Zhengshen Zhang commited on
Upload folder using huggingface_hub
Browse files- logs/calvin_finetune_vlm_freeze_gr1_unfreeze_pure-gr-repeat/2025-04-07/19-58/version1/2025-04-07/19-51/19-51/version_0/events.out.tfevents.1744026758.n117-200-042.123145.0 +3 -0
- logs/calvin_finetune_vlm_freeze_gr1_unfreeze_pure-gr-repeat/2025-04-07/19-58/version1/2025-04-07/19-51/19-51/version_0/hparams.yaml +175 -0
- logs/calvin_finetune_vlm_freeze_gr1_unfreeze_pure-gr-repeat/2025-04-07/19-58/version1/2025-04-07/19-58/19-58/version_0/events.out.tfevents.1744027165.n117-200-042.145464.0 +3 -0
- logs/calvin_finetune_vlm_freeze_gr1_unfreeze_pure-gr-repeat/2025-04-07/19-58/version1/2025-04-07/19-58/19-58/version_0/hparams.yaml +175 -0
- logs/calvin_finetune_vlm_freeze_gr1_unfreeze_pure-gr-repeat/2025-04-07/19-58/version1/2025-04-07/19-58/2025-04-07_19:58:41.783287-project.json +188 -0
logs/calvin_finetune_vlm_freeze_gr1_unfreeze_pure-gr-repeat/2025-04-07/19-58/version1/2025-04-07/19-51/19-51/version_0/events.out.tfevents.1744026758.n117-200-042.123145.0
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:75004aeac84a166e758bed0c27065d6f70c370553a475cd0d3bd898099dbbb61
|
| 3 |
+
size 12878
|
logs/calvin_finetune_vlm_freeze_gr1_unfreeze_pure-gr-repeat/2025-04-07/19-58/version1/2025-04-07/19-51/19-51/version_0/hparams.yaml
ADDED
|
@@ -0,0 +1,175 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
configs:
|
| 2 |
+
robovlm_name: RoboKosMos
|
| 3 |
+
parent: null
|
| 4 |
+
task_name: calvin_finetune_vlm_freeze_gr1_unfreeze_pure-gr
|
| 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 |
+
image_std:
|
| 14 |
+
- 0.26862954
|
| 15 |
+
- 0.26130258
|
| 16 |
+
- 0.27577711
|
| 17 |
+
window_size: 10
|
| 18 |
+
fwd_pred_next_n: 10
|
| 19 |
+
arm_gripper_loss_ratio: 0.01
|
| 20 |
+
cap_loss_ratio: 0.05
|
| 21 |
+
fwd_loss_ratio: 0
|
| 22 |
+
seed: 123
|
| 23 |
+
batch_size: 12
|
| 24 |
+
num_workers: 32
|
| 25 |
+
data_scale: 1
|
| 26 |
+
optimizer: adam
|
| 27 |
+
learning_rate: 2.0e-05
|
| 28 |
+
min_lr_scale: 0.01
|
| 29 |
+
weight_decay: 0
|
| 30 |
+
warmup_epochs: 0.25
|
| 31 |
+
warmup_steps: 0
|
| 32 |
+
warmup_ratio: null
|
| 33 |
+
use_hand_rgb: true
|
| 34 |
+
use_time_causal_attn: false
|
| 35 |
+
use_mim_obs_loss: false
|
| 36 |
+
use_pixel_loss: true
|
| 37 |
+
use_obs_queries: true
|
| 38 |
+
use_vision_resampler: false
|
| 39 |
+
vision_masked_ratio: 0.9
|
| 40 |
+
use_tube_mask: false
|
| 41 |
+
cache_root: runs/cache/kosmos
|
| 42 |
+
model_load_path: null
|
| 43 |
+
model_load_source: torch
|
| 44 |
+
resume: null
|
| 45 |
+
model_path: /mnt/hdfs/public/zhengshen/vlm_ckpts/kosmos-2-patch14-224
|
| 46 |
+
model_config: /mnt/hdfs/public/zhengshen/vlm_ckpts/kosmos-2-patch14-224/config.json
