save all

llamaindex_cloud_rag/create_cloud_index.py

@@ -0,0 +1,28 @@
+from llama_index.core import SimpleDirectoryReader
+from llama_index.readers.dashscope.base import DashScopeParse
+from llama_index.readers.dashscope.utils import ResultType
+from llama_index.indices.managed.dashscope import DashScopeCloudIndex
+
+
+def read_parse_upload_local_documents(dir, num_workers=1):
+    """读取、解析、上传本地文件到百炼数据管理平台。
+
+    Args:
+        dir (str): 本地文件存储的路径。
+        num_workers (int, optional): 执行的并发数。
+
+    Returns:
+        已上传到云端的文件列表
+    """
+    parse = DashScopeParse(result_type=ResultType.DASHSCOPE_DOCMIND)
+    file_extractor = {'.txt': parse, '.docx': parse, ".pdf": parse}  # 设置需要读取解析的文件格式，请根据实际需求调整
+    documents = SimpleDirectoryReader(input_dir=dir, file_extractor=file_extractor).load_data(num_workers=num_workers)
+    
+    return documents
+
+
+if __name__ == '__main__':
+    dir = "./docs/" # 本例中，业务相关文件存储在当前路径下的docs文件夹，请根据实际情况调整。
+    documents = read_parse_upload_local_documents(dir)
+    cloud_index_name = "my_first_index"  # 设置云端知识库索引名称
+    index = DashScopeCloudIndex.from_documents(documents, cloud_index_name, verbose=True) # 创建云端知识库索引

llamaindex_cloud_rag/docs/百炼系列智能音箱产品介绍.txt

@@ -0,0 +1,11 @@
+百炼系列智能音箱产品介绍
+
+踏入智能家居的核心地带，百炼智能音箱系列以其创新设计与智能交互技术，重新定义家庭生活的每一个角落。
+
+百炼EchoSphere S1 —— 智慧家居的指挥官： 设计简约而不失时尚，采用360度全向发声技术，配备高端音频单元，提供深沉低音与清澈高音的完美平衡。支持Wi-Fi与蓝牙5.0双连接，轻松接入您的智能家居生态。内置智能语音助手，只需一声令下，即可控制家中电器、查询天气、播放音乐等。长达20小时的续航能力，让音乐与便利伴您一整天。参考售价：999 - 1299。
+通义VoiceBox V7 —— 生活小秘书： 采用8个远场麦克风阵列，即使在嘈杂环境中也能精准识别您的指令。拥有7英寸触摸屏，不仅能听更能看，查看菜谱、视频通话、浏览新闻一应俱全。4GB内存与64GB存储空间，存储您喜爱的歌曲与节目。内置电池支持8小时连续使用，从厨房到卧室，随心移动。参考售价：1499 - 1799。
+星尘SoundWave SW Pro —— 音乐发烧友的选择： 采用HiFi级音频解码芯片，支持aptX HD与LDAC高品质音频传输，还原录音室级别的音质。独特的声学结构设计，带来震撼的3D环绕声效果。12GB RAM与256GB存储，存储海量音乐资源。AI音效调节，根据环境自动优化音质。内置电池续航高达12小时，户外派对也能尽兴。参考售价：2999 - 3499。
+百炼SmartHalo H1 —— 家庭互动的中心： 独特的环形灯效设计，不仅是一款音箱，更是家居装饰的艺术品。支持多种语音助手，兼容iOS与Android系统，家庭成员均可个性化设置。内置丰富的儿童教育资源，成为孩子成长的好伙伴。5000mAh电池，满足全家人的日常使用需求。参考售价：1299 - 1599。
+百炼MiniBoom MB Plus —— 便携式智能乐趣： 极致小巧，仅手掌大小，却拥有令人惊喜的音质表现。IP67防水防尘设计，无论户外探险还是浴室放松，都能陪伴左右。支持语音控制与触控操作，内置电池续航达10小时。无线串联功能，轻松构建环绕立体声场。参考售价：299 - 399。
+
+每一款百炼智能音箱都是对智能生活美学的诠释，旨在为您的家庭带来前所未有的便捷与愉悦。选择百炼，让智能生活的声音更加悦耳动听。

llamaindex_cloud_rag/rag.py

@@ -0,0 +1,57 @@
+from llama_index.core import Settings
+from llama_index.llms.dashscope import DashScope
+from llama_index.indices.managed.dashscope import DashScopeCloudIndex
+from llama_index.core.postprocessor import SimilarityPostprocessor
+from llama_index.postprocessor.dashscope_rerank import DashScopeRerank
+
+'''
+本例中构建检索引擎时，需要手动设置下列参数，请根据实际效果调整。
+'''
+Settings.llm = DashScope(model_name="qwen-max") # 设置检索引擎生成回答时调用的大模型。
+similarity_top_k = 5  # 检索引擎找到的相似度最高的结果数
+similarity_cutoff = 0.4  # 过滤检索结果时使用的最低相似度阈值
+top_n = 1  # 进行重排后返回语义上相关度最高的结果数
+
+'''
+本例中构建RAG应用时，设置如下问答模板，请根据实际需求调整。
+'''
+init_chat = "\n您好，我是AI助手，可以回答关于百炼系列产品的提问。有什么可以帮您的？（请输入问题，退出请输入'q'）\n> "
+resp_with_no_answer = "很抱歉，知识库未提供相关信息。" + "\n"
+prompt_template = "回答如下问题: {0}\n如果根据提供的信息无法回答，请返回：{1}"
+
+
+'''
+格式化输出。
+'''
+def prettify_rag(resp):
+    output = ""
+    output += "\n回答：{0}\n".format(resp.response)
+    for j in range(len(resp.source_nodes)):
+        output += "\n产品知识库中的相关文本：\n{0}\n".format(resp.source_nodes[j].text)
+    return output
+
+
+'''
+基于云端知识库的向量索引，构建检索引擎，能够接收终端用户的提问，从云端知识库中检索相关的文本片段，再将提问和检索结果合并后输入到大模型，并生成回答。
+RAG应用提供与终端用户的交互界面，如果无法检索到相关的文本片段，或根据检索到的文本片段无法回答终端用户的提问，则返回适当的错误信息。
+'''
+if __name__ == '__main__':
+    index = DashScopeCloudIndex("my_first_index")  # 读取百炼平台上已创建的知识库索引
+    query_engine = index.as_query_engine(  # 构建检索引擎
+        similarity_top_k=similarity_top_k,
+        node_postprocessors=[  # 默认检索结果可能不满足需求，本例中通过加入node_postprocessors对检索结果进行后处理。
+            SimilarityPostprocessor(similarity_cutoff=similarity_cutoff),  # 过滤不满足最低相似度阈值的检索结果。
+            DashScopeRerank(top_n=top_n, model="gte-rerank")  # 对检索结果进行重排，返回语义上相关度最高的结果。
+        ],
+        response_mode="tree_summarize"
+    )
+    while True:
+        user_prompt = input(init_chat)
+        if user_prompt in ['q','Q']:
+            break
+        resp = query_engine.query(prompt_template.format(user_prompt, resp_with_no_answer))
+        if len(resp.source_nodes) == 0:
+            output = resp_with_no_answer  # 如果未找到相关上下文信息，则返回适当的报错信息。
+        else:
+            output = prettify_rag(resp)
+        print(output)

