zhhao1

I want to know what is the termination condition of ctc beam search. It seems that it didn't need the end token in this code.

The paper gave me a lot of inspiration, but I have some questions about the structure. First, why the text encoder use the mask self attention? The paper have written...

During inference, a letter dictionary is needed. Where can is find it?

documentation and training recipes

1

Thank you so much for the great work. I wonder if there are documentation and training examples for quick use.

task_num+1

1

`self.lora_task_embedding.update(nn.ModuleDict({adapter_name: nn.Embedding(self.task_num+1, self.te_dim)}))` 你好，有个问题，这里为什么要self.task_num+1，+1是为了避免意外情况，还是什么含义