tiny_model_full_config.yaml

# This config demostrates specifying full config options in single yaml file
# Optionally, users can add config groups to hold some frequently used configs and
# refer them in `defaults` to make this config more concise as shown in
# tiny_model_mixed_config.yaml

module:
  # equivalent to setting `module/optim: adamw` under `defaults`
  optim:
    _target_: torch.optim.AdamW
    lr: 1.0e-05
    betas:
    - 0.9
    - 0.999
    eps: 1.0e-08
    weight_decay: 0
    amsgrad: false
  # equivalent to setting `module/model: xlmrbase_classifier_tiny` under `defaults`
  model:
    _target_: torchtext.models.RobertaBundle.build_model
    encoder_conf:
      _target_: torchtext.models.RobertaEncoderConf
      vocab_size: 102
      embedding_dim: 8
      ffn_dimension: 8
      padding_idx: 1
      max_seq_len: 128
      num_attention_heads:  1
      num_encoder_layers: 1
      dropout: 0.1
      scaling: null
      normalize_before: False
    head:
      _target_: torchtext.models.RobertaClassificationHead
      num_classes: 2
      input_dim: 8
      inner_dim: 8
      dropout: 0.4
    freeze_encoder: True
    checkpoint: null

# equivalent to setting `datamodule: doc_classification_datamodule` under `defaults`
datamodule:
  _target_: torchrecipes.text.doc_classification.datamodule.doc_classification.DocClassificationDataModule.from_config
  columns:
  - text
  - label
  label_column: label
  batch_size: 16
  num_workers: 0
  drop_last: False
  pin_memory: False
  dataset:
    _target_: torchtext.datasets.sst2.SST2
    root: ~/.torchtext/cache

# equivalent to setting `transform: doc_classification_transform_tiny` under `defaults`
transform:
  transform:
    _target_: torchrecipes.text.doc_classification.transform.doc_classification_text_transform.DocClassificationTextTransform
    vocab_path: https://download.pytorch.org/models/text/xlmr.vocab_example.pt
    spm_model_path: https://download.pytorch.org/models/text/xlmr.sentencepiece_example.bpe.model
  label_transform:
    _target_: torchtext.transforms.LabelToIndex
    label_names:
    - "0"
    - "1"
  num_labels: 2

# equivalent to setting `trainer: cpu` under `defaults`
trainer:
  _target_: pytorch_lightning.trainer.Trainer
  accelerator: cpu
  devices: null
  strategy: null
  max_epochs: 1
  default_root_dir: /tmp/doc_classification/torchrecipes
  enable_checkpointing: true
  fast_dev_run: false
  logger:
    _target_: pytorch_lightning.loggers.TensorBoardLogger
    save_dir: /tmp/torchrecipes/doc_classification/logs