rc-foundry 0.1.5__py3-none-any.whl → 0.1.7__py3-none-any.whl

rf3/configs/datasets/train/na_complex_distillation.yaml

@@ -0,0 +1,50 @@
+# TODO: Inherit from common config with default Transform pipeline
+
+na_complex_distillation:
+  dataset:
+    _target_: atomworks.ml.datasets.StructuralDatasetWrapper
+    save_failed_examples_to_dir: null
+
+    # cif parser
+    cif_parser_args:
+      #assume_residues_all_resolved: true
+      cache_dir: null
+      load_from_cache: false
+      save_to_cache: false
+
+    # metadata parser
+    dataset_parser:
+      _target_: atomworks.ml.datasets.parsers.GenericDFParser
+      pn_unit_iid_colnames: null #[]
+
+    # metadata dataset
+    dataset:
+      _target_: atomworks.ml.datasets.PandasDataset
+      name: tf_distillation
+      id_column: example_id
+      data: ${paths.data.na_complex_distillation_parquet_dir}/transcriptionFactor_distillation_rf3.newDL.csv
+      columns_to_load:
+        - example_id
+        - path
+    transform:
+      _target_: ${datasets.pipeline_target}
+      is_inference: False
+      protein_msa_dirs: [{"dir": "${paths.data.na_complex_distillation_data_dir}/a3m/", "extension": ".a3m", "directory_depth": 1}]
+      rna_msa_dirs: []
+      n_recycles: ${datasets.n_recycles_train}
+      crop_size: ${datasets.crop_size}
+      n_msa: ${datasets.n_msa}
+      diffusion_batch_size: ${datasets.diffusion_batch_size_train}
+      max_atoms_in_crop: ${datasets.max_atoms_in_crop}
+      crop_contiguous_probability: 0.25
+      crop_spatial_probability: 0.75
+      pad_dna_p_skip: 0.0
+      run_confidence_head: ${datasets.run_confidence_head}
+      take_first_chiral_subordering: ${datasets.take_first_chiral_subordering}
+      use_element_for_atom_names_of_atomized_tokens: ${datasets.use_element_for_atom_names_of_atomized_tokens}
+      mirror_prob: 0.0
+      atomization_prob: ${datasets.atomization_prob}
+      ligand_dropout_prob: ${datasets.ligand_dropout_prob}
+      p_unconditional: ${datasets.p_unconditional}
+      p_dropout_atom_level_embeddings: ${datasets.p_dropout_atom_level_embeddings}
+      add_residue_is_paired_feature: ${datasets.add_residue_is_paired_feature}

rf3/configs/datasets/train/pdb/af3_weighted_sampling.yaml

@@ -0,0 +1,8 @@
+weights:
+  _target_: atomworks.ml.samplers.calculate_weights_for_pdb_dataset_df
+  # We do not include beta here, since it is different for interfaces and chains
+  alphas:
+    a_prot: 3.0 # 3 for AF-3
+    a_nuc: 3.0 # 3 for AF-3
+    a_ligand: 1.0 # 1 for AF-3
+    a_loi: 5.0 # 5 for AF-3

