-
Notifications
You must be signed in to change notification settings - Fork 0
Commit
This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository.
- Loading branch information
Showing
150 changed files
with
188 additions
and
4 deletions.
There are no files selected for viewing
File renamed without changes.
File renamed without changes
File renamed without changes
File renamed without changes
File renamed without changes
File renamed without changes
File renamed without changes
File renamed without changes
File renamed without changes
File renamed without changes
File renamed without changes
File renamed without changes
File renamed without changes
File renamed without changes
File renamed without changes
File renamed without changes
File renamed without changes
File renamed without changes
File renamed without changes
File renamed without changes
File renamed without changes
File renamed without changes
File renamed without changes
File renamed without changes
File renamed without changes
File renamed without changes
File renamed without changes
File renamed without changes
File renamed without changes
File renamed without changes
File renamed without changes
File renamed without changes
File renamed without changes
File renamed without changes
File renamed without changes
File renamed without changes
File renamed without changes
File renamed without changes
File renamed without changes
File renamed without changes
File renamed without changes
File renamed without changes
File renamed without changes
File renamed without changes
File renamed without changes
File renamed without changes
File renamed without changes
File renamed without changes
File renamed without changes
File renamed without changes
File renamed without changes
File renamed without changes
File renamed without changes
File renamed without changes
File renamed without changes
File renamed without changes
File renamed without changes
File renamed without changes
File renamed without changes
File renamed without changes
File renamed without changes
File renamed without changes
File renamed without changes
File renamed without changes
File renamed without changes
File renamed without changes.
File renamed without changes
File renamed without changes
File renamed without changes
File renamed without changes
File renamed without changes
File renamed without changes
File renamed without changes
File renamed without changes
File renamed without changes
File renamed without changes
File renamed without changes
File renamed without changes.
File renamed without changes
File renamed without changes
File renamed without changes
File renamed without changes
File renamed without changes
File renamed without changes
File renamed without changes
File renamed without changes
File renamed without changes
File renamed without changes
File renamed without changes
File renamed without changes
File renamed without changes
File renamed without changes
File renamed without changes.
File renamed without changes
File renamed without changes
File renamed without changes
File renamed without changes
File renamed without changes
File renamed without changes
