Add files using upload-large-folder tool

SHA256SUMS

@@ -0,0 +1,8 @@
+2f9345c143cd7576472a0b88815504e4540447e5fcf2c4e927f62f431beeb10c  ./gen/seq_module_unconditional/seq_module_unconditional_ckpt.tar.gz
+938f13187c5e76df0c44c959ae2c6c44a764c2d8c673a2ac104e5cfa3cd05199  ./gen/seq_module_unconditional/config.yaml
+944de5df74846043142ab9e16c135eb8a2ccfdc10b761bde513be423dffb056b  ./gen/sp_module/sp_module_ckpt.tar.gz
+7c0c2b66f020a6bfc3e7fd01bead5b1e7cc679c77a5871fc8a5e5a092e1d12e9  ./gen/sp_module/config.yaml
+d537cc4cc0238f66f9872c19304aed98cfb1d5c9c2f9c72bfd82b2f2a1e6474c  ./gen/seq_module_conditional/seq_module_conditional_ckpt.tar.gz
+a0b1f4ef43dbb617d06edca6124a9755def9f9163853f951d017a87a3986fc16  ./gen/seq_module_conditional/config.yaml
+cf376f406334d3a86c7dc07dfc942df5af16b3a69adc781ce0322c4104851321  ./csp/sp_module_ckpt.tar.gz
+a839101b25a878268b70a564593d098d723bdb43647674b79d5bbd6728b5ff24  ./csp/config.yaml

