如何为一个热编码的基于charachter的输入构建Kerras lstm文本生成模型的输入向量

I am trying to train a Keras lstm model for generation of potry I encode my data as sequence of charachters where each charachter is one hot encoded. I am using many-to-one model so my input is a 2-dimentional array like this one. [[0,1,0], [1,0,0], ... ,[1,0,0]] Where each vector represents one time step, I want to have 512 charachters fead to each of the lstm cells and have the last one predict the same one hot encoded vector with prediction of the next charachter. I am trying to train a Keras lstm model for gen