| data: | |
| name: PerturbationDataModule | |
| kwargs: | |
| toml_config_path: /large_storage/ctc/userspace/aadduri/datasets/basak_tahoe/generalization.toml | |
| embed_key: X_hvg | |
| output_space: gene | |
| pert_rep: onehot | |
| basal_rep: sample | |
| num_workers: 24 | |
| pin_memory: true | |
| n_basal_samples: 1 | |
| basal_mapping_strategy: random | |
| should_yield_control_cells: true | |
| batch_col: plate | |
| pert_col: drugname_drugconc | |
| cell_type_key: cell_name | |
| control_pert: DMSO_TF | |
| map_controls: true | |
| perturbation_features_file: null | |
| store_raw_basal: false | |
| int_counts: false | |
| barcode: true | |
| output_dir: null | |
| debug: true | |
| model: | |
| name: PertSets | |
| checkpoint: null | |
| device: cuda | |
| kwargs: | |
| cell_set_len: 256 | |
| blur: 0.05 | |
| hidden_dim: 1440 | |
| loss: energy | |
| confidence_head: false | |
| n_encoder_layers: 4 | |
| n_decoder_layers: 4 | |
| predict_residual: true | |
| softplus: true | |
| freeze_pert: false | |
| transformer_decoder: false | |
| finetune_vci_decoder: false | |
| residual_decoder: false | |
| decoder_loss_weight: 1.0 | |
| batch_encoder: true | |
| nb_decoder: false | |
| mask_attn: false | |
| use_effect_gating_token: false | |
| use_basal_projection: false | |
| distributional_loss: energy | |
| init_from: null | |
| transformer_backbone_key: llama | |
| transformer_backbone_kwargs: | |
| max_position_embeddings: 256 | |
| hidden_size: 1440 | |
| intermediate_size: 4416 | |
| num_hidden_layers: 4 | |
| num_attention_heads: 12 | |
| num_key_value_heads: 12 | |
| head_dim: 120 | |
| use_cache: false | |
| attention_dropout: 0.0 | |
| hidden_dropout: 0.0 | |
| layer_norm_eps: 1.0e-06 | |
| pad_token_id: 0 | |
| bos_token_id: 1 | |
| eos_token_id: 2 | |
| tie_word_embeddings: false | |
| rotary_dim: 0 | |
| use_rotary_embeddings: false | |
| training: | |
| wandb_track: false | |
| weight_decay: 0.0005 | |
| batch_size: 64 | |
| lr: 0.0001 | |
| max_steps: 400000 | |
| train_seed: 42 | |
| val_freq: 4000 | |
| ckpt_every_n_steps: 4000 | |
| gradient_clip_val: 10 | |
| loss_fn: mse | |
| devices: 1 | |
| strategy: auto | |
| wandb: | |
| entity: your_entity_name | |
| project: state | |
| local_wandb_dir: ./wandb_logs | |
| tags: | |
| - basak_tahoe | |
| name: cs256_batch_encoder | |
| output_dir: /large_storage/ctc/userspace/aadduri/preprint/basak_tahoe | |
| use_wandb: true | |
| overwrite: false | |
| return_adatas: false | |
| pred_adata_path: null | |
| true_adata_path: null | |