带有对话流智能体的 RAG 聊天机器人
本教程解释了如何使用对话流代理构建一个检索增强生成（RAG）应用程序，以您的 OpenSearch 数据作为知识库。
将以 your_ 为前缀的占位符替换为您自己的值。
构建 RAG 对话式搜索的另一种方法是使用 RAG 管道。更多信息，请参阅使用 Cohere Command 模型进行对话式搜索。
先决条件
在本教程中，您将构建一个 RAG 应用程序，它将 OpenSearch 向量索引作为大型语言模型 (LLM) 的知识库。对于数据检索，您将使用语义搜索。有关全面的语义搜索设置，请参阅本教程。
首先，您需要更新集群设置。如果您没有专用的机器学习 (ML) 节点，请设置 "plugins.ml_commons.only_run_on_ml_node": false。为避免触发本机内存断路器，请将 "plugins.ml_commons.native_memory_threshold" 设置为 100%
PUT _cluster/settings
{
"persistent": {
"plugins.ml_commons.only_run_on_ml_node": false,
"plugins.ml_commons.native_memory_threshold": 100,
"plugins.ml_commons.agent_framework_enabled": true
}
}
步骤 1：准备知识库
使用以下步骤准备将补充 LLM 知识的知识库。
步骤 1.1：注册文本嵌入模型
注册一个文本嵌入模型，该模型将文本转换为向量嵌入
POST /_plugins/_ml/models/_register
{
"name": "huggingface/sentence-transformers/all-MiniLM-L12-v2",
"version": "1.0.2",
"model_format": "TORCH_SCRIPT"
}
请记下文本嵌入模型 ID；您将在以下步骤中使用它。
作为替代方案，您可以通过调用 Get Task API 获取模型 ID
GET /_plugins/_ml/tasks/your_task_id
部署模型
POST /_plugins/_ml/models/your_text_embedding_model_id/_deploy
测试模型
POST /_plugins/_ml/models/your_text_embedding_model_id/_predict
{
"text_docs":[ "today is sunny"],
"return_number": true,
"target_response": ["sentence_embedding"]
}
有关在 OpenSearch 集群中使用模型的更多信息，请参阅预训练模型。
步骤 1.2：创建摄入管道
创建一个带有文本嵌入处理器的摄入管道，该处理器可以调用上一步中创建的模型，从文本字段生成嵌入
PUT /_ingest/pipeline/test_population_data_pipeline
{
"description": "text embedding pipeline",
"processors": [
{
"text_embedding": {
"model_id": "your_text_embedding_model_id",
"field_map": {
"population_description": "population_description_embedding"
}
}
}
]
}
有关摄入管道的更多信息，请参阅摄入管道。
步骤 1.3：创建向量索引
创建一个向量索引，并将摄入管道指定为默认管道
PUT test_population_data
{
"mappings": {
"properties": {
"population_description": {
"type": "text"
},
"population_description_embedding": {
"type": "knn_vector",
"dimension": 384
}
}
},
"settings": {
"index": {
"knn.space_type": "cosinesimil",
"default_pipeline": "test_population_data_pipeline",
"knn": "true"
}
}
}
有关向量索引的更多信息，请参阅创建向量索引。
步骤 1.4：摄入数据
将测试数据摄入到向量索引中
POST _bulk
{"index": {"_index": "test_population_data"}}
{"population_description": "Chart and table of population level and growth rate for the Ogden-Layton metro area from 1950 to 2023. United Nations population projections are also included through the year 2035.\nThe current metro area population of Ogden-Layton in 2023 is 750,000, a 1.63% increase from 2022.\nThe metro area population of Ogden-Layton in 2022 was 738,000, a 1.79% increase from 2021.\nThe metro area population of Ogden-Layton in 2021 was 725,000, a 1.97% increase from 2020.\nThe metro area population of Ogden-Layton in 2020 was 711,000, a 2.16% increase from 2019."}
{"index": {"_index": "test_population_data"}}
{"population_description": "Chart and table of population level and growth rate for the New York City metro area from 1950 to 2023. United Nations population projections are also included through the year 2035.\\nThe current metro area population of New York City in 2023 is 18,937,000, a 0.37% increase from 2022.\\nThe metro area population of New York City in 2022 was 18,867,000, a 0.23% increase from 2021.\\nThe metro area population of New York City in 2021 was 18,823,000, a 0.1% increase from 2020.\\nThe metro area population of New York City in 2020 was 18,804,000, a 0.01% decline from 2019."}