llamaindex_cloud_rag/requirements.txt

@@ -0,0 +1,12 @@
+llama-index==0.10.65
+dashscope==1.20.4
+llama-index-embeddings-dashscope==0.1.4
+llama-index-indices-managed-llama-cloud==0.2.7
+llama-index-llms-dashscope==0.1.2
+llama-index-node-parser-dashscope==0.1.2
+llama-index-postprocessor-dashscope-rerank==0.1.4
+llama-index-readers-dashscope==0.1.2
+llama-index-readers-file==0.1.33
+llama-index-readers-llama-parse==0.1.6
+llama-parse==0.4.9
+llama-index-indices-managed-dashscope==0.1.2

local_rag/README.md

@@ -0,0 +1,2 @@
+Run: uvicorn main:app --port 7866
+Then visit 127.0.0.1:7866

local_rag/VectorStore/tmp_abcd/graph_store.json

@@ -0,0 +1 @@
+{"graph_dict": {}}

local_rag/VectorStore/tmp_abcd/image__vector_store.json

@@ -0,0 +1 @@
+{"embedding_dict": {}, "text_id_to_ref_doc_id": {}, "metadata_dict": {}}

local_rag/VectorStore/tmp_abcd/index_store.json

@@ -0,0 +1 @@
+{"index_store/data": {"ab9a86f4-0029-48e5-b823-eccfc6f58622": {"__type__": "vector_store", "__data__": "{\"index_id\": \"ab9a86f4-0029-48e5-b823-eccfc6f58622\", \"summary\": null, \"nodes_dict\": {\"c7e14d46-b930-4a98-acaf-2854ac14c1de\": \"c7e14d46-b930-4a98-acaf-2854ac14c1de\", \"fb1f2a38-146e-4840-8f0d-221033dbc849\": \"fb1f2a38-146e-4840-8f0d-221033dbc849\"}, \"doc_id_dict\": {}, \"embeddings_dict\": {}}"}}}

local_rag/chat.py

@@ -0,0 +1,93 @@
+import os
+from openai import OpenAI
+from llama_index.core import StorageContext,load_index_from_storage,Settings
+from llama_index.embeddings.dashscope import (
+    DashScopeEmbedding,
+    DashScopeTextEmbeddingModels,
+    DashScopeTextEmbeddingType,
+)
+from llama_index.postprocessor.dashscope_rerank import DashScopeRerank
+from create_kb import *
+DB_PATH = "VectorStore"
+TMP_NAME = "tmp_abcd"
+EMBED_MODEL = DashScopeEmbedding(
+    model_name=DashScopeTextEmbeddingModels.TEXT_EMBEDDING_V2,
+    text_type=DashScopeTextEmbeddingType.TEXT_TYPE_DOCUMENT,
+)
+# 若使用本地嵌入模型，请取消以下注释：
+# from langchain_community.embeddings import ModelScopeEmbeddings
+# from llama_index.embeddings.langchain import LangchainEmbedding
+# embeddings = ModelScopeEmbeddings(model_id="modelscope/iic/nlp_gte_sentence-embedding_chinese-large")
+# EMBED_MODEL = LangchainEmbedding(embeddings)
+
+# 设置嵌入模型
+Settings.embed_model = EMBED_MODEL
+
+def get_model_response(multi_modal_input,history,model,temperature,max_tokens,history_round,db_name,similarity_threshold,chunk_cnt):
+    # prompt = multi_modal_input['text']
+    prompt = history[-1][0]
+    tmp_files = multi_modal_input['files']
+    if os.path.exists(os.path.join("File",TMP_NAME)):
+        db_name = TMP_NAME
+    else:
+        if tmp_files:
+            create_tmp_kb(tmp_files)
+            db_name = TMP_NAME
+    # 获取index
+    print(f"prompt:{prompt},tmp_files:{tmp_files},db_name:{db_name}")
+    try:
+        dashscope_rerank = DashScopeRerank(top_n=chunk_cnt,return_documents=True)
+        storage_context = StorageContext.from_defaults(
+            persist_dir=os.path.join(DB_PATH,db_name)
+        )
+        index = load_index_from_storage(storage_context)
+        print("index获取完成")
+        retriever_engine = index.as_retriever(
+            similarity_top_k=20,
+        )
+        # 获取chunk
+        retrieve_chunk = retriever_engine.retrieve(prompt)
+        print(f"原始chunk为：{retrieve_chunk}")
+        try:
+            results = dashscope_rerank.postprocess_nodes(retrieve_chunk, query_str=prompt)
+            print(f"rerank成功，重排后的chunk为：{results}")
+        except:
+            results = retrieve_chunk[:chunk_cnt]
+            print(f"rerank失败，chunk为：{results}")
+        chunk_text = ""
+        chunk_show = ""
+        for i in range(len(results)):
+            if results[i].score >= similarity_threshold:
+                chunk_text = chunk_text + f"## {i+1}:\n {results[i].text}\n"
+                chunk_show = chunk_show + f"## {i+1}:\n {results[i].text}\nscore: {round(results[i].score,2)}\n"
+        print(f"已获取chunk：{chunk_text}")
+        prompt_template = f"请参考以下内容：{chunk_text}，以合适的语气回答用户的问题：{prompt}。如果参考内容中有图片链接也请直接返回。"
+    except Exception as e:
+        print(f"异常信息：{e}")
+        prompt_template = prompt
+        chunk_show = ""
+    history[-1][-1] = ""
+    client = OpenAI(
+        api_key=os.getenv("DASHSCOPE_API_KEY"),
+        base_url="https://dashscope.aliyuncs.com/compatible-mode/v1",
+    )                
+    system_message = {'role': 'system', 'content': 'You are a helpful assistant.'}
+    messages = []
+    history_round = min(len(history),history_round)
+    for i in range(history_round):
+        messages.append({'role': 'user', 'content': history[-history_round+i][0]})
+        messages.append({'role': 'assistant', 'content': history[-history_round+i][1]})
+    messages.append({'role': 'user', 'content': prompt_template})
+    messages = [system_message] + messages
+    completion = client.chat.completions.create(
+        model=model,
+        messages=messages,
+        temperature=temperature,
+        max_tokens=max_tokens,
+        stream=True
+        )
+    assistant_response = ""
+    for chunk in completion:
+        assistant_response += chunk.choices[0].delta.content
+        history[-1][-1] = assistant_response
+        yield history,chunk_show