|
| 47 |
+
tcp_rel: false
|
| 48 |
+
use_clip_norm: false
|
| 49 |
+
use_state: false
|
| 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: false
|
| 57 |
+
bits: -1
|
| 58 |
+
freeze_mm_mlp_adapter: true
|
| 59 |
+
freeze_backbone: true
|
| 60 |
+
freeze_resampler: true
|
| 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: false
|
| 72 |
+
train_act_head: true
|
| 73 |
+
train_decoder_layers: -1
|
| 74 |
+
vision_resampler:
|
| 75 |
+
vis_dim: 1024
|
| 76 |
+
depth: 8
|
| 77 |
+
dim_head: 64
|
| 78 |
+
heads: 8
|
| 79 |
+
num_latents: 64
|
| 80 |
+
act_encoder: null
|
| 81 |
+
act_head:
|
| 82 |
+
type: GR1
|
| 83 |
+
hidden_size: 384
|
| 84 |
+
action_dim: 7
|
| 85 |
+
down_sample: none
|
| 86 |
+
latent: 1
|
| 87 |
+
fwd_pred_next_n: 10
|
| 88 |
+
action_space: continuous
|
| 89 |
+
with_history: true
|
| 90 |
+
history_type: post
|
| 91 |
+
state_dim: 7
|
| 92 |
+
sequence_length: 10
|
| 93 |
+
training_target: act_pred
|
| 94 |
+
img_feat_dim: 768
|
| 95 |
+
lang_feat_dim: 512
|
| 96 |
+
patch_feat_dim: 768
|
| 97 |
+
resampler_depth: 3
|
| 98 |
+
resampler_dim_head: 128
|
| 99 |
+
resampler_heads: 4
|
| 100 |
+
resampler_num_media_embeds: 1
|
| 101 |
+
resampler_num_latents: 9
|
| 102 |
+
without_norm_pix_loss: false
|
| 103 |
+
use_hand_rgb: true
|
| 104 |
+
n_layer: 12
|
| 105 |
+
n_head: 12
|
| 106 |
+
n_inner: 1536
|
| 107 |
+
activation_function: relu
|
| 108 |
+
n_positions: 1024
|
| 109 |
+
resid_pdrop: 0.1
|
| 110 |
+
attn_pdrop: 0.1
|
| 111 |
+
action_mode: ee_rel_state
|
| 112 |
+
clip_backbone: ViT-B/32
|
| 113 |
+
mae_ckpt: /mnt/hdfs/public/zhengshen/vlm_ckpts/mae_pretrain_vit_base.pth
|
| 114 |
+
policy_ckpt: null
|
| 115 |
+
act_pred: true
|
| 116 |
+
fwd_pred: false
|
| 117 |
+
fwd_pred_hand: false
|
| 118 |
+
fwd_head: null
|
| 119 |
+
tokenizer:
|
| 120 |
+
type: AutoProcessor
|
| 121 |
+
pretrained_model_name_or_path: /mnt/hdfs/public/zhengshen/vlm_ckpts/kosmos-2-patch14-224
|
| 122 |
+
tokenizer_type: kosmos
|
| 123 |
+
max_text_len: 256
|
| 124 |
+
additional_special_tokens: null
|
| 125 |
+
vlm:
|
| 126 |
+
type: AutoModelForVision2Seq
|
| 127 |
+
name: kosmos
|
| 128 |
+
pretrained_model_name_or_path: /mnt/hdfs/public/zhengshen/vlm_ckpts/kosmos-2-patch14-224
|
| 129 |
+
trainer:
|
| 130 |
+
accelerator: gpu
|
| 131 |
+
strategy: deepspeed_stage_2
|
| 132 |
+
precision: '16'
|
| 133 |
+
logger:
|
| 134 |
+
- tensorboard
|
| 135 |
+
gradient_clip_val: 1.0
|
| 136 |
+
use_distributed_sampler: false
|
| 137 |
+
log_every_n_steps: 10
|
| 138 |
+
max_epochs: 100
|
| 139 |
+
val_check_interval: null
|
| 140 |
+
check_val_every_n_epoch: 1
|
| 141 |
+
max_steps: -1
|
| 142 |
+
accumulate_grad_batches: 1
|
| 143 |
