Upload 2 files
Browse files
configs/kosmos_ph_post_train_bridge.json
ADDED
@@ -0,0 +1,167 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"robovlm_name": "RoboKosMos",
|
3 |
+
"parent": null,
|
4 |
+
"model": "kosmos",
|
5 |
+
"model_url": "https://huggingface.co/microsoft/kosmos-2-patch14-224",
|
6 |
+
"seq_len": 1,
|
7 |
+
"image_size": 224,
|
8 |
+
"image_mean": [
|
9 |
+
0.48145466,
|
10 |
+
0.4578275,
|
11 |
+
0.40821073
|
12 |
+
],
|
13 |
+
"image_std": [
|
14 |
+
0.26862954,
|
15 |
+
0.26130258,
|
16 |
+
0.27577711
|
17 |
+
],
|
18 |
+
"window_size": 16,
|
19 |
+
"fwd_pred_next_n": 10,
|
20 |
+
"arm_gripper_loss_ratio": 0.01,
|
21 |
+
"cap_loss_ratio": 0.05,
|
22 |
+
"fwd_loss_ratio": 0,
|
23 |
+
"seed": 123,
|
24 |
+
"batch_size": 4,
|
25 |
+
"num_workers": 16,
|
26 |
+
"data_scale": 1,
|
27 |
+
"optimizer": "adam",
|
28 |
+
"learning_rate": 2e-05,
|
29 |
+
"min_lr_scale": 0.01,
|
30 |
+
"weight_decay": 0,
|
31 |
+
"warmup_epochs": 0,
|
32 |
+
"warmup_steps": 0,
|
33 |
+
"warmup_ratio": null,
|
34 |
+
"use_hand_rgb": true,
|
35 |
+
"use_time_causal_attn": false,
|
36 |
+
"use_mim_obs_loss": false,
|
37 |
+
"use_pixel_loss": true,
|
38 |
+
"use_obs_queries": true,
|
39 |
+
"use_vision_resampler": false,
|
40 |
+
"vision_masked_ratio": 0.9,
|
41 |
+
"use_tube_mask": false,
|
42 |
+
"output_root": "runs/checkpoints/oxe_post_train/kosmos",
|
43 |
+
"log_root": "runs/logs/oxe_post_train/kosmos",
|
44 |
+
"cache_root": "runs/cache/oxe_post_train/kosmos",
|
45 |
+
"model_load_path": "runs/checkpoints/oxe_pretrain/kosmos/2024-11-21/01-16/epoch=0-step=50000.pt",
|
46 |
+
"model_load_source": "torch",
|
47 |
+
"resume": null,
|
48 |
+
"model_path": ".vlms/kosmos-2-patch14-224",
|
49 |
+
"model_config": ".vlms/kosmos-2-patch14-224/config.json",
|
50 |
+
"train_setup": {
|
51 |
+
"precision": "16",
|
52 |
+
"predict_action": true,
|
53 |
+
"predict_forward": false,
|
54 |
+
"predict_forward_hand": false,
|
55 |
+
"predict_caption": false,
|
56 |
+
"train_vision": true,
|
57 |
+
"bits": -1,
|
58 |
+
"freeze_mm_mlp_adapter": false,
|
59 |
+
"freeze_backbone": false,
|
60 |
+
"freeze_resampler": false,
|
61 |
+
"tune_mm_mlp_adapter": false,
|
62 |
+
"mm_use_im_start_end": false,
|
63 |
+
"mm_use_im_patch_token": false,
|
64 |
+
"gradient_checkpointing": false,
|
65 |
+
"lora_enable": false,
|
66 |
+
"mm_projector_lr": 0.0001,
|
67 |
+
"lora_r": 64,
|
68 |
+
"lora_alpha": 16,
|
69 |
+
"lora_dropout": 0.05,
|
70 |
+
"lora_bias": "none",
|
71 |
+
"train_text_embedding": true
|
72 |
+
},
|
73 |
+
"vision_resampler": {
|
74 |
+
"vis_dim": 1024,
|
75 |
+
"depth": 8,
|
76 |
+
"dim_head": 64,
|
77 |
+
"heads": 8,
|
78 |
+
"num_latents": 64
|
79 |
+
},
|
80 |
+
"act_encoder": null,
|
81 |
+
"act_head": {
|
82 |
+
"type": "LSTMDecoder",
|
83 |
+
"hidden_size": 1024,
|
84 |
+
"action_dim": 7,
|
85 |
+
"down_sample": "none",
|
86 |
+
"latent": 1,
|
87 |
+
"fwd_pred_next_n": 1,
|
