lm-sys/FastChat

[Feature Request] Add a ctranslate2 model worker

Open

#2,133 创建于 2023年8月1日

在 GitHub 查看
 (2 评论) (1 反应) (1 负责人)Python (38,959 star) (4,736 fork)batch import
enhancementgood first issue

描述

According to some recent analysis on twitter, CTranslate2 can serve LLMs a little faster than vLLM and (maybe?) with a small quality increase. At least for Llama 2.

This could either be a a model worker that's added directly to fastchat OR a doc with extensive documentation on how to write a custom model worker (with mostly working \ implementation code) that anyone can use in their own project. The second option might be best so that people can expand FastChat with custom model workers without having to change the base project too much.

If this is appealing I can get to it at some point.

贡献者指南

技术栈
python
领域
backendmachine 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。
3
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。
3-5 days
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。
blocked
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
PythonFastChat codebaseLLM serving conceptsCTranslate2 library
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。
25
研究方向
Start by examining existing model workers in FastChat, such as those for vLLM, to understand the pattern and integration points. Look at files like 'fastchat/serve/model worker.py' and any associated configurations. Review the conversation in this issue for any additional requirements or direction from maintainers. Investigate the CTranslate2 Python library's documentation and example code to plan the integration. Since the issue has an assignee, check if they have any open branches or related commits.