Make Word.MergeAll sub-quadratic

未关闭
#7,872 0 条评论 0 个 reaction 已指派 0 人 在 GitHub 查看

维护者通常 1 天内回复

还没有人认领这个 Issue。

评估

难度
3/5
预计耗时
1-2 天
新手友好度
42/100
Issue 类型
重构
描述清晰度
基本清楚
活跃度
活跃
技术栈
csharp, machine-learning
领域
performance

调研方向

The slow path is Word.MergeAll in src/Microsoft.ML.Tokenizers/Model/Word.cs, where symbols are removed from a Vec<T>-style list with Remove() while a PriorityQueue<T> drives the merges. Read that method first and find which tests in the Microsoft.ML.Tokenizers test project cover BPE merging so you can confirm the output is unchanged. Done means the removal no longer shifts the list on every merge, the existing tokenizer tests still pass, and the change is benchmarked against the old version.

由索引模型根据 Issue 内容生成。

描述

area-Tokenizers tenet-performance

The current implementation of Word.MergeAll uses Vec<T>.Remove() to remove symbols from the list:

https://github.com/dotnet/machinelearning/blob/81310b23aeee0cc12720edde8794e3cc94afef3e/src/Microsoft.ML.Tokenizers/Model/Word.cs#L187-L194

Based on data I found, it seems it's common that around 75 % of symbols are removed here (though the data conflates multiple steps, so it could be not fully relevant). This makes the method quadratic, likely significantly affecting its performance.

We should consider improving that. (This won't make the method linear, since the use of PriorityQueue<T> leads to at least O(n log n).)

主要语言
C#
星标
9.4k
派生
2k
平均合并
2 天 18 小时
30 天内合并 PR
64

环境准备

从这里开始

  1. 先读完整个 Issue,再读项目的贡献指南。
  2. 在 Issue 下留言说明你要接手 —— 这能避免两个人做同样的事。
  3. Fork 仓库,在一个分支上完成修改。
  4. 提交 Pull Request,并在描述里引用这个 Issue 编号。

dotnet/machinelearning 的其他 Issue

查看 dotnet/machinelearning 的全部 Issue

相似的 Issue

更多 C# Issue

把新 issue 发到你的邮箱

精选适合新手参与的 GitHub issue 摘要。