def get_transformer_kwargs()

in muss/mining/training.py [0:0]

• 41 lines of code
• 2 McCabe index (conditional complexity)

def get_transformer_kwargs(dataset, language, use_access, use_short_name=False):
    kwargs = {
        'dataset': dataset,
        'parametrization_budget': 128,
        'predict_files': get_predict_files(language),
        'train_kwargs': {
            'ngpus': 8,
            'arch': 'bart_large',
            'max_tokens': 4096,
            'truncate_source': True,
            'layernorm_embedding': True,
            'share_all_embeddings': True,
            'share_decoder_input_output_embed': True,
            'required_batch_size_multiple': 1,
            'criterion': 'label_smoothed_cross_entropy',
            'lr': 3e-04,
            'label_smoothing': 0.1,
            'dropout': 0.1,
            'attention_dropout': 0.1,
            'weight_decay': 0.01,
            'optimizer': 'adam',
            'adam_betas': '(0.9, 0.999)',
            'adam_eps': 1e-08,
            'clip_norm': 0.1,
        },
        'preprocessors_kwargs': {
            'SentencePiecePreprocessor': {
                'vocab_size': 32000,
                'input_filepaths': [
                    get_data_filepath(dataset, 'train', 'complex'),
                    get_data_filepath(dataset, 'train', 'simple'),
                ],
            }
            # 'SentencePiecePreprocessor': {'vocab_size': 32000, 'input_filepaths':  [get_dataset_dir('enwiki') / 'all_sentences']}
        },
        'evaluate_kwargs': get_evaluate_kwargs(language),
    }
    if use_access:
        kwargs['preprocessors_kwargs'] = add_dicts(
            get_access_preprocessors_kwargs(language, use_short_name=use_short_name), kwargs['preprocessors_kwargs']
        )
    return kwargs