rf3/configs/datasets/train/pdb/base.yaml

@@ -0,0 +1,32 @@
+dataset:
+  _target_: atomworks.ml.datasets.StructuralDatasetWrapper
+  save_failed_examples_to_dir: ${paths.data.failed_examples_dir}
+  cif_parser_args:
+    cache_dir: null
+    load_from_cache: false
+    save_to_cache: false
+  dataset:
+    _target_: atomworks.ml.datasets.PandasDataset
+    # we will use the example_id as the unique column
+    id_column: example_id
+  transform:
+    # common Transform pipeline components for all PDB datasets
+    _target_: ${datasets.pipeline_target}
+    is_inference: False
+    protein_msa_dirs: ${paths.data.protein_msa_dirs}
+    rna_msa_dirs: ${paths.data.rna_msa_dirs}
+    n_recycles: ${datasets.n_recycles_train}
+    crop_size: ${datasets.crop_size}
+    n_msa: ${datasets.n_msa}
+    diffusion_batch_size: ${datasets.diffusion_batch_size_train}
+    max_atoms_in_crop: ${datasets.max_atoms_in_crop}
+    run_confidence_head: ${datasets.run_confidence_head}
+    p_unconditional: ${datasets.p_unconditional}
+    p_dropout_atom_level_embeddings: ${datasets.p_dropout_atom_level_embeddings}
+    take_first_chiral_subordering: ${datasets.take_first_chiral_subordering}
+    use_element_for_atom_names_of_atomized_tokens: ${datasets.use_element_for_atom_names_of_atomized_tokens}
+    mirror_prob: ${datasets.mirror_prob}
+    atomization_prob: ${datasets.atomization_prob}
+    ligand_dropout_prob: ${datasets.ligand_dropout_prob}
+    add_residue_is_paired_feature: ${datasets.add_residue_is_paired_feature}
+    add_cyclic_bonds: ${datasets.add_cyclic_bonds}

rf3/configs/datasets/train/pdb/plinder.yaml

@@ -0,0 +1,54 @@
+# PLINDER is a subset of the PDB, so we inherit from the base PDB config
+
+defaults:
+  - base
+
+dataset:
+  dataset_parser:
+    _target_: atomworks.ml.datasets.parsers.InterfacesDFParser
+    base_dir: /projects/ml/frozen_pdb_copies/2025_07_13_pdb
+  dataset:
+    name: plinder
+    data: ${paths.data.pdb_data_dir}/interfaces_df_train_plinder.parquet
+    filters:
+      # filters common across all PDB datasets
+      - "deposition_date < '2024-01-01'"
+      - "resolution < 9.0"
+      - "num_polymer_pn_units <= 300" 
+      # interface-specific filters
+      - "~(pn_unit_1_non_polymer_res_names.notnull() and pn_unit_1_non_polymer_res_names.str.contains('${resolve_import:atomworks.constants,AF3_EXCLUDED_LIGANDS_REGEX}', regex=True))"
+      - "~(pn_unit_2_non_polymer_res_names.notnull() and pn_unit_2_non_polymer_res_names.str.contains('${resolve_import:atomworks.constants,AF3_EXCLUDED_LIGANDS_REGEX}', regex=True))"
+    columns_to_load:
+      # columns common across all PDB datasets
+      - example_id
+      - pdb_id
+      - assembly_id
+      - deposition_date
+      - resolution
+      - num_polymer_pn_units
+      - method
+      - n_prot
+      - n_nuc
+      - n_ligand
+      - n_peptide
+      - total_num_atoms_in_unprocessed_assembly
+      # interface specific columns
+      - pn_unit_1_iid
+      - pn_unit_2_iid
+      - pn_unit_1_non_polymer_res_names
+      - pn_unit_2_non_polymer_res_names
+      - is_inter_molecule
+      - all_pn_unit_iids_after_processing
+      - involves_loi
+      - pli_qcov__50__strong__component
+      - pli_qcov__70__strong__component
+      - pli_qcov__50__weak__component
+      - pli_qcov__70__weak__component
+  transform:
+    # interface-specific Transform pipeline parameters
+    crop_contiguous_probability: 0.0
+    crop_spatial_probability: 1.0
+  
+weights:
+  _target_: atomworks.ml.samplers.calculate_weights_by_inverse_cluster_size
+  cluster_column: pli_qcov__50__weak__component # Need to ablate

