RAG 系统快速入门指南

你是否也被 AI 和 AI 搞得有点蒙？我也是！但我们得顺势而为，不然无法坚持做出影响！

这篇博文讲的正是这样一个 AI 的东西，它在科技世界中产生了很大的潜力。不管你是初学者还是专家，不管你从事科技工作还是对科技有兴趣，你都应该了解这个。

在这篇博文中，我会深入讲解检索增强生成（RAG），并使用一个出色的框架 LLMWARE 创建一个快速的 prompt 模型。

让我们开始...3️⃣...2️⃣...1️⃣...🤓

从基础开始

让我们从基础开始，这样你能轻松理解 RAG。

首先，什么是 AI？AI 或人工智能就是制造智能机器的科学和工程学。

在 AI 中，有许多子集。看看下面的图表：

现在让我们讨论另一个庞杂的领域——机器学习（ML）。从上面的图表可以看出，ML 是 AI 的一个子集。ML 专注于构建从数据中学习的计算机系统。因此，ML 是 AI 的一部分，它处理并训练一种称为模型的软件，以对数据做出有用的预测或生成内容。

有趣的事实：LLM 是人工智能（AI）程序的一种，基于机器学习构建。因此，LLM 在海量数据集上训练——这就是"大型"这个名称的由来。

AI
├── ML (Machine Learning)
│   ├── LLM (Large Language Models)
│   └── RAG (Retrieval-Augmented Generation)

RAG 或检索增强生成是一个突破性的 AI 框架（就像 NextJs 是 Js 的框架一样），用于通过让模型基于外部知识来源来改进 LLM 生成响应的质量。

RAG 将 LLM 已有的强大能力扩展到特定领域或组织的内部知识库，完全无需重新训练模型。这是一种经济高效的方法，能改进 LLM 的输出，使其在各种情境中保持相关、准确和有用。

我希望你现在对 RAG 的概念有所了解。为了让概念更清晰，让我们跳到示例部分，在那里我们将创建一个简单的项目，使用 LLMWARE 框架来测试基于 Prompt 的 RAG 模型。

如果你不了解 LLMWARE，请阅读下面的文章。这是构建项目的唯一前置条件！😝

LLMware.ai 🤖: An Ultimate Python Toolkit for Building LLM Apps

Rohan Sharma ・ Aug 29 '24

用 LLMWare.ai 开始 Prompt 模型 🤖

llmware 为构建基于 LLM 的应用程序（如 RAG、Agents）提供了统一的框架，使用小型、专用的模型，可以私下部署，与企业知识来源安全集成，并成本有效地为任何业务流程调整和适配。

在这个例子中，我们将说明：

探索 - 如何在 llmware ModelCatalog 中探索模型。

加载模型 - 如何从目录中加载选定的模型。

Prompt - 如何创建基本 prompt 并运行模型推理。

1️⃣ 安装 llmware（如上所述）。或者在终端中运行这个代码：

 pip3 install llmware

**2️⃣ 考虑到你没有任何测试问题来测试这个项目。因此，你可以使用下面这个：

def hello_world_questions():

    """ This is a set of useful test questions to do a 'hello world' but there is nothing special about the
    questions - please feel free to edit and ask your own queries with your own context passages.

    --if you are using one of the llmware models, please take note that the models have been trained to answer
    based on the information provided, so if you ask a question without passing any context passage, then
    don't be surprised if the model responds with 'Not Found.' """

    test_list = [

    {"query": "What is the total amount of the invoice?",
     "answer": "$22,500.00",
     "context": "Services Vendor Inc. \n100 Elm Street Pleasantville, NY \nTO Alpha Inc. 5900 1st Street "
                "Los Angeles, CA \nDescription Front End Engineering Service $5000.00 \n Back End Engineering"
                " Service $7500.00 \n Quality Assurance Manager $10,000.00 \n Total Amount $22,500.00 \n"
                "Make all checks payable to Services Vendor Inc. Payment is due within 30 days."
                "If you have any questions concerning this invoice, contact Bia Hermes. "
                "THANK YOU FOR YOUR BUSINESS!  INVOICE INVOICE # 0001 DATE 01/01/2022 FOR Alpha Project P.O. # 1000"},

