Upload folder using huggingface_hub
Browse files- configs/kosmos_ph_calvin_abc.json +154 -0
- configs/kosmos_ph_calvin_abcd.json +155 -0
- configs/kosmos_ph_oxe-pretrain.json +159 -0
configs/kosmos_ph_calvin_abc.json
ADDED
|
@@ -0,0 +1,154 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"robovlm_name": "RoboKosMos",
|
| 3 |
+
"parent": null,
|
| 4 |
+
"task_name": "calvin_finetune",
|
| 5 |
+
"model": "kosmos",
|
| 6 |
+
"model_url": "https://huggingface.co/microsoft/kosmos-2-patch14-224",
|
| 7 |
+
"seq_len": 1,
|
| 8 |
+
"image_size": 224,
|
| 9 |
+
"image_mean": [
|
| 10 |
+
0.48145466,
|
| 11 |
+
0.4578275,
|
| 12 |
+
0.40821073
|
| 13 |
+
],
|
| 14 |
+
"image_std": [
|
| 15 |
+
0.26862954,
|
| 16 |
+
0.26130258,
|
| 17 |
+
0.27577711
|
| 18 |
+
],
|
| 19 |
+
"window_size": 16,
|
| 20 |
+
"fwd_pred_next_n": 10,
|
| 21 |
+
"arm_gripper_loss_ratio": 0.01,
|
| 22 |
+
"cap_loss_ratio": 0.05,
|
| 23 |
+
"fwd_loss_ratio": 0,
|
| 24 |
+
"seed": 123,
|
| 25 |
+
"batch_size": 4,
|
| 26 |
+
"num_workers": 16,
|
| 27 |
+
"data_scale": 1,
|
| 28 |
+
"optimizer": "adam",
|
| 29 |
+
"learning_rate": 0.0001,
|
| 30 |
+
"min_lr_scale": 0.01,
|
| 31 |
+
"weight_decay": 0,
|
| 32 |
+
"warmup_epochs": 0.25,
|
| 33 |
+
"warmup_steps": 0,
|
| 34 |
+
"warmup_ratio": null,
|
| 35 |
+
"use_hand_rgb": true,
|
| 36 |
+
"use_time_causal_attn": false,
|
| 37 |
+
"use_mim_obs_loss": false,
|
| 38 |
+
"use_pixel_loss": true,
|
| 39 |
+
"use_obs_queries": true,
|
| 40 |
+
"use_vision_resampler": false,
|
| 41 |
+
"vision_masked_ratio": 0.9,
|
| 42 |
+
"use_tube_mask": false,
|
| 43 |
+
"output_root": "checkpoints/kosmos/calvin_finetune",
|
| 44 |
+
"log_root": "logs/kosmos/calvin_finetune",
|
| 45 |
+
"cache_root": "cache/kosmos",
|
| 46 |
+
"model_load_path": null,
|
| 47 |
+
"model_load_source": "torch",
|
| 48 |
+
"resume": null,
|
| 49 |
+
"model_path": ".vlms/kosmos-2-patch14-224",
|
| 50 |
+
"model_config": ".vlms/kosmos-2-patch14-224/config.json",
|
| 51 |
+
"train_setup": {
|
| 52 |
+
"precision": "16",
|
| 53 |
+
"predict_action": true,
|
| 54 |
+
"predict_forward": false,
|
| 55 |
+
"predict_forward_hand": false,
|
| 56 |
+
"predict_caption": false,
|
| 57 |
+
"train_vision": true,
|
| 58 |
+
"bits": -1,
|
| 59 |
+
"freeze_mm_mlp_adapter": false,
|
| 60 |
+
"freeze_backbone": false,
|
| 61 |
+
"freeze_resampler": false,
|
| 62 |
+
"tune_mm_mlp_adapter": false,
|
| 63 |
+
"mm_use_im_start_end": false,
|
| 64 |
+
"mm_use_im_patch_token": false,
|
| 65 |
+
