Upload stage1/stage1_sparsity_beta/0408_robotwin_liberoplus_stage1_latent_64_sparsity_1e-9/config.yaml with huggingface_hub

stage1/stage1_sparsity_beta/0408_robotwin_liberoplus_stage1_latent_64_sparsity_1e-9/config.yaml

@@ -0,0 +1,62 @@
+datasets:
+  vla_data:
+    data_mix: robotwin_mix_50_libero_plus
+    data_root_dir: /e2e-data/embodied-research-data/RobotData/IPEC-COMMUNITY
+    dataset_py: lerobot_datasets_lam
+    image_size:
+    - 224
+    - 224
+    per_device_batch_size: 32
+    training_stage: stage1_lam
+    video_backend: pyav
+  vla_data_eval:
+    data_mix: libero_all
+    data_root_dir: /e2e-data/embodied-research-data/RobotData/IPEC-COMMUNITY
+    dataset_py: lerobot_datasets_lam_eval
+    image_size:
+    - 224
+    - 224
+    training_stage: stage1_lam
+    video_backend: pyav
+framework:
+  lam:
+    beta: 1.0e-06
+    dec_blocks: 24
+    dropout: 0.0
+    enc_blocks: 24
+    image_channels: 3
+    lambda_l1: 0.001
+    lambda_sparsity: 1.0e-09
+    latent_dim: 64
+    model_dim: 1024
+    num_heads: 16
+    patch_size: 16
+  name: adaworld_sparsity
+output_dir: /e2e-data/evad-tech-vla/zhoulei21/LAM_Project/stage1/stage1_sparsity_beta/0408_robotwin_liberoplus_stage1_latent_64_sparsity_1e-9
+run_id: 0408_robotwin_liberoplus_stage1_latent_64_sparsity_1e-9
+run_root_dir: /e2e-data/evad-tech-vla/zhoulei21/LAM_Project/stage1/stage1_sparsity_beta
+seed: 42
+trainer:
+  eval_interval: 1000
+  freeze_modules: ''
+  gradient_accumulation_steps: 1
+  gradient_clipping: 1.0
+  is_resume: false
+  learning_rate:
+    base: 2.5e-05
+  logging_frequency: 10
+  lr_scheduler_type: constant
+  max_train_steps: 150000
+  num_warmup_steps: 5000
+  optimizer:
+    betas:
+    - 0.9
+    - 0.95
+    eps: 1.0e-08
+    weight_decay: 0.01
+  save_interval: 10000
+  scheduler_specific_kwargs:
+    min_lr: 5.0e-07
+training_stage: stage1_lam
+wandb_entity: your_wandb_entity
+wandb_project: llavavla