{"index": {"_index": "test_population_data"}}
{"population_description": "Chart and table of population level and growth rate for the Chicago metro area from 1950 to 2023. United Nations population projections are also included through the year 2035.\\nThe current metro area population of Chicago in 2023 is 8,937,000, a 0.4% increase from 2022.\\nThe metro area population of Chicago in 2022 was 8,901,000, a 0.27% increase from 2021.\\nThe metro area population of Chicago in 2021 was 8,877,000, a 0.14% increase from 2020.\\nThe metro area population of Chicago in 2020 was 8,865,000, a 0.03% increase from 2019."}
{"index": {"_index": "test_population_data"}}
{"population_description": "Chart and table of population level and growth rate for the Miami metro area from 1950 to 2023. United Nations population projections are also included through the year 2035.\\nThe current metro area population of Miami in 2023 is 6,265,000, a 0.8% increase from 2022.\\nThe metro area population of Miami in 2022 was 6,215,000, a 0.78% increase from 2021.\\nThe metro area population of Miami in 2021 was 6,167,000, a 0.74% increase from 2020.\\nThe metro area population of Miami in 2020 was 6,122,000, a 0.71% increase from 2019."}
{"index": {"_index": "test_population_data"}}
{"population_description": "Chart and table of population level and growth rate for the Austin metro area from 1950 to 2023. United Nations population projections are also included through the year 2035.\\nThe current metro area population of Austin in 2023 is 2,228,000, a 2.39% increase from 2022.\\nThe metro area population of Austin in 2022 was 2,176,000, a 2.79% increase from 2021.\\nThe metro area population of Austin in 2021 was 2,117,000, a 3.12% increase from 2020.\\nThe metro area population of Austin in 2020 was 2,053,000, a 3.43% increase from 2019."}
{"index": {"_index": "test_population_data"}}
{"population_description": "Chart and table of population level and growth rate for the Seattle metro area from 1950 to 2023. United Nations population projections are also included through the year 2035.\\nThe current metro area population of Seattle in 2023 is 3,519,000, a 0.86% increase from 2022.\\nThe metro area population of Seattle in 2022 was 3,489,000, a 0.81% increase from 2021.\\nThe metro area population of Seattle in 2021 was 3,461,000, a 0.82% increase from 2020.\\nThe metro area population of Seattle in 2020 was 3,433,000, a 0.79% increase from 2019."}
步骤 2：准备 LLM
本教程使用 Amazon Bedrock Claude 模型进行对话式搜索。您也可以使用其他 LLM。有关使用外部托管模型的更多信息，请参阅连接到外部托管模型。
步骤 2.1：创建连接器
为 Claude 模型创建连接器
POST /_plugins/_ml/connectors/_create
{
"name": "BedRock Claude instant-v1 Connector ",
"description": "The connector to BedRock service for claude model",
"version": 1,
"protocol": "aws_sigv4",
"parameters": {
"region": "us-east-1",
