LangChain

大语言模型（LLM）正在成为一种变革性技术，使开发人员能够构建以前无法构建的应用程序。但是，单独使用这些LLM通常不足以创建一个真正强大的应用程序——当你可以将它们与其他计算或知识来源相结合时，便可能实现其真正的能力。

LangChain是一个用于开发由语言模型驱动的应用程序的框架，允许开发人员将语言模型连接到其他数据源并与其环境相交互。LangChain旨在帮助开发者在以下六个主要领域，按照复杂性递增的顺序：

• 📃 LLMs and Prompts: 这包括提示管理、提示优化、适用于所有 LLM 的通用界面以及用于处理 LLM 的通用实用程序。
• 🔗 Chains: 链不仅仅是单个 LLM 调用，而是调用序列（无论是对 LLM 还是对不同的实用程序）。 LangChain 为链提供标准接口、与其他工具的大量集成以及用于常见应用程序的端到端链。
• 📚 Data Augmented Generation: 数据增强生成涉及特定类型的链，这些链首先与外部数据源交互以获取数据以用于生成步骤。 这方面的例子包括对长文本的总结和对特定数据源的问答。
• 🤖 Agents: 代理涉及 LLM 做出关于采取哪些行动的决定，采取该行动，看到一个观察，并重复直到完成。LangChain 为代理提供了一个标准接口，可供选择的代理选择，以及端到端代理的示例。
• 🧠 Memory: 内存是链/代理调用之间持久状态的概念。 LangChain 提供了内存的标准接口、内存实现的集合以及使用内存的链/代理的示例。
• 🧐 Evaluation: [BETA] 众所周知，生成模型很难用传统指标进行评估。 评估它们的一种新方法是使用语言模型本身进行评估，LangChain 提供了一些提示/链来协助这一点。

The large language Model (LLM) is becoming a transformative technology that enables developers to build applications that could not be built before. However, using these LLM alone is often not enough to create a really powerful application– when you can combine them with other sources of computing or knowledge, it is possible to achieve its true capabilities.

LangChain is a framework for developing applications driven by language models that allow developers to connect language models to other data sources and interact with their environment. LangChain is designed to help developers in the following six main areas in the order of increasing complexity:

• 📃 LLMs and Prompts: this includes prompt management, prompt optimization, a common interface for all LLM, and a general utility for handling LLM.
• 🔗 Chains: a chain is not just a single LLM call, but a sequence of calls (whether for LLM or for different utilities). LangChain provides a standard interface for chains, extensive integration with other tools, and end-to-end chains for common applications.
• 📚 Data Augmented Generation: data enhancement generation involves specific types of chains that first interact with external data sources to obtain data for use in the generation step. Examples of this include summaries of long texts and questions and answers to specific data sources.
• 🤖 Agents: the agent involves LLM making a decision about what action to take, taking that action, seeing an observation, and repeating it until it is complete. LangChain provides a standard interface for agents, a choice of agents, and an example of an end-to-end agent.
• 🧠 Memory: memory is the concept of persistent state between chain / proxy calls. LangChain provides a standard interface to memory, a collection of memory implementations, and examples of chains / agents that use memory.
• 🧐 Evaluation: [BETA] it is well known that generation models are difficult to evaluate with traditional indicators. A new way to evaluate them is to evaluate them using the language model itself, and LangChain provides some hints / chains to help with this.