params.yaml

TrainingArguments:
  num_train_epochs: 1
  warmup_steps: 100
  per_device_train_batch_size: 1
  weight_decay: 0.01
  logging_steps: 5
  evaluation_strategy: "no"
  # eval_steps: 500
  save_steps: 1e6
  gradient_accumulation_steps: 4




# output_dir=self.config.root_dir,
#             num_train_epochs=1,
#             warmup_steps=100,
#             per_device_train_batch_size=1,
#             per_device_eval_batch_size=1,
#             weight_decay=0.01,
#             logging_steps=5,
#             evaluation_strategy="no",  # no evaluation for now
#             save_steps=1e6,
#             gradient_accumulation_steps=4,
#             fp16=True if torch.cuda.is_available() else False