training_args.json

{
  "model_name_or_path": "Salesforce/codegen-2B-mono",
  "train_file": "/kaggle/working/repo/data_raw/outdated_y+_FINAL.json",
  "valid_file": "/kaggle/working/repo/data_raw/outdated_y+_FINAL.json",
  "output_dir": "/kaggle/working/outputs/prompt_tuning_codegen2b_global",
  "num_virtual_tokens": 20,
  "prompt_init": "random",
  "prompt_init_text": "Generate the next line:",
  "max_input_length": 512,
  "max_target_length": 128,
  "max_seq_length": 640,
  "per_device_train_batch_size": 1,
  "per_device_eval_batch_size": 1,
  "gradient_accumulation_steps": 32,
  "learning_rate": 0.005,
  "num_train_epochs": 9,
  "warmup_ratio": 0.03,
  "weight_decay": 0.0,
  "max_grad_norm": 1.0,
  "logging_steps": 10,
  "eval_steps": 200,
  "save_steps": 200,
  "seed": 42,
  "fp16": true,
  "bf16": false,
  "num_workers": 2,
  "resume_from_checkpoint": null
}