Deleted old configs

configs/architectures/double_subs.yml

-name: shapenet_architectures
-# pretrained: None
-
-# data
-dataset: shapenet
-subclass: chair
-transform: [scaling, warping, linear_max_res]
-position_encoding: intertwined
-resolution: 128
-spatial_dim: 3
-num_vocab: 3
-
-# architecture
-architecture: encoder_multi_decoder
-embedding: [composite, single_conv, substitution, double_substitution]
-head: [composite, single_conv, substitution, double_substitution]
-
-# attention
-attention: basic
-num_positions: 2500
-embed_dim: 256
-num_layers: 8
-num_heads: 4
-
-# training
-loss_function: cross_entropy
-val_loss_function: cross_entropy
-epochs: 500
-warmup_steps: 0.1
-batch_size: 1
-accumulate_grad_batches: 1
-learning_rate: 0.00003
-dropout: 0.0
-
-# hardware
-gpus: 8
-precision: 32
-
-# logging
-log_gpu: 'min_max'
-log_gradient: False
-log_weights_and_biases: False
-log_learning_rate: True
-parameter_search: False
\ No newline at end of file

configs/architectures/encoder_decoder.yml

-name: shapenet_architectures
-# pretrained: None
-
-# data
-dataset: shapenet
-subclass: chair
-transform: [scaling, warping, linear_max_res]
-position_encoding: intertwined
-resolution: 128
-spatial_dim: 3
-num_vocab: 3
-
-# architecture
-architecture: encoder_decoder
-embedding: [composite, substitution]
-head: [substitution]
-
-# attention
-attention: basic
-num_positions: 3072
-embed_dim: 256
-num_layers: 8
-num_heads: 4
-
-# training
-loss_function: cross_entropy
-val_loss_function: cross_entropy
-epochs: 500
-warmup_steps: 0.1
-batch_size: 1
-accumulate_grad_batches: 1
-learning_rate: 0.00003
-dropout: 0.0
-
-# hardware
-gpus: 8
-precision: 32
-
-# logging
-log_gpu: 'min_max'
-log_gradient: False
-log_weights_and_biases: False
-log_learning_rate: True
-parameter_search: False
\ No newline at end of file

configs/architectures/encoder_only.yml

-name: shapenet_architectures
-# pretrained: None
-
-# data
-dataset: shapenet
-subclass: chair
-transform: [scaling, warping, linear_max_res]
-position_encoding: intertwined
-resolution: 128
-spatial_dim: 3
-num_vocab: 3
-
-# architecture
-architecture: encoder_only
-embedding: [composite]
-head: [composite]
-
-# attention
-attention: basic
-num_positions: 3072
-embed_dim: 256
-num_layers: 8
-num_heads: 4
-
-# training
-loss_function: cross_entropy
-val_loss_function: cross_entropy
-epochs: 500
-warmup_steps: 0.1
-batch_size: 1
-accumulate_grad_batches: 1
-learning_rate: 0.00003
-dropout: 0.0
-
-# hardware
-gpus: 8
-precision: 32
-
-# logging
-log_gpu: 'min_max'
-log_gradient: False
-log_weights_and_biases: False
-log_learning_rate: True
-parameter_search: False
\ No newline at end of file

configs/architectures_small/double_subs.yml

-name: shapenet_architectures_small
-# pretrained: None
-
-# data
-dataset: shapenet
-subclass: chair
-transform: [scaling, warping, linear_max_res]
-position_encoding: intertwined
-resolution: 64
-spatial_dim: 3
-num_vocab: 3
-
-# architecture
-architecture: encoder_multi_decoder
-embedding: [composite_C, single_conv, substitution, double_substitution]
-head: [composite_C, single_conv, substitution, double_substitution]
-
-# attention
-attention: basic
-num_positions: 2048
-embed_dim: 256
-num_layers: 8
-num_heads: 4
-
-# training
-loss_function: cross_entropy
-val_loss_function: cross_entropy
-epochs: 369
-warmup_steps: 0.1
-batch_size: 4
-accumulate_grad_batches: 1
-learning_rate: 0.00003
-dropout: 0.1
-
-# hardware
-gpus: 1
-precision: 32
-
-# logging
-log_gpu: 'min_max'
-log_gradient: False
-log_weights_and_biases: False
-log_learning_rate: True
-parameter_search: False
\ No newline at end of file

