Skip to content

Commit

Permalink
update question_answering.ipynb
Browse files Browse the repository at this point in the history
  • Loading branch information
@Dobiichi-Origami
Dobiichi-Origami committed Nov 2, 2023
1 parent 25d7757 commit db31ddb
Show file tree
Hide file tree
Showing 3 changed files with 62 additions and 584 deletions.
Binary file added cookbook/question_anwsering/example_data/中国古代史-明朝.pdf
View file
Binary file not shown.
152 changes: 62 additions & 90 deletions cookbook/question_anwsering/question_answering.ipynb
View file
Original file line number Diff line number Diff line change
Expand Up @@ -43,26 +43,27 @@
"outputs": [],
"source": [
"!pip install langchain\n",
"!pip install chromadb\n",
"!pip install qianfan"
]
},
{
"cell_type": "code",
"execution_count": 1,
"execution_count": null,
"id": "981930ef-a0b4-46f9-b60b-a495117ea38e",
"metadata": {},
"outputs": [],
"source": [
"import os\n",
"\n",
"os.environ['QIANFAN_AK'] = \"your_api_key\"\n",
"os.environ['QIANFAN_SK'] = \"your_secret_key\"\n",
"os.environ['QIANFAN_AK'] = \"your_ak\"\n",
"os.environ['QIANFAN_SK'] = \"your_sk\"\n",
"\n",
"# 此处为 Langsmith 相关功能开关。当且仅当你知道这是做什么用时，可删除注释并设置变量以使用 Langsmith 相关功能\n",
"# os.environ['LANGCHAIN_TRACING_V2'] = \"true\"\n",
"# os.environ['LANGCHAIN_ENDPOINT'] = \"https://api.smith.langchain.com\"\n",
"# os.environ['LANGCHAIN_API_KEY'] = \"your_langchian_api_key\"\n",
"# os.environ['LANGCHAIN_PROJECT'] = \"your_project_name\"\n",
"# os.environ['LANGCHAIN_API_KEY'] = \"LANGCHAIN_API_KEY\"\n",
"# os.environ['LANGCHAIN_PROJECT'] = \"LANGCHAIN_PROJECT\"\n",
"\n",
"\n",
"is_chinese = True\n",
Expand Down Expand Up @@ -107,12 +108,14 @@
"source": [
"## Step 1. Load\n",
"\n",
"指定一个 `DocumentLoader` 来把你指定的非结构化数据加载成 `Documents`。一个 `Document` 是文字（即 `page_content`）和与之相关的元数据的结合体"
"指定一个 `DocumentLoader` 来把你指定的非结构化数据加载成 `Documents`。一个 `Document` 是文字（即 `page_content`）和与之相关的元数据的结合体\n",
"\n",
"此处我们使用 `WebBaseLoader` ，从网页中加载一个 `Documents`。"
]
},
{
"cell_type": "code",
"execution_count": 2,
"execution_count": null,
"id": "cf4d5c72",
"metadata": {},
"outputs": [],
Expand All @@ -123,6 +126,31 @@
"data = loader.load()"
]
},
{
"cell_type": "markdown",
"id": "f3fa0645",
"metadata": {},
"source": [
"Langchain 内提供了非常多样的 Loader ，辅助用户从不同来源读取数据。这些 Loader 都声明在 `langchain.document_loaders` 包中。\n",
"\n",
"对于我们的中文示例，我们还提供了一种从 PDF 读取 `Document` 的演示样例："
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "9c079033",
"metadata": {},
"outputs": [],
"source": [
"! pip install pdfplumber\n",
"from langchain.document_loaders import PDFPlumberLoader\n",
"\n",
"if is_chinese:\n",
" loader = PDFPlumberLoader(\"example_data/中国古代史-明朝.pdf\")\n",
" data = loader.load()"
]
},
{
"cell_type": "markdown",
"id": "fd2cc9a7",
Expand All @@ -135,7 +163,7 @@
},
{
"cell_type": "code",
"execution_count": 3,
"execution_count": null,
"id": "4b11c01d",