+
train_dataset:
|
| 144 |
+
type: DiskCalvinDataset
|
| 145 |
+
data_dir: /opt/tiger/robogen/task_ABC_D/training
|
| 146 |
+
shift_first: false
|
| 147 |
+
model_name: kosmos
|
| 148 |
+
rgb_pad: 10
|
| 149 |
+
gripper_pad: 4
|
| 150 |
+
val_dataset:
|
| 151 |
+
type: DiskCalvinDataset
|
| 152 |
+
data_dir: /opt/tiger/robogen/task_ABC_D/validation
|
| 153 |
+
model_name: kosmos
|
| 154 |
+
norm_action: true
|
| 155 |
+
norm_min: -0.65
|
| 156 |
+
norm_max: 0.65
|
| 157 |
+
raw_config_path: configs/calvin_finetune_hdfs/gr1/vlm_freeze_gr1_unfreeze/finetune_kosmos_cont-gr1-post_full-ft_gr_wd-0_all-use-hand_ws-10_act-10.json
|
| 158 |
+
config: configs/calvin_finetune_hdfs/gr1/vlm_freeze_gr1_unfreeze/finetune_kosmos_cont-gr1-post_full-ft_gr_wd-0_all-use-hand_ws-10_act-10.json
|
| 159 |
+
gpus: 8
|
| 160 |
+
num_nodes: 1
|
| 161 |
+
log_dir: runs/logs/kosmos/calvin_finetune_vlm_freeze_gr1_unfreeze_pure-gr/2025-04-07/19-51
|
| 162 |
+
output_dir: runs/checkpoints/kosmos/calvin_finetune_vlm_freeze_gr1_unfreeze_pure-gr/2025-04-07/19-51
|
| 163 |
+
data_dir: null
|
| 164 |
+
annotation_file: null
|
| 165 |
+
data_subfolder: null
|
| 166 |
+
task_num: null
|
| 167 |
+
exp_name: 19-51
|
| 168 |
+
use_multi_modal_emb: false
|
| 169 |
+
no_video_pretrained_model: false
|
| 170 |
+
finetune: false
|
| 171 |
+
llm:
|
| 172 |
+
type: null
|
| 173 |
+
n_embd: null
|
| 174 |
+
n_layer: null
|
| 175 |
+
n_head: null
|
logs/calvin_finetune_vlm_freeze_gr1_unfreeze_pure-gr-repeat/2025-04-07/19-58/version1/2025-04-07/19-58/19-58/version_0/events.out.tfevents.1744027165.n117-200-042.145464.0
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:8c2646dfb8a9a24afcd2ac0afb834154979bcfae30b5f0d0fc4e3b32a752698e
|
| 3 |
+
size 5883096
|
logs/calvin_finetune_vlm_freeze_gr1_unfreeze_pure-gr-repeat/2025-04-07/19-58/version1/2025-04-07/19-58/19-58/version_0/hparams.yaml
ADDED
|
@@ -0,0 +1,175 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
configs:
|
| 2 |
+
robovlm_name: RoboKosMos
|
| 3 |
+
parent: null
|
| 4 |
+
task_name: calvin_finetune_vlm_freeze_gr1_unfreeze_pure-gr
|
| 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 |
+
image_std:
|
| 14 |
+
- 0.26862954
|
| 15 |
+
- 0.26130258
|
| 16 |
+
- 0.27577711
|
| 17 |
+
window_size: 10
|
| 18 |
+
fwd_pred_next_n: 10
|
| 19 |
+
arm_gripper_loss_ratio: 0.01
|
| 20 |
+
cap_loss_ratio: 0.05
|
| 21 |
+
fwd_loss_ratio: 0
|
| 22 |
+
seed: 123
|
| 23 |
+
batch_size: 4
|
| 24 |
+
num_workers: 32
|
| 25 |
+
data_scale: 1
|
| 26 |
+
optimizer: adam
|
| 27 |
+
learning_rate: 2.0e-05
|
| 28 |
+
min_lr_scale: 0.01
|
| 29 |
+
weight_decay: 0
|
| 30 |
+
warmup_epochs: 0.25
|
| 31 |
+
warmup_steps: 0
|
| 32 |
+
warmup_ratio: null
|
| 33 |
+
