【NLP】提升文本生成多样性的实用方法

比如用T5模型，训练数据是inputText-outputText格式，预测时do_sample=False

# 预测代码from transformers import TFAutoModelForSeq2SeqLM
from transformers import AutoTokenizercheckpoint_local = "./path/"
tokenizer = AutoTokenizer.from_pretrained(checkpoint_local)
model = TFAutoModelForSeq2SeqLM.from_pretrained(checkpoint_local,