"metadata": {},
"outputs": [],
Expand All @@ -158,21 +186,10 @@
},
{
"cell_type": "code",
"execution_count": 4,
"execution_count": null,
"id": "e9c302c8",
"metadata": {},
"outputs": [
{
"name": "stderr",
"output_type": "stream",
"text": [
"[INFO] [09-15 20:50:50] logging.py:55 [t:8485264192]: trying to refresh access_token\n",
"[INFO] [09-15 20:50:50] logging.py:55 [t:8485264192]: sucessfully refresh access_token\n",
"[INFO] [09-15 20:50:50] logging.py:55 [t:8485264192]: requesting llm api endpoint: /embeddings/embedding-v1\n",
"[INFO] [09-15 20:50:51] logging.py:55 [t:8485264192]: requesting llm api endpoint: /embeddings/embedding-v1\n"
]
}
],
"outputs": [],
"source": [
"from langchain.embeddings import QianfanEmbeddingsEndpoint\n",
"from langchain.vectorstores import Chroma\n",
Expand Down Expand Up @@ -224,11 +241,14 @@
"source": [
"from langchain.chains import RetrievalQA\n",
"from langchain.chat_models import QianfanChatEndpoint\n",
"from langchain.prompts import PromptTemplate\n",
"\n",
"QA_CHAIN_PROMPT = PromptTemplate.from_template(CUSTOM_PROMPT_TEMPLATE)\n",
"\n",
"llm = QianfanChatEndpoint()\n",
"retriever=vectorstore.as_retriever(search_type=\"similarity_score_threshold\", search_kwargs={'score_threshold': 0.0})\n",
" \n",
"qa_chain = RetrievalQA.from_chain_type(llm, retriever=retriever)\n",
"qa_chain = RetrievalQA.from_chain_type(llm, retriever=retriever, chain_type_kwargs={\"prompt\": QA_CHAIN_PROMPT})\n",
"qa_chain({\"query\": QUESTION1})"
]
},
Expand All @@ -237,73 +257,48 @@
"id": "f7d52c84",
"metadata": {},
"source": [
"注意，此处不光可以传入 `ChatModel` ，也可以传入一个 `LLM` 对象到 `RetrievalQA` 中"
]
},
{
"cell_type": "markdown",
"id": "fa82f437",
"metadata": {},
"source": [
"#### 自定义 prompt\n",
"在上面的执行中，我们使用了千帆平台上提供的大模型调用，成功就一个问进行了问答。\n",
"\n",
"此外，`RetrievalQA` 链中使用的 `prompt` 参数也是可以定制的。由于 Langchain `RetrievalQA` 链中默认提供的 prompt 是用英语编写的，所以此处我们替换为了我们手动实现的中文 prompt，针对中文语境进行优化。\n",
"\n",
"`RetrievalQA` 链中使用的 `prompt` 参数也是可以定制的。由于 `RetrievalQA` 链中默认提供的 prompt 是用英语编写的，此处我们可以替换为我们手动实现的中文 prompt"
"#### 使用不同的大模型\n",
"\n",
"\n",
"除了使用默认的模型，即 `ERNIE-Bot-turbo` 以外，用户还可以设置上面 `QianfanChatEndpoint` 的 `model` 参数，来指定使用不同的大模型。例如我们想使用 ERNIE-Bot-4 模型时，就可以这么设置："
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "e4fee704",
"id": "05f194c9",
"metadata": {},
"outputs": [],
"source": [
"from langchain.prompts import PromptTemplate\n",
"\n",
"QA_CHAIN_PROMPT = PromptTemplate.from_template(CUSTOM_PROMPT_TEMPLATE)\n",
"llm = QianfanChatEndpoint(model=\"ERNIE-Bot-4\")\n",
"\n",
"qa_chain = RetrievalQA.from_chain_type(llm, retriever=retriever, chain_type_kwargs={\"prompt\": QA_CHAIN_PROMPT})\n",
"result = qa_chain({\"query\": QUESTION1})\n",
"result[\"result\"]"
"qa_chain({\"query\": QUESTION1})"
]
},
{
"cell_type": "markdown",
"id": "c825e9bf-6a56-46e4-8bbb-05441f76cb96",
"id": "214cfd40",
"metadata": {},
"source": [