configs/architectures_small/encoder_decoder.yml

-name: shapenet_architectures_small
-# pretrained: None
-
-# data
-dataset: shapenet
-subclass: chair
-transform: [scaling, warping, linear_max_res]
-position_encoding: intertwined
-resolution: 64
-spatial_dim: 3
-num_vocab: 3
-
-# architecture
-architecture: encoder_decoder
-embedding: [composite_C, substitution]
-head: [substitution]
-
-# attention
-attention: basic
-num_positions: 2048
-embed_dim: 256
-num_layers: 8
-num_heads: 4
-
-# training
-loss_function: cross_entropy
-val_loss_function: cross_entropy
-epochs: 369
-warmup_steps: 0.1
-batch_size: 4
-accumulate_grad_batches: 1
-learning_rate: 0.00003
-dropout: 0.1
-
-# hardware
-gpus: 1
-precision: 32
-
-# logging
-log_gpu: 'min_max'
-log_gradient: False
-log_weights_and_biases: False
-log_learning_rate: True
-parameter_search: False
\ No newline at end of file

configs/architectures_small/encoder_only.yml

-name: shapenet_architectures_small
-# pretrained: None
-
-# data
-dataset: shapenet
-subclass: chair
-transform: [scaling, warping, linear_max_res]
-position_encoding: intertwined
-resolution: 64
-spatial_dim: 3
-num_vocab: 3
-
-# architecture
-architecture: encoder_only
-embedding: [composite_A]
-head: [composite_A]
-
-# attention
-attention: basic
-num_positions: 2048
-embed_dim: 512
-num_layers: 12
-num_heads: 8
-
-# training
-loss_function: cross_entropy
-val_loss_function: cross_entropy
-epochs: 97
-warmup_steps: 0.1
-batch_size: 1
-accumulate_grad_batches: 1
-learning_rate: 0.00003
-dropout: 0.2
-
-# hardware
-gpus: 1
-precision: 32
-
-# logging
-log_gpu: 'min_max'
-log_gradient: False
-log_weights_and_biases: False
-log_learning_rate: True
-parameter_search: False
\ No newline at end of file

configs/dropout/low.yml

-name: shapenet_dropout
-# pretrained: None
-
-# data
-dataset: shapenet
-subclass: chair
-transform: [scaling, warping, linear_max_res]
-position_encoding: intertwined
-resolution: 128
-spatial_dim: 3
-num_vocab: 3
-
-# architecture
-architecture: encoder_multi_decoder
-embedding: [composite, single_conv, substitution]
-head: [composite, single_conv, substitution]
-
-# attention
-attention: basic
-num_positions: 2048
-embed_dim: 256
-num_layers: 8
-num_heads: 4
-
-# training
-loss_function: cross_entropy
-val_loss_function: cross_entropy
-epochs: 500
-warmup_steps: 0.1
-batch_size: 1
-accumulate_grad_batches: 1
-learning_rate: 0.00003
-dropout: 0.1
-
-# hardware
-gpus: 8
-precision: 32
-
-# logging
-log_gpu: 'min_max'
-log_gradient: False
-log_weights_and_biases: False
-log_learning_rate: True
-parameter_search: False
\ No newline at end of file

configs/dropout/medium.yml

-name: shapenet_dropout
-# pretrained: None
-
-# data
-dataset: shapenet
-subclass: chair
-transform: [scaling, warping, linear_max_res]
-position_encoding: intertwined
-resolution: 128
-spatial_dim: 3
-num_vocab: 3
-
-# architecture
-architecture: encoder_multi_decoder
-embedding: [composite, single_conv, substitution]
-head: [composite, single_conv, substitution]
-
-# attention
-attention: basic
-num_positions: 2500
-embed_dim: 256
-num_layers: 8
-num_heads: 4
-
-# training
-loss_function: cross_entropy
-val_loss_function: cross_entropy
-epochs: 500
-warmup_steps: 0.1
-batch_size: 1
-accumulate_grad_batches: 1
-learning_rate: 0.00003
-dropout: 0.2
-
-# hardware
-gpus: 8
-precision: 32
-
-# logging
-log_gpu: 'min_max'
-log_gradient: False
-log_weights_and_biases: False
-log_learning_rate: True
-parameter_search: False
\ No newline at end of file

configs/mnist_debug.yml

-name: mnist_debug
-dataset: mnist
-
-# training
-epochs: 5
-warmup_steps: 0.2
-batch_size: 64
-accumulate_grad_batches: 1
-
-learning_rate: 0.003
-
-# architecture 
-embed_dim: 16
-num_heads: 2
-num_layers: 8
-num_positions: 512
-num_vocab: 3
-tree_depth: 6
-spatial_dim: 2
-
-gpus: 1
-precision: 16