"service_name": "bedrock",
"anthropic_version": "bedrock-2023-05-31",
"max_tokens_to_sample": 8000,
"temperature": 0.0001,
"response_filter": "$.completion"
},
"credential": {
"access_key": "your_aws_access_key",
"secret_key": "your_aws_secret_key",
"session_token": "your_aws_session_token"
},
"actions": [
{
"action_type": "predict",
"method": "POST",
"url": "https://bedrock-runtime.us-east-1.amazonaws.com/model/anthropic.claude-instant-v1/invoke",
"headers": {
"content-type": "application/json",
"x-amz-content-sha256": "required"
},
"request_body": "{\"prompt\":\"${parameters.prompt}\", \"max_tokens_to_sample\":${parameters.max_tokens_to_sample}, \"temperature\":${parameters.temperature}, \"anthropic_version\":\"${parameters.anthropic_version}\" }"
}
]
}
记下连接器 ID；您将使用它来注册模型。
步骤 2.2：注册模型
注册托管在 Amazon Bedrock 上的 Claude 模型
POST /_plugins/_ml/models/_register
{
"name": "Bedrock Claude Instant model",
"function_name": "remote",
"description": "Bedrock Claude instant-v1 model",
"connector_id": "your_LLM_connector_id"
}
请记下 LLM 模型 ID；您将在以下步骤中使用它。
步骤 2.3：部署模型
部署 Claude 模型
POST /_plugins/_ml/models/your_LLM_model_id/_deploy
步骤 2.4：测试模型
要测试模型，请发送一个 Predict API 请求
POST /_plugins/_ml/models/your_LLM_model_id/_predict
{
"parameters": {
"prompt": "\n\nHuman: how are you? \n\nAssistant:"
}
}
步骤 3：注册代理
OpenSearch 提供以下代理类型：flow、conversational_flow 和 conversational。有关代理的更多信息，请参阅代理。
在本教程中，您将使用一个 conversational_flow 代理。该代理包含以下内容：
- 元信息：
name、type和description。 app_type：区分应用程序类型。memory：将用户问题和 LLM 响应存储为对话，以便代理可以从内存中检索对话历史记录并继续相同的对话。tools：定义要使用的工具列表。代理将按顺序运行这些工具。
要注册代理，请发送以下请求
POST /_plugins/_ml/agents/_register
{
"name": "population data analysis agent",
"type": "conversational_flow",
"description": "This is a demo agent for population data analysis",
"app_type": "rag",
"memory": {
"type": "conversation_index"
},
"tools": [
{
"type": "VectorDBTool",
"name": "population_knowledge_base",
"parameters": {
"model_id": "your_text_embedding_model_id",
"index": "test_population_data",
"embedding_field": "population_description_embedding",
"source_field": [
"population_description"
],
"input": "${parameters.question}"
}
},
{
"type": "MLModelTool",
"name": "bedrock_claude_model",
"description": "A general tool to answer any question",
"parameters": {
"model_id": "your_LLM_model_id",
"prompt": "\n\nHuman:You are a professional data analysist. You will always answer question based on the given context first. If the answer is not directly shown in the context, you will analyze the data and find the answer. If you don't know the answer, just say don't know. \n\nContext:\n${parameters.population_knowledge_base.output:-}\n\n${parameters.chat_history:-}\n\nHuman:${parameters.question}\n\nAssistant:"
}
}
]
}
OpenSearch 返回一个代理 ID
{
"agent_id": "fQ75lI0BHcHmo_czdqcJ"
}
记下代理 ID；您将在下一步中使用它。
步骤 4：运行代理
您将运行代理来分析西雅图人口的增长。当您运行此代理时，它将创建一个新对话。稍后，您可以通过提出其他问题来继续此对话。
步骤 4.1：开始新对话
首先，通过向 LLM 提问来开始一个新对话
POST /_plugins/_ml/agents/your_agent_id/_execute
{
"parameters": {
"question": "what's the population increase of Seattle from 2021 to 2023?"