rf3/configs/datasets/train/pdb/train_interface.yaml

@@ -0,0 +1,51 @@
+defaults:
+  - base
+  - af3_weighted_sampling
+
+dataset:
+  dataset_parser:
+    _target_: atomworks.ml.datasets.parsers.InterfacesDFParser
+    base_dir: /projects/ml/frozen_pdb_copies/2025_07_13_pdb
+  dataset:
+    name: interface
+    data: ${paths.data.pdb_data_dir}/interfaces_df_train.parquet
+    filters:
+      # filters common across all PDB datasets
+      - "deposition_date < '2024-01-01'"
+      - "resolution < 9.0"
+      - "num_polymer_pn_units <= 300" 
+      - "cluster.notnull()"
+      # interface specific filters
+      - "~(pn_unit_1_non_polymer_res_names.notnull() and pn_unit_1_non_polymer_res_names.str.contains('${resolve_import:atomworks.constants,AF3_EXCLUDED_LIGANDS_REGEX}', regex=True))"
+      - "~(pn_unit_2_non_polymer_res_names.notnull() and pn_unit_2_non_polymer_res_names.str.contains('${resolve_import:atomworks.constants,AF3_EXCLUDED_LIGANDS_REGEX}', regex=True))"
+      - "is_inter_molecule"
+    columns_to_load:
+      # columns common across all PDB datasets
+      - example_id
+      - pdb_id
+      - assembly_id
+      - deposition_date
+      - resolution
+      - num_polymer_pn_units
+      - method
+      - cluster
+      - n_prot
+      - n_nuc
+      - n_ligand
+      - n_peptide
+      - total_num_atoms_in_unprocessed_assembly
+      # interface specific columns
+      - pn_unit_1_iid
+      - pn_unit_2_iid
+      - pn_unit_1_non_polymer_res_names
+      - pn_unit_2_non_polymer_res_names
+      - is_inter_molecule
+      - all_pn_unit_iids_after_processing
+      - involves_loi
+  transform:
+    # interface-specific Transform pipeline parameters
+    crop_contiguous_probability: 0.0
+    crop_spatial_probability: 1.0
+
+weights:
+  beta: 1.0

rf3/configs/datasets/train/pdb/train_pn_unit.yaml

@@ -0,0 +1,46 @@
+defaults:
+  - base
+  - af3_weighted_sampling
+
+dataset: 
+  dataset_parser:
+    _target_: atomworks.ml.datasets.parsers.PNUnitsDFParser
+    base_dir: /projects/ml/frozen_pdb_copies/2025_07_13_pdb
+  dataset:
+    name: pn_unit
+    data: ${paths.data.pdb_data_dir}/pn_units_df_train.parquet
+    filters:
+        # filters common across all PDB datasets
+      - "deposition_date < '2024-01-01'"
+      - "resolution < 9.0"
+      - "num_polymer_pn_units <= 300" 
+      - "cluster.notnull()"
+      # pn_unit specific filters
+      - "~(q_pn_unit_non_polymer_res_names.notnull() and q_pn_unit_non_polymer_res_names.str.contains('${resolve_import:atomworks.constants,AF3_EXCLUDED_LIGANDS_REGEX}', regex=True))"
+    columns_to_load:
+      # columns common across all PDB datasets
+      - example_id
+      - pdb_id
+      - assembly_id
+      - deposition_date
+      - resolution
+      - num_polymer_pn_units
+      - method
+      - cluster
+      - n_prot
+      - n_nuc
+      - n_ligand
+      - n_peptide
+      - total_num_atoms_in_unprocessed_assembly
+      # pn_unit specific columns
+      - q_pn_unit_iid
+      - q_pn_unit_non_polymer_res_names
+      - all_pn_unit_iids_after_processing
+      - q_pn_unit_is_loi
+  transform:
+    # pn_unit-specific Transform pipeline parameters
+    crop_contiguous_probability: 0.3333333333333333
+    crop_spatial_probability: 0.6666666666666667
+
+weights:
+  beta: 0.5