    {"query": "What was the amount of the trade surplus?",
     "answer": "62.4 billion yen ($416.6 million)",
     "context": "Japan's September trade balance swings into surplus, surprising expectations"
                "Japan recorded a trade surplus of 62.4 billion yen ($416.6 million) for September, "
                "beating expectations from economists polled by Reuters for a trade deficit of 42.5 "
                "billion yen. Data from Japan's customs agency revealed that exports in September "
                "increased 4.3% year on year, while imports slid 16.3% compared to the same period "
                "last year. According to FactSet, exports to Asia fell for the ninth straight month, "
                "which reflected ongoing China weakness. Exports were supported by shipments to "
                "Western markets, FactSet added. — Lim Hui Jie"}
]

    return test_list

**3️⃣ 创建一个 Python 文件，比如 fast_start_rag.py，并粘贴下面的代码：

import time
from llmware.prompts import Prompt
from llmware.models import ModelCatalog

def fast_start_prompting  (model_name):

    """ This is the main example script - it loads the question list, loads the model, and executes the prompts. """

    t0 = time.time()

    # load in the 'hello world' test questions above
    test_list = hello_world_questions()

    print(f"\n > Loading Model: {model_name}...")

    prompter = Prompt().load_model(model_name)

    t1 = time.time()
    print(f"\n > Model {model_name} load time: {t1-t0} seconds")

    for i, entries in enumerate(test_list):
        print(f"\n{i+1}. Query: {entries['query']}")

        #   run the prompt
        output = prompter.prompt_main(entries["query"],
                                      context=entries["context"],
                                      prompt_name="default_with_context",
                                      temperature=0.30)

        #   'output' is a dictionary with two keys - 'llm_response' and 'usage'
        #   --'llm_response' is the output from the model
        #   --'usage' is a dictionary with the usage stats

        llm_response = output["llm_response"].strip("\n")
        print(f"LLM Response: {llm_response}")

        #   note: the 'gold answer' is the answer we provided above in the hello_world question list
        print(f"Gold Answer: {entries['answer']}")

        print(f"LLM Usage: {output['usage']}")

    t2 = time.time()
    print(f"\nTotal processing time: {t2-t1} seconds")

    return 0

if __name__ == "__main__":

    #   Step 1 - we will pick a model from the ModelCatalog

    #   A few useful methods to discover and display a list of available models...

    #   all generative models
    llm_models = ModelCatalog().list_generative_models()

    #   if you only want to see the local models
    llm_local_models = ModelCatalog().list_generative_local_models()

    #   to see only the open source models
    llm_open_source_models = ModelCatalog().list_open_source_models()

    #   we will print out the local models
    for i, models in enumerate(llm_local_models):
        print("models: ", i, models["model_name"], models["model_family"])

    #   for purposes of demo, try a few selected models from the list

    #   each of these pytorch models are ~1b parameters and will run reasonably fast and accurate on CPU
    #   --per note above, may require separate pip3 install of: torch and transformers
    pytorch_generative_models = ["llmware/bling-1b-0.1", "llmware/bling-tiny-llama-v0", "llmware/bling-falcon-1b-0.1"]

    #   bling-answer-tool is 1b parameters quantized
    #   bling-phi-3-gguf is 3.8b parameters quantized
    #   dragon-yi-6b-gguf is 6b parameters quantized
    gguf_generative_models = ["bling-answer-tool", "bling-phi-3-gguf","llmware/dragon-yi-6b-gguf"]

    #   by default, we will select a gguf model requiring no additional imports
    model_name = gguf_generative_models[0]

    #   to swap in a GPT-4 openai model - uncomment these two lines
    #   model_name = "gpt-4"
    #   os.environ["USER_MANAGED_OPENAI_API_KEY"] = "<insert-your-openai-key>"

    fast_start_prompting(model_name)

**4️⃣ 回到终端，运行下面的代码来运行应用程序：

python fast_start_rag.py

虽然代码自我解释（检查注释），但你可能想知道，刚才发生了什么！你可能有很多问题。但等等！我有专门的解释部分，特别是给视觉学习者。请查看这个链接，Prompt Models (Ex. 3): Fast Start to RAG (2024)。如果你想了解更多，那就看一下这个播放列表：

Fast Start to RAG (2024 updates) - YouTube

学习如何在这个易于跟随的分步教程系列中掌握 RAG 的基础知识

检索增强生成（RAG）是优化大语言模型输出的过程，使其在生成响应之前引用其训练数据来源之外的权威知识库。

如果你仍然有任何问题，请在评论区留言。或者，你可以通过以下链接加入 LLMWare 官方 Discord 频道：https://discord.com/invite/fCztJQeV7J

感谢！你是最美的人！继续学习，继续努力。祝你有美好的一天！！💝

一些评论可能仅对已登录的访问者可见。登录以查看所有评论。

如需进一步操作，你可以考虑屏蔽此人和/或举报滥用。