}
}
响应包含 LLM 生成的答案
{
"inference_results": [
{
"output": [
{
"name": "memory_id",
"result": "gQ75lI0BHcHmo_cz2acL"
},
{
"name": "parent_message_id",
"result": "gg75lI0BHcHmo_cz2acZ"
},
{
"name": "bedrock_claude_model",
"result": """ Based on the context given:
- The metro area population of Seattle in 2021 was 3,461,000
- The current metro area population of Seattle in 2023 is 3,519,000
- So the population increase of Seattle from 2021 to 2023 is 3,519,000 - 3,461,000 = 58,000"""
}
]
}
]
}
响应包含以下字段
memory_id是内存（对话）的标识符，它将单个对话中的所有消息分组。请记下此 ID；您将在下一步中使用它。parent_message_id是人类与 LLM 之间当前消息（一个问题/答案）的标识符。一个内存可以包含多条消息。
要获取内存详情，请调用 Get Memory API
GET /_plugins/_ml/memory/gQ75lI0BHcHmo_cz2acL
要获取内存中的所有消息，请调用 Get Messages API
GET /_plugins/_ml/memory/gQ75lI0BHcHmo_cz2acL/messages
要获取消息详情，请调用 Get Message API
GET /_plugins/_ml/memory/message/gg75lI0BHcHmo_cz2acZ
为了调试目的，您可以通过调用 Get Message Traces API 来获取消息的追踪数据。
GET /_plugins/_ml/memory/message/gg75lI0BHcHmo_cz2acZ/traces
4.2 通过提问新问题来继续对话
要继续相同的对话，请提供上一步中的内存 ID。
此外，您可以提供以下参数
message_history_limit：指定您希望代理在新问答回合中包含多少条历史消息。prompt：使用此参数自定义 LLM 提示。例如，以下示例添加了一个新指令always learn useful information from chat history和一个新参数next_action
POST /_plugins/_ml/agents/your_agent_id/_execute
{
"parameters": {
"question": "What's the population of New York City in 2023?",
"next_action": "then compare with Seattle population of 2023",
"memory_id": "gQ75lI0BHcHmo_cz2acL",
"message_history_limit": 5,
"prompt": "\n\nHuman:You are a professional data analysist. You will always answer question based on the given context first. If the answer is not directly shown in the context, you will analyze the data and find the answer. If you don't know the answer, just say don't know. \n\nContext:\n${parameters.population_knowledge_base.output:-}\n\n${parameters.chat_history:-}\n\nHuman:always learn useful information from chat history\nHuman:${parameters.question}, ${parameters.next_action}\n\nAssistant:"
}
}
响应包含 LLM 生成的答案
{
"inference_results": [
{
"output": [
{
"name": "memory_id",
"result": "gQ75lI0BHcHmo_cz2acL"
},
{
"name": "parent_message_id",
"result": "wQ4JlY0BHcHmo_cz8Kc-"
},
{
"name": "bedrock_claude_model",
"result": """ Based on the context given:
- The current metro area population of New York City in 2023 is 18,937,000
- The current metro area population of Seattle in 2023 is 3,519,000
- So the population of New York City in 2023 (18,937,000) is much higher than the population of Seattle in 2023 (3,519,000)"""
}
]
}
]
}
如果您知道代理应使用哪个工具来执行特定的 Predict API 请求，您可以在执行代理时指定该工具。例如，如果您想将上述答案翻译成中文，则无需从知识库中检索任何数据。要仅运行 Claude 模型，请在 selected_tools 参数中指定 bedrock_claude_model 工具
POST /_plugins/_ml/agents/your_agent_id/_execute
{
"parameters": {
"question": "Translate last answer into Chinese?",
"selected_tools": ["bedrock_claude_model"]
}
}
代理将按照 selected_tools 中定义的新顺序逐个运行工具。
配置多个知识库
您可以为代理配置多个知识库。例如，如果您同时拥有产品描述和评论数据，则可以使用以下两个工具配置代理
{
"name": "My product agent",
"type": "conversational_flow",
"description": "This is an agent with product description and comments knowledge bases.",
"memory": {
"type": "conversation_index"
},
"app_type": "rag",
"tools": [
{
"type": "VectorDBTool",
"name": "product_description_vectordb",
"parameters": {
"model_id": "your_embedding_model_id",
"index": "product_description_data",
"embedding_field": "product_description_embedding",
"source_field": [
"product_description"
],
"input": "${parameters.question}"
}
},
{
"type": "VectorDBTool",
"name": "product_comments_vectordb",
"parameters": {
"model_id": "your_embedding_model_id",
"index": "product_comments_data",