"gradient_checkpointing": false,
|
| 66 |
+
"lora_enable": false,
|
| 67 |
+
"mm_projector_lr": 0.0001,
|
| 68 |
+
"lora_r": 64,
|
| 69 |
+
"lora_alpha": 16,
|
| 70 |
+
"lora_dropout": 0.05,
|
| 71 |
+
"lora_bias": "none",
|
| 72 |
+
"train_text_embedding": true
|
| 73 |
+
},
|
| 74 |
+
"vision_resampler": {
|
| 75 |
+
"vis_dim": 1024,
|
| 76 |
+
"depth": 8,
|
| 77 |
+
"dim_head": 64,
|
| 78 |
+
"heads": 8,
|
| 79 |
+
"num_latents": 64
|
| 80 |
+
},
|
| 81 |
+
"act_encoder": null,
|
| 82 |
+
"act_head": {
|
| 83 |
+
"type": "LSTMDecoder",
|
| 84 |
+
"hidden_size": 1024,
|
| 85 |
+
"action_dim": 7,
|
| 86 |
+
"down_sample": "none",
|
| 87 |
+
"latent": 1,
|
| 88 |
+
"fwd_pred_next_n": 1,
|
| 89 |
+
"window_size": 1,
|
| 90 |
+
"action_space": "continuous",
|
| 91 |
+
"with_history": true,
|
| 92 |
+
"history_type": "post"
|
| 93 |
+
},
|
| 94 |
+
"fwd_head": null,
|
| 95 |
+
"tokenizer": {
|
| 96 |
+
"type": "AutoProcessor",
|
| 97 |
+
"pretrained_model_name_or_path": ".vlms/kosmos-2-patch14-224",
|
| 98 |
+
"tokenizer_type": "kosmos",
|
| 99 |
+
"max_text_len": 256,
|
| 100 |
+
"additional_special_tokens": null
|
| 101 |
+
},
|
| 102 |
+
"trainer": {
|
| 103 |
+
"accelerator": "gpu",
|
| 104 |
+
"strategy": "deepspeed_stage_2",
|
| 105 |
+
"precision": "16",
|
| 106 |
+
"logger": [
|
| 107 |
+
"tensorboard"
|
| 108 |
+
],
|
| 109 |
+
"gradient_clip_val": 1.0,
|
| 110 |
+
"use_distributed_sampler": false,
|
| 111 |
+
"log_every_n_steps": 10,
|
| 112 |
+
"max_epochs": 5,
|
| 113 |
+
"val_check_interval": null,
|
| 114 |
+
"check_val_every_n_epoch": 1,
|
| 115 |
+
"max_steps": -1,
|
| 116 |
+
"accumulate_grad_batches": 1
|
| 117 |
+
},
|
| 118 |
+
"train_dataset": {
|
| 119 |
+
"type": "DiskCalvinDataset",
|
| 120 |
+
"data_dir": "datasets/calvin_data/task_ABC_D/training",
|
| 121 |
+
"shift_first": false,
|
| 122 |
+
"model_name": "kosmos",
|
| 123 |
+
"rgb_pad": 10,
|
| 124 |
+
"gripper_pad": 4
|
| 125 |
+
},
|
| 126 |
+
"val_dataset": {
|
| 127 |
+
"type": "DiskCalvinDataset",
|
| 128 |
+
"data_dir": "datasets/calvin_data/task_ABC_D/validation",
|
| 129 |
+
"model_name": "kosmos"
|
| 130 |
+
},
|
| 131 |
+
"norm_action": true,
|
| 132 |
+
"norm_min": -0.65,
|
| 133 |
+
"norm_max": 0.65,
|
| 134 |
+
"raw_config_path": "configs/kosmos/finetune_kosmos_cont-lstm-post_full-ft_text_vision_wd=0_hist=16_act=10_use-hand_aug-shift_act-norm_lr-1e-4_abc.json",
|
| 135 |
+
"config": "configs/kosmos/finetune_kosmos_cont-lstm-post_full-ft_text_vision_wd=0_hist=16_act=10_use-hand_aug-shift_act-norm_lr-1e-4_abc.json",
