Update app.py

app.py

@@ -1,9 +1,13 @@
 import os
 import multiprocessing
 import concurrent.futures
-from langchain_community.document_loaders import TextLoader, DirectoryLoader
+from langchain.document_loaders import TextLoader, DirectoryLoader
 from langchain.text_splitter import RecursiveCharacterTextSplitter
-from langchain_community.embeddings import HuggingFaceEmbeddings
+from langchain.vectorstores import FAISS
+from sentence_transformers import SentenceTransformer
+import faiss
+import torch
+import numpy as np
 from transformers import AutoModelForCausalLM, AutoTokenizer, TextIteratorStreamer, BitsAndBytesConfig
 from datetime import datetime
 import json
@@ -11,30 +15,11 @@ import gradio as gr
 import re
 from threading import Thread
 from transformers.agents import Tool, HfEngine, ReactJsonAgent
-from huggingface_hub import InferenceClient
-import logging
-import torch
-import numpy as np
-import faiss
-import warnings
-
-# Suppress specific warnings
-warnings.filterwarnings("ignore", category=FutureWarning)
-
-logging.basicConfig(level=logging.INFO)
-logger = logging.getLogger(__name__)
-
-try:
-    from langchain_community.vectorstores import FAISS
-except ImportError:
-    logger.error("Failed to import FAISS. Make sure it's installed correctly.")
-    logger.info("You can try: pip install faiss-cpu --no-cache")
-    FAISS = None
 
 class DocumentRetrievalAndGeneration:
     def __init__(self, embedding_model_name, lm_model_id, data_folder):
         self.all_splits = self.load_documents(data_folder)
-        self.embeddings = HuggingFaceEmbeddings(model_name=embedding_model_name)
+        self.embeddings = SentenceTransformer(embedding_model_name)
         self.gpu_index = self.create_faiss_index()
         self.tokenizer, self.model = self.initialize_llm(lm_model_id)
         self.retriever_tool = self.create_retriever_tool()
@@ -45,15 +30,17 @@ class DocumentRetrievalAndGeneration:
         documents = loader.load()
         text_splitter = RecursiveCharacterTextSplitter(chunk_size=5000, chunk_overlap=250)
         all_splits = text_splitter.split_documents(documents)
-        logger.info(f'Loaded {len(documents)} documents')
-        logger.info(f"Split into {len(all_splits)} chunks")
+        print('Length of documents:', len(documents))
+        print("LEN of all_splits", len(all_splits))
+        for i in range(3):
+            print(all_splits[i].page_content)
         return all_splits
 
     def create_faiss_index(self):
         all_texts = [split.page_content for split in self.all_splits]
-        embeddings = self.embeddings.embed_documents(all_texts)
-        index = faiss.IndexFlatL2(len(embeddings[0]))
-        index.add(np.array(embeddings))
+        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
@@ -74,6 +61,33 @@ class DocumentRetrievalAndGeneration:
         )
         return tokenizer, model
 
+    def generate_response_with_timeout(self, input_ids, max_new_tokens=1000):
+        try:
+            streamer = TextIteratorStreamer(self.tokenizer, timeout=60.0, skip_prompt=True, skip_special_tokens=True)
+            generate_kwargs = dict(
+                input_ids=input_ids,
+                max_new_tokens=max_new_tokens,
+                do_sample=True,
+                top_p=1.0,
+                top_k=20,
+                temperature=0.8,
+                repetition_penalty=1.2,
+                eos_token_id=[128001, 128008, 128009],
+                streamer=streamer,
+            )
+            
+            thread = Thread(target=self.model.generate, kwargs=generate_kwargs)
+            thread.start()
+            
+            generated_text = ""
+            for new_text in streamer:
+                generated_text += new_text
+            
+            return generated_text
+        except Exception as e:
+            print(f"Error in generate_response_with_timeout: {str(e)}")
+            return "Text generation process encountered an error"
+
     def create_retriever_tool(self):
         class RetrieverTool(Tool):
             name = "retriever"
@@ -92,15 +106,13 @@ class DocumentRetrievalAndGeneration:
 