local_rag/create_kb.py

@@ -0,0 +1,134 @@
+#####################################
+######       创建知识库         #######
+#####################################
+import gradio as gr
+import os
+import shutil
+from llama_index.core import VectorStoreIndex,Settings,SimpleDirectoryReader
+from llama_index.embeddings.dashscope import (
+    DashScopeEmbedding,
+    DashScopeTextEmbeddingModels,
+    DashScopeTextEmbeddingType,
+)
+from llama_index.core.schema import TextNode
+from upload_file import *
+DB_PATH = "VectorStore"
+STRUCTURED_FILE_PATH = "File/Structured"
+UNSTRUCTURED_FILE_PATH = "File/Unstructured"
+TMP_NAME = "tmp_abcd"
+EMBED_MODEL = DashScopeEmbedding(
+    model_name=DashScopeTextEmbeddingModels.TEXT_EMBEDDING_V2,
+    text_type=DashScopeTextEmbeddingType.TEXT_TYPE_DOCUMENT,
+)
+# 若使用本地嵌入模型，请取消以下注释：
+# from langchain_community.embeddings import ModelScopeEmbeddings
+# from llama_index.embeddings.langchain import LangchainEmbedding
+# embeddings = ModelScopeEmbeddings(model_id="modelscope/iic/nlp_gte_sentence-embedding_chinese-large")
+# EMBED_MODEL = LangchainEmbedding(embeddings)
+
+
+# 设置嵌入模型
+Settings.embed_model = EMBED_MODEL
+# 刷新知识库
+def refresh_knowledge_base():
+    return os.listdir(DB_PATH)
+
+# 创建非结构化向量数据库
+def create_unstructured_db(db_name:str,label_name:list):
+    print(f"知识库名称为：{db_name}，类目名称为：{label_name}")
+    if label_name is None:
+        gr.Info("没有选择类目")
+    elif len(db_name) == 0:
+        gr.Info("没有命名知识库")
+    # 判断是否存在同名向量数据库
+    elif db_name in os.listdir(DB_PATH):
+        gr.Info("知识库已存在，请换个名字或删除原来知识库再创建")
+    else:
+        gr.Info("正在创建知识库，请等待知识库创建成功信息显示后前往RAG问答")
+        documents = []
+        for label in label_name:
+            label_path = os.path.join(UNSTRUCTURED_FILE_PATH,label)
+            documents.extend(SimpleDirectoryReader(label_path).load_data())
+        index = VectorStoreIndex.from_documents(
+            documents
+        )
+        db_path = os.path.join(DB_PATH,db_name)
+        if not os.path.exists(db_path):
+            os.mkdir(db_path)
+            index.storage_context.persist(db_path)
+        elif os.path.exists(db_path):
+            pass
+        gr.Info("知识库创建成功，可前往RAG问答进行提问")
+    
+# 创建结构化向量数据库
+def create_structured_db(db_name:str,data_table:list):
+    print(f"知识库名称为：{db_name}，数据表名称为：{data_table}")
+    if data_table is None:
+        gr.Info("没有选择数据表")
+    elif len(db_name) == 0:
+        gr.Info("没有命名知识库")
+    # 判断是否存在同名向量数据库
+    elif db_name in os.listdir(DB_PATH):
+        gr.Info("知识库已存在，请换个名字或删除原来知识库再创建")
+    else:
+        gr.Info("正在创建知识库，请等待知识库创建成功信息显示后前往RAG问答")
+        documents = []
+        for label in data_table:
+            label_path = os.path.join(STRUCTURED_FILE_PATH,label)
+            documents.extend(SimpleDirectoryReader(label_path).load_data())
+        # index = VectorStoreIndex.from_documents(
+        #     documents
+        # )
+        nodes = []
+        for doc in documents:
+            doc_content = doc.get_content().split('\n')
+            for chunk in doc_content:
+                node = TextNode(text=chunk)
+                node.metadata = {'source': doc.get_doc_id(),'file_name':doc.metadata['file_name']}
+                nodes = nodes + [node]
+        index = VectorStoreIndex(nodes)
+        db_path = os.path.join(DB_PATH,db_name)
+        if not os.path.exists(db_path):
+            os.mkdir(db_path)
+        index.storage_context.persist(db_path)
+        gr.Info("知识库创建成功，可前往RAG问答进行提问")
+
+
+# 删除指定名称知识库
+def delete_db(db_name:str):
+    if db_name is not None:
+        folder_path = os.path.join(DB_PATH, db_name)
+        if os.path.exists(folder_path):
+            shutil.rmtree(folder_path)
+            gr.Info(f"已成功删除{db_name}知识库")
+            print(f"已成功删除{db_name}知识库")
+        else:
+            gr.Info(f"{db_name}知识库不存在")
+            print(f"{db_name}知识库不存在")
+
+# 实时更新知识库列表
+def update_knowledge_base():
+    return gr.update(choices=os.listdir(DB_PATH))
+
+# 临时文件创建知识库
+def create_tmp_kb(files):
+    if not os.path.exists(os.path.join("File",TMP_NAME)):
+        os.mkdir(os.path.join("File",TMP_NAME))
+    for file in files:
+        file_name = os.path.basename(file)
+        shutil.move(file,os.path.join("File",TMP_NAME,file_name))
+    documents = SimpleDirectoryReader(os.path.join("File",TMP_NAME)).load_data()
+    index = VectorStoreIndex.from_documents(
+        documents
+    )
+    db_path = os.path.join(DB_PATH,TMP_NAME)
+    if not os.path.exists(db_path):
+        os.mkdir(db_path)
+    index.storage_context.persist(db_path)
+
+# 清除tmp文件夹下内容 
+def clear_tmp():
+    if os.path.exists(os.path.join("File",TMP_NAME)):
+        shutil.rmtree(os.path.join("File",TMP_NAME))
+    if os.path.exists(os.path.join(DB_PATH,TMP_NAME)):
+        shutil.rmtree(os.path.join(DB_PATH,TMP_NAME))