|
| 136 |
+
"gpus": 8,
|
| 137 |
+
"num_nodes": 4,
|
| 138 |
+
"log_dir": "logs/kosmos/calvin_finetune/2024-11-20/04-00",
|
| 139 |
+
"output_dir": "checkpoints/kosmos/calvin_finetune/2024-11-20/04-00",
|
| 140 |
+
"data_dir": null,
|
| 141 |
+
"annotation_file": null,
|
| 142 |
+
"data_subfolder": null,
|
| 143 |
+
"task_num": null,
|
| 144 |
+
"exp_name": "04-00",
|
| 145 |
+
"use_multi_modal_emb": false,
|
| 146 |
+
"no_video_pretrained_model": false,
|
| 147 |
+
"finetune": false,
|
| 148 |
+
"llm": {
|
| 149 |
+
"type": null,
|
| 150 |
+
"n_embd": null,
|
| 151 |
+
"n_layer": null,
|
| 152 |
+
"n_head": null
|
| 153 |
+
}
|
| 154 |
+
}
|
configs/kosmos_ph_calvin_abcd.json
ADDED
|
@@ -0,0 +1,155 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"robovlm_name": "RoboKosMos",
|
| 3 |
+
"parent": null,
|
| 4 |
+
"task_name": "calvin_finetune",
|
| 5 |
+
"model": "kosmos",
|
| 6 |
+
"model_url": "https://huggingface.co/microsoft/kosmos-2-patch14-224",
|
| 7 |
+
"seq_len": 1,
|
| 8 |
+
"image_size": 224,
|
| 9 |
+
"image_mean": [
|
| 10 |
+
0.48145466,
|
| 11 |
+
0.4578275,
|
| 12 |
+
0.40821073
|
| 13 |
+
],
|
| 14 |
+
"image_std": [
|
| 15 |
+
0.26862954,
|
| 16 |
+
0.26130258,
|
| 17 |
+
0.27577711
|
| 18 |
+
],
|
| 19 |
+
"window_size": 16,
|
| 20 |
+
"fwd_pred_next_n": 10,
|
| 21 |
+
"arm_gripper_loss_ratio": 0.01,
|
| 22 |
+
"cap_loss_ratio": 0.05,
|
| 23 |
+
"fwd_loss_ratio": 0,
|
| 24 |
+
"seed": 123,
|
| 25 |
+
"batch_size": 4,
|
| 26 |
+
"num_workers": 16,
|
| 27 |
+
"data_scale": 1,
|
| 28 |
+
"optimizer": "adam",
|
| 29 |
+
"learning_rate": 0.0001,
|
| 30 |
+
"min_lr_scale": 0.01,
|
| 31 |
+
"weight_decay": 0,
|
| 32 |
+
"warmup_epochs": 0.25,
|
| 33 |
+
"warmup_steps": 0,
|
| 34 |
+
"warmup_ratio": null,
|
| 35 |
+
"use_hand_rgb": true,
|
| 36 |
+
"use_time_causal_attn": false,
|
| 37 |
+
"use_mim_obs_loss": false,
|
| 38 |
+
"use_pixel_loss": true,
|
| 39 |
+
"use_obs_queries": true,
|
| 40 |
+
"use_vision_resampler": false,
|
| 41 |
+
"vision_masked_ratio": 0.9,
|
| 42 |
+
"use_tube_mask": false,
|
| 43 |
+
"output_root": "checkpoints/kosmos/calvin_finetune",
|
| 44 |
+
"log_root": "logs/kosmos/calvin_finetune",
|
| 45 |
+
"cache_root": "cache/kosmos",
|
| 46 |
+
"model_load_path": null,
|
| 47 |
+
"model_load_source": "torch",
|
| 48 |
+
"resume": null,
|
| 49 |
+
"model_path": "/mnt/bn/robotics-data-lxh-lq/LLaVA/kosmos-2-patch14-224",
|
| 50 |
+
"model_config": "/mnt/bn/robotics-data-lxh-lq/LLaVA/kosmos-2-patch14-224/config.json",
