Update app.py

app.py

@@ -1,24 +1,21 @@
 import os
-import multiprocessing
-import concurrent.futures
-from langchain_community.document_loaders import TextLoader, DirectoryLoader
-from langchain.text_splitter import RecursiveCharacterTextSplitter
-from langchain_community.vectorstores import FAISS 
-from sentence_transformers import SentenceTransformer
-import faiss
+import torch.cuda
 import numpy as np
-from datetime import datetime
-import json
+import faiss
 import gradio as gr
 import re
-from threading import Thread
 from openai import OpenAI
+from langchain_community.document_loaders import TextLoader, DirectoryLoader
+from langchain.text_splitter import RecursiveCharacterTextSplitter
+from langchain_community.vectorstores import FAISS 
+from sentence_transformers import SentenceTransformer
 
 class MultiAgentRAG:
     def __init__(self, embedding_model_name, openai_model_id, data_folder, api_key=None):
+        self.use_gpu = torch.cuda.is_available()
         self.all_splits = self.load_documents(data_folder)
         self.embeddings = SentenceTransformer(embedding_model_name)
-        self.gpu_index = self.create_faiss_index()
+        self.faiss_index = self.create_faiss_index()
         self.openai_client = OpenAI(api_key=api_key or os.environ.get("OPENAI_API_KEY"))
         self.openai_model_id = openai_model_id
 
@@ -27,10 +24,6 @@ class MultiAgentRAG:
         documents = loader.load()
         text_splitter = RecursiveCharacterTextSplitter(chunk_size=5000, chunk_overlap=250)
         all_splits = text_splitter.split_documents(documents)
-        print('Length of documents:', len(documents))
-        print("LEN of all_splits", len(all_splits))
-        for i in range(min(3, len(all_splits))):
-            print(all_splits[i].page_content)
         return all_splits
 
     def create_faiss_index(self):
@@ -38,9 +31,12 @@ class MultiAgentRAG:
         embeddings = self.embeddings.encode(all_texts, convert_to_tensor=True).cpu().numpy()
         index = faiss.IndexFlatL2(embeddings.shape[1])
         index.add(embeddings)
-        gpu_resource = faiss.StandardGpuResources()
-        gpu_index = faiss.index_cpu_to_gpu(gpu_resource, 0, index)
-        return gpu_index
+        try:
+            gpu_resource = faiss.StandardGpuResources()
+            gpu_index = faiss.index_cpu_to_gpu(gpu_resource, 0, index)
+            return gpu_index
+        except:
+            return index
 
     def generate_openai_response(self, messages, max_tokens=1000):
         try:
@@ -54,88 +50,49 @@ class MultiAgentRAG:
                 presence_penalty=0
             )
             return response.choices[0].message.content
-        except Exception as e:
-            print(f"Error in generate_openai_response: {str(e)}")
+        except:
             return "Text generation process encountered an error"
 
     def retrieval_agent(self, query):
         query_embedding = self.embeddings.encode(query, convert_to_tensor=True).cpu().numpy()
-        distances, indices = self.gpu_index.search(np.array([query_embedding]), k=3)
+        distances, indices = self.faiss_index.search(np.array([query_embedding]), k=3)
         content = ""
-        for idx, distance in zip(indices[0], distances[0]):
-            content += "-" * 50 + "\n"
+        for idx in indices[0]:
             content += self.all_splits[idx].page_content + "\n"
         return content
 
     def grading_agent(self, query, retrieved_content):
         messages = [
-            {"role": "system", "content": "You are an expert at evaluating the relevance of retrieved content to a query."},
-            {"role": "user", "content": f"""
-            Evaluate the relevance of the following retrieved content to the given query:
-            
-            Query: {query}
-            
-            Retrieved Content:
-            {retrieved_content}
-            
-            Rate the relevance on a scale of 1-10 and explain your rating:
-            """}
+            {"role": "system", "content": "You are an expert at evaluating relevance."},
+            {"role": "user", "content": f"Query: {query}\nRetrieved Content:\n{retrieved_content}\nRate the relevance on a scale of 1-10."}
         ]
-        
         grading_response = self.generate_openai_response(messages)
-        
-        # Extract the numerical rating from the response
         match = re.search(r'\b([1-9]|10)\b', grading_response)
-        rating = int(match.group()) if match else 5  # Default to 5 if no rating found
+        rating = int(match.group()) if match else 5
         return rating, grading_response
 
     def query_rewrite_agent(self, original_query):
         messages = [
-            {"role": "system", "content": "You are an expert at rewriting queries to improve information retrieval results."},
-            {"role": "user", "content": f"""
-            The following query did not yield relevant results. Please rewrite it to potentially improve retrieval:
-            
-            Original Query: {original_query}
-            
-            Rewritten Query:
-            """}
+            {"role": "system", "content": "You are an expert at rewriting queries."},
+            {"role": "user", "content": f"Original Query: {original_query}\nRewritten Query:"}
         ]
-        
-        rewritten_query = self.generate_openai_response(messages)
-        return rewritten_query.strip()
+        return self.generate_openai_response(messages).strip()
 
