Update app.py

app.py

@@ -14,62 +14,85 @@ from langchain.chains import RetrievalQA
 from langchain_community.llms import HuggingFaceHub
 
 # define constants
-EMB_MODEL1 = 'BAAI/bge-base-en-v1.5'
+# Embedding models
+EMB_MODEL_bge = 'BAAI/bge-base-en-v1.5'
+EMB_MODEL_gtr_t5 = 'sentence-transformers/gtr-t5-base'
+EMB_MODEL_e5 = 'intfloat/e5-large-v2'
+# Chat app model
 MISTRAL_MODEL1 = 'mistralai/Mixtral-8x7B-Instruct-v0.1'
 HF_MODEL1 = 'HuggingFaceH4/zephyr-7b-beta'
 # define paths
 vector_path = 'faiss_index'
+vector_path_2 = 'faiss_index_2'
+vector_path_e5 = 'faiss_index_3'
 hf_token = os.environ["HUGGINGFACEHUB_API_TOKEN"]
 
-def respond(message, history #,
-    #system_message,
-    #max_tokens,
-    #temperature,
-    #top_p
-    ):
+def respond(message, history):
     
     # Initialize your embedding model
-    embedding_model = HuggingFaceEmbeddings(model_name=EMB_MODEL1)
+    embedding_model_bge = HuggingFaceEmbeddings(model_name=EMB_MODEL_bge)
+    embedding_model_gtr_t5 = HuggingFaceEmbeddings(model_name=EMB_MODEL_gtr_t5)
+    embedding_model_e5 = HuggingFaceEmbeddings(model_name=EMB_MODEL_e5)
     
     # Load FAISS from relative path
-    if os.path.exists("faiss_index"):
-        vectordb = FAISS.load_local(vector_path, embedding_model, allow_dangerous_deserialization=True)
-    else:
-        raise FileNotFoundError("FAISS index not found in Space. Please upload it to faiss_index/")
+    vectordb_bge = FAISS.load_local(vector_path_bge, embedding_model_bge, allow_dangerous_deserialization=True)
+    vectordb_gtr_t5 = FAISS.load_local(vector_path_gtr_t5, embedding_model_gtr_t5, allow_dangerous_deserialization=True)
+    vectordb_e5 = FAISS.load_local(vector_path_e5, embedding_model_e5, allow_dangerous_deserialization=True)
     
     # define retriever object
-    retriever = vectordb.as_retriever(search_type="similarity", search_kwargs={"k": 5})
+    retriever_bge = vectordb_bge.as_retriever(search_type="similarity", search_kwargs={"k": 5})
+    retriever_gtr_t5 = vectordb_gtr_t5.as_retriever(search_type="similarity", search_kwargs={"k": 5})
+    retriever_e5 = vectordb_e5.as_retriever(search_type="similarity", search_kwargs={"k": 5})
 
     # initialse chatbot llm
     llm = HuggingFaceHub(
         repo_id=MISTRAL_MODEL1, 
         huggingfacehub_api_token=hf_token,
-        model_kwargs={"temperature": 0.5, "max_new_tokens": 512}
+        model_kwargs={"temperature": 0.7, "max_new_tokens": 512}
         )
     
     # create a RAG pipeline
-    qa_chain = RetrievalQA.from_chain_type(llm=llm, retriever=retriever)
+    qa_chain_bge = RetrievalQA.from_chain_type(llm=llm, retriever=retriever_bge)
+    qa_chain_gtr_t5 = RetrievalQA.from_chain_type(llm=llm, retriever=retriever_gtr_t5)
+    qa_chain_e5 = RetrievalQA.from_chain_type(llm=llm, retriever=retriever_e5)
+    
     #generate results
-    result = qa_chain.invoke(message)
-    responce = result['result']
+    responce_bge = qa_chain_bge.invoke(message)['result']
+    responce_gtr_t5 = qa_chain_gtr_t5.invoke(message)['result']
+    responce_e5 = qa_chain_e5.invoke(message)['result']
 
     # remove the top instructions
     instruction_prefix = (
     "Use the following pieces of context to answer the question at the end. If you don't know the answer, just say that you don't know, don't try to make up an answer."
     )
-    if responce.strip().startswith(instruction_prefix):
-        responce = responce.strip()[len(instruction_prefix):].strip()
-
+    if responce_bge.strip().startswith(instruction_prefix):
+        responce_bge = responce_bge.strip()[len(instruction_prefix):].strip()
+    if responce_gtr_t5.strip().startswith(instruction_prefix):
+        responce_gtr_t5 = responce_gtr_t5.strip()[len(instruction_prefix):].strip()
+    if responce_e5.strip().startswith(instruction_prefix):
+        responce_e5 = responce_e5.strip()[len(instruction_prefix):].strip()
+        
     # Split question, Helpful Answer and Reason
-    match = re.search(
-        r"^(.*?)(?:\n+)?Question:\s*(.*?)(?:\n+)?Helpful Answer:\s*(.*)", responce, re.DOTALL
-    )
-    if match:
-        original_text = match.group(1).strip()
-        question = match.group(2).strip()
-        answer = match.group(3).strip()
+    match_bge = re.search(r"^(.*?)(?:\n+)?Question:\s*(.*?)(?:\n+)?Helpful Answer:\s*(.*)", responce_bge, re.DOTALL)
+    match_gtr_t5 = re.search(r"^(.*?)(?:\n+)?Question:\s*(.*?)(?:\n+)?Helpful Answer:\s*(.*)", responce_gtr_t5, re.DOTALL)
+    match_e5 = re.search(r"^(.*?)(?:\n+)?Question:\s*(.*?)(?:\n+)?Helpful Answer:\s*(.*)", responce_e5, re.DOTALL)
+    
+    if match_bge:
+        #original_text_bge = match_bge.group(1).strip()
+        question_bge = match_bge.group(2).strip()
+        answer_bge = match_bge.group(3).strip()
+        
+    if match_gtr_t5:
+        #original_text_gtr_t5 = match_gtr_t5.group(1).strip()
+        #question_gtr_t5 = match_gtr_t5.group(2).strip()
+        answer_gtr_t5 = match_gtr_t5.group(3).strip()
+        
+    if match_e5:
+        #original_text_e5 = match_e5.group(1).strip()
+        #question_e5 = match_e5.group(2).strip()
+        answer_e5 = match_e5.group(3).strip()
 
-    formatted_responce = f'Question:{question}\nHelpful Answer:\n{answer}\nReasoning:\n{original_text}'
+    formatted_responce = f'Question:{question_bge}\nHelpful Answer Type 1:\n{answer_bge}\nHelpful Answer Type 2:\n{answer_gtr_t5}\nHelpful Answer Type 3:\n{answer_e5}'
     yield formatted_responce