88 |
+
"window_size": 1,
|
89 |
+
"action_space": "continuous",
|
90 |
+
"with_history": true,
|
91 |
+
"history_type": "post"
|
92 |
+
},
|
93 |
+
"fwd_head": null,
|
94 |
+
"tokenizer": {
|
95 |
+
"type": "AutoProcessor",
|
96 |
+
"pretrained_model_name_or_path": ".vlms/kosmos-2-patch14-224",
|
97 |
+
"tokenizer_type": "kosmos",
|
98 |
+
"max_text_len": 256,
|
99 |
+
"additional_special_tokens": null
|
100 |
+
},
|
101 |
+
"vlm": {
|
102 |
+
"type": "AutoModelForVision2Seq",
|
103 |
+
"name": "kosmos",
|
104 |
+
"pretrained_model_name_or_path": ".vlms/kosmos-2-patch14-224"
|
105 |
+
},
|
106 |
+
"trainer": {
|
107 |
+
"accelerator": "gpu",
|
108 |
+
"strategy": "deepspeed_stage_2",
|
109 |
+
"precision": "16",
|
110 |
+
"logger": [
|
111 |
+
"tensorboard"
|
112 |
+
],
|
113 |
+
"gradient_clip_val": 1.0,
|
114 |
+
"use_distributed_sampler": false,
|
115 |
+
"log_every_n_steps": 10,
|
116 |
+
"max_epochs": 5,
|
117 |
+
"val_check_interval": 10000,
|
118 |
+
"check_val_every_n_epoch": null,
|
119 |
+
"max_steps": 50000,
|
120 |
+
"accumulate_grad_batches": 1,
|
121 |
+
"limit_val_batches": 100
|
122 |
+
},
|
123 |
+
"train_dataset": {
|
124 |
+
"type": "OpenVLADataset",
|
125 |
+
"data_root_dir": "datasets/open-x-embodiment",
|
126 |
+
"model_name": "kosmos",
|
127 |
+
"image_aug": true,
|
128 |
+
"mode": "train",
|
129 |
+
"data_mix": "rt_1",
|
130 |
+
"window_sample": "sliding",
|
131 |
+
"organize_type": "interleave",
|
132 |
+
"shuffle_buffer_size": 51200,
|
133 |
+
"train": true
|
134 |
+
},
|
135 |
+
"val_dataset": {
|
136 |
+
"type": "OpenVLADataset",
|
137 |
+
"data_root_dir": "datasets/open-x-embodiment",
|
138 |
+
"model_name": "kosmos",
|
139 |
+
"mode": "train",
|
140 |
+
"data_mix": "rt_1",
|
141 |
+
"window_sample": "sliding",
|
142 |
+
"organize_type": "interleave",
|
143 |
+
"shuffle_buffer_size": 1000,
|
144 |
+
"train": false,
|
145 |
+
"filter_langs": true
|
146 |
+
},
|
147 |
+
"raw_config_path": "configs/kosmos/oxe_post_train/finetune_kosmos_cont-lstm-post_full-ft_text_vision_wd=0_hist=16_act=10_use-hand_aug-shift_act-norm_lr-1e-5_bs-128_warm-0_oxe-pre_bridge.json",
|
148 |
+
"config": "configs/kosmos/oxe_post_train/finetune_kosmos_cont-lstm-post_full-ft_text_vision_wd=0_hist=16_act=10_use-hand_aug-shift_act-norm_lr-1e-5_bs-128_warm-0_oxe-pre_bridge.json",
|
149 |
+
"gpus": 8,
|
150 |
+
"num_nodes": 4,
|
151 |
+
"log_dir": "runs/logs/oxe_post_train/kosmos/2024-11-24/17-09",
|
152 |
+
"output_dir": "runs/logs/oxe_post_train/kosmos/2024-11-24/17-09",
|
153 |
+
"data_dir": null,
|
154 |
+
"annotation_file": null,
|
155 |
+
"data_subfolder": null,
|
156 |
+
"task_num": null,
|
157 |
+
"exp_name": "17-09",
|
158 |
+
"use_multi_modal_emb": false,
|
159 |
+
"no_video_pretrained_model": false,
|
160 |
+
"finetune": false,
|
161 |
+
"llm": {
|
162 |
+
"type": null,
|
163 |
+
"n_embd": null,
|
164 |
+
"n_layer": null,
|
165 |
+
"n_head": null
|
166 |
+
}
|