rf3/configs/datasets/train/rna_monomer_distillation.yaml

@@ -0,0 +1,56 @@
+# TODO: Inherit from common config with default Transform pipeline
+
+rna_monomer_distillation:
+  dataset:
+    _target_: atomworks.ml.datasets.StructuralDatasetWrapper
+    save_failed_examples_to_dir: ${paths.data.failed_examples_dir}
+
+    # cif parser arguments
+    cif_parser_args:
+      cache_dir: null
+      load_from_cache: False
+      save_to_cache: False 
+
+    # metadata parser
+    dataset_parser:
+      _target_: atomworks.ml.datasets.parsers.GenericDFParser
+      pn_unit_iid_colnames: null
+
+    # metadata dataset
+    dataset:
+      _target_: atomworks.ml.datasets.PandasDataset
+      name: rna_monomer_distillation
+      id_column: example_id
+      data: /projects/ml/afavor/rna_distillation/rna_distillation_filtered_df.parquet
+      columns_to_load:
+        - example_id
+        - path
+        - cluster_id
+        - seq_hash
+        - overall_plddt
+        - overall_pde
+        - overall_pae
+
+    transform:
+      _target_: ${datasets.pipeline_target}
+      is_inference: False
+      protein_msa_dirs: []
+      rna_msa_dirs: [{"dir": "/projects/ml/afavor/rna_distillation/all_MSAs_renamed", "extension": ".afa", "directory_depth": 2}]
+      n_recycles: ${datasets.n_recycles_train}
+      crop_size: ${datasets.crop_size}
+      n_msa: ${datasets.n_msa}
+      diffusion_batch_size: ${datasets.diffusion_batch_size_train}
+      max_atoms_in_crop: ${datasets.max_atoms_in_crop}
+      crop_contiguous_probability: 1.0
+      crop_spatial_probability: 0.0
+      pad_dna_p_skip: 0.0
+      b_factor_min: 0.6
+      run_confidence_head: ${datasets.run_confidence_head}
+      take_first_chiral_subordering: ${datasets.take_first_chiral_subordering}
+      use_element_for_atom_names_of_atomized_tokens: ${datasets.use_element_for_atom_names_of_atomized_tokens}
+      mirror_prob: 0.0
+      atomization_prob: ${datasets.atomization_prob}
+      ligand_dropout_prob: ${datasets.ligand_dropout_prob}
+      p_unconditional: ${datasets.p_unconditional}
+      p_dropout_atom_level_embeddings: ${datasets.p_dropout_atom_level_embeddings}
+      add_residue_is_paired_feature: ${datasets.add_residue_is_paired_feature}

rf3/configs/datasets/val/af3_ab_set.yaml

@@ -0,0 +1,11 @@
+defaults:
+  - base
+
+dataset:
+  dataset_parser:
+    _target_: atomworks.ml.datasets.parsers.ValidationDFParserLikeAF3
+    base_dir: /projects/ml/frozen_pdb_copies/2025_07_13_pdb
+  dataset:
+    _target_: atomworks.ml.datasets.PandasDataset
+    name: af3_validation
+    data: /net/scratch/rib7/rf3_ab_splits/entry_level_val_df.parquet

rf3/configs/datasets/val/af3_validation.yaml

@@ -0,0 +1,11 @@
+defaults:
+  - base
+
+dataset:
+  dataset_parser:
+    _target_: atomworks.ml.datasets.parsers.ValidationDFParserLikeAF3
+    base_dir: /projects/ml/frozen_pdb_copies/2025_07_13_pdb
+  dataset:
+    _target_: atomworks.ml.datasets.PandasDataset
+    name: af3_validation
+    data: ${paths.data.pdb_data_dir}/entry_level_val_df.parquet

