deep-learning - 为什么运行 run_squad.py 后看不到结果？

Question

我正在使用以下评论从 BERT 预测我的自定义数据集：

python run_squad.py --model_type bert --model_name_or_path salti/bert-base-multilingual-cased-finetuned-squad --output_dir models/salti/bert-base-multilingual-cased-finetuned-squad --data_dir data/squad --predict_file chatbot_dataset_new.json --do_eval --version_2_with_negative --do_lower_case --per_gpu_eval_batch_size 12 --max_seq_length 384 --doc_stride 128 --vocab_file=salti/bert-base-multilingual-cased-finetuned-squad/vocab.txt --bert_config_file=salti/bert-base-multilingual-cased-finetuned-squad/config.json --do_predict True

但是，预测完成后我没有看到任何结果。发生了什么？我能看到的唯一几行是：

INFO:tensorflow:Writing predictions to: models/salti/bert-base-multilingual-cased-finetuned-squad/predictions.json
I0518 22:48:42.233764 140135444539200 run_squad.py:745] Writing predictions to: models/salti/bert-base-multilingual-cased-finetuned-squad/predictions.json
INFO:tensorflow:Writing nbest to: models/salti/bert-base-multilingual-cased-finetuned-squad/nbest_predictions.json
I0518 22:48:42.234008 140135444539200 run_squad.py:746] Writing nbest to: models/salti/bert-base-multilingual-cased-finetuned-squad/nbest_predictions.json

我想看到类似的东西

Results = {
    # a) scores averaged over all examples in the dev set
    'exact': 66.25958056093658,         
    'f1': 69.66994428499025,            
    'total': 11873,  # number of examples in the dev set
    
    # b) scores averaged over only positive examples (have answers)
    'HasAns_exact': 68.91025641025641,  
    'HasAns_f1': 75.74076391627662,     
    'HasAns_total': 5928, # number of positive examples
    
    # c) scores averaged over only negative examples (no answers)
    'NoAns_exact': 63.61648444070648, 
    'NoAns_f1': 63.61648444070648, 
    'NoAns_total': 5945, # number of negative examples
    
    # d) given probabilities of no-answer for each example, what would the best scores and thresholds be?
    'best_exact': 66.25958056093658, 
    'best_exact_thresh': 0.0, 
    'best_f1': 69.66994428499046, 
    'best_f1_thresh': 0.0
}

任何人都可以帮助我吗？

标签： deep-learningbert-language-model