update

main.py

@@ -3,6 +3,7 @@ from fastapi import FastAPI
 from service.vector_store import vector_store
 from pydantic import BaseModel
 from fastapi.responses import StreamingResponse
+from service.api import api
 
 app = FastAPI()
 
@@ -11,16 +12,13 @@ def read_root():
     return {"Hello": "World!"}
 
 @app.get("/api/search")
-def vector_search(inputs: str):
+def search(inputs: str):
     return vector_store.search(inputs)
 
-
-
-async def fake_video_streamer():
-    for i in range(10):
-        yield "some fake video bytes"
-        time.sleep(0.5)
+@app.get("/api/suggestion")
+def get_suggestion(inputs: str):
+    return api.get_suggestion(inputs)
 
 @app.get("/api/answer")
-async def answer(inputs: str):
-    return StreamingResponse(fake_video_streamer())
+async def get_answer(inputs: str):
+    return StreamingResponse(api.get_answer(inputs))

service/api.py

@@ -0,0 +1,51 @@
+import time
+from typing import Any
+from .llm import search_bot
+from .vector_store import vector_store
+
+
+class API():
+
+    HF_SPACE="JR818/Jarvis-Backend"
+    PROJECT='filecoin'
+
+    def __init__(self) -> None:
+        pass
+
+    def get_suggestion(self,query):
+        if not query:
+            return []
+        docs = vector_store.marginal_search(query)
+        res = []
+        for doc in docs:
+            name = doc.metadata.get('title_llm','')
+            url = doc.metadata.get('source','')
+            if name == '' or url == '':
+                continue
+            res.append({"name":name,"url":url})
+        return res
+        
+    
+    def get_answer(self,query):
+        docs = vector_store.search(query)
+        source = []
+        context = []
+        for doc in docs:
+            if doc.metadata['score']<0.2:
+                continue
+            if 'title_llm' in doc.metadata:
+                source.append({
+                    "name":doc.metadata.get('title_llm',''),
+                    "url":doc.metadata.get('source',''),
+                    "description":doc.metadata.get('description_llm','')
+                })
+            context.append(doc.metadata.get('description',''))
+        yield source
+        context = "\n".join(context)
+        answer = search_bot(query,context,stream=True)
+        text = ""
+        for i in answer:
+            text += i
+            yield i
+
+api = API()

service/llm.py

@@ -0,0 +1,74 @@
+from langchain.chat_models import ChatOpenAI
+from langchain.schema import StrOutputParser
+from langchain.schema.runnable import RunnablePassthrough
+from langchain.prompts import ChatPromptTemplate
+
+class Bot():
+
+    def __init__(self,template,model='gpt-3.5-turbo-1106'):
+        self.template = template
+
+        self.llm = ChatOpenAI(
+            model=model,
+            streaming=True,
+            temperature=0,
+        )
+        self.prompt = ChatPromptTemplate.from_messages(
+            [
+                ("system", self.template),
+                ("human", "{question}"),
+            ]
+        )
+        self.runner = self.prompt | self.llm |StrOutputParser()
+
+    def __call__(self,question,context,stream=False):
+        prompt = self.prompt.format(question=question,context=context)
+        inputs = {"context":context,"question":question}
+        if stream:
+            res = self.runner.stream(inputs)
+        else:
+            res = self.runner.invoke(inputs)
+        return res
+    
+    def custom_call(self,**kwargs):
+        return self.runner.invoke(kwargs)
+
+
+class SearchBot(Bot):
+    TEMPLATE="""\
+        You are an expert tasked with summarzing search result and answering questions about FileCoin.
+
+        Generate a comprehensive and informative answer of 80 words or less for the \
+        given question based solely on the provided search results in the context. You must \
+        only use information from the provided search results. 
+
+        Anything between the following `context`  html blocks is the search result retrieved from a knowledge \
+        bank. 
+
+        <context>
+            {context} 
+        <context/>
+    """
+
+    def __init__(self):
+        super().__init__(self.TEMPLATE)
+
+class SummerizeBot(Bot):
+    TEMPLATE = """
+        You are an expert tasked with summarzing description for documents.
+        
+        Generate a comprehensive and informative description of 50 words or less based on the context.
+
+        Anything between the following `context` html blocks is the content of the documents. 
+        <context>
+            {context} 
+        <context/>
+        
+        Remember to return the description only.
+    """
+
+    def __init__(self):
+        super().__init__(template=self.TEMPLATE)
+
+search_bot = SearchBot()
+summerize_bot = SummerizeBot()