000
245
- rlds_traj_threads: 8
246
- rlds_read_threads: 8
247
- lerobot_episode_index_start: 353
248
- lerobot_episode_index_end: 765
249
- device_eval_batch_size: null
250
- subset_num_batches: 64
251
- max_examples: null
252
- max_new_tokens: 448
253
- mm_evaluator: null
254
- save_dir: null
255
- save_to_checkpoint_dir: false
256
- eval_name: null
257
- skip_if_metrics_cached: true
258
- eval_interval: 0
259
- inf_eval_interval: -1
260
- inf_evaluators: []
261
- save_folder: /vast/users/xiaodan/zhangjian/checkpoints/dobot_tidy_up_the_desk_lerobot_1760321094_MolmoE-7B-10131629-5000_openai_seq768_flow_matching_three_images_resize_proprio-8_ft_ah_fullyft_llm_bs288
262
- remote_save_folder: null
263
- canceled_check_interval: 50
264
- save_interval: 500
265
- save_interval_unsharded: 500
266
- save_interval_ephemeral: null
267
- save_interval_action_head: 500
268
- save_num_checkpoints_to_keep: 1
269
- save_num_unsharded_checkpoints_to_keep: 1
270
- save_num_action_head_checkpoints_to_keep: 2
271
- save_overwrite: true
272
- force_save_unsharded: false
273
- no_pre_train_checkpoint: true
274
- initial_model_checkpoint: /vast/users/xiaodan/zhangjian/molmo_data/Molmo-7B-10131629-5000
275
- load_model_config: null
276
- checkpoint_dir: /vast/users/xiaodan/zhangjian/molmo_data/Molmo-7B-10131629-5000
277
- load_path: null
278
- load_path_sharded_checkpointer: null
279
- reset_optimizer_state: false
280
- reset_trainer_state: false
281
- save_dataloader_state: false
282
- reset_dataloader_state: false
283
- keep_lr_on_load: true
284
- sharded_checkpointer: torch_legacy
285
- max_duration: 500000
286
- global_train_batch_size: 288
287
- device_train_batch_size: 36
288
- device_train_microbatch_size: 36
289
- device_eval_batch_size: 4
290
- eval_subset_num_batches: -1
291
- eval_on_load: false
292
- device_inf_eval_batch_size: 16
293
- inf_eval_subset_num_batches: -1
294
- device_train_grad_accum: 1
295
- max_grad_norm: 1.0
296
- multi_component_grad_norm: true
297
- batch_divisor: global_batch
298
- max_grad_norm_ratio: null
299
- precision: amp_bf16
300
- wandb:
301
- project: a1-vla-realmachine
302
- entity: demo0
303
- group: null
304
- name: dobot_tidy_up_the_desk_lerobot_1760321094_MolmoE-7B-10131629-5000_openai_seq768_flow_matching_three_images_resize_proprio-8_ft_ah_fullyft_llm_bs288_20251103_163413
305
- tags:
306
- - watching
307
- log_artifacts: false
308
- rank_zero_only: true
309
- log_interval: 1
310
- speed_monitor:
311
- window_size: 20
312
- gpu_flops_available: null
313
- console_log_interval: 1
314
- gen1_gc_interval: 1
315
- compile: null
316
- fsdp:
317
- use_orig_params: true
318
- sharding_strategy: FULL_SHARD
319
- wrapping_strategy: by_block_and_size
320
- precision: float
321
- hybrid_sharding_num_model_replicas: null
322
- softmax_auxiliary_loss: true
323
- softmax_auxiliary_loss_scale: 0.0001
324
- time_limit: null
325
- extra_steps_after_cancel: 10
326
- python_profiling: false
327
- torch_profiling: false
328
- stop_at: 500000
329
- stop_after: null
330
- activation_checkpointing: whole_layer
331
- fused_loss: null
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
dobot_tidy_up_the_desk_lerobot_1760321094_MolmoE-7B-10131629-5000_openai_seq768_flow_matching_three_images_resize_proprio-8_ft_ah_fullyft_llm_bs288/step6500-unsharded/model.pt DELETED
@@ -1,3 +0,0 @@
1
- version https://git-lfs.github.com/spec/v1
2
- oid sha256:9ec2c89856b10f7f3db9f1613e37a4aeef1987f51b834278719b353b7c676809
3
- size 33843178215
 
 
 
 
dobot_tidy_up_the_desk_lerobot_1760321094_MolmoE-7B-10131629-5000_openai_seq768_flow_matching_three_images_resize_proprio-8_ft_ah_fullyft_llm_bs288/{step3000-unsharded → step7000-unsharded}/config.yaml RENAMED
File without changes
dobot_tidy_up_the_desk_lerobot_1760321094_MolmoE-7B-10131629-5000_openai_seq768_flow_matching_three_images_resize_proprio-8_ft_ah_fullyft_llm_bs288/{step3000-unsharded → step7000-unsharded}/model.pt RENAMED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:6c281f72e3b1b6a19215fa847f603da82f4ef64ccf741b4d67f9f11209c6cf9b
3
  size 33843178215
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:8efd8451efae8edc3c4fee1a2c0c1811cd143024c85ec46484c38de3903bec13
3
  size 33843178215