csp/config.yaml

@@ -0,0 +1,220 @@
+preprocess:
+  task: csp
+  num_cpus: 60
+  lmdb_dir: ${paths.data_dir}/lmdb
+  split_dir: ${paths.data_dir}/splits
+  metal_dir: ${paths.data_dir}/metals
+  seq_dir: ${paths.data_dir}/seqs
+  filter:
+    num_cpus: 16
+    max_bbs: 20
+    max_atoms: 200
+    max_cps: 20
+    prop_list:
+    - working_capacity_vacuum_swing [mmol/g]
+    - working_capacity_temperature_swing [mmol/g]
+  mof_matcher:
+    optimizer:
+      steps: 200
+      popsize: 20
+      maxiter: 20
+    tolerance:
+      ltol: 0.2
+      stol: 0.5
+      angle_tol: 5.0
+  mof_checker:
+    use_matched_coords: true
+data:
+  train_sample_limit: null
+  val_sample_limit: null
+  dataset_prefix: MetalOxo_final
+  lmdb_dir: ${paths.data_dir}/lmdb/${experiment.task}
+  split_dir: ${paths.data_dir}/splits
+  rot_symmetry: true
+  corrupt_trans: ${experiment.training.corrupt_trans}
+  corrupt_rots: ${experiment.training.corrupt_rots}
+  corrupt_lattice: ${experiment.training.corrupt_lattice}
+  corrupt_torsions: ${experiment.training.corrupt_torsions}
+  t_sample_dist: uniform
+  t_uniform:
+    t_min: 0.0
+    t_max: 1.0
+  t_logit_normal:
+    mean: -0.5
+    std: 1.0
+  lattice:
+    lognormal:
+      loc: ${lattice.lognormal.loc}
+      scale: ${lattice.lognormal.scale}
+    uniform:
+      low: 60.0
+      high: 120.0
+      eps: 0.1
+  loader:
+    sampler: dynamic
+    overfit:
+      num_samples: 100000
+    dynamic:
+      max_num_atoms: 1500
+      max_batch_size: null
+    num_workers: 32
+    prefetch_factor: 10
+    batch_size:
+      train: 128
+      valid: 64
+      predict: 20
+experiment:
+  task: csp
+  project: mofflow_2
+  name: 146M-csp
+  debug: false
+  seed: 123
+  num_devices: 8
+  warm_start: null
+  warm_start_cfg_override: true
+  visualize_freq: 1000
+  use_ema: false
+  ema:
+    decay: 0.999
+    apply_ema_every_n_steps: 1
+    start_step: 0
+    evaluate_ema_weights_instead: false
+  training:
+    corrupt_trans: true
+    corrupt_rots: true
+    corrupt_lattice: true
+    corrupt_torsions: true
+    translation_loss_weight: 3.0
+    rotation_loss_weight: 1.0
+    lattice_loss_weight: 1.0
+    torsion_angle_loss_weight: 1.0
+    torsion_norm_loss_weight: 0.005
+    t_normalize_clip: 0.9
+  optimizer:
+    lr: 1.0e-05
+    betas:
+    - 0.9
+    - 0.98
+    eps: 1.0e-08
+    weight_decay: 0.0
+  lr_scheduler_type: null
+  lr_scheduler:
+    linear_warmup:
+      warmup_epochs: 10
+      min_lr: 1.0e-07
+    reduce_on_plateau:
+      factor: 0.6
+      patience: 10
+      min_lr: 1.0e-07
+  trainer:
+    overfit_batches: 0
+    min_epochs: 10
+    max_epochs: 200
+    accelerator: gpu
+    log_every_n_steps: 1
+    deterministic: false
+    strategy: ddp
+    val_check_interval: 1.0
+    check_val_every_n_epoch: 1
+    accumulate_grad_batches: 1
+    gradient_clip_val: 10.0
+  checkpointer:
+    dirpath: ${paths.ckpt_dir}
+    filename: epoch_{epoch}-step_{step}-loss_{train/loss:.4f}
+    auto_insert_metric_name: false
+    save_last: true
+    save_top_k: 10
+    monitor: step
+    mode: max
+model:
+  init_mode: 1in5out
+  corrupt_trans: ${experiment.training.corrupt_trans}
+  corrupt_rots: ${experiment.training.corrupt_rots}
+  corrupt_lattice: ${experiment.training.corrupt_lattice}
+  corrupt_torsions: ${experiment.training.corrupt_torsions}
+  initialize_cfg:
+    max_atom_types: 100
+    atom_type_dim: 64
+    atom_feats_dim: 39
+    time_embed_dim: 64
+    coords_feats_dim: 64
+    node_embed_dim: ${model.node_embed_dim}
+    add_pos_embed: true
+  node_embed_dim: 1024
+  interaction_cfg:
+    num_layers: 10
+    max_radius: 50
+    max_neighbors: 130
+    dist_embed_dim: 128
+    trans_encoder_cfg:
+      ln_mode: pre
+      init_mode: ${model.init_mode}
+      dropout: 0.0
+      ffn_bias: false
+      norm: rmsnorm
+      attention:
+        node_embed_dim: ${model.node_embed_dim}
+        dist_embed_dim: ${model.interaction_cfg.dist_embed_dim}
+        no_heads: 16
+      encoder:
+        ffn_dim: ${eval:'4 * ${model.node_embed_dim}'}
+  pool_cfg:
+    node_embed_dim: ${model.node_embed_dim}
+    max_radius: .inf
+    max_neighbors: 1000
+    gaussian:
+      max_radius: 20
+      dist_embed_dim: 128
+    no_heads: 16
+  output_model_cfg:
+    torsion:
+      normalize: false
+      node_embed_dim: ${model.node_embed_dim}
+      max_radius: 5
+      max_neighbors: 24
+      gaussian:
+        max_radius: 5
+        dist_embed_dim: 128
+      no_heads: 16
+    rotation:
+      node_embed_dim: ${model.node_embed_dim}
+      use_svd: false
+paths:
+  root_dir: ${oc.env:PROJECT_ROOT}
+  log_dir: ${paths.root_dir}/logs/${experiment.project}/${experiment.name}
+  logger_dir: ${paths.root_dir}/logs
+  ckpt_dir: ${paths.log_dir}/ckpt
+  inference_dir: ${paths.log_dir}/inference
+  data_dir: ${paths.root_dir}/data
+lattice:
+  lognormal:
+    loc:
+    - 2.5542190074920654
+    - 2.754688262939453
+    - 2.9632623195648193
+    scale:
+    - 0.37387514114379883
+    - 0.30108675360679626
+    - 0.31256887316703796
+logger: wandb
+wandb:
+  name: ${experiment.name}
+  project: ${experiment.project}
+  save_dir: ${paths.logger_dir}
+wandb_watch:
+  log: all
+  log_freq: 500
+tensorboard:
+  name: tensorboard
+  version: ${experiment.project}/${experiment.name}
+  save_dir: ${paths.logger_dir}
+  log_graph: false
+  default_hp_metric: true
+callbacks:
+  tensorboard:
+    log_freq: 1
+    log_grad_hist: false
+    log_weight_hist: false
+    log_grad_norm: true
+    log_weight_norm: true
+    log_norm_input: true