local_rag/html_string.py

@@ -0,0 +1,150 @@
+main_html = """<!DOCTYPE html>
+<html lang="zh">
+<head>
+    <meta charset="UTF-8">
+    <meta name="viewport" content="width=device-width, initial-scale=1.0">
+    <title>阿里云本地RAG解决方案</title>
+    <link rel="stylesheet" href="https://fonts.googleapis.com/icon?family=Material+Icons">
+    <style>
+        body {
+            font-family: Arial, sans-serif;
+            background-color: #f5f5f5;
+            margin: 0;
+            padding: 0;
+            display: flex;
+            flex-direction: column;
+            align-items: center;
+        }
+        header {
+            background-color: #2196f3;
+            color: white;
+            width: 100%;
+            padding: 1.5em;
+            text-align: center;
+            box-shadow: 0 2px 4px rgba(0, 0, 0, 0.1);
+        }
+        main {
+            margin: 2em;
+            box-shadow: 0 2px 4px rgba(0, 0, 0, 0.1);
+            background-color: white;
+            border-radius: 8px;
+            overflow: hidden;
+            width: 90%;
+            max-width: 800px;
+            padding: 2em;
+        }
+        h1 {
+            color: #333;
+        }
+        p {
+            color: #666;
+            font-size: 1.1em;
+        }
+        ul {
+            list-style-type: none;
+            padding: 0;
+        }
+        ul li {
+            background-color: #2196f3;
+            margin: 0.5em 0;
+            padding: 1em;
+            border-radius: 4px;
+            transition: background-color 0.3s;
+        }
+        ul li a {
+            color: white;
+            text-decoration: none;
+            display: flex;
+            align-items: center;
+        }
+        ul li:hover {
+            background-color: #1976d2;
+        }
+        .material-icons {
+            margin-right: 0.5em;
+        }
+    </style>
+</head>
+<body>
+    <header>
+        <h1>阿里云本地RAG解决方案</h1>
+    </header>
+    <main>
+        <p>如果您需要基于上传的文档与模型直接对话，请直接访问<a href="/chat">RAG问答</a>，并在输入框位置上传文件，就可以开始对话了。（此次上传的数据在页面刷新后无法保留，若您希望可以持久使用、维护知识库，请创建知识库）。</p>
+        <p>如果您需要创建或更新知识库，请按照<a href="/upload_data">上传数据</a>、<a href="/create_knowledge_base">创建知识库</a>操作，在<a href="/chat">RAG问答</a>中的“知识库选择”位置选择您需要使用的知识库。</p>
+        <p>如果您需要基于已创建好的知识库进行问答，请直接访问<a href="/chat">RAG问答</a>，在“加载知识库”处选择您已创建的知识库。</p>
+        <ul>
+            <li><a href="/upload_data"><span class="material-icons"></span> 1. 上传数据</a></li>
+            <li><a href="/create_knowledge_base"><span class="material-icons"></span> 2. 创建知识库</a></li>
+            <li><a href="/chat"><span class="material-icons"></span> 3. RAG问答</a></li>
+        </ul>
+    </main>
+</body>
+</html>"""
+
+plain_html = """<!DOCTYPE html>
+<html lang="zh">
+    <head>
+        <title>RAG问答</title>
+        <link href="https://fonts.googleapis.com/icon?family=Material+Icons" rel="stylesheet">
+        <style>
+        .links-container {
+            display: flex;
+            justify-content: center; /* 在容器中居中分布子元素 */
+            list-style-type: none; /* 去掉ul默认的列表样式 */
+            padding: 0; /* 去掉ul默认的内边距 */
+            margin: 0; /* 去掉ul默认的外边距 */
+        }
+        .links-container li {
+            margin: 0 5px; /* 每个li元素的左右留出一些空间 */
+            padding: 10px 15px; /* 添加内边距 */
+            border: 1px solid #ccc; /* 添加边框 */
+            border-radius: 5px; /* 添加圆角 */
+            background-color: #f9f9f9; /* 背景颜色 */
+            transition: background-color 0.3s; /* 背景颜色变化的过渡效果 */
+            display: flex; /* 使用flex布局 */
+            align-items: center; /* 垂直居中对齐 */
+            height: 50px; /* 设置固定高度，确保一致 */
+        }
+        .links-container li:hover {
+            background-color: #e0e0e0; /* 悬停时的背景颜色 */
+        }
+        .links-container a {
+            text-decoration: none !important; /* 去掉链接的下划线 */
+            color: #333; /* 链接颜色 */
+            font-family: Arial, sans-serif; /* 字体 */
+            font-size: 14px; /* 字体大小 */
+            display: flex; /* 使用flex布局 */
+            align-items: center; /* 垂直居中对齐 */
+            height: 100%; /* 确保链接高度与父元素一致 */
+        }
+        .material-icons {
+            font-size: 20px; /* 图标大小 */
+            margin-right: 8px; /* 图标和文字间的间距 */
+            text-decoration: none; /* 确保图标没有下划线 */
+        }
+
+        /* 深色模式样式 */
+        @media (prefers-color-scheme: dark) {
+            .links-container li {
+                background-color: #333; /* 深色模式下的背景颜色 */
+                border-color: #555; /* 深色模式下的边框颜色 */
+            }
+            .links-container li:hover {
+                background-color: #555; /* 深色模式下悬停时的背景颜色 */
+            }
+            .links-container a {
+                color: #f9f9f9; /* 深色模式下的文字颜色 */
+            }
+        }
+        </style>
+    </head>
+    <body>
+        <ul class="links-container">
+            <li><a href="/"><span class="material-icons">home</span> 主页</a></li>
+            <li><a href="/upload_data"><span class="material-icons">cloud_upload</span> 上传数据</a></li>
+            <li><a href="/create_knowledge_base"><span class="material-icons">library_add</span> 创建知识库</a></li>
+            <li><a href="/chat"><span class="material-icons">question_answer</span> RAG问答</a></li>
+        </ul>
+    </body>
+</html>"""