167 |
+
}
|
configs/kosmos_ph_post_train_google_robot.json
ADDED
@@ -0,0 +1,167 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"robovlm_name": "RoboKosMos",
|
3 |
+
"parent": null,
|
4 |
+
"model": "kosmos",
|
5 |
+
"model_url": "https://huggingface.co/microsoft/kosmos-2-patch14-224",
|
6 |
+
"seq_len": 1,
|
7 |
+
"image_size": 224,
|
8 |
+
"image_mean": [
|
9 |
+
0.48145466,
|
10 |
+
0.4578275,
|
11 |
+
0.40821073
|
12 |
+
],
|
13 |
+
"image_std": [
|
14 |
+
0.26862954,
|
15 |
+
0.26130258,
|
16 |
+
0.27577711
|
17 |
+
],
|
18 |
+
"window_size": 16,
|
19 |
+
"fwd_pred_next_n": 10,
|
20 |
+
"arm_gripper_loss_ratio": 0.01,
|
21 |
+
"cap_loss_ratio": 0.05,
|
22 |
+
"fwd_loss_ratio": 0,
|
23 |
+
"seed": 123,
|
24 |
+
"batch_size": 4,
|
25 |
+
"num_workers": 16,
|
26 |
+
"data_scale": 1,
|
27 |
+
"optimizer": "adam",
|
28 |
+
"learning_rate": 2e-05,
|
29 |
+
"min_lr_scale": 0.01,
|
30 |
+
"weight_decay": 0,
|
31 |
+
"warmup_epochs": 0,
|
32 |
+
"warmup_steps": 0,
|
33 |
+
"warmup_ratio": null,
|
34 |
+
"use_hand_rgb": true,
|
35 |
+
"use_time_causal_attn": false,
|
36 |
+
"use_mim_obs_loss": false,
|
37 |
+
"use_pixel_loss": true,
|
38 |
+
"use_obs_queries": true,
|
39 |
+
"use_vision_resampler": false,
|
40 |
+
"vision_masked_ratio": 0.9,
|
41 |
+
"use_tube_mask": false,
|
42 |
+
"output_root": "runs/checkpoints/oxe_post_train",
|
43 |
+
"log_root": "runs/logs/oxe_post_train",
|
44 |
+
"cache_root": "runs/cache/oxe_post_train",
|
45 |
+
"model_load_path": "runs/checkpoints/oxe_pretrain/kosmos/2024-11-21/01-16/epoch=0-step=50000.pt",
|
46 |
+
"model_load_source": "torch",
|
47 |
+
"resume": null,
|
48 |
+
"model_path": ".vlms/kosmos-2-patch14-224",
|
49 |
+
"model_config": ".vlms/kosmos-2-patch14-224/config.json",
|
50 |
+
"train_setup": {
|
51 |
+
"precision": "16",
|
52 |
+
"predict_action": true,
|
53 |
+
"predict_forward": false,
|
54 |
+
"predict_forward_hand": false,
|
55 |
+
"predict_caption": false,
|
56 |
+
"train_vision": true,
|
57 |
+
"bits": -1,
|
58 |
+
"freeze_mm_mlp_adapter": false,
|
59 |
+
"freeze_backbone": false,
|
60 |
+
"freeze_resampler": false,
|
61 |
+
"tune_mm_mlp_adapter": false,
|
62 |
+
"mm_use_im_start_end": false,
|
63 |
+
"mm_use_im_patch_token": false,
|
64 |
+
"gradient_checkpointing": false,
|
65 |
+
"lora_enable": false,
|
66 |
+
"mm_projector_lr": 0.0001,
|
67 |
+
"lora_r": 64,
|
68 |
+
"lora_alpha": 16,
|
69 |
+
"lora_dropout": 0.05,
|
70 |
+
"lora_bias": "none",
|
71 |
+
"train_text_embedding": true
|
72 |
+
},
|
73 |
+
"vision_resampler": {
|
74 |
+
"vis_dim": 1024,
|
75 |
+
"depth": 8,
|
76 |
+
"dim_head": 64,
|
77 |
+
"heads": 8,
|
78 |
+
"num_latents": 64
|
79 |
+
},
|
80 |
+
"act_encoder": null,
|
81 |
+
"act_head": {
|
82 |
+
"type": "LSTMDecoder",
|
83 |
+
"hidden_size": 1024,
|
84 |
+
"action_dim": 7,
|
85 |
+
"down_sample": "none",
|
86 |
+
"latent": 1,
|
87 |
+
"fwd_pred_next_n": 1,
|
88 |
+
"window_size": 1,
|
89 |
+
