model_config.json

{
  "model_path": "meta-llama/Llama-3.2-3B-Instruct",
  "train_data_path": "data/train_target_indirect_biased_order1.json",
  "val_data_path": "data/train_target_true_eval.json",
  "batch_size": 2,
  "weight_decay": 0.0,
  "learning_rate": 2e-05,
  "num_train_epochs": 5,
  "max_train_steps": null,
  "gradient_accumulation_steps": 1,
  "lr_scheduler_type": "constant",
  "num_warmup_steps": 0.0,
  "logfile": "exp/llama32_3B_instruct_contaminate_mmlupro_with_indirect_eval_order1/log.txt",
  "outputdir": "exp/llama32_3B_instruct_contaminate_mmlupro_with_indirect_eval_order1",
  "log_interval": 200,
  "save_interval": 20000,
  "master_port": "12355",
  "lora_config": "./config/lora_config.json",
  "losstype": "mse",
  "load_from": "",
  "unlearnmode": false,
  "probetype": "",
  "probelayer": -1,
  "alpha": 1.0
}