|
| 51 |
+
"train_setup": {
|
| 52 |
+
"precision": "16",
|
| 53 |
+
"predict_action": true,
|
| 54 |
+
"predict_forward": false,
|
| 55 |
+
"predict_forward_hand": false,
|
| 56 |
+
"predict_caption": false,
|
| 57 |
+
"train_vision": true,
|
| 58 |
+
"bits": -1,
|
| 59 |
+
"freeze_mm_mlp_adapter": false,
|
| 60 |
+
"freeze_backbone": false,
|
| 61 |
+
"freeze_resampler": false,
|
| 62 |
+
"tune_mm_mlp_adapter": false,
|
| 63 |
+
"mm_use_im_start_end": false,
|
| 64 |
+
"mm_use_im_patch_token": false,
|
| 65 |
+
"gradient_checkpointing": false,
|
| 66 |
+
"lora_enable": false,
|
| 67 |
+
"mm_projector_lr": 0.0001,
|
| 68 |
+
"lora_r": 64,
|
| 69 |
+
"lora_alpha": 16,
|
| 70 |
+
"lora_dropout": 0.05,
|
| 71 |
+
"lora_bias": "none",
|
| 72 |
+
"train_text_embedding": true
|
| 73 |
+
},
|
| 74 |
+
"vision_resampler": {
|
| 75 |
+
"vis_dim": 1024,
|
| 76 |
+
"depth": 8,
|
| 77 |
+
"dim_head": 64,
|
| 78 |
+
"heads": 8,
|
| 79 |
+
"num_latents": 64
|
| 80 |
+
},
|
| 81 |
+
"act_encoder": null,
|
| 82 |
+
"act_head": {
|
| 83 |
+
"type": "LSTMDecoder",
|
| 84 |
+
"hidden_size": 1024,
|
| 85 |
+
"action_dim": 7,
|
| 86 |
+
"down_sample": "none",
|
| 87 |
+
"latent": 1,
|
| 88 |
+
"fwd_pred_next_n": 1,
|
| 89 |
+
"window_size": 1,
|
| 90 |
+
"action_space": "continuous",
|
| 91 |
+
"with_history": true,
|
| 92 |
+
"history_type": "post"
|
| 93 |
+
},
|
| 94 |
+
"fwd_head": null,
|
| 95 |
+
"tokenizer": {
|
| 96 |
+
"type": "AutoProcessor",
|
| 97 |
+
"pretrained_model_name_or_path": "/mnt/bn/robotics-data-lxh-lq/LLaVA/kosmos-2-patch14-224",
|
| 98 |
+
"tokenizer_type": "kosmos",
|
| 99 |
+
"max_text_len": 256,
|
| 100 |
+
"additional_special_tokens": null
|
| 101 |
+
},
|
| 102 |
+
"trainer": {
|
| 103 |
+
"accelerator": "gpu",
|
| 104 |
+
"strategy": "deepspeed_stage_2",
|
| 105 |
+
"precision": "16",
|
| 106 |
+
"logger": [
|
| 107 |
+
"tensorboard"
|
| 108 |
+
],
|
| 109 |
+
"gradient_clip_val": 1.0,
|
| 110 |
+
"use_distributed_sampler": false,
|
| 111 |
+
"log_every_n_steps": 10,
|
| 112 |
+
"max_epochs": 5,
|
| 113 |
+
"val_check_interval": null,
|
| 114 |
+
"check_val_every_n_epoch": 1,
|
| 115 |
+
"max_steps": -1,
|
| 116 |
+
"accumulate_grad_batches": 1
|
| 117 |
+
},
|
| 118 |
+
"train_dataset": {
|
| 119 |
+
"type": "DiskCalvinDataset",
|
| 120 |
+
"data_dir": "/mnt/bn/robotics/manipulation_data/calvin_data/task_ABCD_D/training",
|
| 121 |
+
"shift_first": false,
|
| 122 |
+
"model_name": "kosmos",
|
| 123 |
+
