简介:
小李哥将继续每天介绍一个基于亚马逊云科技AWS云计算平台的全球前沿AI技术解决方案,帮助大家快速了解国际上最热门的云计算平台亚马逊云科技AWS AI最佳实践,并应用到自己的日常工作里。
本次介绍用当下热门的LangChain框架的基本知识和学习亚马逊云科技热门AI模型管理服务Amazon SageMaker实操,手把手开发生成式AI应用(对话机器人和代码生成器)。本文将通过利用LangChain的提示词工程PromptTemplate模块、问答链load_qa_chain模块、RetrievalQA问答检索模块、Document文档模块、Embedding向量化模块、VectorstoreIndexCreator 索引模块、Conversation对话模块,编写一步一步的细节代码和展示LangChain应用实际样例,0基础学会AI核心技能。网页聊天AI机器人界面如下。
本方案使用 GPT-J 6B Embedding FP16作为向量模型,使用Falcon 7B Instruct BF16作为大语言模型。本架构设计全部采用了云原生Serverless架构,提供可扩展和安全的AI应用解决方案。本方案架构图如下:
什么是LangChain?
LangChain 是一个专为开发者设计的开源库,旨在简化和增强语言模型的开发和部署。随着生成式AI技术的发展,如何有效地利用这些强大的语言模型成为了开发者关注的焦点。LangChain 提供了一套工具集,帮助开发者轻松地集成和管理语言模型,支持从模型选择到部署的一系列操作,使开发过程更加高效和简洁。
LangChain的应用场景
智能客服和对话系统
LangChain 可以用来构建智能客服和对话系统,通过接入强大的语言模型,能够更好地理解和回应用户的自然语言输入。这种应用场景在电商、金融等领域非常受欢迎,能够有效提升用户体验,减少人工客服的工作量。
内容生成和自动化写作
对于需要大量内容、文案生成的行业,如媒体、市场营销等,LangChain 可以帮助开发自动化写作工具。通过定制化的语言模型,系统可以根据输入的关键词或主题自动生成高质量的文章、新闻稿或广告文案,大大提高内容生产效率。
数据分析和报告生成
在数据密集型行业,如金融和市场分析,LangChain 可以用来开发数据分析和报告生成工具。这些工具能够从海量数据中提取关键信息,并生成易于理解的分析报告,帮助决策者快速获取所需的洞见。
语言翻译和内容本地化调整
LangChain 还可以用于构建自动翻译和内容本地化系统,帮助出海企业跨越语言障碍,进入全球市场。通过结合先进的语言模型和翻译技术,可以实现高质量的文本翻译,同时保留原文的语义和风格。
本方案包括的内容:
使用 Amazon SageMaker JumpStart 部署AI基础模型(FMs)生成内容
使用 SageMaker Studio 中的Jupyter Notebook来学习 LangChain 库结合基础模型的使用。
洞察和测试利用LangChain的不同场景的生成式 AI 应用解决方案。
项目搭建具体步骤:
1. 首先我们登录亚马逊云科技控制台,进入SageMaker服务中的Studio
2. 点击“Open”,打开SageMaker中的Jupyter Notebook
3. 打开“SageMaker JumpStart”下拉栏,选择“GPT-J 6B Embedding FP16”向量模型
4. 为模型配置底层服务器资源类型,我们选择“ml.g5.2xlarge”,复制模型节点名称并点击“Deploy”部署
5.按相同方式,我们部署另外的模型“Falcon 7B Embedding FP16”
6. 我们可以在SageMaker服务下的Endpoints界面查看部署状态,“Inservice”则表述部署完成
7. 接下来我们在SageMaker Studio中新建一个Jupyter Notebook,利用LangChain优化大模型的回复
8. 首先我们在Cell中复制以下代码,安装LangChain和相关依赖
import warnings
warnings.filterwarnings('ignore')
!pip install --upgrade pip
!pip install --upgrade sagemaker --quiet
!pip install ipywidgets==7.0.0 --quiet
!pip install langchain==0.1.11 --quiet
!pip install faiss-cpu --quiet
9. 接下来我们导入依赖并初始化模型训练客户端
import time
import boto3, json
from typing import Any, Dict, List, Optional
session = boto3.Session()
sagemaker_runtime_client = session.client("sagemaker-runtime")
10. 接下来我们开始创建一个AI问答机器人。首先定义文档内容、LangChain提示词模板PromptTemplate、和LangChain输入、输出格式处理函数ContentHandler。
from typing import Dict
from langchain.chains.question_answering import load_qa_chain
from langchain_core.prompts import PromptTemplate
from langchain_community.llms import SagemakerEndpoint
from langchain_community.llms.sagemaker_endpoint import LLMContentHandler
from langchain.docstore.document import Document
example_doc_1 = """
Amazon Bedrock is a fully managed service that offers a choice of high-performing foundation models (FMs) from
leading AI companies like AI21 Labs, Anthropic, Cohere, Meta, Mistral AI, Stability AI, and Amazon via a single
API, along with a broad set of capabilities you need to build generative AI applications with security, privacy,
and responsible AI. Using Amazon Bedrock, you can easily experiment with and evaluate top FMs for your use case,
privately customize them with your data using techniques such as fine-tuning and Retrieval Augmented Generation
(RAG), and build agents that execute tasks using your enterprise systems and data sources. Since Amazon Bedrock
is serverless, you don't have to manage any infrastructure, and you can securely integrate and deploy generative
AI capabilities into your applications using the AWS services you are already familiar with.
