change the position of initialize the embedding model

app.py

@@ -7,9 +7,20 @@ from langchain_community.vectorstores.faiss import FAISS
 from langchain_huggingface import HuggingFaceEmbeddings
 from huggingface_hub import snapshot_download
 
+def create_embedding_model(model_name):
+    """Create embedding model instance"""
+    return HuggingFaceEmbeddings(
+        model_name=model_name,
+        model_kwargs={'trust_remote_code': True}
+    )
+
 class RAGChatbot:
     def __init__(self):
+        # First create embeddings directly
+        self.embeddings = create_embedding_model('intfloat/multilingual-e5-large-instruct')
+        # Then initialize other models
         self.init_models()
+        # Finally initialize vector store
         self.init_vector_store()
         
         self.background_prompt = '''
@@ -68,41 +79,10 @@ class RAGChatbot:
         Now, please guide me step by step to describe the legal issues I am facing, according to the above requirements.
         '''
 
-    import gradio as gr
-import os
-import spaces
-from transformers import AutoModelForCausalLM, AutoTokenizer, TextIteratorStreamer
-from threading import Thread
-from langchain_community.vectorstores.faiss import FAISS
-from langchain_huggingface import HuggingFaceEmbeddings
-from huggingface_hub import snapshot_download
-
-class RAGChatbot:
-    def __init__(self):
-        # First initialize models to create embeddings
-        self.init_models()
-        # Then initialize vector store which uses embeddings
-        self.init_vector_store()
-        
-        self.background_prompt = '''
-        As an AI legal assistant, you are a highly trained expert in U.S. and Canadian law...
-        [rest of your existing background prompt]
-        '''
-
-    @spaces.GPU(duration=120)
+    @spaces.GPU
     def init_models(self):
-        """Initialize the LLM and embedding models"""
-        print("Initializing models...")
-        
-        # Embedding model initialization first
-        print("Loading embedding model...")
-        self.embeddings = HuggingFaceEmbeddings(
-            model_name='intfloat/multilingual-e5-large-instruct',
-            model_kwargs={'trust_remote_code': True}
-        )
-        
-        # LLM initialization
-        print("Loading LLM model...")
+        """Initialize the LLM model"""
+        print("Initializing LLM model...")
         self.llm_model_name = 'StevenChen16/llama3-8b-Lawyer'
         self.tokenizer = AutoTokenizer.from_pretrained(self.llm_model_name)
         self.model = AutoModelForCausalLM.from_pretrained(
@@ -113,7 +93,7 @@ class RAGChatbot:
             self.tokenizer.eos_token_id,
             self.tokenizer.convert_tokens_to_ids("<|eot_id|>")
         ]
-        print("Models initialized successfully")
+        print("LLM model initialized successfully")
 
     def init_vector_store(self):
         """Load vector store from HuggingFace Hub"""
@@ -128,7 +108,7 @@ class RAGChatbot:
             print("Loading vector store...")
             # Load the vector store from downloaded files
             self.vector_store = FAISS.load_local(
-                folder_path=repo_path,  # Specify the parameter name explicitly
+                folder_path=repo_path,
                 embeddings=self.embeddings,
                 allow_dangerous_deserialization=True
             )