"rgb_pad": 10,
|
| 124 |
+
"gripper_pad": 4
|
| 125 |
+
},
|
| 126 |
+
"val_dataset": {
|
| 127 |
+
"type": "DiskCalvinDataset",
|
| 128 |
+
"data_dir": "/mnt/bn/robotics/manipulation_data/calvin_data/task_ABCD_D/validation",
|
| 129 |
+
"model_name": "kosmos"
|
| 130 |
+
},
|
| 131 |
+
"norm_action": true,
|
| 132 |
+
"norm_min": -0.65,
|
| 133 |
+
"norm_max": 0.65,
|
| 134 |
+
"scheduler": "cosine",
|
| 135 |
+
"raw_config_path": "configs/kosmos/finetune_kosmos_cont-lstm-post_full-ft_text_vision_wd=0_hist=16_act=10_use-hand_aug-shift_act-norm_lr-1e-4.json",
|
| 136 |
+
"config": "configs/kosmos/finetune_kosmos_cont-lstm-post_full-ft_text_vision_wd=0_hist=16_act=10_use-hand_aug-shift_act-norm_lr-1e-4.json",
|
| 137 |
+
"gpus": 8,
|
| 138 |
+
"num_nodes": 4,
|
| 139 |
+
"log_dir": "logs/kosmos/calvin_finetune/2024-11-20/04-00",
|
| 140 |
+
"output_dir": "checkpoints/kosmos/calvin_finetune/2024-11-20/04-00",
|
| 141 |
+
"data_dir": null,
|
| 142 |
+
"annotation_file": null,
|
| 143 |
+
"data_subfolder": null,
|
| 144 |
+
"task_num": null,
|
| 145 |
+
"exp_name": "17-54",
|
| 146 |
+
"use_multi_modal_emb": false,
|
| 147 |
+
"no_video_pretrained_model": false,
|
| 148 |
+
"finetune": false,
|
| 149 |
+
"llm": {
|
| 150 |
+
"type": null,
|
| 151 |
+
"n_embd": null,
|
| 152 |
+
"n_layer": null,
|
| 153 |
+
"n_head": null
|
| 154 |
+
}
|
| 155 |
+
}
|
configs/kosmos_ph_oxe-pretrain.json
ADDED
|
@@ -0,0 +1,159 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"parent": null,
|
| 3 |
+
"model": "kosmos",
|
| 4 |
+
"seq_len": 1,
|
| 5 |
+
"image_size": 224,
|
| 6 |
+
"image_mean": [
|
| 7 |
+
0.48145466,
|
| 8 |
+
0.4578275,
|
| 9 |
+
0.40821073
|
| 10 |
+
],
|
| 11 |
+
"image_std": [
|
| 12 |
+
0.26862954,
|
| 13 |
+
0.26130258,
|
| 14 |
+
0.27577711
|
| 15 |
+
],
|
| 16 |
+
"window_size": 16,
|
| 17 |
+
"fwd_pred_next_n": 10,
|
| 18 |
+
"arm_gripper_loss_ratio": 0.01,
|
| 19 |
+
"cap_loss_ratio": 0.05,
|
| 20 |
+
"fwd_loss_ratio": 0,
|
| 21 |
+
"seed": 123,
|
| 22 |
+
"batch_size": 4,
|
| 23 |
+
"num_workers": 16,
|
| 24 |
+
"data_scale": 1,
|
| 25 |
+
"optimizer": "adam",
|
| 26 |
+
"learning_rate": 2e-05,
|
| 27 |
+
"min_lr_scale": 0.01,
|
| 28 |
+
"weight_decay": 0,
|
| 29 |
+
"warmup_epochs": 0,
|
| 30 |
+
"warmup_steps": 5000,
|
| 31 |
+
"warmup_ratio": null,
|
| 32 |
+
"use_hand_rgb": true,
|
| 33 |
+
"use_time_causal_attn": false,
|
| 34 |
+
"use_mim_obs_loss": false,
|
| 35 |
+
"use_pixel_loss": true,
|
| 36 |
+
"use_obs_queries": true,
|
| 37 |
+
