Upload config.yaml with huggingface_hub

config.yaml

@@ -1,7 +1,7 @@
 data:
   name: PerturbationDataModule
   kwargs:
-    toml_config_path: /large_storage/ctc/userspace/aadduri/datasets/tahoe_5_holdout/generalization.toml
+    toml_config_path: /large_storage/ctc/userspace/aadduri/datasets/basak_tahoe/generalization.toml
     embed_key: X_hvg
     output_space: gene
     pert_rep: onehot
@@ -11,7 +11,7 @@ data:
     n_basal_samples: 1
     basal_mapping_strategy: random
     should_yield_control_cells: true
-    batch_col: sample
+    batch_col: plate
     pert_col: drugname_drugconc
     cell_type_key: cell_name
     control_pert: DMSO_TF
@@ -19,6 +19,7 @@ data:
     perturbation_features_file: null
     store_raw_basal: false
     int_counts: false
+    barcode: true
   output_dir: null
   debug: true
 model:
@@ -28,7 +29,7 @@ model:
   kwargs:
     cell_set_len: 256
     blur: 0.05
-    hidden_dim: 1488
+    hidden_dim: 1440
     loss: energy
     confidence_head: false
     n_encoder_layers: 4
@@ -40,21 +41,22 @@ model:
     finetune_vci_decoder: false
     residual_decoder: false
     decoder_loss_weight: 1.0
-    batch_encoder: false
+    batch_encoder: true
     nb_decoder: false
     mask_attn: false
     use_effect_gating_token: false
+    use_basal_projection: false
     distributional_loss: energy
-    init_from: /large_storage/ctc/userspace/aadduri/preprint/tahoe_scaling_cs256/cs256/checkpoints/last.ckpt
+    init_from: null
     transformer_backbone_key: llama
     transformer_backbone_kwargs:
       max_position_embeddings: 256
-      hidden_size: 1488
-      intermediate_size: 5952
-      num_hidden_layers: 6
+      hidden_size: 1440
+      intermediate_size: 4416
+      num_hidden_layers: 4
       num_attention_heads: 12
       num_key_value_heads: 12
-      head_dim: 124
+      head_dim: 120
       use_cache: false
       attention_dropout: 0.0
       hidden_dropout: 0.0
@@ -70,20 +72,22 @@ training:
   weight_decay: 0.0005
   batch_size: 64
   lr: 0.0001
-  max_steps: 10
+  max_steps: 400000
   train_seed: 42
-  val_freq: 100
-  ckpt_every_n_steps: 1
+  val_freq: 4000
+  ckpt_every_n_steps: 4000
   gradient_clip_val: 10
   loss_fn: mse
+  devices: 1
+  strategy: auto
 wandb:
-  entity: arcinstitute
-  project: vci1
-  local_wandb_dir: /large_storage/ctc/userspace/aadduri/wandb_dir/vci1
+  entity: your_entity_name
+  project: state
+  local_wandb_dir: ./wandb_logs
   tags:
-  - tahoe_scaling
-name: cs256
-output_dir: /large_storage/ctc/userspace/aadduri/preprint/tahoe_fig2_release
+  - basak_tahoe
+name: cs256_batch_encoder
+output_dir: /large_storage/ctc/userspace/aadduri/preprint/basak_tahoe
 use_wandb: true
 overwrite: false
 return_adatas: false