examples/nlp/glue_benchmark/glue_benchmark_config.yaml

# GLUE Benchmark with pre-trained BERT models
supported_tasks: &supported_tasks ['cola', 'sst-2', 'mrpc', 'sts-b', 'qqp', 'mnli', 'qnli', 'rte', 'wnli']

trainer:
  devices: 1 # the number of gpus, 0 for CPU
  num_nodes: 1
  max_epochs: 3
  max_steps: -1 # precedence over max_epochs
  accumulate_grad_batches: 1 # accumulates grads every k batches
  precision: 16
  accelerator: gpu
  strategy: ddp
  enable_checkpointing: False  # Provided by exp_manager
  logger: False  # Provided by exp_manager

model:
  task_name: &task_name mrpc # choose from: ["cola", "sst-2", "mrpc", "sts-b", "qqp", "mnli", "qnli", "rte", "wnli"] GLUE task name, MNLI includes both matched and mismatched dev sets
  supported_tasks: *supported_tasks
  output_dir: null # dir to write write predictions
  nemo_path: null # filename to save the model and associated artifacts to .nemo file
  dataset:
    data_dir: ??? # /path/to/data
    max_seq_length: 128
    use_cache: true

    # shared across dataloaders:
    num_workers:  2
    pin_memory: false
    drop_last: false

  train_ds:
    ds_item: 'train.tsv'
    shuffle: true
    num_samples: -1
    batch_size: 32

  validation_ds:
    ds_item: 'dev.tsv' # for MNLI 'dev_matched.tsv' and 'dev_mismatched.tsv' will de used
    shuffle: false
    num_samples: -1
    batch_size: 32

  tokenizer:
      tokenizer_name: ${model.language_model.pretrained_model_name} # or sentencepiece
      vocab_file: null # path to vocab file
      tokenizer_model: null # only used if tokenizer is sentencepiece
      special_tokens: null # only necessary for adding transformer/bert-specific special tokens to tokenizer if the tokenizer does not already have these inherently.

  language_model:
    pretrained_model_name: bert-base-uncased
    lm_checkpoint: null
    config_file: null # json file, precedence over config
    config: null

  optim:
    name: adam
    lr: 5e-5
    weight_decay: 0.00

    sched:
      name: WarmupAnnealing
      # Scheduler params
      warmup_steps: null
      warmup_ratio: 0.1
      last_epoch: -1

      # pytorch lightning args
      monitor: val_loss
      reduce_on_plateau: false

exp_manager:
  exp_dir: null  # exp_dir for your experiment, if None, defaults to "./NeMo_experiments"
  name: *task_name # The name of your model
  create_tensorboard_logger: True  # Whether you want exp_manger to create a tb logger
  create_checkpoint_callback: True  # Whether you want exp_manager to create a modelcheckpoint callback

hydra:
  run:
    dir: .
  job_logging:
    root:
      handlers: null