init

.gitignore

@@ -0,0 +1,4 @@
+/venv
+/__pycache__
+/.idea
+*/__pycache__

README.md

@@ -1,2 +1,151 @@
 # PassageSummary
-PassageSummary是一个基于Gpt工作的Api，提供文章总结，话题提取等若干总结功能
+PassageSummary是一个基于Gpt工作的Api，提供文章总结，话题提取等若干总结功能。
+
+# 约束
+
+ApiService提供基本的文本缓存服务，支持对于完全相同的文本的缓存。此外，在ApiService返回异常的时候，会返回`errno`和`message`两个参数，其分别表示全局唯一错误码和错误信息。 
+
+ApiService的返回内容均为Json格式，其中token参数表示OpenAI的ApiKey，使用`gpt-3.5-turbo`模型。 
+
+# Api
+
+得到ApiService的工作状态：
+
+```
+GET /
+{
+ "message":"This server is working normally."
+}
+```
+
+上传文本：
+
+```
+POST /passage
+请求：
+{
+ "content":"This is a content",
+ "token":"sk-ss"
+}
+
+返回：
+正常返回：
+{
+ "hash":"d622ac64268ce69eef0f3dc8277d06a9182f71c7"
+}
+错误返回：
+{
+ "errno":1,
+ "message":"文本转换异常"
+}
+```
+
+询问文本：
+
+```
+POST /passage/{hash}
+请求：
+{
+ "action":"ask",
+ "param":"这篇文章主要描述了什么？",
+ "token":"sk-xxx"
+}
+
+返回：
+正常返回：
+{
+ "content":"这篇文章主要讲述了...."
+}
+错误返回：
+{
+ "errno":2,
+ "message":"文章内容不存在..."
+}
+```
+
+得到文章话题：
+
+```
+POST /passage/{hash}
+请求:
+{
+ "action":"topic",
+ "token":"sk-xxx"
+}
+
+返回：
+正常返回：
+{
+ "topics":
+ [
+ {
+ "topic":"原神怎么你了？",
+ "relative":"0.2"
+ }
+ ]
+}
+//topic表示的是话题，relative表示话题相关度
+错误返回：
+{
+ "errno":2,
+ "message":"文章内容不存在..."
+}
+```
+
+判断文章与话题的相关度：
+
+```
+POST /passage/{hash}
+请求:
+{
+ "action":"getTopicRelative",
+ "param":"原神,原批",
+ "token":"sk-xxx"
+}
+
+返回：
+正常返回：
+{
+ "topics":
+ [
+ {
+ "topic":"原神",
+ "relative":"0.2"
+ },
+ {
+ "topic":"原批",
+ "relative":"0.9"
+ }
+ ]
+}
+//topic表示的是话题，relative表示话题相关度
+错误返回：
+{
+ "errno":2,
+ "message":"文章内容不存在..."
+}
+```
+
+总结文章：
+
+```
+POST /passage/{hash}
+请求:
+{
+ "action":"summary",
+ "token":"sk-xxx"
+}
+
+返回：
+正常返回：
+{
+ "content":"这篇文章讲述了一个原批转换为星批的故事。"
+}
+
+错误返回：
+{
+ "errno":2,
+ "message":"文章内容不存在..."
+}
+```
+

main.py

@@ -0,0 +1,31 @@
+import os
+from models import request
+from passages import savePassage
+from passages import passageAnalysis
+from fastapi import FastAPI
+
+# 初始化，创建缓存目录：
+if not os.path.exists('cache'):
+ os.mkdir('cache')
+
+app = FastAPI()
+
+
+@app.get("/")
+async def root():
+ return {"message": "This server is working normally."}
+
+
+@app.post("/passage")
+async def say_hello(req: request.SavePassageRequest):
+ return savePassage.save_passage(req.content, req.token)
+
+
+@app.post("/passage/{hash}")
+async def action(hash: str, req: request.PassageRequest):
+ if not os.path.exists(os.path.join('cache', hash)):
+ return {
+ "errno": 10001,
+ "message": "hash对应的文件不存在，或者是文件读取异常"
+ }
+ return passageAnalysis.dispatch_action(req, hash)

models/request.py

@@ -0,0 +1,12 @@
+from pydantic import BaseModel
+
+
+class SavePassageRequest(BaseModel):
+ content: str
+ token: str
+
+
+class PassageRequest(BaseModel):
+ action: str
+ param: object | None
+ token: str

