File size: 1,789 Bytes
550665c |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 |
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
# SketchModelArguments
sketch_model_name: google/electra-base-discriminator
sketch_architectures: ElectraForSequenceClassification
# IntensiveModelArguments
intensive_model_name: google/electra-base-discriminator
intensive_model_mode: transfer
intensive_architectures: ElectraForQuestionAnsweringAVPool
TrainingArguments:
# report_to: wandb
run_name: squadv2-electra-base-sketch,squadv2-electra-base-intensive
output_dir: outputs
overwrite_output_dir: False
learning_rate: 2e-5
evaluation_strategy: epoch
save_strategy: steps # Save checkpoints every specified number of steps
# save_steps: 5000 # Save model checkpoints every 5000 steps
save_steps: 5000
save_total_limit: 2 # Maximum number of checkpoints to keep
# load_best_model_at_end: True # Disable to avoid loading the best model at the end
# no need to specify checkpoint_dir, it defaults to output_dir
# no need to specify logging_dir, it defaults to output_dir
per_device_train_batch_size: 512
per_device_eval_batch_size: 512
num_train_epochs: 10.0
# no need to specify metric_for_best_model for resuming from checkpoints
no_cuda: False
fp16: True
warmup_ratio: 0.1
weight_decay: 0.01
|