data:
  type: triplets
  TRAIN_DATA_DIR: /home/user/Projects/SPLADE-PT-BR/splade/data/pt/triplets
  VALIDATION_DATA_DIR: /home/user/Projects/SPLADE-PT-BR/splade/data/pt/val_retrieval
  QREL_PATH: /home/user/Projects/SPLADE-PT-BR/splade/data/pt/val_retrieval/qrel.json
train:
  model:
    _target_: splade.models.transformer_rep.Splade
    model_type_or_dir: neuralmind/bert-base-portuguese-cased
config:
  lr: 2.0e-05
  seed: 123
  gradient_accumulation_steps: 4
  weight_decay: 0.01
  validation_metrics:
  - MRR@10
  pretrained_no_yaml_config: false
  nb_iterations: 150000
  train_batch_size: 8
  eval_batch_size: 16
  index_retrieval_batch_size: 16
  record_frequency: 1000
  train_monitoring_freq: 500
  warmup_steps: 6000
  max_length: 256
  fp16: true
  matching_type: splade
  monitoring_ckpt: true
  tokenizer_type: neuralmind/bert-base-portuguese-cased
  loss: InBatchPairwiseNLL
  checkpoint_dir: experiments/pt/checkpoint
  index_dir: experiments/pt/index
  out_dir: experiments/pt/out
  regularization:
    FLOPS:
      lambda_q: 0.0003
      lambda_d: 0.0001
      T: 50000
index: {}
retrieve_evaluate: {}
flops: {}
init_dict:
  model_type_or_dir: neuralmind/bert-base-portuguese-cased
  fp16: true