rf3/configs/datasets/val/base.yaml

@@ -0,0 +1,32 @@
+dataset:
+  _target_: atomworks.ml.datasets.StructuralDatasetWrapper
+  save_failed_examples_to_dir: ${paths.data.failed_examples_dir}
+  cif_parser_args:
+    cache_dir: null
+    load_from_cache: False
+    save_to_cache: False
+  dataset:
+    _target_: atomworks.ml.datasets.PandasDataset
+    # we will use the example_id as the unique column
+    id_column: example_id
+  # return all keys (do not subset)
+  transform:
+    # common Transform pipeline components for all PDB datasets
+    _target_: ${datasets.pipeline_target}
+    is_inference: True
+    protein_msa_dirs: ${paths.data.protein_msa_dirs}
+    rna_msa_dirs: ${paths.data.rna_msa_dirs}
+    n_recycles: ${datasets.n_recycles_validation}
+    crop_size: null # do not crop for inference
+    n_msa: ${datasets.n_msa}
+    diffusion_batch_size: ${datasets.diffusion_batch_size_inference}
+    max_atoms_in_crop: null # do not crop for inference
+    return_atom_array: True # return atom array for inference
+    run_confidence_head: ${datasets.run_confidence_head}
+    p_unconditional: 1.0 # unconditional for inference, unless explicitly overridden
+    p_dropout_atom_level_embeddings: 0.0 # always use embeddings in inference
+    take_first_chiral_subordering: ${datasets.take_first_chiral_subordering}
+    use_element_for_atom_names_of_atomized_tokens: ${datasets.use_element_for_atom_names_of_atomized_tokens}
+    add_residue_is_paired_feature: ${datasets.add_residue_is_paired_feature}
+
+key_to_balance: ${datasets.key_to_balance}

rf3/configs/datasets/val/runs_and_poses.yaml

@@ -0,0 +1,12 @@
+defaults:
+  - base
+
+dataset:
+  dataset_parser:
+    _target_: atomworks.ml.datasets.parsers.ValidationDFParserLikeAF3
+  dataset:
+    _target_: atomworks.ml.datasets.PandasDataset
+    name: af3_validation
+    data: /projects/ml/datahub/dfs/af3_splits/2024_12_16/runs_n_poses_entry_level_df.parquet
+    filters:
+      - "n_tokens_total < 1000" # Subset to reasonably-sized examples for efficiency

rf3/configs/debug/default.yaml

@@ -0,0 +1,66 @@
+# @package _global_
+
+defaults:
+  - override /logger: null
+
+# default debugging setup, runs 1 full epoch
+# other debugging configs can inherit from this one
+
+# overwrite task name so debugging logs are stored in separate folder
+task_name: "debug"
+
+extras:
+  ignore_warnings: False
+  enforce_tags: False
+
+# sets level of all command line loggers to 'DEBUG'
+# https://hydra.cc/docs/tutorials/basic/running_your_app/logging/
+hydra:
+  job_logging:
+    root:
+      level: DEBUG
+  # use the below to also set hydra loggers to 'DEBUG'
+  verbose: True
+
+# Print example ID before forward pass
+callbacks:
+  print_example_id_before_forward_pass:
+      _target_: foundry.callbacks.train_logging.PrintExampleIDBeforeForwardPassCallback
+  timing_logging:
+      _target_: foundry.callbacks.timing_logging.TimingCallback
+      log_every_n: 5
+
+dataloader:
+  train:
+    dataloader_params: 
+      batch_size: 1
+      num_workers: 0  # debuggers don't like multiprocessing -- work on main thread
+      pin_memory: False  # disable gpu memory pin
+      prefetch_factor: null # must be null for num_workers=0
+    n_fallback_retries: 0 # disable fallback retries for debugging
+
+  val:
+    dataloader_params:
+      batch_size: 1
+      num_workers: 0
+      pin_memory: False
+      prefetch_factor: null # must be null for num_workers=0
+
+datasets:
+  crop_size: 100  # set small crop size for quick debugging
+  diffusion_batch_size_train: 1
+  diffusion_batch_size_inference: 2
+  n_recycles_train: 1
+  n_recycles_validation: 1
+  n_msa: 128
+  key_to_balance: null # otherwise big examples will be processed first
+
+trainer:
+  devices_per_node: 1
+  limit_train_batches: 2
+  limit_val_batches: 1
+  validate_every_n_epochs: 1
+
+# Set tags to help identify debugging runs
+tags:
+  - debug