"use_vision_resampler": false,
|
| 38 |
+
"vision_masked_ratio": 0.9,
|
| 39 |
+
"use_tube_mask": false,
|
| 40 |
+
"output_root": "/mnt/bn/robotics-data-lxh-lq-v2/checkpoints/video_pretrain_manipulation/kosmos/calvin_finetune",
|
| 41 |
+
"log_root": "/mnt/bn/robotics-data-lxh-lq-v2/logs/video_pretrain_manipulation/kosmos/calvin_finetune",
|
| 42 |
+
"cache_root": "/mnt/bn/robotics-data-lxh-lq-v2/cache/video_pretrain_manipulation/kosmos",
|
| 43 |
+
"model_load_path": null,
|
| 44 |
+
"model_load_source": "torch",
|
| 45 |
+
"resume": null,
|
| 46 |
+
"model_path": "/mnt/bn/robotics-data-lxh-lq/LLaVA/kosmos-2-patch14-224",
|
| 47 |
+
"model_config": "/mnt/bn/robotics-data-lxh-lq/LLaVA/kosmos-2-patch14-224/config.json",
|
| 48 |
+
"train_setup": {
|
| 49 |
+
"precision": "16-mixed",
|
| 50 |
+
"predict_action": true,
|
| 51 |
+
"predict_forward": false,
|
| 52 |
+
"predict_forward_hand": false,
|
| 53 |
+
"predict_caption": false,
|
| 54 |
+
"train_vision": true,
|
| 55 |
+
"bits": -1,
|
| 56 |
+
"freeze_mm_mlp_adapter": false,
|
| 57 |
+
"freeze_backbone": false,
|
| 58 |
+
"freeze_resampler": false,
|
| 59 |
+
"tune_mm_mlp_adapter": false,
|
| 60 |
+
"mm_use_im_start_end": false,
|
| 61 |
+
"mm_use_im_patch_token": false,
|
| 62 |
+
"gradient_checkpointing": false,
|
| 63 |
+
"lora_enable": false,
|
| 64 |
+
"mm_projector_lr": 0.0001,
|
| 65 |
+
"lora_r": 64,
|
| 66 |
+
"lora_alpha": 16,
|
| 67 |
+
"lora_dropout": 0.05,
|
| 68 |
+
"lora_bias": "none",
|
| 69 |
+
"train_text_embedding": true
|
| 70 |
+
},
|
| 71 |
+
"vision_resampler": {
|
| 72 |
+
"vis_dim": 1024,
|
| 73 |
+
"depth": 8,
|
| 74 |
+
"dim_head": 64,
|
| 75 |
+
"heads": 8,
|
| 76 |
+
"num_latents": 64
|
| 77 |
+
},
|
| 78 |
+
"act_encoder": null,
|
| 79 |
+
"act_head": {
|
| 80 |
+
"type": "LSTMDecoder",
|
| 81 |
+
"hidden_size": 1024,
|
| 82 |
+
"action_dim": 7,
|
| 83 |
+
"down_sample": "none",
|
| 84 |
+
"latent": 1,
|
| 85 |
+
"fwd_pred_next_n": 1,
|
| 86 |
+
"window_size": 1,
|
| 87 |
+
"action_space": "continuous",
|
| 88 |
+
"with_history": true,
|
| 89 |
+
"history_type": "post"
|
| 90 |
+
},
|
| 91 |
+
"fwd_head": null,
|
| 92 |
+
"tokenizer": {
|
| 93 |
+
"type": "AutoProcessor",
|
| 94 |
+
"pretrained_model_name_or_path": "/mnt/bn/robotics-data-lxh-lq/LLaVA/kosmos-2-patch14-224",
|
| 95 |
+
"tokenizer_type": "kosmos",
|
| 96 |
+
"max_text_len": 256,
|
| 97 |
+
"additional_special_tokens": null
|
| 98 |
+
},
|
| 99 |
+
"trainer": {
|
| 100 |
+
"accelerator": "gpu",
|
| 101 |
+