passages/passageAnalysis.py

@@ -0,0 +1,124 @@
+import os
+import json
+from langchain import OpenAI
+from models import request
+from llama_index import (
+ GPTSimpleVectorIndex,
+ PromptHelper,
+ LLMPredictor,
+ QuestionAnswerPrompt,
+ ServiceContext
+)
+
+
+def dispatch_action(req: request.PassageRequest, hash: str):
+ path = os.path.join('cache', hash)
+ vector = os.path.join(path, 'index.json')
+ match req.action:
+ case "ask":
+ return ask(vector, str(req.param), req.token)
+ case "topic":
+ return get_topics(vector, req.token)
+ case "getTopicRelative":
+ return get_topic_relative(vector, str(req.param), req.token)
+ case "summary":
+ return summary(vector, req.token)
+
+
+def ask(vector: str, ask_question: str, token: str):
+ response = common_ask(vector, ask_question, token)
+ if response.response is None:
+ return {
+ "errno": 10002,
+ "message": "Gpt未返回信息，请检查Token是否有效！"
+ }
+ return {
+ "content": response.response
+ }
+
+
+def summary(vector: str, token: str):
+ response = common_ask(vector, "Summary this passage in Chinese", token)
+ if response.response is None:
+ return {
+ "errno": 10002,
+ "message": "Gpt未返回信息，请检查Token是否有效！"
+ }
+ return {
+ "content": response.response
+ }
+
+
+def get_topics(vector: str, token: str):
+ response = common_ask(vector, "Analysis this passage, getting the topic or key word of it. "
+ "Returning {{xxx#relative}}. 'xxx' is the the topic or key word of the passage and "
+ "relative is a num between 0 and 1 presenting the closeness of "
+ "the topic or key word and the text. "
+ "For example, returning '{{Minecraft#0.2}},{{Game#0.8}}'", token)
+ return get_topic_with_relative(response)
+
+
+def get_topic_relative(vector: str, key_word: str, token: str):
+ response = common_ask(vector, "Analysis this passage, getting the relative of the topic or key word "
+ "with the passage. "
+ "Returning {{xxx#relative}}. 'xxx' is the the topic or key word given and "
+ "relative is a num between 0 and 1 presenting the closeness of "
+ "the topic or key word and the text. "
+ "For example, giving 'TopicA,TopicB' returning "
+ "'{{TopicA#0.2}},{{TopicB#0.8}}'. Now the giving keyword is " + key_word, token)
+ return get_topic_with_relative(response)
+
+
+def get_topic_with_relative(response):
+ if response.response is None:
+ return {
+ "errno": 10002,
+ "message": "Gpt未返回信息，请检查Token是否有效！"
+ }
+ topic = []
+ topics = response.response.split(",")
+ for i in topics:
+ cts = i.split("#")
+ if len(cts) != 2:
+ return {
+ "errno": 10003,
+ "message": "Gpt返回无效信息，请尝试重新请求或舍弃请求."
+ }
+ temp = {
+ "topic": str(cts[0]).replace("{{", "").replace("\n", "").replace("{", ""),
+ "relative": str(cts[1]).replace("}}", "").replace("}", "")
+ }
+ topic.append(temp)
+ return {
+ "content": topic
+ }
+
+
+def common_ask(vector: str, ask_question: str, token: str, prompt: str = "Please answer the question with the context "
+ "information"):
+ llm_predictor, prompt_helper = prepare_llama_para(token)
+
+ service_context = ServiceContext.from_defaults(llm_predictor=llm_predictor, prompt_helper=prompt_helper)
+
+ qa_prompt_impl = (
+ "We have provided context information below. \n"
+ "---------------------\n"
+ "{context_str}"
+ "\n---------------------\n"
+ f"{prompt}: {{query_str}}\n"
+ )
+ qa_prompt = QuestionAnswerPrompt(qa_prompt_impl)
+ index = GPTSimpleVectorIndex.load_from_disk(vector, service_context=service_context)
+ response = index.query(ask_question, response_mode="compact", text_qa_template=qa_prompt)
+ return response
+
+
+def prepare_llama_para(token):
+ os.environ["OPENAI_API_KEY"] = token
+ max_input_size = 4096
+ num_outputs = 1024
+ max_chunk_overlap = 20
+ chunk_size_limit = 1000
+ llm_predictor = LLMPredictor(llm=OpenAI(temperature=0, model_name="text-davinci-003", max_tokens=num_outputs))
+ prompt_helper = PromptHelper(max_input_size, num_outputs, max_chunk_overlap, chunk_size_limit=chunk_size_limit)
+ return llm_predictor, prompt_helper

passages/savePassage.py

@@ -0,0 +1,40 @@
+import os
+from passages import passageAnalysis
+from llama_index import (
+ GPTSimpleVectorIndex,
+ SimpleDirectoryReader,
+ LLMPredictor,
+ ServiceContext,
+ PromptHelper
+)
+
+
+def save_passage(content: str, token: str):
+ name = hash(str)
+ res = {
+ "hash": name
+ }
+
+ # permanently cache
+ dir_path = os.path.join('cache', str(name))
+ if not os.path.exists(dir_path):
+ os.mkdir(dir_path)
+ else:
+ return res
+ file_name = os.path.join(dir_path, 'raw')
+ index_name = os.path.join(dir_path, 'index.json')
+ with open(file_name, "w") as file:
+ file.write(content)
+ llm_predictor, prompt_helper = passageAnalysis.prepare_llama_para(token)
+ documents = SimpleDirectoryReader(dir_path).load_data()
+ service_context = ServiceContext.from_defaults(llm_predictor=llm_predictor, prompt_helper=prompt_helper)
+ index = GPTSimpleVectorIndex.from_documents(
+ documents, service_context=service_context
+ )
+ index.save_to_disk(index_name)
+ return res
+
+
+def get_passage_content(hash: str):
+ with open(hash, "r") as file:
+ return file.read()