书生大模型包括一系列开源的高质量的大模型，同时为了便利开发者和使用者，提供了全栈的开发工具。
GitHub地址为：https://github.com/internLM/
书生大模型全链路开源体系总共分三大部分

书生大模型全链路开源体系

模型

• InternLM: 一系列的基础大模型和对话大模型.
• InternLM-Math: 强大的、专业的数学大模型.
• InternLM-XComposer: 基于InternLM开发的，支持视觉-语言复合数据的大模型.

InternLM

GitHub地址：https://github.com/InternLM/InternLM
书生基础大模型有以下版本

InternLM3

InternLM2.5

InternLM-Math

InternLM2-Math-Plus

GitHub地::https://github.com/InternLM/InternLM-Math

InternLM2-Math-Base

InternLM-XComposer

书生.浦语-灵笔
GitHub地址：https://github.com/InternLM/InternLM-XComposer

工具链

• InternEvo: 支持大模型预训练和微调的轻量级框架。
• XTuner: 高效的大模型微调工具，支持多种大模型和多种调优方法。
• LMDeploy: 用来压缩、部署和使用大模型的工具。
• Lagent: 让用户高效率的开发agent的轻量级框架.
• AgentLego: 扩展和增强agent的一组类库和工具
• OpenCompass: 大模型评测平台.
• OpenAOE: 大模型比对工具.

InternEvo

XTuner

GitHub地址：https://github.com/InternLM/InternEvo/
支持多种大模型的预训练和微调

• InternLM
• InternLM2
• InternLM3
• Llama2
• Qwen2
• Baichuan2
• gemma
• Qwen2-MoE
• Mixtral

LMDeploy

GitHub地址：https://github.com/InternLM/lmdeploy

LLMs

Llama (7B - 65B)
Llama2 (7B - 70B)
Llama3 (8B, 70B)
Llama3.1 (8B, 70B)
Llama3.2 (1B, 3B)
InternLM (7B - 20B)
InternLM2 (7B - 20B)
InternLM3 (8B)
InternLM2.5 (7B)
Qwen (1.8B - 72B)
Qwen1.5 (0.5B - 110B)
Qwen1.5 - MoE (0.5B - 72B)
Qwen2 (0.5B - 72B)
Qwen2-MoE (57BA14B)
Qwen2.5 (0.5B - 32B)
Baichuan (7B)
Baichuan2 (7B-13B)
Code Llama (7B - 34B)
ChatGLM2 (6B)
GLM4 (9B)
CodeGeeX4 (9B)
Falcon (7B - 180B)
YI (6B-34B)
Mistral (7B)
DeepSeek-MoE (16B)
DeepSeek-V2 (16B, 236B)
DeepSeek-V2.5 (236B)
Mixtral (8x7B, 8x22B)
Gemma (2B - 7B)
Dbrx (132B)
StarCoder2 (3B - 15B)
Phi-3-mini (3.8B)
Phi-3.5-mini (3.8B)
Phi-3.5-MoE (16x3.8B)
MiniCPM3 (4B)

Lagent

GitHub地址：https://github.com/InternLM/lagent
高效、轻量级的开发工具，大大提高agent的开发效率。

lagent

AgentLego

提供了多种类库，支持开发强大的智能体
General ability

• Calculator: Calculate by Python interpreter.
• GoogleSearch: Search on Google.

Speech related

• TextToSpeech: Speak the input text into audio.
• SpeechToText: Transcribe an audio into text.

Image-processing related

• ImageDescription: Describe the input image.
• OCR: Recognize the text from a photo.
• VQA: Answer the question according to the image.
• HumanBodyPose: Estimate the pose or keypoints of human in an image.
• HumanFaceLandmark: Estimate the landmark or keypoints of human faces in an image.
• ImageToCanny: Extract the edge image from an image.
• ImageToDepth: Generate the depth image of an image.
• ImageToScribble: Generate a sketch scribble of an image.
• ObjectDetection: Detect all objects in the image.
• TextToBbox: Detect specific objects described by the given text in the image.
• Segment Anything series
  • SegmentAnything: Segment all items in the image.
  • SegmentObject: Segment the certain objects in the image according to the given object name.

AIGC related

• TextToImage: Generate an image from the input text.
• ImageExpansion: Expand the peripheral area of an image based on its content.
• ObjectRemove: Remove the certain objects in the image.
• ObjectReplace: Replace the certain objects in the image.
• ImageStylization: Modify an image according to the instructions.
• ControlNet series
  • CannyTextToImage: Generate an image from a canny edge image and a description.
  • DepthTextToImage: Generate an image from a depth image and a description.
  • PoseToImage: Generate an image from a human pose image and a description.
  • ScribbleTextToImage: Generate an image from a sketch scribble image and a description.
• ImageBind series
  • AudioToImage: Generate an image according to audio.
  • ThermalToImage: Generate an image according a thermal image.
  • AudioImageToImage: Generate am image according to a audio and image.
  • AudioTextToImage: Generate an image from a audio and text prompt.

OpenCompass

GitHub地址： https://github.com/open-compass/opencompass

image.png

OpenAOE

What can you get from OpenAOE?

OpenAOE can:

1. return one or more LLMs' answers at the same time by a single prompt.
2. provide access to commercial LLM APIs, with default support for gpt3.5, gpt4, Google Palm, Minimax, Claude, Spark, etc., and also support user-defined access to other large model APIs. (API keys need to be prepared in advanced)
3. provide access to open-source LLM APIs. ( We recommend to use LMDeploy to deploy with one click)
4. provide backend APIs and a WEB-UI to meet the needs of different requirements.

应用

• HuixiangDou茴香豆: 企业级知识库搭建工具.
• MindSearch: 基于大模型的互联网搜索引擎.

HuixiangDou茴香豆

GitHub地址：https://github.com/InternLM/HuixiangDou
HuixiangDou1 is a professional knowledge assistant based on LLM.

Advantages:

1. Design three-stage pipelines of preprocess, rejection and response
   • chat_in_group copes with group chat scenario, answer user questions without message flooding, see 2401.08772, 2405.02817, Hybrid Retrieval and Precision Report
   • chat_with_repo for real-time streaming chat
2. No training required, with CPU-only, 2G, 10G, 20G and 80G configuration
3. Offers a complete suite of Web, Android, and pipeline source code, industrial-grade and commercially viable

Check out the scenes in which HuixiangDou are running and join WeChat Group to try AI assistant inside.

If this helps you, please give it a star

MindSearch

GitHub地址： https://github.com/InternLM/MindSearch
MindSearch 是一个开源的 AI 搜索引擎框架，具有与 Perplexity.ai Pro 相同的性能。您可以轻松部署它来构建您自己的搜索引擎，可以使用闭源 LLM（如 GPT、Claude）或开源 LLM（InternLM2.5 系列模型经过专门优化，能够在 MindSearch 框架中提供卓越的性能；其他开源模型没做过具体测试）。其拥有以下特性：

• 🤔 任何想知道的问题：MindSearch 通过搜索解决你在生活中遇到的各种问题
• 📚 深度知识探索：MindSearch 通过数百网页的浏览，提供更广泛、深层次的答案
• 🔍 透明的解决方案路径：MindSearch 提供了思考路径、搜索关键词等完整的内容，提高回复的可信度和可用性。
• 💻 多种用户界面：为用户提供各种接口，包括 React、Gradio、Streamlit 和本地调试。根据需要选择任意类型。
• 🧠 动态图构建过程：MindSearch 将用户查询分解为图中的子问题节点，并根据 WebSearcher 的搜索结果逐步扩展图