local_rag/main.py

@@ -0,0 +1,116 @@
+from fastapi import FastAPI
+from fastapi.responses import HTMLResponse
+import gradio as gr
+import os
+from html_string import main_html,plain_html
+from upload_file import *
+from create_kb import *
+from chat import get_model_response
+def user(user_message, history):
+    print(user_message)
+    return {'text': '','files': user_message['files']}, history + [[user_message['text'], None]]
+
+#####################################
+######       gradio界面       #######
+#####################################
+
+def get_chat_block():
+    with gr.Blocks(theme=gr.themes.Base(),css=".gradio_container { background-color: #f0f0f0; }") as chat:
+        gr.HTML(plain_html)
+        with gr.Row():     
+            with gr.Column(scale=10):
+                chatbot = gr.Chatbot(label="Chatbot",height=750,avatar_images=("images/user.jpeg","images/tongyi.png"))
+                with gr.Row():
+                    # 
+                    input_message = gr.MultimodalTextbox(label="请输入",file_types=[".xlsx",".csv",".docx",".pdf",".txt"],scale=7)
+                    clear_btn = gr.ClearButton(chatbot,input_message,scale=1)
+            # 模型与知识库参数
+            with gr.Column(scale=5):
+                knowledge_base =gr.Dropdown(choices=os.listdir(DB_PATH),label="加载知识库",interactive=True,scale=2)
+                with gr.Accordion(label="召回文本段",open=False):
+                    chunk_text = gr.Textbox(label="召回文本段",interactive=False,scale=5,lines=10)
+                with gr.Accordion(label="模型设置",open=True):
+                    model =gr.Dropdown(choices=['qwen-max','qwen-plus','qwen-turbo'],label="选择模型",interactive=True,value="qwen-max",scale=2)
+                    temperature = gr.Slider(maximum=2,minimum=0,interactive=True,label="温度参数",step=0.01,value=0.85,scale=2)
+                    max_tokens = gr.Slider(maximum=8000,minimum=0,interactive=True,label="最大回复长度",step=50,value=1024,scale=2)
+                    history_round = gr.Slider(maximum=30,minimum=1,interactive=True,label="携带上下文轮数",step=1,value=3,scale=2)
+                with gr.Accordion(label="RAG参数设置",open=True):
+                    chunk_cnt = gr.Slider(maximum=20,minimum=1,interactive=True,label="选择召回片段数",step=1,value=5,scale=2)
+                    similarity_threshold = gr.Slider(maximum=1,minimum=0,interactive=True,label="相似度阈值",step=0.01,value=0.2,scale=2)
+        input_message.submit(fn=user,inputs=[input_message,chatbot],outputs=[input_message,chatbot],queue=False).then(
+            fn=get_model_response,inputs=[input_message,chatbot,model,temperature,max_tokens,history_round,knowledge_base,similarity_threshold,chunk_cnt],outputs=[chatbot,chunk_text]
+            )
+        chat.load(update_knowledge_base,[],knowledge_base)
+        chat.load(clear_tmp)
+    return chat
+
+
+def get_upload_block():
+    with gr.Blocks(theme=gr.themes.Base()) as upload:
+        gr.HTML(plain_html)
+        with gr.Tab("非结构化数据"):
+            with gr.Accordion(label="新建类目",open=True):
+                with gr.Column(scale=2):
+                    unstructured_file = gr.Files(file_types=["pdf","docx","txt"])
+                    with gr.Row():
+                        new_label = gr.Textbox(label="类目名称",placeholder="请输入类目名称",scale=5)
+                        create_label_btn = gr.Button("新建类目",variant="primary",scale=1)
+            with gr.Accordion(label="管理类目",open=False):
+                with gr.Row():
+                    data_label =gr.Dropdown(choices=os.listdir(UNSTRUCTURED_FILE_PATH),label="管理类目",interactive=True,scale=8,multiselect=True)
+                    delete_label_btn = gr.Button("删除类目",variant="stop",scale=1)
+        with gr.Tab("结构化数据"):
+            with gr.Accordion(label="新建数据表",open=True):
+                with gr.Column(scale=2):
+                    structured_file = gr.Files(file_types=["xlsx","csv"])
+                    with gr.Row():
+                        new_label_1 = gr.Textbox(label="数据表名称",placeholder="请输入数据表名称",scale=5)
