# lightning.pytorch==2.3.3
seed_everything: 0
trainer:
  precision: bf16-mixed
  max_steps: 50000
data:
  class_path: lightning_ir.LightningIRDataModule
  init_args:
    num_workers: 1
    train_batch_size: 64
    shuffle_train: true
    train_dataset:
      class_path: lightning_ir.RunDataset
      init_args:
        run_path_or_id: msmarco-passage/train/rank-distillm/set-encoder
        depth: 100
        sample_size: 8
        sampling_strategy: log_random
        targets: score
        normalize_targets: false
model:
  class_path: lightning_ir.BiEncoderModule
  init_args:
    model_name_or_path: bert-base-uncased
    config:
      class_path: lightning_ir.ColConfig
      init_args:
        similarity_function: dot
        query_expansion: true
        attend_to_query_expanded_tokens: true
        query_mask_scoring_tokens: null
        doc_mask_scoring_tokens: punctuation
        query_aggregation_function: mean
        normalize: false
        add_marker_tokens: false
        embedding_dim: 128
        projection: linear
        query_pooling_strategy: mean
        doc_expansion: false
        attend_to_doc_expanded_tokens: false
        doc_pooling_strategy: mean
        sparsification: null
        query_length: 32
        doc_length: 256
    loss_functions:
    - class_path: lightning_ir.SupervisedMarginMSE
    - class_path: lightning_ir.KLDivergence
    - class_path: lightning_ir.InBatchCrossEntropy
      init_args:
        pos_sampling_technique: first
        neg_sampling_technique: first
        max_num_neg_samples: 8
optimizer:
  class_path: torch.optim.AdamW
  init_args:
    lr: 2.0e-05
lr_scheduler:
  class_path: lightning_ir.LinearLRSchedulerWithLinearWarmup
  init_args:
    num_warmup_steps: 5000
    final_value: 0.02
    num_delay_steps: 0