diff --git a/scripts/configs/pizeoelectric.yaml b/scripts/configs/pizeoelectric.yaml
deleted file mode 100644
index ebae2c0..0000000
--- a/scripts/configs/pizeoelectric.yaml
+++ /dev/null
@@ -1,116 +0,0 @@
-seed_everything: 35
-log_level: info
-
-data:
-  root: .
-  tensor_target_name: piezoelectric_tensor_total
-  tensor_target_formula: ijk=ikj
-  trainset_filename: /Users/mjwen.admin/Documents/Dataset/di_pizeoelectric_tensor/piezoelectric_tensors_n20.json
-  valset_filename: /Users/mjwen.admin/Documents/Dataset/di_pizeoelectric_tensor/piezoelectric_tensors_n20.json
-  testset_filename: /Users/mjwen.admin/Documents/Dataset/di_pizeoelectric_tensor/piezoelectric_tensors_n20.json
-  r_cut: 5.0
-  reuse: false
-  loader_kwargs:
-    batch_size: 32
-    shuffle: true
-
-model:
-  ##########
-  # embedding
-  ##########
-
-  # atom species embedding
-  species_embedding_dim: 16
-
-  # spherical harmonics embedding of edge direction
-  irreps_edge_sh: 0e + 1o + 2e + 3o
-
-  # radial edge distance embedding
-  radial_basis_type: bessel
-  num_radial_basis: 8
-  radial_basis_start: 0.
-  radial_basis_end: 5.
-
-  ##########
-  # message passing conv layers
-  ##########
-  num_layers: 3
-
-  # radial network
-  invariant_layers: 2 # number of radial layers
-  invariant_neurons: 32 # number of hidden neurons in radial function
-
-  # Average number of neighbors used for normalization. Options:
-  # 1. `auto` to determine it automatically, by setting it to average number
-  #    of neighbors of the training set
-  # 2. float or int provided here.
-  # 3. `null` to not use it
-  average_num_neighbors: auto
-
-  # point convolution
-  conv_layer_irreps: 32x0o+32x0e + 16x1o+16x1e + 4x2o+4x2e + 2x3o+2x3e
-  nonlinearity_type: gate
-  normalization: batch
-  resnet: true
-
-  ##########
-  # output
-  ##########
-
-  conv_to_output_hidden_irreps_out: 16x1o + 4x2o + 2x3o
-
-  # output_format and output_formula should be used together.
-  # - output_format (can be `irreps` or `cartesian`) determines what the loss
-  #   function will be on (either on the irreps space or the cartesian space).
-  # - output_formula gives what the cartesian formula of the tensor is.
-  #   For example, ijkl=jikl=klij specifies a forth-rank elasticity tensor.
-  output_format: irreps
-  output_formula: ijk=ikj
-
-  # pooling node feats to graph feats
-  reduce: mean
-
-trainer:
-  max_epochs: 10 # number of maximum training epochs
-  num_nodes: 1
-  accelerator: cpu
-  devices: 1
-
-  callbacks:
-    - class_path: pytorch_lightning.callbacks.ModelCheckpoint
-      init_args:
-        monitor: val/score
-        mode: min
-        save_top_k: 3
-        save_last: true
-        verbose: false
-    - class_path: pytorch_lightning.callbacks.EarlyStopping
-      init_args:
-        monitor: val/score
-        mode: min
-        patience: 150
-        min_delta: 0
-        verbose: true
-    - class_path: pytorch_lightning.callbacks.ModelSummary
-      init_args:
-        max_depth: -1
-
-  #logger:
-  #  class_path: pytorch_lightning.loggers.wandb.WandbLogger
-  #  init_args:
-  #    save_dir: matten_logs
-  #    project: matten_proj
-
-optimizer:
-  class_path: torch.optim.Adam
-  init_args:
-    lr: 0.01
-    weight_decay: 0.00001
-
-lr_scheduler:
-  class_path: torch.optim.lr_scheduler.ReduceLROnPlateau
-  init_args:
-    mode: min
-    factor: 0.5
-    patience: 50
-    verbose: true