"strategy": "deepspeed_stage_2",
|
| 102 |
+
"precision": "16-mixed",
|
| 103 |
+
"logger": [
|
| 104 |
+
"tensorboard"
|
| 105 |
+
],
|
| 106 |
+
"gradient_clip_val": 1.0,
|
| 107 |
+
"use_distributed_sampler": false,
|
| 108 |
+
"log_every_n_steps": 10,
|
| 109 |
+
"max_epochs": 5,
|
| 110 |
+
"val_check_interval": 10000,
|
| 111 |
+
"check_val_every_n_epoch": null,
|
| 112 |
+
"max_steps": 100000,
|
| 113 |
+
"accumulate_grad_batches": 1,
|
| 114 |
+
"limit_val_batches": 1000
|
| 115 |
+
},
|
| 116 |
+
"train_dataset": {
|
| 117 |
+
"type": "OpenVLADataset",
|
| 118 |
+
"data_root_dir": "/mnt/bn/robotics-data-lxh-lq/openvla/datasets/open-x-embodiment",
|
| 119 |
+
"model_name": "kosmos",
|
| 120 |
+
"image_aug": true,
|
| 121 |
+
"mode": "train",
|
| 122 |
+
"data_mix": "oxe_magic_soup",
|
| 123 |
+
"window_sample": "sliding",
|
| 124 |
+
"organize_type": "interleave",
|
| 125 |
+
"shuffle_buffer_size": 51200,
|
| 126 |
+
"train": true
|
| 127 |
+
},
|
| 128 |
+
"val_dataset": {
|
| 129 |
+
"type": "OpenVLADataset",
|
| 130 |
+
"data_root_dir": "/mnt/bn/robotics-data-lxh-lq/openvla/datasets/open-x-embodiment",
|
| 131 |
+
"model_name": "kosmos",
|
| 132 |
+
"mode": "train",
|
| 133 |
+
"data_mix": "rt_1",
|
| 134 |
+
"window_sample": "sliding",
|
| 135 |
+
"organize_type": "interleave",
|
| 136 |
+
"shuffle_buffer_size": 10000,
|
| 137 |
+
"train": false
|
| 138 |
+
},
|
| 139 |
+
"raw_config_path": "configs/kosmos/oxe_new/finetune_kosmos_cont-lstm-post_full-ft_text_vision_wd=0_hist=16_act=10_use-hand_aug-shift_act-norm_lr-2e-5_oxe.json",
|
| 140 |
+
"config": "configs/kosmos/oxe_new/finetune_kosmos_cont-lstm-post_full-ft_text_vision_wd=0_hist=16_act=10_use-hand_aug-shift_act-norm_lr-2e-5_oxe.json",
|
| 141 |
+
"gpus": 8,
|
| 142 |
+
"num_nodes": 4,
|
| 143 |
+
"log_dir": "/mnt/bn/robotics-data-lxh-lq-v2/logs/video_pretrain_manipulation/kosmos/calvin_finetune/2024-11-21/01-16",
|
| 144 |
+
"output_dir": "/mnt/bn/robotics-data-lxh-lq-v2/checkpoints/video_pretrain_manipulation/kosmos/calvin_finetune/2024-11-21/01-16",
|
| 145 |
+
"data_dir": null,
|
| 146 |
+
"annotation_file": null,
|
| 147 |
+
"data_subfolder": null,
|
| 148 |
+
"task_num": null,
|
| 149 |
+
"exp_name": "01-16",
|
| 150 |
+
"use_multi_modal_emb": false,
|
| 151 |
+
"no_video_pretrained_model": false,
|
| 152 |
+
"finetune": false,
|
| 153 |
+
"llm": {
|
| 154 |
+
"type": null,
|
| 155 |
+
"n_embd": null,
|
| 156 |
+
"n_layer": null,
|
| 157 |
+
"n_head": null
|
| 158 |
+
}
|
| 159 |
+
}
|