liyouli

I also encountered this problem, when I used a pre-trained model XLNetForQuestionAnsweringSimple to predict a QA task. # result = evaluate(args, model, tokenizer, prefix=global_step) File "run_factoid_pytorch.py", line 718, in evaluate...

如果是英文的文本的话可以在预处理的时候将多个空格都处理成一个空格：# here !!! fix the problem ``` merged_multi_span_data = [] for p in data['data'][0]['paragraphs']: for ques in p['qas']: p['context'] = " ".join(p['context'].split()) # here !!! current_example = {"id": len(merged_multi_span_data) +...

Thank you for your timely response!

后面解决这个问题了，是因为之前训练的时候由于机器的磁盘空间不足，所以模型保存载挂载的afs目录下，读取模型的时候不支持直接afs直接读取，从afs目录拷贝到磁盘上就行了； For anyone who meets this error, copy the model into your local disk, this problem will be fixed.

I encountered the same problem.