54、Exploring Sequence-to-Sequence Architectures: From Machine Translation to Decoders

雪落无声360

于 2025-09-04 11:17:37 发布

阅读量18

点赞数

CC 4.0 BY-SA版权

分类专栏：自然语言处理实战指南文章标签： machine translation sequence-to-sequence Transformer

本文链接：https://blog.csdn.net/agile9scrum/article/details/151201156

自然语言处理实战指南专栏收录该内容

55 篇文章 ¥69.90 ¥99.00

订阅专栏

超级会员免费看

Exploring Sequence-to-Sequence Architectures: From Machine Translation to Decoders

1. Machine Translation Basics

In the realm of machine translation, we often start by creating a translator object. We first format the source and target characters. For instance, we select a mini - batch of 32 samples, convert lists of character strings to lists of tensors, and pad them to get tensors.

src_batch = pad_sequence(seqs2tensors( 
    train_src_seqs[:32], token2idx), 
    batch_first=True, padding_value=0) 
tgt_batch = pad_sequence(seqs2tensors( 
    train_tgt_seqs[:32], token2idx), 
    batch_first=True, padding_value=0)

We then create padding masks to remove padding symbols from the attention:

了解本专栏