use_hand_rgb: true
|
| 34 |
+
use_time_causal_attn: false
|
| 35 |
+
use_mim_obs_loss: false
|
| 36 |
+
use_pixel_loss: true
|
| 37 |
+
use_obs_queries: true
|
| 38 |
+
use_vision_resampler: false
|
| 39 |
+
vision_masked_ratio: 0.9
|
| 40 |
+
use_tube_mask: false
|
| 41 |
+
cache_root: runs/cache/kosmos
|
| 42 |
+
model_load_path: null
|
| 43 |
+
model_load_source: torch
|
| 44 |
+
resume: null
|
| 45 |
+
model_path: /mnt/hdfs/public/zhengshen/vlm_ckpts/kosmos-2-patch14-224
|
| 46 |
+
model_config: /mnt/hdfs/public/zhengshen/vlm_ckpts/kosmos-2-patch14-224/config.json
|
| 47 |
+
tcp_rel: false
|
| 48 |
+
use_clip_norm: false
|
| 49 |
+
use_state: false
|
| 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: false
|
| 57 |
+
bits: -1
|
| 58 |
+
freeze_mm_mlp_adapter: true
|
| 59 |
+
freeze_backbone: true
|
| 60 |
+
freeze_resampler: true
|
| 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: false
|
| 72 |
+
train_act_head: true
|
| 73 |
+
train_decoder_layers: -1
|
| 74 |
+
vision_resampler:
|
| 75 |
+
vis_dim: 1024
|
| 76 |
+
depth: 8
|
| 77 |
+
dim_head: 64
|
| 78 |
+
heads: 8
|
| 79 |
+
num_latents: 64
|
| 80 |
+
act_encoder: null
|
| 81 |
+
act_head:
|
| 82 |
+
type: GR1
|
| 83 |
+
hidden_size: 384
|
| 84 |
+
action_dim: 7
|
| 85 |
+
down_sample: none
|
| 86 |
+
latent: 1
|
| 87 |
+
fwd_pred_next_n: 10
|
| 88 |
+
action_space: continuous
|
| 89 |
+
with_history: true
|
| 90 |
+
history_type: post
|
| 91 |
+
state_dim: 7
|
| 92 |
+
sequence_length: 10
|
| 93 |
+
training_target: act_pred
|
| 94 |
+
img_feat_dim: 768
|
| 95 |
+
lang_feat_dim: 512
|
| 96 |
+
patch_feat_dim: 768
|
| 97 |
+
resampler_depth: 3
|
| 98 |
+
resampler_dim_head: 128
|
| 99 |
+
resampler_heads: 4
|
| 100 |
+
resampler_num_media_embeds: 1
|
| 101 |
+
resampler_num_latents: 9
|
| 102 |
+
without_norm_pix_loss: false
|
| 103 |
+
use_hand_rgb: true
|
| 104 |
+
n_layer: 12
|
| 105 |
+
n_head: 12
|
| 106 |
+
n_inner: 1536
|
| 107 |
+
activation_function: relu
|
| 108 |
+
n_positions: 1024
|
| 109 |
+
resid_pdrop: 0.1
|
| 110 |
+
attn_pdrop: 0.1
|
| 111 |
+
action_mode: ee_rel_state
|
| 112 |
+
clip_backbone: ViT-B/32
|
| 113 |
+
mae_ckpt: /mnt/hdfs/public/zhengshen/vlm_ckpts/mae_pretrain_vit_base.pth
|
| 114 |
+
policy_ckpt: null
|
| 115 |
+
act_pred: true
|
| 116 |
+
fwd_pred: false
|
| 117 |
+
fwd_pred_hand: false
|
| 118 |
+
fwd_head: null
|
| 119 |
+
tokenizer:
|
| 120 |
+
type: AutoProcessor
|
| 121 |
+
pretrained_model_name_or_path: /mnt/hdfs/public/zhengshen/vlm_ckpts/kosmos-2-patch14-224
|
| 122 |
+
tokenizer_type: kosmos
|
| 123 |
+
max_text_len: 256
|
| 124 |
+
additional_special_tokens: null
|
| 125 |
+
vlm:
|
| 126 |
+
type: AutoModelForVision2Seq
|
| 127 |
+
name: kosmos