configs/mnist_s.yml

-name: mnist_s
-dataset: mnist
-
-# training
-epochs: 200
-warmup_steps: 0.1
-batch_size: 4 
-accumulate_grad_batches: 16
-
-learning_rate: 0.0001
-
-# architecture 
-embed_dim: 512
-num_heads: 8
-num_layers: 24
-num_positions: 512
-num_vocab: 3
-tree_depth: 6
-spatial_dim: 2
-
-gpus: 1
-precision: 16

configs/mnist_xs.yml

-name: mnist_xs
-dataset: mnist
-
-# training
-epochs: 100
-warmup_steps: 0.1
-batch_size: 16
-accumulate_grad_batches: 4
-
-learning_rate: 0.0003
-
-# architecture 
-embed_dim: 64
-num_heads: 4
-num_layers: 16
-num_positions: 512
-num_vocab: 3
-tree_depth: 6
-spatial_dim: 2
-
-gpus: 1
-precision: 16

configs/mnist_xxs.yml

-name: mnist_xxs
-dataset: mnist
-
-# training
-epochs: 30
-warmup_steps: 0.1
-batch_size: 64
-accumulate_grad_batches: 1
-
-learning_rate: 0.003
-
-# architecture 
-embed_dim: 16
-num_heads: 2
-num_layers: 8
-num_positions: 512
-num_vocab: 3
-tree_depth: 6
-spatial_dim: 2
-
-gpus: 1
-precision: 16
-

configs/num_heads/low.yml

-name: shapenet_num_heads
-# pretrained: None
-
-# data
-dataset: shapenet
-subclass: chair
-transform: [scaling, warping, linear_max_res]
-position_encoding: intertwined
-resolution: 128
-spatial_dim: 3
-num_vocab: 3
-
-# architecture
-architecture: encoder_multi_decoder
-embedding: [composite, single_conv, substitution]
-head: [composite, single_conv, substitution]
-
-# attention
-attention: basic
-num_positions: 2500
-embed_dim: 256
-num_layers: 8
-num_heads: 1
-
-# training
-loss_function: cross_entropy
-val_loss_function: cross_entropy
-epochs: 500
-warmup_steps: 0.1
-batch_size: 1
-accumulate_grad_batches: 1
-learning_rate: 0.00003
-dropout: 0.0
-
-# hardware
-gpus: 8
-precision: 32
-
-# logging
-log_gpu: 'min_max'
-log_gradient: False
-log_weights_and_biases: False
-log_learning_rate: True
-parameter_search: False
\ No newline at end of file

configs/num_heads/medium.yml

-name: shapenet_num_heads
-# pretrained: None
-
-# data
-dataset: shapenet
-subclass: chair
-transform: [scaling, warping, linear_max_res]
-position_encoding: intertwined
-resolution: 128
-spatial_dim: 3
-num_vocab: 3
-
-# architecture
-architecture: encoder_multi_decoder
-embedding: [composite, single_conv, substitution]
-head: [composite, single_conv, substitution]
-
-# attention
-attention: basic
-num_positions: 2500
-embed_dim: 256
-num_layers: 8
-num_heads: 2
-
-# training
-loss_function: cross_entropy
-val_loss_function: cross_entropy
-epochs: 500
-warmup_steps: 0.1
-batch_size: 1
-accumulate_grad_batches: 1
-learning_rate: 0.00003
-dropout: 0.0
-
-# hardware
-gpus: 8
-precision: 32
-
-# logging
-log_gpu: 'min_max'
-log_gradient: False
-log_weights_and_biases: False
-log_learning_rate: True
-parameter_search: False
\ No newline at end of file

configs/shapenet_autoencoder.yml

-name: shapenet_autoencoder
-
-# data
-dataset: shapenet
-subclass: chair
-position_encoding: intertwined
-transform: [linear_max_res]
-resolution: 32
-spatial_dim: 3
-num_vocab: 3
-
-# architecture
-embedding: double_substitution
-head: double_substitution
-architecture: autoencoder
-attention: none
-num_positions: 0
-embed_dim: 256
-num_layers: 0
-num_heads: 0
-
-# training
-loss_function: cross_entropy
-val_loss_function: cross_entropy
-epochs: 10
-warmup_steps: 0.1
-batch_size: 1
-accumulate_grad_batches: 1
-learning_rate: 0.001
-
-# hardware
-gpus: 1
-precision: 32
-
-# logging
-log_gpu: 'min_max'
-log_learning_rate: True
-log_gradient: False
-log_weights_and_biases: False
-parameter_search: False
\ No newline at end of file