csp/sp_module_ckpt.tar.gz

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cf376f406334d3a86c7dc07dfc942df5af16b3a69adc781ce0322c4104851321
+size 1582384815

gen/seq_module_conditional/config.yaml

@@ -0,0 +1,102 @@
+data:
+  train_sample_limit: null
+  val_sample_limit: null
+  seq_dir: ${paths.data_dir}/seqs
+  dataset_prefix: mof_sequence
+  vocab_path: ${data.seq_dir}/vocab.json
+  max_len: null
+  loader:
+    num_workers: 16
+    prefetch_factor: 10
+    max_tokens: 8000
+    max_batch_size: null
+model:
+  conditional: true
+  max_seq_len: 2048
+  attention:
+    dim: 1024
+    depth: 6
+    heads: 8
+    rotary_pos_emb: true
+    attn_flash: true
+    use_scalenorm: true
+experiment:
+  project: mofflow_2
+  name: 63M-seq_cond
+  debug: false
+  seed: 123
+  num_devices: 4
+  warm_start: null
+  warm_start_cfg_override: true
+  sample_seq_freq: 500
+  use_ema: false
+  ema:
+    decay: 0.999
+    apply_ema_every_n_steps: 1
+    start_step: 0
+    evaluate_ema_weights_instead: true
+  optimizer:
+    lr: 0.0003
+    betas:
+    - 0.9
+    - 0.999
+    eps: 1.0e-08
+    weight_decay: 0.0
+  lr_scheduler_type: null
+  lr_scheduler:
+    linear_warmup:
+      warmup_epochs: 10
+      min_lr: 1.0e-07
+    reduce_on_plateau:
+      factor: 0.6
+      patience: 10
+      min_lr: 1.0e-06
+  trainer:
+    overfit_batches: 0
+    min_epochs: 10
+    max_epochs: 20
+    accelerator: gpu
+    log_every_n_steps: 1
+    deterministic: false
+    strategy: ddp
+    val_check_interval: 1.0
+    check_val_every_n_epoch: 1
+    accumulate_grad_batches: 1
+    gradient_clip_val: 1.0
+  checkpointer:
+    dirpath: ${paths.ckpt_dir}
+    monitor: step
+    save_last: true
+    save_top_k: -1
+    filename: epoch_{epoch}-step_{step}-loss_{valid/loss:.4f}
+    auto_insert_metric_name: false
+    mode: max
+logger: wandb
+wandb:
+  name: ${experiment.name}
+  project: ${experiment.project}
+  save_dir: ${paths.logger_dir}
+wandb_watch:
+  log: all
+  log_freq: 500
+tensorboard:
+  name: tensorboard
+  version: ${experiment.project}/${experiment.name}
+  save_dir: ${paths.logger_dir}
+  log_graph: false
+  default_hp_metric: true
+callbacks:
+  tensorboard:
+    log_freq: ${experiment.trainer.log_every_n_steps}
+    log_grad_hist: false
+    log_weight_hist: false
+    log_grad_norm: true
+    log_weight_norm: true
+    log_norm_input: true
+paths:
+  root_dir: ${oc.env:PROJECT_ROOT}
+  log_dir: ${paths.root_dir}/logs/${experiment.project}/${experiment.name}
+  logger_dir: ${paths.root_dir}/logs
+  ckpt_dir: ${paths.log_dir}/ckpt
+  inference_dir: ${paths.log_dir}/inference
+  data_dir: ${paths.root_dir}/data