|
| 128 |
+
pretrained_model_name_or_path: /mnt/hdfs/public/zhengshen/vlm_ckpts/kosmos-2-patch14-224
|
| 129 |
+
trainer:
|
| 130 |
+
accelerator: gpu
|
| 131 |
+
strategy: deepspeed_stage_2
|
| 132 |
+
precision: '16'
|
| 133 |
+
logger:
|
| 134 |
+
- tensorboard
|
| 135 |
+
gradient_clip_val: 1.0
|
| 136 |
+
use_distributed_sampler: false
|
| 137 |
+
log_every_n_steps: 10
|
| 138 |
+
max_epochs: 100
|
| 139 |
+
val_check_interval: null
|
| 140 |
+
check_val_every_n_epoch: 1
|
| 141 |
+
max_steps: -1
|
| 142 |
+
accumulate_grad_batches: 1
|
| 143 |
+
train_dataset:
|
| 144 |
+
type: DiskCalvinDataset
|
| 145 |
+
data_dir: /opt/tiger/robogen/task_ABC_D/training
|
| 146 |
+
shift_first: false
|
| 147 |
+
model_name: kosmos
|
| 148 |
+
rgb_pad: 10
|
| 149 |
+
gripper_pad: 4
|
| 150 |
+
val_dataset:
|
| 151 |
+
type: DiskCalvinDataset
|
| 152 |
+
data_dir: /opt/tiger/robogen/task_ABC_D/validation
|
| 153 |
+
model_name: kosmos
|
| 154 |
+
norm_action: true
|
| 155 |
+
norm_min: -0.65
|
| 156 |
+
norm_max: 0.65
|
| 157 |
+
raw_config_path: configs/calvin_finetune_hdfs/gr1/vlm_freeze_gr1_unfreeze/finetune_kosmos_cont-gr1-post_full-ft_gr_wd-0_all-use-hand_ws-10_act-10.json
|
| 158 |
+
config: configs/calvin_finetune_hdfs/gr1/vlm_freeze_gr1_unfreeze/finetune_kosmos_cont-gr1-post_full-ft_gr_wd-0_all-use-hand_ws-10_act-10.json
|
| 159 |
+
gpus: 8
|
| 160 |
+
num_nodes: 1
|
| 161 |
+
log_dir: runs/logs/kosmos/calvin_finetune_vlm_freeze_gr1_unfreeze_pure-gr/2025-04-07/19-58
|
| 162 |
+
output_dir: runs/checkpoints/kosmos/calvin_finetune_vlm_freeze_gr1_unfreeze_pure-gr/2025-04-07/19-58
|
| 163 |
+
data_dir: null
|
| 164 |
+
annotation_file: null
|
| 165 |
+
data_subfolder: null
|
| 166 |
+
task_num: null
|
| 167 |
+
exp_name: 19-58
|
| 168 |
+
use_multi_modal_emb: false
|
| 169 |
+
no_video_pretrained_model: false
|
| 170 |
+
finetune: false
|
| 171 |
+
llm:
|
| 172 |
+
type: null
|
| 173 |
+
n_embd: null
|
| 174 |
+
n_layer: null
|
| 175 |
+
n_head: null
|
logs/calvin_finetune_vlm_freeze_gr1_unfreeze_pure-gr-repeat/2025-04-07/19-58/version1/2025-04-07/19-58/2025-04-07_19:58:41.783287-project.json
ADDED
|
@@ -0,0 +1,188 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"robovlm_name": "RoboKosMos",
|
| 3 |
+
"parent": null,
|
| 4 |
+
"task_name": "calvin_finetune_vlm_freeze_gr1_unfreeze_pure-gr",
|
| 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": 10,
|
| 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": 32,
|
| 27 |
+
"data_scale": 1,
|
| 28 |
+
"optimizer": "adam",
|
| 29 |
+
"learning_rate": 2e-05,
|
| 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 |
+
"cache_root": "runs/cache/kosmos",
|
| 44 |
+
"model_load_path": null,
|
| 45 |
+
