push embedding-model

app.py

@@ -11,6 +11,8 @@ from langchain.prompts import PromptTemplate
 from langchain.chains import ConversationalRetrievalChain
 from langchain_community.vectorstores import Qdrant
 from langchain.memory import ConversationBufferMemory
+from langchain_huggingface import HuggingFaceEmbeddings
+from langchain.retrievers.multi_query import MultiQueryRetriever
 
 import chainlit as cl
 
@@ -27,13 +29,13 @@ for pdf_path in pdf_paths:
     documents.extend(loader.load())
 
 text_splitter = RecursiveCharacterTextSplitter(
-    chunk_size=1000,
-    chunk_overlap=200,
+    chunk_size=2000,
+    chunk_overlap=100,
 )
 
 docs = text_splitter.split_documents(documents)
 
-embedding = OpenAIEmbeddings(model="text-embedding-3-small")
+embedding = HuggingFaceEmbeddings(model_name="XicoC/midterm-finetuned-arctic")
 
 vectorstore = Qdrant.from_documents(
     documents=docs,
@@ -78,6 +80,11 @@ llm = ChatOpenAI(
     streaming=True,
 )
 
+retriever_llm = ChatOpenAI(model="gpt-4o-mini", temperature=0)
+multiquery_retriever = MultiQueryRetriever.from_llm(
+    retriever=retriever, llm=retriever_llm
+)
+
 
 @cl.on_chat_start
 async def start_chat():
@@ -87,7 +94,7 @@ async def start_chat():
 
     qa = ConversationalRetrievalChain.from_llm(
         llm,
-        retriever=retriever,
+        retriever=multiquery_retriever,
         memory=memory,
         combine_docs_chain_kwargs={"prompt": PROMPT},
         return_source_documents=True,

requirements.txt

@@ -8,4 +8,5 @@ langchain_openai==0.1.1
 pypdf
 chromadb
 qdrant-client
-importlib-metadata<7.0,>=6.0
+importlib-metadata<7.0,>=6.0
+sentence-transformers