2026-04-20 11:06:55,458 - INFO - ===== Training Configuration =====
2026-04-20 11:06:55,460 - INFO - model_name           : microsoft/graphcodebert-base
2026-04-20 11:06:55,464 - INFO - output_dir           : output_checkpoints/graphcodebert-mixcode-fft
2026-04-20 11:06:55,466 - INFO - num_epochs           : 4
2026-04-20 11:06:55,467 - INFO - batch_size           : 64
2026-04-20 11:06:55,468 - INFO - learning_rate        : 5e-05
2026-04-20 11:06:55,471 - INFO - max_length           : 512
2026-04-20 11:06:55,472 - INFO - num_labels           : 2
2026-04-20 11:06:55,474 - INFO - use_wandb            : True
2026-04-20 11:06:55,475 - INFO - freeze_base          : True
2026-04-20 11:06:55,476 - INFO - loss_type            : r-drop
2026-04-20 11:06:55,478 - INFO - focal_alpha          : 1.0
2026-04-20 11:06:55,480 - INFO - focal_gamma          : 2.0
2026-04-20 11:06:55,482 - INFO - r_drop_alpha         : 6.0
2026-04-20 11:06:55,483 - INFO - infonce_temperature  : 0.07
2026-04-20 11:06:55,485 - INFO - infonce_weight       : 0.5
2026-04-20 11:06:55,487 - INFO - seed                 : 42
2026-04-20 11:06:55,488 - INFO - resume_from_checkpoint : None
2026-04-20 11:06:55,490 - INFO - save_steps           : 50
2026-04-20 11:06:55,491 - INFO - eval_steps           : 1000
2026-04-20 11:06:55,494 - INFO - label_smoothing      : 0.3
2026-04-20 11:06:55,499 - INFO - adversarial_epsilon  : 0.5
2026-04-20 11:06:55,504 - INFO - use_swa              : True
2026-04-20 11:06:55,508 - INFO - swa_start_epoch      : 2
2026-04-20 11:06:55,510 - INFO - swa_lr               : 1e-05
2026-04-20 11:06:55,511 - INFO - data_augmentation    : True
2026-04-20 11:06:55,512 - INFO - aug_rename_prob      : 0.6
2026-04-20 11:06:55,514 - INFO - aug_format_prob      : 0.6
2026-04-20 11:06:55,516 - INFO - hidden_dropout_prob  : 0.3
2026-04-20 11:06:55,518 - INFO - attention_probs_dropout_prob : 0.3
2026-04-20 11:06:55,519 - INFO - classifier_dropout   : 0.3
2026-04-20 11:06:55,521 - INFO - mixup_alpha          : 1.0
2026-04-20 11:06:55,523 - INFO - low_pass_keep_ratio  : 0.5
2026-04-20 11:06:55,524 - INFO - freq_consistency_weight : 0.2
2026-04-20 11:06:55,526 - INFO - =================================
2026-04-20 11:06:59,772 - INFO - Model placed on cuda
2026-04-20 11:06:59,789 - INFO - ===== Model Architecture =====
2026-04-20 11:06:59,794 - INFO - 
RobertaForSequenceClassification(
  (roberta): RobertaModel(
    (embeddings): RobertaEmbeddings(
      (word_embeddings): Embedding(50265, 768, padding_idx=1)
      (position_embeddings): Embedding(514, 768, padding_idx=1)
      (token_type_embeddings): Embedding(1, 768)
      (LayerNorm): LayerNorm((768,), eps=1e-05, elementwise_affine=True)
      (dropout): Dropout(p=0.1, inplace=False)
    )
    (encoder): RobertaEncoder(
      (layer): ModuleList(
        (0-11): 12 x RobertaLayer(
          (attention): RobertaAttention(
            (self): RobertaSdpaSelfAttention(
              (query): Linear(in_features=768, out_features=768, bias=True)
              (key): Linear(in_features=768, out_features=768, bias=True)
              (value): Linear(in_features=768, out_features=768, bias=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
            (output): RobertaSelfOutput(
              (dense): Linear(in_features=768, out_features=768, bias=True)
              (LayerNorm): LayerNorm((768,), eps=1e-05, elementwise_affine=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
          )
          (intermediate): RobertaIntermediate(
            (dense): Linear(in_features=768, out_features=3072, bias=True)
            (intermediate_act_fn): GELUActivation()
          )
          (output): RobertaOutput(
            (dense): Linear(in_features=3072, out_features=768, bias=True)
            (LayerNorm): LayerNorm((768,), eps=1e-05, elementwise_affine=True)
            (dropout): Dropout(p=0.1, inplace=False)
          )
        )
      )
    )
  )
  (classifier): RobertaClassificationHead(
    (dense): Linear(in_features=768, out_features=768, bias=True)
    (dropout): Dropout(p=0.3, inplace=False)
    (out_proj): Linear(in_features=768, out_features=2, bias=True)
  )
)
2026-04-20 11:06:59,803 - INFO - ===== Parameter Summary =====
2026-04-20 11:06:59,809 - INFO - Total Parameters:         124,647,170
2026-04-20 11:06:59,810 - INFO - Trainable Parameters:     592,130
2026-04-20 11:06:59,811 - INFO - Non-trainable Parameters: 124,055,040
2026-04-20 11:06:59,813 - INFO - ===== Tokenizer Summary =====
2026-04-20 11:06:59,908 - INFO - Vocab size: 50265 | Special tokens: ['<s>', '</s>', '<unk>', '<pad>', '<mask>']
2026-04-20 11:06:59,912 - INFO - ===== End of Architecture Log =====
2026-04-20 11:06:59,917 - INFO - Data augmentation enabled (rename=0.6, format=0.6)
2026-04-20 11:07:10,050 - INFO - === Starting training with MixCode + FFT low-pass consistency ===