"model_load_source": "torch",
|
| 46 |
+
"resume": null,
|
| 47 |
+
"model_path": "/mnt/hdfs/public/zhengshen/vlm_ckpts/kosmos-2-patch14-224",
|
| 48 |
+
"model_config": "/mnt/hdfs/public/zhengshen/vlm_ckpts/kosmos-2-patch14-224/config.json",
|
| 49 |
+
"tcp_rel": false,
|
| 50 |
+
"use_clip_norm": false,
|
| 51 |
+
"use_state": false,
|
| 52 |
+
"train_setup": {
|
| 53 |
+
"precision": "16",
|
| 54 |
+
"predict_action": true,
|
| 55 |
+
"predict_forward": false,
|
| 56 |
+
"predict_forward_hand": false,
|
| 57 |
+
"predict_caption": false,
|
| 58 |
+
"train_vision": false,
|
| 59 |
+
"bits": -1,
|
| 60 |
+
"freeze_mm_mlp_adapter": true,
|
| 61 |
+
"freeze_backbone": true,
|
| 62 |
+
"freeze_resampler": true,
|
| 63 |
+
"tune_mm_mlp_adapter": false,
|
| 64 |
+
"mm_use_im_start_end": false,
|
| 65 |
+
"mm_use_im_patch_token": false,
|
| 66 |
+
"gradient_checkpointing": false,
|
| 67 |
+
"lora_enable": false,
|
| 68 |
+
"mm_projector_lr": 0.0001,
|
| 69 |
+
"lora_r": 64,
|
| 70 |
+
"lora_alpha": 16,
|
| 71 |
+
"lora_dropout": 0.05,
|
| 72 |
+
"lora_bias": "none",
|
| 73 |
+
"train_text_embedding": false,
|
| 74 |
+
"train_act_head": true,
|
| 75 |
+
"train_decoder_layers": -1
|
| 76 |
+
},
|
| 77 |
+
"vision_resampler": {
|
| 78 |
+
"vis_dim": 1024,
|
| 79 |
+
"depth": 8,
|
| 80 |
+
"dim_head": 64,
|
| 81 |
+
"heads": 8,
|
| 82 |
+
"num_latents": 64
|
| 83 |
+
},
|
| 84 |
+
"act_encoder": null,
|
| 85 |
+
"act_head": {
|
| 86 |
+
"type": "GR1",
|
| 87 |
+
"hidden_size": 384,
|
| 88 |
+
"action_dim": 7,
|
| 89 |
+
"down_sample": "none",
|
| 90 |
+
"latent": 1,
|
| 91 |
+
"fwd_pred_next_n": 10,
|
| 92 |
+
"action_space": "continuous",
|
| 93 |
+
"with_history": true,
|
| 94 |
+
"history_type": "post",
|
| 95 |
+
"state_dim": 7,
|
| 96 |
+
"sequence_length": 10,
|
| 97 |
+
"training_target": "act_pred",
|
| 98 |
+
"img_feat_dim": 768,
|
| 99 |
+
"lang_feat_dim": 512,
|
| 100 |
+
"patch_feat_dim": 768,
|
| 101 |
+
"resampler_depth": 3,
|
| 102 |
+
"resampler_dim_head": 128,
|
| 103 |
+
"resampler_heads": 4,
|
| 104 |
+
"resampler_num_media_embeds": 1,
|
| 105 |
+
"resampler_num_latents": 9,
|
| 106 |
+
"without_norm_pix_loss": false,
|
| 107 |
+
"use_hand_rgb": true,
|
| 108 |
+
"n_layer": 12,
|
| 109 |
+
"n_head": 12,
|
| 110 |
+
"n_inner": 1536,
|
| 111 |
+
"activation_function": "relu",
|
| 112 |
+
"n_positions": 1024,
|
| 113 |
+
"resid_pdrop": 0.1,
|
| 114 |
+
"attn_pdrop": 0.1,
|
| 115 |
+
"action_mode": "ee_rel_state",
|
| 116 |
+
"clip_backbone": "ViT-B/32",
|
| 117 |
+
"mae_ckpt": "/mnt/hdfs/public/zhengshen/vlm_ckpts/mae_pretrain_vit_base.pth",
|
| 118 |
+
"policy_ckpt": null,
|
| 119 |
+
"act_pred": true,
|
| 120 |
+
"fwd_pred": false,
|
| 121 |
+
"fwd_pred_hand": false