"embedding_field": "product_comment_embedding",
"source_field": [
"product_comment"
],
"input": "${parameters.question}"
}
},
{
"type": "MLModelTool",
"description": "A general tool to answer any question",
"parameters": {
"model_id": "",
"prompt": "\n\nHuman:You are a professional product recommendation engine. You will always recommend product based on the given context. If you don't have enough context, you will ask Human to provide more information. If you don't see any related product to recommend, just say we don't have such product. \n\n Context:\n${parameters.product_description_vectordb.output}\n\n${parameters.product_comments_vectordb.output}\n\nHuman:${parameters.question}\n\nAssistant:"
}
}
]
}
当您运行代理时，代理将查询产品描述和评论数据，然后将查询结果和问题发送给 LLM。
要查询特定的知识库，请在 selected_tools 中指定。例如，如果问题仅与产品评论相关，您只能从 product_comments_vectordb 中检索信息。
POST /_plugins/_ml/agents/your_agent_id/_execute
{
"parameters": {
"question": "What feature people like the most for Amazon Echo Dot",
"selected_tools": ["product_comments_vectordb", "MLModelTool"]
}
}
在索引上运行查询
使用 SearchIndexTool 在任何索引上运行任何 OpenSearch 查询。
设置：注册代理
POST /_plugins/_ml/agents/_register
{
"name": "Demo agent",
"type": "conversational_flow",
"description": "This agent supports running any search query",
"memory": {
"type": "conversation_index"
},
"app_type": "rag",
"tools": [
{
"type": "SearchIndexTool",
"parameters": {
"input": "{\"index\": \"${parameters.index}\", \"query\": ${parameters.query} }"
}
},
{
"type": "MLModelTool",
"description": "A general tool to answer any question",
"parameters": {
"model_id": "your_llm_model_id",
"prompt": "\n\nHuman:You are a professional data analysist. You will always answer question based on the given context first. If the answer is not directly shown in the context, you will analyze the data and find the answer. If you don't know the answer, just say don't know. \n\n Context:\n${parameters.SearchIndexTool.output:-}\n\nHuman:${parameters.question}\n\nAssistant:"
}
}
]
}
运行 BM25 查询
POST /_plugins/_ml/agents/your_agent_id/_execute
{
"parameters": {
"question": "what's the population increase of Seattle from 2021 to 2023?",
"index": "test_population_data",
"query": {
"query": {
"match": {
"population_description": "${parameters.question}"
}
},
"size": 2,
"_source": "population_description"
}
}
}
仅暴露 question 参数
要仅暴露 question 参数，请如下定义代理
POST /_plugins/_ml/agents/_register
{
"name": "Demo agent",
"type": "conversational_flow",
"description": "This is a test agent support running any search query",
"memory": {
"type": "conversation_index"
},
"app_type": "rag",
"tools": [
{
"type": "SearchIndexTool",
"parameters": {
"input": "{\"index\": \"${parameters.index}\", \"query\": ${parameters.query} }",
"index": "test_population_data",
"query": {
"query": {
"match": {
"population_description": "${parameters.question}"
}
},
"size": 2,
"_source": "population_description"
}
}
},
{
"type": "MLModelTool",
"description": "A general tool to answer any question",
"parameters": {
"model_id": "your_llm_model_id",
"prompt": "\n\nHuman:You are a professional data analyst. You will always answer question based on the given context first. If the answer is not directly shown in the context, you will analyze the data and find the answer. If you don't know the answer, just say don't know. \n\n Context:\n${parameters.SearchIndexTool.output:-}\n\nHuman:${parameters.question}\n\nAssistant:"
}
}
]
}
现在您可以只指定 question 参数来运行代理
POST /_plugins/_ml/agents/your_agent_id/_execute
{
"parameters": {
"question": "what's the population increase of Seattle from 2021 to 2023?"
