dzungpham/graphcodebert-code-classification

12026-04-28 11:00:57,792 - INFO - ===== Training Configuration =====22026-04-28 11:00:57,793 - INFO - model_name           : microsoft/graphcodebert-base32026-04-28 11:00:57,793 - INFO - output_dir           : output_checkpoints/graphcodebert-vanilla/42026-04-28 11:00:57,794 - INFO - num_epochs           : 352026-04-28 11:00:57,796 - INFO - max_steps            : -162026-04-28 11:00:57,796 - INFO - batch_size           : 25672026-04-28 11:00:57,797 - INFO - learning_rate        : 2e-0582026-04-28 11:00:57,798 - INFO - max_length           : 51292026-04-28 11:00:57,799 - INFO - num_labels           : 2102026-04-28 11:00:57,800 - INFO - use_wandb            : True112026-04-28 11:00:57,800 - INFO - freeze_base          : True122026-04-28 11:00:57,801 - INFO - loss_type            : ce132026-04-28 11:00:57,801 - INFO - focal_alpha          : 1.0142026-04-28 11:00:57,802 - INFO - focal_gamma          : 2.0152026-04-28 11:00:57,803 - INFO - r_drop_alpha         : 6.0162026-04-28 11:00:57,803 - INFO - infonce_temperature  : 0.07172026-04-28 11:00:57,805 - INFO - infonce_weight       : 0.5182026-04-28 11:00:57,806 - INFO - seed                 : 42192026-04-28 11:00:57,806 - INFO - wandb_run_name       : graphcodebert-vanilla202026-04-28 11:00:57,808 - INFO - resume_from_checkpoint : None212026-04-28 11:00:57,808 - INFO - save_steps           : 100222026-04-28 11:00:57,809 - INFO - eval_steps           : 50232026-04-28 11:00:57,810 - INFO - logging_steps        : 5242026-04-28 11:00:57,810 - INFO - label_smoothing      : 0252026-04-28 11:00:57,811 - INFO - adversarial_epsilon  : 0262026-04-28 11:00:57,812 - INFO - use_swa              : False272026-04-28 11:00:57,813 - INFO - swa_start_epoch      : 0282026-04-28 11:00:57,814 - INFO - swa_lr               : 1e-05292026-04-28 11:00:57,815 - INFO - data_augmentation    : False302026-04-28 11:00:57,816 - INFO - aug_rename_prob      : 0.0312026-04-28 11:00:57,817 - INFO - aug_format_prob      : 0.0322026-04-28 11:00:57,817 - INFO - mixup_alpha          : 0.0332026-04-28 11:00:57,818 - INFO - low_pass_keep_ratio  : 0.5342026-04-28 11:00:57,819 - INFO - freq_consistency_weight : 0.0352026-04-28 11:00:57,819 - INFO - hidden_dropout_prob  : 0.3362026-04-28 11:00:57,820 - INFO - attention_probs_dropout_prob : 0.3372026-04-28 11:00:57,821 - INFO - classifier_dropout   : 0.3382026-04-28 11:00:57,822 - INFO - =================================392026-04-28 11:00:58,827 - INFO - Model placed on cuda402026-04-28 11:00:58,830 - INFO - ===== Model Architecture =====412026-04-28 11:00:58,832 - INFO - 42RobertaForSequenceClassification(43  (roberta): RobertaModel(44    (embeddings): RobertaEmbeddings(45      (word_embeddings): Embedding(50265, 768, padding_idx=1)46      (position_embeddings): Embedding(514, 768, padding_idx=1)47      (token_type_embeddings): Embedding(1, 768)48      (LayerNorm): LayerNorm((768,), eps=1e-05, elementwise_affine=True)49      (dropout): Dropout(p=0.3, inplace=False)50    )51    (encoder): RobertaEncoder(52      (layer): ModuleList(53        (0-11): 12 x RobertaLayer(54          (attention): RobertaAttention(55            (self): RobertaSdpaSelfAttention(56              (query): Linear(in_features=768, out_features=768, bias=True)57              (key): Linear(in_features=768, out_features=768, bias=True)58              (value): Linear(in_features=768, out_features=768, bias=True)59              (dropout): Dropout(p=0.3, inplace=False)60            )61            (output): RobertaSelfOutput(62              (dense): Linear(in_features=768, out_features=768, bias=True)63              (LayerNorm): LayerNorm((768,), eps=1e-05, elementwise_affine=True)64              (dropout): Dropout(p=0.3, inplace=False)65            )66          )67          (intermediate): RobertaIntermediate(68            (dense): Linear(in_features=768, out_features=3072, bias=True)69            (intermediate_act_fn): GELUActivation()70          )71          (output): RobertaOutput(72            (dense): Linear(in_features=3072, out_features=768, bias=True)73            (LayerNorm): LayerNorm((768,), eps=1e-05, elementwise_affine=True)74            (dropout): Dropout(p=0.3, inplace=False)75          )76        )77      )78    )79  )80  (classifier): RobertaClassificationHead(81    (dense): Linear(in_features=768, out_features=768, bias=True)82    (dropout): Dropout(p=0.3, inplace=False)83    (out_proj): Linear(in_features=768, out_features=2, bias=True)84  )85)862026-04-28 11:00:58,834 - INFO - ===== Parameter Summary =====872026-04-28 11:00:58,834 - INFO - Total Parameters:         124,647,170882026-04-28 11:00:58,835 - INFO - Trainable Parameters:     592,130892026-04-28 11:00:58,836 - INFO - Non-trainable Parameters: 124,055,040902026-04-28 11:00:58,836 - INFO - ===== Tokenizer Summary =====912026-04-28 11:00:58,850 - INFO - Vocab size: 50265 | Special tokens: ['<s>', '</s>', '<unk>', '<pad>', '<mask>']922026-04-28 11:00:58,851 - INFO - ===== End of Architecture Log =====932026-04-28 11:04:20,159 - INFO - === Starting training with MixCode + FFT low-pass consistency ===942026-04-28 12:00:21,592 - INFO - Training completed successfully.952026-04-28 12:00:23,561 - INFO - Final model saved to output_checkpoints/graphcodebert-vanilla/final_model96