configs/shapenet_concat_split.yml

-name: shapenet_concat_split
-
-# data
-dataset: shapenet
-subclass: chair
-resolution: 32
-spatial_dim: 3
-num_vocab: 3
-
-# architecture
-embedding: concat_A
-head: split_A
-architecture: encoder_decoder
-attention: basic
-num_positions: 1024
-embed_dim: 256
-num_layers: 8
-num_heads: 4
-
-# training
-loss_function: cross_entropy
-epochs: 100
-warmup_steps: 0.1
-batch_size: 16
-accumulate_grad_batches: 1
-learning_rate: 0.0003
-
-# hardware
-gpus: 4
-precision: 32
-
-# logging
-log_gpu: 'min_max'
-log_learning_rate: True
-log_gradient: True
-log_weights_and_biases: False
-parameter_search: False
\ No newline at end of file

configs/shapenet_debug.yml

-name: shapenet_debug
-# pretrained: None
-
-# data
-dataset: shapenet
-subclass: chair
-transform: [scaling, warping, linear_max_res]
-position_encoding: intertwined
-resolution: 64
-spatial_dim: 3
-num_vocab: 3
-
-# architecture
-architecture: encoder_only
-embedding: [composite_A]
-head: [composite_A]
-
-# attention
-attention: basic
-num_positions: 2048
-embed_dim: 512
-num_layers: 8
-num_heads: 8
-
-# training
-loss_function: cross_entropy
-val_loss_function: cross_entropy
-epochs: 97
-warmup_steps: 0.1
-batch_size: 1
-accumulate_grad_batches: 1
-learning_rate: 0.00003
-dropout: 0.2
-
-# hardware
-gpus: 1
-precision: 32
-
-# logging
-log_gpu: 'min_max'
-log_gradient: False
-log_weights_and_biases: False
-log_learning_rate: True
-parameter_search: False
\ No newline at end of file

configs/shapenet_intertwined.yml

-name: shapenet_intertwined
-
-# data
-dataset: shapenet
-subclass: chair
-position_encoding: intertwined
-resolution: 64
-spatial_dim: 3
-num_vocab: 3
-
-# architecture
-architecture: encoder_only
-embedding: [composite_A]
-head: [composite_A]
-
-# attention
-attention: basic
-num_positions: 4096
-embed_dim: 256
-num_layers: 8
-num_heads: 4
-
-# training
-loss_function: cross_entropy
-val_loss_function: cross_entropy
-epochs: 100
-warmup_steps: 0.1
-batch_size: 1
-accumulate_grad_batches: 1
-learning_rate: 0.0003
-
-# hardware
-gpus: 8
-precision: 32
-
-# logging
-log_gpu: 'min_max'
-log_gradient: False
-log_weights_and_biases: False
-log_learning_rate: True
-parameter_search: False
\ No newline at end of file

configs/shapenet_layerwise.yml

-name: shapenet_layerwise
-pretrained: logs/shapenet_layerwise/version_2/checkpoints/last.ckpt
-
-# data
-dataset: shapenet
-subclass: chair
-transform: [scaling, warping, linear_max_128]
-position_encoding: centered
-resolution: 128
-spatial_dim: 3
-num_vocab: 3
-
-# architecture
-architecture: encoder_only
-embedding: [composite]
-head: [composite]
-
-# attention
-attention: basic
-num_positions: 4096
-embed_dim: 256
-num_layers: 8
-num_heads: 4
-
-# training
-loss_function: cross_entropy
-val_loss_function: cross_entropy
-epochs: 100
-warmup_steps: 0.1
-batch_size: 1
-accumulate_grad_batches: 1
-learning_rate: 0.00003
-
-# hardware
-gpus: 8
-precision: 32
-
-# logging
-log_gpu: 'min_max'
-log_gradient: False
-log_weights_and_biases: False
-log_learning_rate: True
-parameter_search: False
\ No newline at end of file

configs/shapenet_long.yml

-name: shapenet_long
-
-# data
-dataset: shapenet
-subclass: chair
-position_encoding: intertwined
-transform: [scaling, warping, linear_max_res]
-resolution: 256
-spatial_dim: 3
-num_vocab: 3
-
-# architecture
-architecture: encoder_multi_decoder
-embedding: [composite, single_conv, substitution, double_substitution]
-head: [composite, single_conv, substitution, double_substitution]
-
-# attention
-attention: basic
-num_positions: 2500
-embed_dim: 256
-num_layers: 4
-num_heads: 4
-
-# training
-loss_function: cross_entropy
-val_loss_function: cross_entropy
-epochs: 500
-warmup_steps: 0.1
-batch_size: 1
-accumulate_grad_batches: 1
-learning_rate: 0.00003
-dropout: 0.0
-
-# hardware
-gpus: 8
-precision: 32
-
-# logging
-log_gpu: 'min_max'
-log_gradient: False
-log_weights_and_biases: False
-log_learning_rate: True
-parameter_search: False
\ No newline at end of file