|
| 122 |
+
},
|
| 123 |
+
"fwd_head": null,
|
| 124 |
+
"tokenizer": {
|
| 125 |
+
"type": "AutoProcessor",
|
| 126 |
+
"pretrained_model_name_or_path": "/mnt/hdfs/public/zhengshen/vlm_ckpts/kosmos-2-patch14-224",
|
| 127 |
+
"tokenizer_type": "kosmos",
|
| 128 |
+
"max_text_len": 256,
|
| 129 |
+
"additional_special_tokens": null
|
| 130 |
+
},
|
| 131 |
+
"vlm": {
|
| 132 |
+
"type": "AutoModelForVision2Seq",
|
| 133 |
+
"name": "kosmos",
|
| 134 |
+
"pretrained_model_name_or_path": "/mnt/hdfs/public/zhengshen/vlm_ckpts/kosmos-2-patch14-224"
|
| 135 |
+
},
|
| 136 |
+
"trainer": {
|
| 137 |
+
"accelerator": "gpu",
|
| 138 |
+
"strategy": "deepspeed_stage_2",
|
| 139 |
+
"precision": "16",
|
| 140 |
+
"logger": [
|
| 141 |
+
"tensorboard"
|
| 142 |
+
],
|
| 143 |
+
"gradient_clip_val": 1.0,
|
| 144 |
+
"use_distributed_sampler": false,
|
| 145 |
+
"log_every_n_steps": 10,
|
| 146 |
+
"max_epochs": 100,
|
| 147 |
+
"val_check_interval": null,
|
| 148 |
+
"check_val_every_n_epoch": 1,
|
| 149 |
+
"max_steps": -1,
|
| 150 |
+
"accumulate_grad_batches": 1
|
| 151 |
+
},
|
| 152 |
+
"train_dataset": {
|
| 153 |
+
"type": "DiskCalvinDataset",
|
| 154 |
+
"data_dir": "/opt/tiger/robogen/task_ABC_D/training",
|
| 155 |
+
"shift_first": false,
|
| 156 |
+
"model_name": "kosmos",
|
| 157 |
+
"rgb_pad": 10,
|
| 158 |
+
"gripper_pad": 4
|
| 159 |
+
},
|
| 160 |
+
"val_dataset": {
|
| 161 |
+
"type": "DiskCalvinDataset",
|
| 162 |
+
"data_dir": "/opt/tiger/robogen/task_ABC_D/validation",
|
| 163 |
+
"model_name": "kosmos"
|
| 164 |
+
},
|
| 165 |
+
"norm_action": true,
|
| 166 |
+
"norm_min": -0.65,
|
| 167 |
+
"norm_max": 0.65,
|
| 168 |
+
"raw_config_path": "configs/calvin_finetune_hdfs/gr1/vlm_freeze_gr1_unfreeze/finetune_kosmos_cont-gr1-post_full-ft_gr_wd-0_all-use-hand_ws-10_act-10.json",
|
| 169 |
+
"config": "configs/calvin_finetune_hdfs/gr1/vlm_freeze_gr1_unfreeze/finetune_kosmos_cont-gr1-post_full-ft_gr_wd-0_all-use-hand_ws-10_act-10.json",
|
| 170 |
+
"gpus": 8,
|
| 171 |
+
"num_nodes": 1,
|
| 172 |
+
"log_dir": "runs/logs/kosmos/calvin_finetune_vlm_freeze_gr1_unfreeze_pure-gr/2025-04-07/19-58",
|
| 173 |
+
"output_dir": "runs/checkpoints/kosmos/calvin_finetune_vlm_freeze_gr1_unfreeze_pure-gr/2025-04-07/19-58",
|
| 174 |
+
"data_dir": null,
|
| 175 |
+
"annotation_file": null,
|
| 176 |
+
"data_subfolder": null,
|
| 177 |
+
"task_num": null,
|
| 178 |
+
"exp_name": "19-58",
|
| 179 |
+
"use_multi_modal_emb": false,
|
| 180 |
+
"no_video_pretrained_model": false,
|
| 181 |
+
"finetune": false,
|
| 182 |
+
"llm": {
|
| 183 |
+
"type": null,
|
| 184 |
+
"n_embd": null,
|
| 185 |
+
"n_layer": null,
|
| 186 |
+
"n_head": null
|
| 187 |
+
}
|
| 188 |
+
}
|