rf3/configs/debug/train_specific_examples.yaml

@@ -0,0 +1,21 @@
+# @package _global_
+
+# See: https://hydra.cc/docs/patterns/configuring_experiments/
+
+# to execute this experiment run:
+# python train.py +debug=train_single_example [any other arguments] 
+
+defaults:
+  - default
+  - gpu
+
+datasets: 
+  # you can add specific example IDs here to load a subset of the dataset (only training supported; PR's welcome to generalize to validation)
+  subset_to_example_ids:
+    - "{['pdb', 'pn_units']}{3px1}{1}{['A_3']}"
+  val: null
+
+tags:
+  - debug
+  - train
+  - specific-examples

rf3/configs/experiment/pretrained/rf3.yaml

@@ -0,0 +1,50 @@
+# @package _global_
+
+name: rf3
+project: rf3
+
+tags:
+  # list of tags to add to the run ( & on wandb to easily find & filter runs)
+  - full
+
+defaults:
+  - override /datasets: pdb_and_distillation
+  - override /model: rf3
+  - override /trainer: rf3
+
+ckpt_config:
+  _target_: foundry.utils.weights.CheckpointConfig
+  path: /net/software/containers/versions/modelhub_inference/ckpts/rf3-w-conf-run10-ep922-remapped.ckpt
+  reset_optimizer: true
+
+model:
+  lr_scheduler:
+    base_lr: 0.9e-3 # 1/2 of original learning rate (1.8e-3)
+  net:
+    feature_initializer:
+      input_feature_embedder:
+        atom_attention_encoder:
+          c_atom_1d_features: 393 # 392 + 1 has_atom_level_embedding = 393
+          atom_1d_features: 
+             - ref_pos                                                                             
+             - ref_charge                                                                          
+             - ref_mask                                                                            
+             - ref_element                                                                         
+             - ref_atom_name_chars                                                                 
+             - ref_pos_ground_truth 
+             - has_atom_level_embedding
+          use_atom_level_embedding: true
+          atom_level_embedding_dim: 384
+    diffusion_module:
+      atom_attention_encoder:
+        c_atom_1d_features: 393 # 392 + 1 has_atom_level_embedding = 393
+        atom_1d_features:
+          - ref_pos                                                                             
+          - ref_charge                                                                          
+          - ref_mask                                                                            
+          - ref_element                                                                         
+          - ref_atom_name_chars                                                                 
+          - ref_pos_ground_truth 
+          - has_atom_level_embedding
+        use_atom_level_embedding: true
+        atom_level_embedding_dim: 384

rf3/configs/experiment/pretrained/rf3_with_confidence.yaml

@@ -0,0 +1,13 @@
+# @package _global_
+
+name: rf3-with-confidence
+
+# For explanation of the "override" syntax, see: https://hydra.cc/docs/upgrades/1.0_to_1.1/defaults_list_override/
+defaults:
+  - pretrained/rf3
+  - override /model: rf3_with_confidence
+  - override /trainer: rf3_with_confidence
+  - _self_
+
+datasets:
+  run_confidence_head: true

rf3/configs/experiment/quick-rf3-with-confidence.yaml

@@ -0,0 +1,15 @@
+# @package _global_
+
+# Experiment that loads a small dataset for quick testing
+
+name: quick-rf3-with-confidence
+
+# For explanation of the "override" syntax, see: https://hydra.cc/docs/upgrades/1.0_to_1.1/defaults_list_override/
+defaults:
+  - quick-rf3
+  - override /model: rf3_with_confidence
+  - override /trainer: rf3_with_confidence
+  - _self_
+
+datasets:
+  run_confidence_head: true

