fourier-spectral-norm-classifier/checkpoint-1500/config_hyperparams.json

{
  "train_config": {
    "TEST": false,
    "model_name": "/kaggle/input/models/dzung271828/microsoft-graphcodebert-base/transformers/default/1",
    "output_dir": "training/fourier-spectral-norm-classifier/",
    "num_epochs": 5,
    "max_steps": -1,
    "batch_size": 512,
    "learning_rate": 1e-06,
    "max_length": 512,
    "num_labels": 2,
    "use_wandb": false,
    "freeze_base": true,
    "loss_type": "ce",
    "focal_alpha": 1.0,
    "focal_gamma": 2.0,
    "r_drop_alpha": 6.0,
    "infonce_temperature": 0.07,
    "infonce_weight": 0.5,
    "seed": 42,
    "resume_from_checkpoint": null,
    "save_steps": 500,
    "eval_steps": 500,
    "logging_steps": 5,
    "label_smoothing": 0.5,
    "adversarial_epsilon": 0.5,
    "use_swa": true,
    "swa_start_epoch": 0,
    "swa_lr": 1e-06,
    "data_augmentation": true,
    "aug_rename_prob": 0.7,
    "aug_format_prob": 0.7,
    "weight_decay": 0.1,
    "mixup_alpha": 1.0,
    "low_pass_keep_ratio": 0.5,
    "freq_consistency_weight": 0.2,
    "use_mixcode": true,
    "use_fgm": true,
    "fgm_freq": 5,
    "use_r_drop": true,
    "use_freq_consistency_loss": true,
    "use_attn_spectral": false,
    "attn_spectral_weight": 0.1,
    "attn_spectral_cutoff_ratio": 0.25,
    "hidden_dropout_prob": 0.3,
    "attention_probs_dropout_prob": 0.3,
    "classifier_dropout": 0.4,
    "device": "cuda",
    "torch_compile": true,
    "cache_dir": "./tokenized_cache",
    "use_swa_actual": true,
    "use_fgm_actual": true,
    "use_r_drop_actual": true,
    "use_mixcode_actual": true,
    "use_attn_spectral_actual": false,
    "use_freq_consistency_loss_actual": true,
    "use_spectral_norm": true
  },
  "training_arguments": {
    "output_dir": "training/fourier-spectral-norm-classifier/",
    "num_train_epochs": 5,
    "per_device_train_batch_size": 512,
    "per_device_eval_batch_size": 1024,
    "learning_rate": 1e-06,
    "warmup_steps": 488,
    "weight_decay": 0.1,
    "logging_steps": 5,
    "eval_steps": 500,
    "save_steps": 500,
    "metric_for_best_model": "macro_f1",
    "greater_is_better": true,
    "save_total_limit": 5,
    "fp16": false,
    "seed": 42
  },
  "training_state": {
    "global_step": 1500,
    "epoch": 1.5353121801432958,
    "best_metric": 0.6724504812400831,
    "best_model_checkpoint": "training/fourier-spectral-norm-classifier/checkpoint-1000"
  }
}