plants.yaml

defaults:
  - default

# Training dataset: Angiosperm 16 genomes
dataset_name: kuleshov-group/Angiosperm_16_genomes

# Batch size configuration
batch_size: 2048 # Total effective batch size
per_device_batch_size: 128 # Batch size per device (adjust based on GPU memory)
num_workers: 8

# Evaluation datasets
evals:
  - name: maize_af
    dataset_name: plantcad/maize-allele-frequency
    dataset_config: null
    split: validation
    genome_url: https://ftp.ensemblgenomes.ebi.ac.uk/pub/plants/release-62/fasta/zea_mays/dna/Zea_mays.Zm-B73-REFERENCE-NAM-5.0.dna_sm.toplevel.fa.gz
    filter_name: none
    window_size: 512
    batch_size: 128
    label_column: AF # Allele frequency column
    transform: identity # No transform for regression (default: identity)
    metrics: [pearson, spearman] # Correlation metrics for regression task