"""
docs = [
Document(
page_content=example_doc_1,
)
]
prompt_template = """Use the following pieces of context to answer the question at the end.
{context}
Question: {question}
Answer:"""
PROMPT = PromptTemplate(
template=prompt_template, input_variables=["context", "question"]
)
class ContentHandler(LLMContentHandler):
content_type = "application/json"
accepts = "application/json"
def transform_input(self, prompt: str, model_kwargs: Dict) -> bytes:
input_str = json.dumps({"inputs": prompt, "parameters": model_kwargs})
# print(input_str)
return input_str.encode("utf-8")
def transform_output(self, output: bytes) -> str:
response_json = json.loads(output.read().decode("utf-8"))
# print(response_json)
return response_json[0]["generated_text"]
content_handler = ContentHandler()
11. 定义模型参数parameters,构建调用亚马逊云科技SageMaker的问答链load_qa_chain,并利用样例问题对问LangChain问答链测试
embedding_endpoint_name = "jumpstart-dft-<REPLACE WITH EMBEDDING ENDPOINT NAME>"
instruct_endpoint_name = "jumpstart-dft-<REPLACE WITH INSTRUCT ENDPOINT NAME>"
parameters = {
"max_length": 200,
"num_return_sequences": 1,
"top_k": 250,
"top_p": 0.95,
"do_sample": False,
"temperature": 1,
}
chain = load_qa_chain(
llm = SagemakerEndpoint(
endpoint_name=instruct_endpoint_name,
region_name='us-east-1',
model_kwargs=parameters,
content_handler=content_handler,
),
prompt=PROMPT,
)
question = "What is Amazon Bedrock?"
chain.invoke({"input_documents": docs, "question": question}, return_only_outputs=True)
12. 下面我们将外挂知识库,在LangChain中结合RAG增强我们AI问答机器人的回复效果,得到更准确的答案。首先我们利用LangChain的SagemakerEndpointEmbeddings类定义文本向量化的规则,再用EmbeddingsContentHandler定义向量化的输入和输出格式,模型使用了SageMaker上向量模型GPT-J 6B Embedding FP16。
# Wrap the SageMaker endpoint for the embedding model into langchain_community.embeddings.SagemakerEndpointEmbeddings.
# This code defines a custom subclass of SagemakerEndpointEmbeddings to handle document embeddings by using a SageMaker endpoint.
# The code also defines a custom content handler for input and output formatting.
from langchain_community.embeddings import SagemakerEndpointEmbeddings
from langchain_community.embeddings.sagemaker_endpoint import EmbeddingsContentHandler
class SagemakerEndpointEmbeddingsJumpStart(SagemakerEndpointEmbeddings):
def embed_documents(self, texts: List[str], chunk_size: int = 5) -> List[List[float]]:
"""Compute doc embeddings using a SageMaker Inference Endpoint.
Args:
texts: The list of texts to embed.
chunk_size: The chunk size defines how many input texts will
be grouped together as request. If None, will use the
chunk size specified by the class.
Returns:
List of embeddings, one for each text.