}
}
运行向量搜索
POST /_plugins/_ml/agents/your_agent_id/_execute
{
"parameters": {
"question": "what's the population increase of Seattle from 2021 to 2023??",
"index": "test_population_data",
"query": {
"query": {
"neural": {
"population_description_embedding": {
"query_text": "${parameters.question}",
"model_id": "your_embedding_model_id",
"k": 10
}
}
},
"size": 2,
"_source": ["population_description"]
}
}
}
要暴露 question 参数，请参阅仅暴露 question 参数。
运行混合搜索查询
混合搜索结合了关键词搜索和向量搜索以提高搜索相关性。更多信息，请参阅混合搜索。
配置搜索管道
PUT /_search/pipeline/nlp-search-pipeline
{
"description": "Post processor for hybrid search",
"phase_results_processors": [
{
"normalization-processor": {
"normalization": {
"technique": "min_max"
},
"combination": {
"technique": "arithmetic_mean",
"parameters": {
"weights": [
0.3,
0.7
]
}
}
}
}
]
}
使用混合查询运行代理
POST /_plugins/_ml/agents/your_agent_id/_execute
{
"parameters": {
"question": "what's the population increase of Seattle from 2021 to 2023??",
"index": "test_population_data",
"query": {
"_source": {
"exclude": [
"population_description_embedding"
]
},
"size": 2,
"query": {
"hybrid": {
"queries": [
{
"match": {
"population_description": {
"query": "${parameters.question}"
}
}
},
{
"neural": {
"population_description_embedding": {
"query_text": "${parameters.question}",
"model_id": "your_embedding_model_id",
"k": 10
}
}
}
]
}
}
}
}
}
要暴露 question 参数，请参阅仅暴露 question 参数。
自然语言查询
PPLTool 可以将自然语言查询 (NLQ) 转换为 Piped Processing Language (PPL) 并执行生成的 PPL 查询。
设置
开始之前，请前往 OpenSearch Dashboards 主页，选择 Add sample data，然后添加 Sample eCommerce orders。
步骤 1：使用 PPLTool 注册代理
PPLTool 具有以下参数
model_type（枚举）：CLAUDE、OPENAI或FINETUNE。execute（布尔值）：如果为true，则执行生成的 PPL 查询。input（字符串）：您必须提供index和question作为输入。
在本教程中，您将使用 Bedrock Claude，因此将 model_type 设置为 CLAUDE
POST /_plugins/_ml/agents/_register
{
"name": "Demo agent for NLQ",
"type": "conversational_flow",
"description": "This is a test flow agent for NLQ",
"memory": {
"type": "conversation_index"
},
"app_type": "rag",
"tools": [
{
"type": "PPLTool",
"parameters": {
"model_id": "your_ppl_model_id",
"model_type": "CLAUDE",
"execute": true,
"input": "{\"index\": \"${parameters.index}\", \"question\": ${parameters.question} }"
}
},
{
"type": "MLModelTool",
"description": "A general tool to answer any question",
"parameters": {
"model_id": "your_llm_model_id",
"prompt": "\n\nHuman:You are a professional data analysist. You will always answer question based on the given context first. If the answer is not directly shown in the context, you will analyze the data and find the answer. If you don't know the answer, just say don't know. \n\n Context:\n${parameters.PPLTool.output:-}\n\nHuman:${parameters.question}\n\nAssistant:"
}
}
]
}
步骤 2：使用 NLQ 运行代理
运行代理
POST /_plugins/_ml/agents/your_agent_id/_execute
{
"parameters": {
"question": "How many orders do I have in last week",
"index": "opensearch_dashboards_sample_data_ecommerce"
}
}
响应包含 LLM 生成的答案
{
"inference_results": [
{
"output": [
{
"name": "memory_id",
"result": "sqIioI0BJhBwrVXYeYOM"
},
{
"name": "parent_message_id",
"result": "s6IioI0BJhBwrVXYeYOW"
},
{
"name": "MLModelTool",
"result": " Based on the given context, the number of orders in the last week is 3992. The data shows a query that counts the number of orders where the order date is greater than 1 week ago. The query result shows the count as 3992."
}
]
}
]
}
更多信息，请通过调用 Get Message Traces API 获取追踪数据。
GET _plugins/_ml/memory/message/s6IioI0BJhBwrVXYeYOW/traces