Create app.py

app.py

@@ -0,0 +1,86 @@
+from langchain_core.prompts import ChatPromptTemplate
+from langchain_core.output_parsers import StrOutputParser
+from langchain_core.runnables import RunnablePassthrough
+from langchain_huggingface.embeddings import HuggingFaceEmbeddings
+from langchain.retrievers.document_compressors import EmbeddingsFilter
+from langchain.retrievers import ContextualCompressionRetriever
+from langchain.retrievers import EnsembleRetriever
+from langchain_community.vectorstores import FAISS
+from langchain_groq import ChatGroq
+from langchain import hub
+import pickle
+import os
+
+GROQ_API_KEY="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"
+os.environ['GROQ_API_KEY'] = GROQ_API_KEY
+
+
+embed_model = HuggingFaceEmbeddings(model_name="Alibaba-NLP/gte-multilingual-base", model_kwargs={"trust_remote_code":True, "device": "cuda"})
+llm = ChatGroq(
+    model="llama-3.1-8b-instant",  
+    temperature=0.0,
+    max_tokens=1024, 
+    max_retries=2
+)
+
+excel_vectorstore = FAISS.load_local(folder_path="./faiss_excel_doc_index", embeddings=embed_model, allow_dangerous_deserialization=True)
+word_vectorstore = FAISS.load_local(folder_path="./faiss_word_doc_index", embeddings=embed_model, allow_dangerous_deserialization=True)
+excel_vectorstore.merge_from(word_vectorstore)
+combined_vectorstore = excel_vectorstore
+
+with open('combined_keyword_retriever.pkl', 'rb') as f:
+    combined_keyword_retriever = pickle.load(f)
+    combined_keyword_retriever.k = 10
+
+semantic_retriever = combined_vectorstore.as_retriever(search_type="mmr", search_kwargs={'k': 10, 'lambda_mult': 0.25})
+
+
+# initialize the ensemble retriever
+ensemble_retriever = EnsembleRetriever(
+    retrievers=[combined_keyword_retriever, semantic_retriever], weights=[0.5, 0.5]
+)
+
+
+embeddings_filter = EmbeddingsFilter(embeddings=embed_model, similarity_threshold=0.6)
+compression_retriever = ContextualCompressionRetriever(
+    base_compressor=embeddings_filter, base_retriever=ensemble_retriever
+)
+
+prompt = hub.pull("rlm/rag-prompt")
+
+
+def format_docs(docs):
+    return "\n\n".join(doc.page_content for doc in docs)
+
+
+rag_chain = (
+    {"context": compression_retriever | format_docs, "question": RunnablePassthrough()}
+    | prompt
+    | llm
+    | StrOutputParser()
+)
+
+
+
+import gradio as gr
+import spaces
+# import torch
+
+# zero = torch.Tensor([0]).cuda()
+
+@spaces.GPU
+def get_response(question, history):
+    print(question)
+
+    # for chunk in rag_chain.stream(question):
+    #     yield chunk
+    respose =  rag_chain.invoke(question)
+    print(respose)
+    
+    return respose
+
+with gr.Blocks() as demo:
+    chatbot = gr.Chatbot(placeholder="<strong>ADAFSA-RAG Chatbot</strong>")
+    gr.ChatInterface(fn=get_response, chatbot=chatbot)
+    
+demo.launch()