"action_space": "continuous",
|
90 |
+
"with_history": true,
|
91 |
+
"history_type": "post"
|
92 |
+
},
|
93 |
+
"fwd_head": null,
|
94 |
+
"tokenizer": {
|
95 |
+
"type": "AutoProcessor",
|
96 |
+
"pretrained_model_name_or_path": ".vlms/kosmos-2-patch14-224",
|
97 |
+
"tokenizer_type": "kosmos",
|
98 |
+
"max_text_len": 256,
|
99 |
+
"additional_special_tokens": null
|
100 |
+
},
|
101 |
+
"vlm": {
|
102 |
+
"type": "AutoModelForVision2Seq",
|
103 |
+
"name": "kosmos",
|
104 |
+
"pretrained_model_name_or_path": ".vlms/kosmos-2-patch14-224"
|
105 |
+
},
|
106 |
+
"trainer": {
|
107 |
+
"accelerator": "gpu",
|
108 |
+
"strategy": "deepspeed_stage_2",
|
109 |
+
"precision": "16",
|
110 |
+
"logger": [
|
111 |
+
"tensorboard"
|
112 |
+
],
|
113 |
+
"gradient_clip_val": 1.0,
|
114 |
+
"use_distributed_sampler": false,
|
115 |
+
"log_every_n_steps": 10,
|
116 |
+
"max_epochs": 5,
|
117 |
+
"val_check_interval": 10000,
|
118 |
+
"check_val_every_n_epoch": null,
|
119 |
+
"max_steps": 50000,
|
120 |
+
"accumulate_grad_batches": 1,
|
121 |
+
"limit_val_batches": 100
|
122 |
+
},
|
123 |
+
"train_dataset": {
|
124 |
+
"type": "OpenVLADataset",
|
125 |
+
"data_root_dir": "datasets/open-x-embodiment",
|
126 |
+
"model_name": "kosmos",
|
127 |
+
"image_aug": true,
|
128 |
+
"mode": "train",
|
129 |
+
"data_mix": "rt_1",
|
130 |
+
"window_sample": "sliding",
|
131 |
+
"organize_type": "interleave",
|
132 |
+
"shuffle_buffer_size": 51200,
|
133 |
+
"train": true
|
134 |
+
},
|
135 |
+
"val_dataset": {
|
136 |
+
"type": "OpenVLADataset",
|
137 |
+
"data_root_dir": "datasets/open-x-embodiment",
|
138 |
+
"model_name": "kosmos",
|
139 |
+
"mode": "train",
|
140 |
+
"data_mix": "rt_1",
|
141 |
+
"window_sample": "sliding",
|
142 |
+
"organize_type": "interleave",
|
143 |
+
"shuffle_buffer_size": 1000,
|
144 |
+
"train": false,
|
145 |
+
"filter_langs": true
|
146 |
+
},
|
147 |
+
"raw_config_path": "configs/kosmos/oxe_post_train/finetune_kosmos_cont-lstm-post_full-ft_text_vision_wd=0_hist=16_act=10_use-hand_aug-shift_act-norm_lr-1e-5_bs-128_warm-0_oxe-pre_rt.json",
|
148 |
+
"config": "configs/kosmos/oxe_post_train/finetune_kosmos_cont-lstm-post_full-ft_text_vision_wd=0_hist=16_act=10_use-hand_aug-shift_act-norm_lr-1e-5_bs-128_warm-0_oxe-pre_rt.json",
|
149 |
+
"gpus": 8,
|
150 |
+
"num_nodes": 4,
|
151 |
+
"log_dir": "runs/logs/oxe_post_train/kosmos/calvin_finetune/2024-11-24/16-11",
|
152 |
+
"output_dir": "runs/logs/oxe_post_train/kosmos/calvin_finetune/2024-11-24/16-11",
|
153 |
+
"data_dir": null,
|
154 |
+
"annotation_file": null,
|
155 |
+
"data_subfolder": null,
|
156 |
+
"task_num": null,
|
157 |
+
"exp_name": "16-11",
|
158 |
+
"use_multi_modal_emb": false,
|
159 |
+
"no_video_pretrained_model": false,
|
160 |
+
"finetune": false,
|
161 |
+
"llm": {
|
162 |
+
"type": null,
|
163 |
+
"n_embd": null,
|
164 |
+
"n_layer": null,
|
165 |
+
"n_head": null
|
166 |
+
}
|
167 |
+
}
|