command:
  - python3
  - ${program}
  - --use_auth_token
  - --group_by_length
  - --overwrite_output_dir
  - --fp16
  - --do_lower_case
  - --do_eval
  - --do_train
  - --fuse_loss_wer
  - ${args}
method: random
metric:
  goal: minimize
  name: eval/wer
parameters:
  config_path:
    value: conf/conformer_transducer_bpe_xlarge.yaml
  dataset_config_name:
    value: release3
  dataset_name:
    value: LIUM/tedlium
  eval_split_name:
    value: validation
  evaluation_strategy:
    value: steps
  eval_steps:
    value: 2000
  fused_batch_size:
    value: 8
  learning_rate:
    values:
      - 1e-1
      - 3e-2
      - 1e-2
      - 3e-3
      - 1e-3
      - 3e-4
      - 1e-4
  logging_steps:
    value: 25
  model_name_or_path:
    value: stt_en_conformer_transducer_xlarge
  max_steps:
    value: 8000
  output_dir:
    value: ./sweep_output_dir
  per_device_eval_batch_size:
    value: 4
  per_device_train_batch_size:
    value: 8
  preprocessing_num_workers:
    value: 4
  save_strategy:
    value: "no"
  tokenizer_path:
    value: tokenizer
  train_split_name:
    value: train
  vocab_size:
    value: 1024
  warmup_steps:
    value: 500
  wandb_project:
    value: rnnt-debug-tedlium
  freeze_encoder:
    values:
      - true
      - false
  add_adapter:
    values:
      - true
      - false
  unfreeze_encoder:
    values:
      - true
      - false
  length_column_name:
    value: input_lengths
program: run_speech_recognition_rnnt.py
project: rnnt-debug-tedlium