File renamed without changes
File renamed without changes
File renamed without changes
File renamed without changes
File renamed without changes
File renamed without changes
File renamed without changes
File renamed without changes
File renamed without changes
File renamed without changes
File renamed without changes
File renamed without changes
File renamed without changes
File renamed without changes
File renamed without changes
File renamed without changes
File renamed without changes
File renamed without changes
File renamed without changes
File renamed without changes
File renamed without changes
File renamed without changes
File renamed without changes
File renamed without changes
File renamed without changes
File renamed without changes
File renamed without changes
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Original file line number | Diff line number | Diff line change |
---|---|---|
@@ -0,0 +1,184 @@ | ||
# 书生大模型全链路开源体系 | ||
|
||
<img src="./assets/image-20241107224830935.png" alt="image-20241107224830935" /> | ||
|
||
## 一、书生·浦语大模型开源开放体系介绍 | ||
|
||
- 汪周谦老师介绍了书生葡语大模型开源开放体系,强调这是一个全面的课程,旨在帮助学生快速了解书生·浦语这个IP。 | ||
|
||
- 课程内容包括人文课程,帮助学生了解书生·浦语的历史和发展,同时也介绍最近半年的新进展。 | ||
|
||
## 二、书生·浦语的开源之路 | ||
|
||
![image-20241107225143251](./assets/image-20241107225143251.png) | ||
|
||
- 从2023年7月6日开始,书生·浦语的大模型率先免费开源,并发布了全链条的开源工具体系。 | ||
|
||
- 2023年9月底,发布了InternLM-20B的重量级模型,适合中小企业和科研机构使用。 | ||
|
||
- 2024年1月,发布了InternLM2,其性能超越了同量级的开源模型,达到了7B甚至70B的性能水平。 | ||
|
||
- 2024年7月初,发布了InternLM2.5,性能获得了质的飞跃。 | ||
|
||
## 三、书生·浦语2.5的性能特点 | ||
|
||
![image-20241107225500781](./assets/image-20241107225500781.png) | ||
|
||
- 书生·浦语2.5在推理能力、上下文记忆和自主规划等方面都有显著提升。 | ||
|
||
- 推理能力方面,书生·浦语2.5的推理性能比上一代提升了20%,达到了领先水平。 | ||
|
||
- 上下文记忆方面,书生·浦语2.5的上下文记忆达到了100万级别,超过了GPT4o的128K级别。 | ||
|
||
- 自主规划和搜索方面,书生·浦语2.5具有完成复杂任务的能力,能够解锁复杂任务。 | ||
|
||
## 四、书生·浦语开源体系的核心思想和技术路线 | ||
|
||
![image-20241107225815805](./assets/image-20241107225815805.png) | ||
|
||
|
||
|
||
- 书生·浦语开源体系的核心思想是不断反馈,通过不断优化数据和模型,提高性能。 | ||
|
||
- 数据驱动和数据质量驱动是书生·浦语开源体系的关键,高质量的合成数据是通过规则构造、模型扩充和反馈生成等方式实现的。 | ||
|
||
- 书生·浦语开源社区还提供了Label LLM开源项目,方便用户对NLP任务进行标注。 | ||
|
||
## 五、书生·浦语2.5的推理能力 | ||
|
||
![image-20241107230014634](./assets/image-20241107230014634.png) | ||
|
||
- 书生·浦语2.5在推理能力方面表现出色,即使是在没有进行太多提示词工程的情况下,也能准确回答问题。 | ||
|
||
- 书生·浦语2.5的推理能力在大海捞针实验中表现优异,能够在超长背景知识中准确定位任何位置的信息。 | ||
- 书生·浦语2.5的推理能力在实际应用中表现出色,能够处理复杂的文本分析和问题回答任务。 | ||
|
||
## 六、书生·浦语2.5原生支持超长上下文的功能 | ||
|
||
- 原生支持超长上下文的功能是未来能够替代RAG的方向之一。 | ||
|
||
- 在一些更通用的场景下,原生可以替代RAG。 | ||
|
||
- 基于规划和搜索解决复杂问题的能力也是原生的重要优势。 | ||
|
||
![image-20241107230418963](./assets/image-20241107230418963.png) | ||
|
||
## 七、Mind Search项目基础介绍 | ||
|
||
![image-20241107230640898](./assets/image-20241107230640898.png) | ||
|
||
- Mind Search项目是一个基于网页浏览的项目。 | ||
|
||
- Mind Search的体验非常好,熟悉后日常的调研和搜索工作都会离不开它。 | ||
|
||
## 八、书生·浦语开源模型谱系 | ||
|
||
![image-20241107230844113](./assets/image-20241107230844113.png) | ||
|
||
- 书生·浦语开源模型谱系包括1.8B参数的模型、7B参数的模型和20亿参数的模型。 | ||
|
||
- 1.8B参数的模型可以在端侧运行,如手机和边缘设备;7B参数的模型适用于轻量级研究和应用;20B参数的模型则在生产环境中表现出色。 | ||
|
||
- 20B参数的模型出现了“涌现”现象,即能够回答未见过的东西。 | ||
|
||
## 九、书生·浦语全链条开源生态 | ||
|
||
