| RetroDataModelArguments: | |
| # DataArguments | |
| max_seq_length: 512 | |
| max_answer_length: 30 | |
| doc_stride: 128 | |
| return_token_type_ids: True | |
| pad_to_max_length: True | |
| preprocessing_num_workers: 5 | |
| overwrite_cache: False | |
| version_2_with_negative: True | |
| null_score_diff_threshold: 0.0 | |
| rear_threshold: 0.0 | |
| n_best_size: 20 | |
| use_choice_logits: False | |
| start_n_top: -1 | |
| end_n_top: -1 | |
| beta1: 1 | |
| beta2: 1 | |
| best_cof: 1 | |
| # ModelArguments | |
| use_auth_token: False | |
| # SketchModelArguments | |
| sketch_revision: en-electra-large-sketch | |
| sketch_model_name: jinmang2/retro-reader | |
| # sketch_model_mode: transfer | |
| sketch_architectures: ElectraForSequenceClassification | |
| # IntensiveModelArguments | |
| intensive_revision: en-electra-large-intensive | |
| intensive_model_name: jinmang2/retro-reader | |
| # intensive_model_mode: transfer | |
| intensive_architectures: ElectraForQuestionAnsweringAVPool | |
| TrainingArguments: | |
| output_dir: outputs | |
| no_cuda: True # If you want to use cuda, | |
| # change `no_cuda` to False and `fp16` to True | |
| per_device_train_batch_size: 1 | |
| per_device_eval_batch_size: 12 |