rf3/configs/experiment/quick-rf3.yaml

@@ -0,0 +1,61 @@
+# @package _global_
+
+# Experiment that loads a small dataset for quick testing
+
+name: quick-rf3
+
+# For explanation of the "override" syntax, see: https://hydra.cc/docs/upgrades/1.0_to_1.1/defaults_list_override/
+defaults:
+  - pretrained/rf3
+  - override /datasets: pdb_only
+
+tags:
+  # list of tags to add to the run ( & on wandb to easily find & filter runs)
+  - quick
+
+project: test
+
+paths:
+  data:
+    pdb_data_dir: /projects/ml/datahub/dfs/af3_splits/2024_12_16
+
+trainer:
+  limit_train_batches: 4
+  limit_val_batches: 4
+
+datasets:
+  train:
+    pdb:
+      # We must adjust the probability, since we set the monomer distillation dataset to null
+      probability: 1.0
+      sub_datasets:
+        interface:
+          dataset:
+            dataset:
+              # A small dataframe that loads quickly
+              data: /projects/ml/datahub/dfs/pdb/test_dfs/interfaces_df.parquet
+              filters:
+                - "total_num_atoms_in_unprocessed_assembly <= 3000" 
+                - "cluster.notnull()"
+        pn_unit:
+          dataset:
+            dataset:
+              # A small dataframe that loads quickly
+              data: /projects/ml/datahub/dfs/pdb/test_dfs/pn_units_df.parquet
+              filters:
+                - "total_num_atoms_in_unprocessed_assembly <= 3000" 
+                - "cluster.notnull()"
+  val: 
+    af3_validation:
+      dataset:
+        dataset:
+          filters:
+            - "n_tokens_total < 500"
+            # (We often want to debug with a ligand)
+            - "interfaces_to_score.str.contains('protein-ligand')"
+            - example_id  in ["{['validation']}{7psi}{1}{[]}", "{['validation']}{7lo1}{2}{[]}", "{['validation']}{6zg9}{1}{[]}", "{['validation']}{7vhy}{1}{[]}"]
+
+model:
+  net:
+    inference_sampler:
+      num_timesteps: 50

rf3/configs/hydra/default.yaml

@@ -0,0 +1,18 @@
+# https://hydra.cc/docs/configure_hydra/intro/
+
+# enable color logging (requires `colorlog` to be installed)
+# defaults:
+#   - override hydra_logging: colorlog
+#   - override job_logging: colorlog
+
+
+# output directory, generated dynamically on each run
+run:
+  dir: ${paths.log_dir}/${task_name}/${name}/${now:%Y-%m-%d}_${now:%H-%M}_JOB_${oc.env:SLURM_JOB_ID,default}
+
+# ... this is where the log file is written (i.e. the programs output)
+job_logging:
+  handlers:
+    file:
+      # Incorporates fix from https://github.com/facebookresearch/hydra/pull/2242
+      filename: ${hydra.runtime.output_dir}/experiment.log

rf3/configs/hydra/no_logging.yaml

@@ -0,0 +1,7 @@
+defaults:
+  - override job_logging: disabled
+  - override hydra_logging: disabled
+
+output_subdir: null  
+run:  
+  dir: .

rf3/configs/inference.yaml

@@ -0,0 +1,7 @@
+# @package _global_
+# ^ The "package" determines where the content of the config is placed in the output config
+# For more information about overriding configs, see: https://hydra.cc/docs/advanced/overriding_packages/#overriding-packages-using-the-defaults-list
+
+defaults:
+  - inference_engine: rf3
+  - _self_

rf3/configs/inference_engine/base.yaml

@@ -0,0 +1,23 @@
+# @package _global_
+
+defaults:
+  - /hydra: no_logging
+
+# Parameters for RF3InferenceEngine.__init__()
+ckpt_path: ???
+num_nodes: 1
+devices_per_node: 1
+compress_outputs: false
+
+# Parameters for RF3InferenceEngine.run()
+inputs: ???
+out_dir: ???
+dump_predictions: true
+dump_trajectories: false
+one_model_per_file: false
+annotate_b_factor_with_plddt: true
+sharding_pattern: null
+skip_existing: false
+template_selection: null
+ground_truth_conformer_selection: null
+cyclic_chains: []