+                        create_label_btn_1 = gr.Button("新建数据表",variant="primary",scale=1)
+            with gr.Accordion(label="管理数据表",open=False):
+                with gr.Row():
+                    data_label_1 =gr.Dropdown(choices=os.listdir(STRUCTURED_FILE_PATH),label="管理数据表",interactive=True,scale=8,multiselect=True)
+                    delete_data_table_btn = gr.Button("删除数据表",variant="stop",scale=1)
+        delete_label_btn.click(delete_label,inputs=[data_label]).then(fn=update_label,outputs=[data_label])
+        create_label_btn.click(fn=upload_unstructured_file,inputs=[unstructured_file,new_label]).then(fn=update_label,outputs=[data_label])
+        delete_data_table_btn.click(delete_data_table,inputs=[data_label_1]).then(fn=update_datatable,outputs=[data_label_1])
+        create_label_btn_1.click(fn=upload_structured_file,inputs=[structured_file,new_label_1]).then(fn=update_datatable,outputs=[data_label_1])
+        upload.load(update_label,[],data_label)
+        upload.load(update_datatable,[],data_label_1)
+    return upload
+
+def get_knowledge_base_block():
+    with gr.Blocks(theme=gr.themes.Base()) as knowledge:
+        gr.HTML(plain_html)
+        # 非结构化数据知识库
+        with gr.Tab("非结构化数据"):
+            with gr.Row():
+                data_label_2 =gr.Dropdown(choices=os.listdir(UNSTRUCTURED_FILE_PATH),label="选择类目",interactive=True,scale=2,multiselect=True)
+                knowledge_base_name = gr.Textbox(label="知识库名称",placeholder="请输入知识库名称",scale=2)
+                create_knowledge_base_btn = gr.Button("确认创建知识库",variant="primary",scale=1)
+        # 结构化数据知识库
+        with gr.Tab("结构化数据"):
+            with gr.Row():
+                data_label_3 =gr.Dropdown(choices=os.listdir(STRUCTURED_FILE_PATH),label="选择数据表",interactive=True,scale=2,multiselect=True)
+                knowledge_base_name_1 = gr.Textbox(label="知识库名称",placeholder="请输入知识库名称",scale=2)
+                create_knowledge_base_btn_1 = gr.Button("确认创建知识库",variant="primary",scale=1)
+        with gr.Row():
+            knowledge_base =gr.Dropdown(choices=os.listdir(DB_PATH),label="管理知识库",interactive=True,scale=4)
+            delete_db_btn = gr.Button("删除知识库",variant="stop",scale=1)
+        create_knowledge_base_btn.click(fn=create_unstructured_db,inputs=[knowledge_base_name,data_label_2]).then(update_knowledge_base,outputs=[knowledge_base])
+        delete_db_btn.click(delete_db,inputs=[knowledge_base]).then(update_knowledge_base,outputs=[knowledge_base])
+        create_knowledge_base_btn_1.click(fn=create_structured_db,inputs=[knowledge_base_name_1,data_label_3]).then(update_knowledge_base,outputs=[knowledge_base])
+        knowledge.load(update_knowledge_base,[],knowledge_base)
+        knowledge.load(update_label,[],data_label_2)
+        knowledge.load(update_datatable,[],data_label_3)
+    return knowledge
+
+app = FastAPI()
+@app.get("/", response_class=HTMLResponse)
+def read_main():
+    html_content = main_html
+    return HTMLResponse(content=html_content)
+
+
+app = gr.mount_gradio_app(app, get_chat_block(), path="/chat")
+app = gr.mount_gradio_app(app, get_upload_block(), path="/upload_data")
+app = gr.mount_gradio_app(app, get_knowledge_base_block(), path="/create_knowledge_base")

local_rag/requirements.txt

@@ -0,0 +1,26 @@
+gradio==4.32.0
+faiss-cpu==1.8.0.post1
+dashscope==1.20.4
+openai==1.55.3
+httpx==0.27.0
+llama-index-vector-stores-faiss==0.1.2
+llama-index-embeddings-dashscope==0.1.4
+llama-index-readers-file==0.1.33
+matplotlib==3.9.3
+docx2txt==0.8
+openpyxl==3.1.5
+llama-index-core==0.10.67
+uvicorn==0.30.6
+fastapi==0.112.0
+llama-index-postprocessor-dashscope-rerank-custom==0.1.0
+simplejson==3.19.3
+pydantic==2.10.6
+# modelscope==1.18.0
+# langchain_community==0.2.16
+# transformers==4.44.2
+# llama_index.embeddings.huggingface==0.2.3
+# llama-index-embeddings-langchain==0.1.2
+# datasets==2.21.0
+# oss2==2.19.0
+# sortedcontainers==2.4.0
+# addict==2.4.0s