"我们还可以选择把数据存储到千帆 Smith 的 Prompt Hub 上。需要搭配使用使用千帆 Smith SDK。\n",
"\n",
"一个例子如下所示"
"或者如果你已经在千帆平台上购买了资源并部署了自己的大模型服务，千帆 Langchain 组件还提供了 `endpoint` 参数，让你能够在 Langchian 中调用自己微调的大模型。有条件的用户可以取消注释并修改下列代码进行体验。"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "a896060f-ebc4-4236-a4ad-32960601c6e8",
"id": "b8456042",
"metadata": {},
"outputs": [],
"source": [
"# !pip install qianfansmithhub"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "aef8e734-ba54-48ae-b959-1898618f2d90",
"metadata": {},
"outputs": [],
"source": [
"# RAG prompt\n",
"# from qianfansmithhub import hub\n",
"# QA_CHAIN_PROMPT_HUB = hub.pull(\"rlm/rag-prompt\")\n",
"\n",
"# qa_chain = RetrievalQA.from_chain_type(\n",
"# llm,\n",
"# retriever=vectorstore.as_retriever(),\n",
"# chain_type_kwargs={\"prompt\": QA_CHAIN_PROMPT_HUB}\n",
"# )\n",
"# result = qa_chain({\"query\": question})\n",
"# result[\"result\"]"
"# llm = QianfanChatEndpoint(endpoint=\"your_service_endpoint\")\n",
"\n",
"# qa_chain = RetrievalQA.from_chain_type(llm, retriever=retriever, chain_type_kwargs={\"prompt\": QA_CHAIN_PROMPT})\n",
"# qa_chain({\"query\": QUESTION1})"
]
},
{
Expand All @@ -325,35 +320,12 @@
"source": [
"from langchain.chains import RetrievalQA\n",
"\n",
"qa_chain = RetrievalQA.from_chain_type(llm, retriever=retriever, return_source_documents=True)\n",
"qa_chain = RetrievalQA.from_chain_type(llm, retriever=retriever, chain_type_kwargs={\"prompt\": QA_CHAIN_PROMPT}, return_source_documents=True)\n",
"result = qa_chain({\"query\": QUESTION1})\n",
"len(result['source_documents'])\n",
"result['source_documents']"
]
},
{
"cell_type": "markdown",
"id": "1b600236",
"metadata": {},
"source": [
"#### 返回引用\n",
"\n",
"或者使用 `RetrievalQAWithSourcesChain` 来构造 Chain，以支持在返回的结果内包含结果的引用源。"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "948f6d19",
"metadata": {},
"outputs": [],
"source": [
"from langchain.chains import RetrievalQAWithSourcesChain\n",
"\n",
"qa_chain = RetrievalQAWithSourcesChain.from_chain_type(llm,retriever=retriever)\n",
"qa_chain({\"question\": QUESTION1})"
]
},
{
"cell_type": "markdown",
"id": "4380e478-e8ae-404b-9577-6b15475a6562",
Expand All @@ -375,7 +347,7 @@
"from langchain.chains import ConversationalRetrievalChain\n",
"\n",
"memory = ConversationSummaryMemory(llm=llm,memory_key=\"chat_history\",return_messages=True)\n",
"qa = ConversationalRetrievalChain.from_llm(llm, retriever=retriever, memory=memory)\n",
"qa = ConversationalRetrievalChain.from_llm(llm, retriever=retriever, memory=memory, combine_docs_chain_kwargs={\"prompt\": QA_CHAIN_PROMPT})\n",
"qa(QUESTION1)"
]
},
Expand Down Expand Up @@ -406,7 +378,7 @@
"name": "python",
"nbconvert_exporter": "python",
"pygments_lexer": "ipython3",
"version": "3.10.12"
"version": "3.10.13"
}
},
"nbformat": 4,
Expand Down
Loading

0 comments on commit db31ddb

Please sign in to comment.