gen/seq_module_conditional/seq_module_conditional_ckpt.tar.gz

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d537cc4cc0238f66f9872c19304aed98cfb1d5c9c2f9c72bfd82b2f2a1e6474c
+size 829792726

gen/seq_module_unconditional/config.yaml

@@ -0,0 +1,102 @@
+data:
+  train_sample_limit: null
+  val_sample_limit: null
+  seq_dir: ${paths.data_dir}/seqs
+  dataset_prefix: mof_sequence
+  vocab_path: ${data.seq_dir}/vocab.json
+  max_len: null
+  loader:
+    num_workers: 16
+    prefetch_factor: 10
+    max_tokens: 8000
+    max_batch_size: null
+model:
+  conditional: false
+  max_seq_len: 2048
+  attention:
+    dim: 1024
+    depth: 6
+    heads: 8
+    rotary_pos_emb: true
+    attn_flash: true
+    use_scalenorm: true
+experiment:
+  project: mofflow_2
+  name: 63M-seq
+  debug: false
+  seed: 123
+  num_devices: 8
+  warm_start: null
+  warm_start_cfg_override: true
+  sample_seq_freq: 500
+  use_ema: false
+  ema:
+    decay: 0.999
+    apply_ema_every_n_steps: 1
+    start_step: 0
+    evaluate_ema_weights_instead: true
+  optimizer:
+    lr: 0.0003
+    betas:
+    - 0.9
+    - 0.999
+    eps: 1.0e-08
+    weight_decay: 0.0
+  lr_scheduler_type: null
+  lr_scheduler:
+    linear_warmup:
+      warmup_epochs: 10
+      min_lr: 1.0e-07
+    reduce_on_plateau:
+      factor: 0.6
+      patience: 10
+      min_lr: 1.0e-06
+  trainer:
+    overfit_batches: 0
+    min_epochs: 10
+    max_epochs: 20
+    accelerator: gpu
+    log_every_n_steps: 1
+    deterministic: false
+    strategy: ddp
+    val_check_interval: 1.0
+    check_val_every_n_epoch: 1
+    accumulate_grad_batches: 1
+    gradient_clip_val: 1.0
+  checkpointer:
+    dirpath: ${paths.ckpt_dir}
+    monitor: step
+    save_last: true
+    save_top_k: 20
+    filename: epoch_{epoch}-step_{step}-loss_{train/loss:.4f}
+    auto_insert_metric_name: false
+    mode: max
+logger: wandb
+wandb:
+  name: ${experiment.name}
+  project: ${experiment.project}
+  save_dir: ${paths.logger_dir}
+wandb_watch:
+  log: all
+  log_freq: 500
+tensorboard:
+  name: tensorboard
+  version: ${experiment.project}/${experiment.name}
+  save_dir: ${paths.logger_dir}
+  log_graph: false
+  default_hp_metric: true
+callbacks:
+  tensorboard:
+    log_freq: ${experiment.trainer.log_every_n_steps}
+    log_grad_hist: false
+    log_weight_hist: false
+    log_grad_norm: true
+    log_weight_norm: true
+    log_norm_input: true
+paths:
+  root_dir: ${oc.env:PROJECT_ROOT}
+  log_dir: ${paths.root_dir}/logs/${experiment.project}/${experiment.name}
+  logger_dir: ${paths.root_dir}/logs
+  ckpt_dir: ${paths.log_dir}/ckpt
+  inference_dir: ${paths.log_dir}/inference
+  data_dir: ${paths.root_dir}/data