local_rag/upload_file.py

@@ -0,0 +1,107 @@
+#####################################
+#######       上传文件         #######
+#####################################
+import gradio as gr
+import os
+import shutil
+import pandas as pd
+STRUCTURED_FILE_PATH = "File/Structured"
+UNSTRUCTURED_FILE_PATH = "File/Unstructured"
+# 刷新非结构化类目
+def refresh_label():
+    return os.listdir(UNSTRUCTURED_FILE_PATH)
+
+# 刷新结构化数据表
+def refresh_data_table():
+    return os.listdir(STRUCTURED_FILE_PATH)
+
+# 上传非结构化数据
+def upload_unstructured_file(files,label_name):
+    if files is None:
+        gr.Info("请上传文件")
+    elif len(label_name) == 0:
+        gr.Info("请输入类目名称")
+    # 判断类目是否存在
+    elif label_name in os.listdir(UNSTRUCTURED_FILE_PATH):
+        gr.Info(f"{label_name}类目已存在")
+    else:
+        try:
+            if not os.path.exists(os.path.join(UNSTRUCTURED_FILE_PATH,label_name)):
+                os.mkdir(os.path.join(UNSTRUCTURED_FILE_PATH,label_name))
+            for file in files:
+                print(file)
+                file_path = file.name
+                file_name = os.path.basename(file_path)
+                destination_file_path = os.path.join(UNSTRUCTURED_FILE_PATH,label_name,file_name)
+                shutil.move(file_path,destination_file_path)
+            gr.Info(f"文件已上传至{label_name}类目中，请前往创建知识库")
+        except:
+            gr.Info(f"请勿重复上传")
+
+# 上传结构化数据
+def upload_structured_file(files,label_name):
+    if files is None:
+        gr.Info("请上传文件")
+    elif len(label_name) == 0:
+        gr.Info("请输入数据表名称")
+    # 判断数据表是否存在
+    elif label_name in os.listdir(STRUCTURED_FILE_PATH):
+        gr.Info(f"{label_name}数据表已存在")
+    else:
+        try:
+            if not os.path.exists(os.path.join(STRUCTURED_FILE_PATH,label_name)):
+                os.mkdir(os.path.join(STRUCTURED_FILE_PATH,label_name))
+            for file in files:
+                file_path = file.name
+                file_name = os.path.basename(file_path)
+                destination_file_path = os.path.join(STRUCTURED_FILE_PATH,label_name,file_name)
+                shutil.move(file_path,destination_file_path)
+                if os.path.splitext(destination_file_path)[1] == ".xlsx":
+                    df = pd.read_excel(destination_file_path)
+                elif os.path.splitext(destination_file_path)[1] == ".csv":
+                    df = pd.read_csv(destination_file_path)
+                txt_file_name = os.path.splitext(file_name)[0]+'.txt'
+                columns = df.columns
+                with open(os.path.join(STRUCTURED_FILE_PATH,label_name,txt_file_name),"w") as file:
+                    for idx,row in df.iterrows():
+                        file.write("【")
+                        info = []
+                        for col in columns:
+                            info.append(f"{col}:{row[col]}")
+                        infos = ",".join(info)
+                        file.write(infos)
+                        if idx != len(df)-1:
+                            file.write("】\n")
+                        else:
+                            file.write("】")
+                os.remove(destination_file_path)
+            gr.Info(f"文件已上传至{label_name}数据表中，请前往创建知识库")
+        except:
+            gr.Info(f"请勿重复上传")
+
+# 实时更新结构化数据表
+def update_datatable():
+    return gr.update(choices=os.listdir(STRUCTURED_FILE_PATH))
+
+
+# 实时更新非结构化类目
+def update_label():
+    return gr.update(choices=os.listdir(UNSTRUCTURED_FILE_PATH))
+
+# 删除类目
+def delete_label(label_name):
+    if label_name is not None:
+        for label in label_name:
+            folder_path = os.path.join(UNSTRUCTURED_FILE_PATH,label)
+            if os.path.exists(folder_path):
+                shutil.rmtree(folder_path)
+                gr.Info(f"{label}类目已删除")
+
+# 删除数据表
+def delete_data_table(table_name):
+    if table_name is not None:
+        for table in table_name:
+            folder_path = os.path.join(STRUCTURED_FILE_PATH,table)
+            if os.path.exists(folder_path):
+                shutil.rmtree(folder_path)
+                gr.Info(f"{table}数据表已删除")

test/compatible/deepseek.py

@@ -0,0 +1,23 @@
+import os
+from openai import OpenAI
+
+client = OpenAI(
+    # 若没有配置环境变量，请用阿里云百炼API Key将下行替换为：api_key="sk-xxx",
+    api_key=os.getenv("DASHSCOPE_API_KEY"),  # 如何获取API Key：https://help.aliyun.com/zh/model-studio/developer-reference/get-api-key
+    base_url="https://dashscope.aliyuncs.com/compatible-mode/v1"
+)
+
+completion = client.chat.completions.create(
+    model="deepseek-r1",  # 此处以 deepseek-r1 为例，可按需更换模型名称。
+    messages=[
+        {'role': 'user', 'content': '9.9和9.11谁大'}
+    ]
+)
+
+# 通过reasoning_content字段打印思考过程
+print("思考过程：")
+print(completion.choices[0].message.reasoning_content)
+
+# 通过content字段打印最终答案
+print("最终答案：")
+print(completion.choices[0].message.content)

test/dashscope/deepseek.py

@@ -0,0 +1,20 @@
+import os
+import dashscope
+
+messages = [
+    {'role': 'user', 'content': '你是谁？'}
+]
+
+response = dashscope.Generation.call(
+    # 若没有配置环境变量，请用阿里云百炼API Key将下行替换为：api_key="sk-xxx",
+    api_key=os.getenv('DASHSCOPE_API_KEY'),
+    model="deepseek-r1",  # 此处以 deepseek-r1 为例，可按需更换模型名称。
+    messages=messages,
+    # result_format参数不可以设置为"text"。
+    result_format='message'
+)
+
+print("=" * 20 + "思考过程" + "=" * 20)
+print(response.output.choices[0].message.reasoning_content)
+print("=" * 20 + "最终答案" + "=" * 20)
+print(response.output.choices[0].message.content)

