facebookresearch/fairseq

Add an option to disable appending eos to the end of each document while preprocessing

Open

#5,615 创建于 2025年5月26日

在 GitHub 查看
 (0 评论) (0 反应) (0 负责人)Python (29,107 star) (6,224 fork)batch import
enhancementhelp wantedneeds triage

描述

🚀 Feature Request

Add an option to disable appending EOS to the end of each document while preprocessing

Motivation

There are instances where adding EOS is not preferred like when packing data to sequence length.

Pitch

Proposing a change to add an argument to handle this

贡献者指南

技术栈
pythonpytorch
领域
machine learning
议题类型
feature
难度面向新贡献者的预计实现难度，1 表示很小改动，5 表示专家级工作。
2
预计时间有经验贡献者完成调查、实现、测试并准备 pull request 的粗略时间范围。
1-3 hours
活动状态议题当前的可参与程度：新鲜、活跃、陈旧、阻塞或等待维护者输入。
fresh
清晰度议题是否清楚说明期望改动、验收标准和下一步。
clear
前置要求
PythonPyTorchfairseq basics
新手友好度1-100 的估计分数，表示该议题对首次贡献者的友好程度。
70
研究方向
Look for where EOS is appended to documents during preprocessing. Typical locations are in fairseq's dictionary construction (e.g., `fairseq/data/dictionary.py`) or in the binarization scripts (e.g., `fairseq cli/preprocess.py`). The solution involves adding a new argument (e.g., ` no append eos`) to the relevant preprocessing function or CLI tool and conditionally skipping the addition of EOS tokens.
Add an option to disable appending eos to the end of each document while preprocessing · facebookresearch/fairseq#5615 | Good First Issue