     def generation_agent(self, query, retrieved_content):
         messages = [
-            {"role": "system", "content": "You are a knowledgeable assistant with access to a comprehensive database."},
-            {"role": "user", "content": f"""
-            I need you to answer my question and provide related information in a specific format.
-            I have provided five relatable json files {retrieved_content}, choose the most suitable chunks for answering the query.
-            RETURN ONLY SOLUTION without additional comments, sign-offs, retrived chunks, refrence to any Ticket or extra phrases. Be direct and to the point.
-            IF THERE IS NO ANSWER RELATABLE IN RETRIEVED CHUNKS, RETURN "NO SOLUTION AVAILABLE".
-            DO NOT GIVE REFRENCE TO ANY CHUNKS OR TICKETS, BE ON POINT.
-            
-            Here's my question:
-            Query: {query}
-            Solution==>
-            """}
+            {"role": "system", "content": "You are a knowledgeable assistant."},
+            {"role": "user", "content": f"Query: {query}\nSolution==>"}
         ]
-        
         return self.generate_openai_response(messages)
 
     def run_multi_agent_rag(self, query):
-        max_iterations = 3
-        for i in range(max_iterations):
+        for _ in range(3):
             retrieved_content = self.retrieval_agent(query)
-            
             relevance_score, grading_explanation = self.grading_agent(query, retrieved_content)
-            
-            if relevance_score >= 7:                 
-                answer = self.generation_agent(query, retrieved_content)
-                return answer, retrieved_content, grading_explanation
-            else:
-                query = self.query_rewrite_agent(query)
-        
-        return "Unable to find a relevant answer after multiple attempts.", "", "Low relevance across all attempts."
+            if relevance_score >= 7:
+                return self.generation_agent(query, retrieved_content), retrieved_content, grading_explanation
+            query = self.query_rewrite_agent(query)
+        return "Unable to find a relevant answer.", "", "Low relevance across all attempts."
 
     def qa_infer_gradio(self, query):
         answer, retrieved_content, grading_explanation = self.run_multi_agent_rag(query)
@@ -143,25 +100,14 @@ class MultiAgentRAG:
 
 def launch_interface(doc_retrieval_gen):
     css_code = """
-        .gradio-container {
-            background-color: #daccdb;
-        }
-        button {
-            background-color: #927fc7;
-            color: black;
-            border: 1px solid black;
-            padding: 10px;
-            margin-right: 10px;
-            font-size: 16px;
-            font-weight: bold;
-        }
+        .gradio-container { background-color: #daccdb; }
+        button { background-color: #927fc7; color: black; border: 1px solid black; padding: 10px; margin-right: 10px; font-size: 16px; font-weight: bold; }
     """
     EXAMPLES = [
         "On which devices can the VIP and CSI2 modules operate simultaneously?", 
         "I'm using Code Composer Studio 5.4.0.00091 and enabled FPv4SPD16 floating point support for CortexM4 in TDA2. However, after building the project, the .asm file shows --float_support=vfplib instead of FPv4SPD16. Why is this happening?", 
-        "Could you clarify the maximum number of cameras that can be connected simultaneously to the video input ports on the TDA2x SoC, considering it supports up to 10 multiplexed input ports and includes 3 dedicated video input modules?"
+        "Could you clarify the maximum number of cameras that can be connected simultaneously to the video input ports on the TDA2x SoC?"
     ]
-
     interface = gr.Interface(
         fn=doc_retrieval_gen.qa_infer_gradio,
         inputs=[gr.Textbox(label="QUERY", placeholder="Enter your query here")],
@@ -172,13 +118,14 @@ def launch_interface(doc_retrieval_gen):
         css=css_code,
         title="TI E2E FORUM Multi-Agent RAG"
     )
-
     interface.launch(debug=True)
 
 if __name__ == "__main__":
     embedding_model_name = 'flax-sentence-embeddings/all_datasets_v3_MiniLM-L12'
-    openai_model_id = "gpt-4-turbo" 
+    openai_model_id = "gpt-4-turbo"
     data_folder = 'sample_embedding_folder2'
-
-    multi_agent_rag = MultiAgentRAG(embedding_model_name, openai_model_id, data_folder)
-    launch_interface(multi_agent_rag)
+    try:
+        multi_agent_rag = MultiAgentRAG(embedding_model_name, openai_model_id, data_folder)
+        launch_interface(multi_agent_rag)
+    except Exception as e:
+        print(f"Error initializing Multi-Agent RAG: {str(e)}")