dataset:
  N: 512
  P: 512
  S: 512
  cellxgene:
    ds_type: h5ad
    filter: false
    num_datasets: 1139
    train: /large_storage/ctc/userspace/aadduri/data/auxillary/esm_cellxgene_train.csv
    val: /large_storage/ctc/userspace/aadduri/data/auxillary/esm_cellxgene_val.csv
  cellxgene-tahoe:
    ds_type: filtered_h5ad
    filter: true
    filter_by_species: null
    num_datasets: 1139
    train: /large_storage/ctc/userspace/aadduri/data/auxillary/esm_tahoe_cellxgene_train_filtered.csv
    val: /large_storage/ctc/userspace/aadduri/data/auxillary/esm_tahoe_cellxgene_val_filtered.csv
  chrom_token_right_idx: 2
  cls_token_idx: 3
  current: scbasecamp-cellxgene-tahoe-filtered
  name: vci
  num_cells: 36238464
  num_train_workers: 32
  num_val_workers: 8
  overrides:
    rpe1_top5000_variable: /large_storage/ctc/datasets/vci/validation/rpe1_top5000_variable.h5ad
  pad_length: 2048
  pad_token_idx: 0
  scbasecamp-cellxgene-tahoe:
    ds_type: filtered_h5ad
    filter: true
    filter_by_species: null
    num_datasets: 15700
    train: /large_storage/ctc/userspace/aadduri/data/auxillary/esm_basecount_tahoe_cellxgene_train.csv
    val: /large_storage/ctc/userspace/aadduri/data/auxillary/esm_basecount_tahoe_cellxgene_val.csv
  scbasecamp-cellxgene-tahoe-filtered:
    ds_type: filtered_h5ad
    filter: true
    filter_by_species: null
    num_datasets: 14420
    train: /large_storage/ctc/userspace/aadduri/data/auxillary/esm_basecount_tahoe_cellxgene_train_filtered.csv
    val: /large_storage/ctc/userspace/aadduri/data/auxillary/esm_basecount_tahoe_cellxgene_val_filtered.csv
  seed: 42
embeddings:
  current: esm2-cellxgene-basecamp-tahoe
  esm2-cellxgene:
    all_embeddings: /large_storage/ctc/userspace/aadduri/data/auxillary/Homo_sapiens.GRCh38.gene_symbol_to_embedding_ESM2.pt
    ds_emb_mapping: /large_storage/ctc/userspace/aadduri/data/auxillary/esm_cellxgene_ds_mapping.torch
    num: 19790
    size: 5120
    valid_genes_masks: null
  esm2-cellxgene-basecamp-tahoe:
    all_embeddings: /large_storage/ctc/userspace/aadduri/data/auxillary/Homo_sapiens.GRCh38.gene_symbol_to_embedding_ESM2.pt
    # ds_emb_mapping: /home/aadduri/vci_pretrain/gene_embidx_mapping_cross.torch
    ds_emb_mapping: /large_storage/ctc/userspace/aadduri/data/auxillary/esm_basecount_tahoe_cellxgene_ds_mapping.torch
    num: 19790
    size: 5120
    valid_genes_masks: /large_storage/ctc/userspace/aadduri/data/auxillary/esm_basecount_tahoe_cellxgene_valid_masks.torch
  esm2-cellxgene-tahoe:
    all_embeddings: /large_storage/ctc/userspace/aadduri/data/auxillary/Homo_sapiens.GRCh38.gene_symbol_to_embedding_ESM2.pt
    ds_emb_mapping: /large_storage/ctc/userspace/aadduri/data/auxillary/esm_basecount_tahoe_cellxgene_ds_mapping.torch
    num: 19790
    size: 5120
    valid_genes_masks: /large_storage/ctc/userspace/aadduri/data/auxillary/esm_basecount_tahoe_cellxgene_valid_masks.torch
experiment:
  checkpoint:
    every_n_train_steps: 1000
    monitor: trainer/train_loss
    path: /data/checkpoints
    save_top_k: 4
  compiled: false
  ddp_timeout: 3600
  deaware: false
  limit_val_batches: 100
  local: local
  name: vci_1.5.0_600M_basecount_tahoe_cxg
  num_epochs: 16
  num_gpus_per_node: 8
  num_nodes: 2
  port: 12400
  profile:
    enable_profiler: false
    max_steps: 110
    profile_steps:
    - 10
    - 100
  val_check_interval: 1000
loss:
  apply_normalization: false
  kernel: energy
  name: tabular
  uniformity: false
model:
  batch_size: 48
  batch_tabular_loss: false
  counts: true
  d_hid: 2048
  dataset_correction: true
  dropout: 0.1
  ema: false
  ema_decay: 0.999
  ema_update_interval: 1000
  emsize: 2048
  name: vci
  nhead: 16
  nlayers: 16
  num_downsample: 1
  output_dim: 2048
  rda: true
  sample_rda: false
  use_flash_attention: true
  variable_masking: true
optimizer:
  end: 1.0
  gradient_accumulation_steps: 8
  max_grad_norm: 0.8
  max_lr: 1.0e-05
  reset_lr_on_restart: false
  start: 0.01
  weight_decay: 0.01
  zclip: true
task:
  mask: 0.2
tokenizer:
  token_dim: 5120
validations:
  diff_exp:
    dataset: /large_storage/ctc/datasets/cellxgene/processed/rpe1_top5000_variable.h5ad
    dataset_name: rpe1_top5000_variable
    enable: false
    eval_interval_multiple: 10
    method: null
    obs_filter_label: non-targeting
    obs_pert_col: gene
    top_k_rank: 200
  perturbation:
    ctrl_label: non-targeting
    dataset: /large_storage/ctc/datasets/vci/validation/replogle_perturbation.h5ad
    dataset_name: replogle_perturbation
    enable: false
    eval_interval_multiple: 10
    pert_col: gene
wandb:
  enable: true
  project: vci