| backbone_config_path: /notebooks/models/exp222 | |
| best_model_path: /notebooks/models/exp222/models/fold_0_2023_best.pth | |
| checkpoint_path: /notebooks/models/exp222/chkp/fold_0_2023_chkp.pth | |
| config_path: /notebooks/models/exp222/config.yaml | |
| criterion: | |
| criterion_type: BCEWithLogitsLoss | |
| mcrmse_loss: | |
| weights: | |
| - 0.5 | |
| - 0.5 | |
| mse_loss: | |
| reduction: mean | |
| rmse_loss: | |
| eps: 1.0e-09 | |
| reduction: mean | |
| smooth_l1_loss: | |
| beta: 0.1 | |
| reduction: mean | |
| data_dir: /notebooks/data | |
| dataset: | |
| bucket_batch_sampler: | |
| bucket_size: 400 | |
| noise_factor: 0.2 | |
| folds: true | |
| labels: | |
| - generated | |
| max_length: 256 | |
| sampler_type: StratifiedBatchSampler | |
| train_batch_size: 48 | |
| train_sources: | |
| - daigt | |
| - persuade | |
| - persuade_gpt | |
| - persuade_humanized_1 | |
| - persuade_gpt_patially_rewritten | |
| - persuade_gpt_patially_rewritten_05 | |
| - persuade_humanized_easy_1 | |
| - daigt_gpt_patially_rewritten | |
| - llama-mistral-partially-r | |
| - moth | |
| - books | |
| - neural-chat-7b | |
| - nbroad | |
| valid_batch_size: 48 | |
| valid_sources: | |
| - none | |
| debug: false | |
| exp_name: exp222_seed2023 | |
| external_dir: /notebooks/data/external | |
| fold: 0 | |
| interim_dir: /notebooks/data/interim | |
| log_path: /notebooks/models/exp222/logs/fold-0.log | |
| logger: | |
| job_type: training | |
| project: DAIGT-AIE | |
| train_print_frequency: 100 | |
| use_wandb: true | |
| valid_print_frequency: 100 | |
| model: | |
| architecture_type: CustomModel | |
| attention_dropout: 0.1 | |
| backbone_type: microsoft/deberta-v3-large | |
| dropout: 0.05 | |
| freeze_embeddings: false | |
| freeze_n_layers: 0 | |
| gem_pooling: | |
| eps: 1.0e-06 | |
| p: 3 | |
| gradient_checkpointing: false | |
| load_embeddings: true | |
| load_head: true | |
| load_n_layers: 24 | |
| load_parts: false | |
| pooling_type: MeanPooling | |
| reinitialize_n_layers: 0 | |
| state_from_model: None | |
| models_dir: /notebooks/models | |
| optimizer: | |
| beta1: 0.9 | |
| beta2: 0.999 | |
| decoder_lr: 2.0e-05 | |
| embeddings_lr: 2.0e-05 | |
| encoder_lr: 2.0e-05 | |
| eps: 1.0e-06 | |
| group_lr_multiplier: 1 | |
| n_groups: 1 | |
| weight_decay: 0.01 | |
| processed_dir: /notebooks/data/processed | |
| raw_dir: /notebooks/data/raw | |
| run_dir: /notebooks/models/exp222 | |
| run_id: exp222_seed2023_fold0 | |
| run_name: exp222_seed2023_fold0 | |
| scheduler: | |
| cosine_schedule_with_warmup: | |
| n_cycles: 0.5 | |
| n_warmup_steps: 0 | |
| type: cosine_schedule_with_warmup | |
| seed: 2023 | |
| tokenizer: null | |
| tokenizer_path: /notebooks/models/exp222/tokenizer | |
| training: | |
| apex: true | |
| epochs: 4 | |
| evaluate_n_times_per_epoch: 1 | |
| gradient_accumulation_steps: 1 | |
| max_grad_norm: 1000 | |