test/dashscope/qwen.py

@@ -0,0 +1,15 @@
+import os
+import dashscope
+
+messages = [
+    {'role': 'system', 'content': 'You are a helpful assistant.'},
+    {'role': 'user', 'content': '你是谁？'}
+]
+response = dashscope.Generation.call(
+    # 若没有配置环境变量，请用百炼API Key将下行替换为：api_key="sk-xxx",
+    api_key=os.getenv('DASHSCOPE_API_KEY'),
+    model="qwen-plus", # 此处以qwen-plus为例，可按需更换模型名称。模型列表：https://help.aliyun.com/zh/model-studio/getting-started/models
+    messages=messages,
+    result_format='message'
+    )
+print(response)

test/langchain/compatible/qwen.py

@@ -0,0 +1,14 @@
+from langchain_openai import ChatOpenAI
+import os
+
+chatLLM = ChatOpenAI(
+    api_key=os.getenv("DASHSCOPE_API_KEY"),
+    base_url="https://dashscope.aliyuncs.com/compatible-mode/v1",
+    model="qwen-plus",  # 此处以qwen-plus为例，您可按需更换模型名称。模型列表：https://help.aliyun.com/zh/model-studio/getting-started/models
+    # other params...
+)
+messages = [
+    {"role": "system", "content": "You are a helpful assistant."},
+    {"role": "user", "content": "你是谁？"}]
+response = chatLLM.invoke(messages)
+print(response.model_dump_json())

test/langchain/dashscope/qwen.py

@@ -0,0 +1,11 @@
+from langchain_community.chat_models.tongyi import ChatTongyi
+from langchain_core.messages import HumanMessage
+
+chatLLM = ChatTongyi(
+    model="qwen-max",   # 此处以qwen-max为例，您可按需更换模型名称。模型列表：https://help.aliyun.com/zh/model-studio/getting-started/models
+    streaming=True,
+    # other params...
+)
+res = chatLLM.stream([HumanMessage(content="hi")], streaming=True)
+for r in res:
+    print("chat resp:", r.content)

test/langchain/deepseek.py

@@ -0,0 +1,15 @@
+import getpass
+import os
+
+if not os.environ.get("DEEPSEEK_API_KEY"):
+  os.environ["DEEPSEEK_API_KEY"] = os.environ['DASHSCOPE_API_KEY']
+
+from langchain.chat_models import init_chat_model
+
+model = init_chat_model("deepseek-chat", model_provider="deepseek")
+
+messages = [
+    {"role": "system", "content": "You are a helpful assistant."},
+    {"role": "user", "content": "你是谁？"}]
+response = model.invoke(messages)
+print(response.model_dump_json())

test/llamaindex/dashscope_parse.py

@@ -0,0 +1,25 @@
+import os
+
+from llama_index.readers.dashscope.base import DashScopeParse
+from llama_index.readers.dashscope.utils import ResultType
+
+# 设置业务空间 ID 将决定文档解析结果在”创建知识库“步骤中上传到哪个业务空间
+os.environ['DASHSCOPE_WORKSPACE_ID'] = "<Your Workspace id, Default workspace is empty.>"
+
+# 第一种方式：使用文档解析器解析一个或多个文件
+file = [
+    # 需要解析的文件，支持pdf,doc,docx
+]
+# 解析文件
+parse = DashScopeParse(result_type=ResultType.DASHSCOPE_DOCMIND)
+documents = parse.load_data(file_path=file)
+
+# 第二种方式：使用文档解析器解析一个文件夹内指定类型的文件
+from llama_index.core import SimpleDirectoryReader
+parse = DashScopeParse(result_type=ResultType.DASHSCOPE_DOCMIND)
+# 定义不同文档类型的解析器
+file_extractor = {".pdf": parse, '.doc': parse, '.docx': parse}
+# 读取文件夹，提取和解析文件信息
+documents = SimpleDirectoryReader(
+    "your_folder", file_extractor=file_extractor
+).load_data(num_workers=1)

test/llamaindex/data/test1.py

@@ -0,0 +1,7 @@
+from llama_index.core import VectorStoreIndex, SimpleDirectoryReader
+
+documents = SimpleDirectoryReader("data").load_data()
+index = VectorStoreIndex.from_documents(documents)
+query_engine = index.as_query_engine()
+response = query_engine.query("Some question about the data should go here")
+print(response)

test/llamaindex/env

@@ -0,0 +1,2 @@
+# MacOS/Linux
+export OPENAI_API_KEY=sk-e2a05bbcfac84e53b73f98acef15a009

test/llamaindex/test1.py

@@ -0,0 +1,7 @@
+from llama_index.core import VectorStoreIndex, SimpleDirectoryReader
+
+documents = SimpleDirectoryReader("data").load_data()
+index = VectorStoreIndex.from_documents(documents)
+query_engine = index.as_query_engine()
+response = query_engine.query("Some question about the data should go here")
+print(response)

test/test_pydantic/test.py

@@ -0,0 +1,34 @@
+# 正确示例：显式类型注解
+import os
+
+from pydantic import BaseModel
+
+class User(BaseModel):
+    id: int  # 显式指定int类型
+    name: str  # 显式指定str类型
+    is_active: bool = True  # 带默认值的类型注解
+
+# 错误示例：缺少类型注解
+# class BadUser(BaseModel):
+#     id = 1  # 缺少类型注解，Pydantic 2.9+将报错
+#     name = "John"  # 缺少类型注解，Pydantic 2.9+将报错
+
+os.environ['app_port'] = '8888'
+
+# 环境变量自动映射示例
+from pydantic_settings import BaseSettings, SettingsConfigDict
+
+
+class AppConfig(BaseSettings):
+    host: str = "localhost"
+    port: int = 8000
+
+    model_config = SettingsConfigDict(
+        env_prefix="APP_",  # 环境变量前缀
+        case_sensitive=False  # 不区分大小写
+    )
+
+
+# 当环境变量存在APP_PORT=8080时
+config = AppConfig()
+print(config.port)  # 输出: 8080 (而非默认的8000)