![image-20241107231124402](./assets/image-20241107231124402.png) | ||
|
||
- 书生·浦语全链条开源生态包括数据、预训练、微调、部署、评测和应用的全链路。 | ||
|
||
- 所有的工具都开源了,实现了方案的透明化。 | ||
|
||
- 书生·万卷是一个开源的预训练语料库,个人用户和企业用户都可以使用。 | ||
|
||
## 十、预训练框架和微调框架 | ||
|
||
![image-20241107232304811](./assets/image-20241107232304811.png) | ||
|
||
- 预训练框架主要是进行显存优化、分布式训练和分布式训练之间的通信优化。 | ||
|
||
- 微调框架XTUNER支持市面上各种其他厂牌的开源模型,包括增量预训练、指令微调、多模态微调和对齐。 | ||
|
||
- XTUNER框架的数据格式兼容一些已经开源的数据格式。 | ||
|
||
![image-20241107232403217](./assets/image-20241107232403217.png) | ||
|
||
- 开源数据集的格式通常需要转化为统一的格式,以便于使用。 | ||
|
||
- 自定义数据集可以直接按照统一的格式进行制作。 | ||
|
||
- 这种格式转换工作对于使用开源数据集非常重要。 | ||
|
||
- 优化加速使用了flash attention技术,该技术无需手动设置即可开启。 | ||
|
||
- 除此之外,还使用了deepspeed zero、Pytorch FSDP、sequence parallel等优化加速方式。 | ||
|
||
- 支持的主要算法有两种:QLoRA的微调和LoRA的微调。 | ||
|
||
- 这两种方法是目前科研和企业产品中使用最广泛的微调方法。 | ||
|
||
- 全量参数微调也支持,但需要在强大的计算设备上进行。 | ||
|
||
## 十一、XTuner的微调效果 | ||
|
||
![image-20241107233042860](./assets/image-20241107233042860.png) | ||
|
||
- 使用XTuner的微调框架可以运行原先无法运行的大参数模型。 | ||
|
||
- 显存管理是XTuner的一个优势,可以有效回收显存,避免在反向传播过程中浪费显存。 | ||
|
||
## 十二、OpenCompass评测体系 | ||
|
||
![image-20241107233219114](./assets/image-20241107233219114.png) | ||
|
||
- OpenCompass评测体系广泛应用于头部大模型企业和科研机构,是唯一获得Meta官方推荐的国产大模型评测体系。 | ||
|
||
![image-20241107233316723](./assets/image-20241107233316723.png) | ||
|
||
## 十三、LMDeploy部署框架 | ||
|
||
![image-20241107233355604](./assets/image-20241107233355604.png) | ||
|
||
- LMDeploy部署框架支持多种开源模型和国产大模型的推理接口。 | ||
|
||
- 推理接口包括PYTHON、RESTFUL、gRPC,量化引擎包括基于权重的量化和KV Cache。 | ||
|
||
- 服务方面支持类OpenAI的服务和Gradio、Triton推理服务。 | ||
|
||
![image-20241107233554087](./assets/image-20241107233554087.png) | ||
|
||
## 十四、智能体框架Legent | ||
|
||
![image-20241107233607667](./assets/image-20241107233607667.png) | ||
|
||
![image-20241107233700820](./assets/image-20241107233700820.png) | ||
|
||
- Legent框架支持多种大语言模型,包括本地的InternLM和GPT。 | ||
|
||
- Legend框架能够可视化展示模型的思维路径和结果,增强了用户体验。 | ||
|
||
## 十五、Mind Search智能体 | ||
|
||
![image-20241107233853334](./assets/image-20241107233853334.png) | ||
|
||
- Mind Search智能体是一个基于AI和搜索结果的搜索引擎。 | ||
|
||
- 可以通过可视化的方式展示搜索过程和结果。 | ||
|
||
- 设计理念独特,能够模拟人脑的思维逻辑,提供详细的搜索路径和最终报告。 | ||
|
||
## 十六、茴香豆企业级知识库构建工具 | ||
|
||
![image-20241107233958593](./assets/image-20241107233958593.png) | ||
|
||
- 茴香豆工具支持检索增强生成(RAG)、知识图谱(KG)和可解释行为。 | ||
|
||
- 工具开源免费商用,已经构建了1500多个知识库和500多个用户群。 | ||
|
||
- 支持七种文档格式更新,无需重新部署,扩展性强。 | ||
|
||
![image-20241107234021990](./assets/image-20241107234021990.png) | ||
|
||
|
||
|
||
**The End.** |
Oops, something went wrong.
Oops, something went wrong.
Oops, something went wrong.
Oops, something went wrong.
Oops, something went wrong.
Oops, something went wrong.
Oops, something went wrong.
Oops, something went wrong.
Oops, something went wrong.
Oops, something went wrong.
Oops, something went wrong.
Oops, something went wrong.
Oops, something went wrong.
Oops, something went wrong.
Oops, something went wrong.
Oops, something went wrong.
Oops, something went wrong.
Oops, something went wrong.
Oops, something went wrong.
Oops, something went wrong.
Oops, something went wrong.
Oops, something went wrong.
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters