File size: 1,581 Bytes
c73d727 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 | # Training configuration for the OmicsFM checkpoint in this folder.
# Extracted verbatim from best_model.ckpt.
checkpoint: bulk_transcriptomics
training_run: bulk_trans
epoch: 53
val_loss: 2.6126632893882027
vocabulary_size: 20272
config:
data:
train_path: data/bulk_trans/protein_mapped/train.h5ad
valid_path: data/bulk_trans/protein_mapped/valid.h5ad
test_path: data/bulk_trans/protein_mapped/test.h5ad
num_bins: 10
detect_groups: false
cache_build_workers: 1
transfer:
checkpoint_path: C:\Users\sander\OneDrive\Bureaublad\Projects\prot_GPT\code\protgpt\model\bulk_trans\best_model.ckpt
freeze_transformer: false
freeze_epochs: 0
device: cuda
model:
d_model: 256
n_heads: 8
n_layers: 6
d_ff: 1024
dropout: 0.1
input_features: 1024
feature_context_ratio: 0.85
use_esmc: false
fasta_path: fasta/human_proteome_canonical_31032022.fasta
esmc_model: esmc_600m
esmc_cache: data/esmc_human_cache.pt
absent_species:
enabled: false
input_absent: 50
absent_context_ratio: 0.5
protein_groups:
enabled: false
max_group_size: 20
input_groups: 256
group_context_ratio: 0.2
encoder: mean
attention:
n_heads: 4
n_encoder_layers: 1
training:
batch_size: 64
learning_rate: 0.0001
weight_decay: 1.0e-05
max_epochs: 200
val_frequency: 0.25
early_stopping_patience: 5000
early_stopping_delta: 0.001
gradient_clip: 1.0
num_workers: 0
output:
save_dir: model
experiment_name: bulk_trans
|