hparams.yaml

experiment_group: training
run_name: roberta-base_2024-04-23T15-11-06
seed: 42
model:
  name: roberta-base
  revision: null
  seed: 42
  base_model: roberta-base
estimator:
  accelerator: gpu
  precision: bf16-true
  deterministic: true
  tf32_mode: high
  convert_to_bettertransformer: false
fit:
  max_epochs: 20
  min_epochs: null
  optimizer_kwargs:
    name: adamw
    lr: 3.0e-05
    init_kwargs:
      fused: true
  scheduler_kwargs:
    name: constant_schedule_with_warmup
    num_warmup_steps: 2000
  log_interval: 100
  enable_progress_bar: true
  limit_train_batches: null
data:
  batch_size: 32
  eval_batch_size: 128
  shuffle: true
  replacement: false
  data_seed: 42
  drop_last: false
  num_workers: 8
  pin_memory: true
  persistent_workers: false
  multiprocessing_context: null
  max_length: 512
root_path: /home/pl487/coreset-project
data_path: /home/pl487/coreset-project/data/processed
dataset: mnli
dataset_split: train
evaluation: null
loggers:
  tensorboard:
    _target_: energizer.loggers.TensorBoardLogger
    root_dir: ./
    name: tb_logs
    version: null
callbacks:
  timer:
    _target_: energizer.active_learning.callbacks.Timer
  lr_monitor:
    _target_: energizer.callbacks.lr_monitor.LearningRateMonitor
  model_checkpoint:
    _target_: energizer.callbacks.model_checkpoint.ModelCheckpoint
    dirpath: .checkpoints
    stage: train
    frequency: 1:epoch
user:
  id: pl487