gen/seq_module_unconditional/seq_module_unconditional_ckpt.tar.gz

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2f9345c143cd7576472a0b88815504e4540447e5fcf2c4e927f62f431beeb10c
+size 690673315

gen/sp_module/config.yaml

@@ -0,0 +1,220 @@
+preprocess:
+  task: csp
+  num_cpus: 60
+  lmdb_dir: ${paths.data_dir}/lmdb
+  split_dir: ${paths.data_dir}/splits
+  metal_dir: ${paths.data_dir}/metals
+  seq_dir: ${paths.data_dir}/seqs
+  filter:
+    num_cpus: 16
+    max_bbs: 20
+    max_atoms: 200
+    max_cps: 20
+    prop_list:
+    - working_capacity_vacuum_swing [mmol/g]
+    - working_capacity_temperature_swing [mmol/g]
+  mof_matcher:
+    optimizer:
+      steps: 200
+      popsize: 20
+      maxiter: 20
+    tolerance:
+      ltol: 0.2
+      stol: 0.5
+      angle_tol: 5.0
+  mof_checker:
+    use_matched_coords: true
+data:
+  train_sample_limit: null
+  val_sample_limit: null
+  dataset_prefix: MetalOxo_final
+  lmdb_dir: ${paths.data_dir}/lmdb/${experiment.task}
+  split_dir: ${paths.data_dir}/splits
+  rot_symmetry: true
+  corrupt_trans: ${experiment.training.corrupt_trans}
+  corrupt_rots: ${experiment.training.corrupt_rots}
+  corrupt_lattice: ${experiment.training.corrupt_lattice}
+  corrupt_torsions: ${experiment.training.corrupt_torsions}
+  t_sample_dist: uniform
+  t_uniform:
+    t_min: 0.0
+    t_max: 1.0
+  t_logit_normal:
+    mean: -0.5
+    std: 1.0
+  lattice:
+    lognormal:
+      loc: ${lattice.lognormal.loc}
+      scale: ${lattice.lognormal.scale}
+    uniform:
+      low: 60.0
+      high: 120.0
+      eps: 0.1
+  loader:
+    sampler: dynamic
+    overfit:
+      num_samples: 100000
+    dynamic:
+      max_num_atoms: 1500
+      max_batch_size: null
+    num_workers: 32
+    prefetch_factor: 10
+    batch_size:
+      train: 128
+      valid: 64
+      predict: 20
+experiment:
+  task: gen
+  project: mofflow_2
+  name: 146M-gen
+  debug: false
+  seed: 123
+  num_devices: 8
+  warm_start: null
+  warm_start_cfg_override: true
+  visualize_freq: 1000
+  use_ema: false
+  ema:
+    decay: 0.999
+    apply_ema_every_n_steps: 1
+    start_step: 0
+    evaluate_ema_weights_instead: false
+  training:
+    corrupt_trans: true
+    corrupt_rots: true
+    corrupt_lattice: true
+    corrupt_torsions: true
+    translation_loss_weight: 3.0
+    rotation_loss_weight: 1.0
+    lattice_loss_weight: 1.0
+    torsion_angle_loss_weight: 1.0
+    torsion_norm_loss_weight: 0.005
+    t_normalize_clip: 0.9
+  optimizer:
+    lr: 1.0e-05
+    betas:
+    - 0.9
+    - 0.98
+    eps: 1.0e-08
+    weight_decay: 0.0
+  lr_scheduler_type: null
+  lr_scheduler:
+    linear_warmup:
+      warmup_epochs: 10
+      min_lr: 1.0e-07
+    reduce_on_plateau:
+      factor: 0.6
+      patience: 10
+      min_lr: 1.0e-07
+  trainer:
+    overfit_batches: 0
+    min_epochs: 10
+    max_epochs: 200
+    accelerator: gpu
+    log_every_n_steps: 1
+    deterministic: false
+    strategy: ddp
+    val_check_interval: 1.0