"""
results = []
_chunk_size = len(texts) if chunk_size > len(texts) else chunk_size
for i in range(0, len(texts), _chunk_size):
response = self._embedding_func(texts[i : i + _chunk_size])
print
results.extend(response)
return results
class ContentHandler(EmbeddingsContentHandler):
content_type = "application/json"
accepts = "application/json"
def transform_input(self, prompt: str, model_kwargs={}) -> bytes:
input_str = json.dumps({"text_inputs": prompt, **model_kwargs})
return input_str.encode("utf-8")
def transform_output(self, output: bytes) -> str:
response_json = json.loads(output.read().decode("utf-8"))
embeddings = response_json["embedding"]
return embeddings
content_handler = ContentHandler()
embeddings = SagemakerEndpointEmbeddingsJumpStart(
endpoint_name=embedding_endpoint_name,
region_name='us-east-1',
content_handler=content_handler,
)
13. 接下来我们预处理数据,将开源的FAQ数据集转换成DataFrame用于接下来的测试。
import glob
import os
import pandas as pd
all_files = glob.glob(os.path.join("rag_data/", "*.csv"))
df_knowledge = pd.concat(
(pd.read_csv(f, header=None, names=["Question", "Answer"]) for f in all_files),
axis=0,
ignore_index=True,
)
df_knowledge.drop(["Question"], axis=1, inplace=True)
df_knowledge.head(5)
14. 下面我们导入必要的依赖,并用LangChain去读取FAQ数据集中的答案部分
# Set up the necessary imports for building the Q&A app.
from langchain.chains import RetrievalQA
from langchain.indexes.vectorstore import VectorstoreIndexCreator
from langchain_community.document_loaders.text import TextLoader
from langchain_community.vectorstores.faiss import FAISS
from langchain_text_splitters import CharacterTextSplitter
from langchain_core.prompts import PromptTemplate
from langchain.chains.question_answering import load_qa_chain
from langchain_community.document_loaders import DataFrameLoader
# Load the processed data into LangChain.
loader = DataFrameLoader(df_knowledge,page_content_column="Answer")
documents = loader.load()
15.接下来我们利用LangChain调用向量模型将FAQ数据集的答案部分向量化并保存,使用大语言模型对我们的实例问题进行相关性搜索得到搜索结果,这里用到了FAISS库加速RAG搜索效率。
question = "What is Amazon SageMaker?"
index_creator = VectorstoreIndexCreator(
vectorstore_cls=FAISS,
embedding=embeddings,
text_splitter=CharacterTextSplitter(chunk_size=300, chunk_overlap=0),
)
index = index_creator.from_loaders([loader])
index.query(question=question, llm=sm_llm)
16. 接下来我们介绍两个应用该方案的现实场景。
16.1 使用LangChain和LLM进行代码生成,样例代码如下,生成的代码需求是: "takes a list of numbers and returns the sum of all even numbers",“根据一个数字列表并返回所有偶数之和。”
from langchain import PromptTemplate, LLMChain
# Define SageMaker LLM
llm = sm_llm
# Define the prompt template
prompt_template = """
Write a Python function that {task_description}.
"""
# Create the prompt template instance
prompt = PromptTemplate(
input_variables=["task_description"],
template=prompt_template,
)
# Create the LLMChain instance with the prompt and LLM
llm_chain = LLMChain(prompt=prompt, llm=llm)
# Generate code based on the task description
task_description = "takes a list of numbers and returns the sum of all even numbers"
output = llm_chain.invoke(input=task_description)
print(output["text"])
我们可以看到我们利用LangChain与大模型交互,根据我们的需求生成了代码。
16.2 利用LangChain ConversationChain与大模型交互构建AI对话机器人。我们这里使用 conversation.predict
方法处理用户输入的内容,利用LangChain与大模型交互生成对话回复。
from langchain.memory import ConversationBufferMemory
from langchain import ConversationChain
parameters = {
"max_length": 50,
"num_return_sequences": 1,
"top_k": 10,
"top_p": 0.01,
"do_sample": False,
"temperature": 0, }
sm_llm = SagemakerEndpoint(
endpoint_name=instruct_endpoint_name,
region_name='us-east-1',
model_kwargs=parameters,
content_handler=content_handler,
)
# Initialize the LLM (in this case, OpenAI's GPT-3)
llm = sm_llm
# Initialize the memory
memory = ConversationBufferMemory()
# Create the conversation chain with memory
conversation = ConversationChain(
llm=llm,
memory=memory,
verbose=False
)
def chat_with_ai():
print("Hi! I'm an AI assistant. How can I help you today?")
while True:
human_input = input("Human: ")
if human_input.lower() == 'exit':
print("Assistant: Goodbye!")
break # Exit the loop if user types 'exit'
# Process input through the conversation chain
response = conversation.predict(input=human_input, stop="\n\n")
# Print the AI's response
print(f"Assistant: {response}")
# Call the function to start chatting
chat_with_ai()
以上就是在亚马逊云科技上使用LangChain和大语言AI模型搭建云原生AI对话机器人,和生成代码的步骤。欢迎大家关注小李哥,未来获取更多国际前沿的生成式AI开发方案!