+    check_val_every_n_epoch: 1
+    accumulate_grad_batches: 1
+    gradient_clip_val: 10.0
+  checkpointer:
+    dirpath: ${paths.ckpt_dir}
+    filename: epoch_{epoch}-step_{step}-loss_{train/loss:.4f}
+    auto_insert_metric_name: false
+    save_last: true
+    save_top_k: 10
+    monitor: step
+    mode: max
+model:
+  init_mode: 1in5out
+  corrupt_trans: ${experiment.training.corrupt_trans}
+  corrupt_rots: ${experiment.training.corrupt_rots}
+  corrupt_lattice: ${experiment.training.corrupt_lattice}
+  corrupt_torsions: ${experiment.training.corrupt_torsions}
+  initialize_cfg:
+    max_atom_types: 100
+    atom_type_dim: 64
+    atom_feats_dim: 39
+    time_embed_dim: 64
+    coords_feats_dim: 64
+    node_embed_dim: ${model.node_embed_dim}
+    add_pos_embed: true
+  node_embed_dim: 1024
+  interaction_cfg:
+    num_layers: 10
+    max_radius: 50
+    max_neighbors: 130
+    dist_embed_dim: 128
+    trans_encoder_cfg:
+      ln_mode: pre
+      init_mode: ${model.init_mode}
+      dropout: 0.0
+      ffn_bias: false
+      norm: rmsnorm
+      attention:
+        node_embed_dim: ${model.node_embed_dim}
+        dist_embed_dim: ${model.interaction_cfg.dist_embed_dim}
+        no_heads: 16
+      encoder:
+        ffn_dim: ${eval:'4 * ${model.node_embed_dim}'}
+  pool_cfg:
+    node_embed_dim: ${model.node_embed_dim}
+    max_radius: .inf
+    max_neighbors: 1000
+    gaussian:
+      max_radius: 20
+      dist_embed_dim: 128
+    no_heads: 16
+  output_model_cfg:
+    torsion:
+      normalize: false
+      node_embed_dim: ${model.node_embed_dim}
+      max_radius: 5
+      max_neighbors: 24
+      gaussian:
+        max_radius: 5
+        dist_embed_dim: 128
+      no_heads: 16
+    rotation:
+      node_embed_dim: ${model.node_embed_dim}
+      use_svd: false
+paths:
+  root_dir: ${oc.env:PROJECT_ROOT}
+  log_dir: ${paths.root_dir}/logs/${experiment.project}/${experiment.name}
+  logger_dir: ${paths.root_dir}/logs
+  ckpt_dir: ${paths.log_dir}/ckpt
+  inference_dir: ${paths.log_dir}/inference
+  data_dir: ${paths.root_dir}/data
+lattice:
+  lognormal:
+    loc:
+    - 2.5542190074920654
+    - 2.754688262939453
+    - 2.9632623195648193
+    scale:
+    - 0.37387514114379883
+    - 0.30108675360679626
+    - 0.31256887316703796
+logger: wandb
+wandb:
+  name: ${experiment.name}
+  project: ${experiment.project}
+  save_dir: ${paths.logger_dir}
+wandb_watch:
+  log: all
+  log_freq: 500
+tensorboard:
+  name: tensorboard
+  version: ${experiment.project}/${experiment.name}
+  save_dir: ${paths.logger_dir}
+  log_graph: false
+  default_hp_metric: true
+callbacks:
+  tensorboard:
+    log_freq: 1
+    log_grad_hist: false
+    log_weight_hist: false
+    log_grad_norm: true
+    log_weight_norm: true
+    log_norm_input: true

gen/sp_module/sp_module_ckpt.tar.gz

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:944de5df74846043142ab9e16c135eb8a2ccfdc10b761bde513be423dffb056b
+size 1582456820