             def forward(self, query: str) -> str:
                 similarityThreshold = 1
-                query_embedding = self.parent.embeddings.embed_query(query)
+                query_embedding = self.parent.embeddings.encode(query, convert_to_tensor=True).cpu().numpy()
                 distances, indices = self.parent.gpu_index.search(np.array([query_embedding]), k=3)
                 content = ""
-                filtered_results = []
                 for idx, distance in zip(indices[0], distances[0]):
                     if distance <= similarityThreshold:
-                        filtered_results.append(idx)
-                    content += "-" * 50 + "\n"
-                    content += self.parent.all_splits[idx].page_content + "\n"
+                        content += "-" * 50 + "\n"
+                        content += self.parent.all_splits[idx].page_content + "\n"
                 return content
 
         return RetrieverTool(self)
@@ -109,33 +121,6 @@ class DocumentRetrievalAndGeneration:
         llm_engine = HfEngine("meta-llama/Meta-Llama-3.1-8B-Instruct")
         return ReactJsonAgent(tools=[self.retriever_tool], llm_engine=llm_engine, max_iterations=4, verbose=2)
 
-    def generate_response_with_timeout(self, input_ids, max_new_tokens=1000):
-        try:
-            streamer = TextIteratorStreamer(self.tokenizer, timeout=60.0, skip_prompt=True, skip_special_tokens=True)
-            generate_kwargs = dict(
-                input_ids=input_ids,
-                max_new_tokens=max_new_tokens,
-                do_sample=True,
-                top_p=1.0,
-                top_k=20,
-                temperature=0.8,
-                repetition_penalty=1.2,
-                eos_token_id=[128001, 128008, 128009],
-                streamer=streamer,
-            )
-            
-            thread = Thread(target=self.model.generate, kwargs=generate_kwargs)
-            thread.start()
-            
-            generated_text = ""
-            for new_text in streamer:
-                generated_text += new_text
-            
-            return generated_text
-        except Exception as e:
-            logger.error(f"Error in generate_response_with_timeout: {str(e)}")
-            return "Text generation process encountered an error"
-
     def run_agentic_rag(self, question: str) -> str:
         enhanced_question = f"""Using the information in your knowledge base, accessible with the 'retriever' tool,
 give a comprehensive answer to the question below.
@@ -149,33 +134,64 @@ Question:
 
         return self.agent.run(enhanced_question)
 
-    def run_standard_rag(self, question: str) -> str:
-        context = self.retriever_tool(query=question)
+    def query_and_generate_response(self, query):
+        # Standard RAG
+        similarityThreshold = 1
+        query_embedding = self.embeddings.encode(query, convert_to_tensor=True).cpu().numpy()
+        distances, indices = self.gpu_index.search(np.array([query_embedding]), k=3)
+        print("Distance", distances, "indices", indices)
+        content = ""
+        filtered_results = []
+        for idx, distance in zip(indices[0], distances[0]):
+            if distance <= similarityThreshold:
+                filtered_results.append(idx)
+            for i in filtered_results:
+                print(self.all_splits[i].page_content)
+            content += "-" * 50 + "\n"
+            content += self.all_splits[idx].page_content + "\n"
+            print("CHUNK", idx)
+            print("Distance:", distance)
+            print("indices:", indices)
+            print(self.all_splits[idx].page_content)
+            print("############################")
 
         conversation = [
             {"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 {context}, choose the most suitable chunks for answering the query.
+            I have provided five relatable json files {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: {question}
+            Query: {query}
             Solution==>
             """}
         ]
         input_ids = self.tokenizer.apply_chat_template(conversation, return_tensors="pt").to(self.model.device)
         
-        return self.generate_response_with_timeout(input_ids)
+        start_time = datetime.now()
+        standard_response = self.generate_response_with_timeout(input_ids)
+        elapsed_time = datetime.now() - start_time
 
-    def query_and_generate_response(self, query):
-        agentic_answer = self.run_agentic_rag(query)
-        standard_answer = self.run_standard_rag(query)
-        
-        combined_answer = f"Agentic RAG Answer:\n{agentic_answer}\n\nStandard RAG Answer:\n{standard_answer}"
-        return combined_answer, ""  # Return empty string for 'content' as it's not used in this implementation
+        print("Generated standard response:", standard_response)
+        print("Time elapsed:", elapsed_time)
+        print("Device in use:", self.model.device)
+
+        standard_solution_text = standard_response.strip()
+        if "Solution:" in standard_solution_text:
+            standard_solution_text = standard_solution_text.split("Solution:", 1)[1].strip()
+
+        # Post-processing to remove "assistant" prefix
+        standard_solution_text = re.sub(r'^assistant\s*', '', standard_solution_text, flags=re.IGNORECASE)
+        standard_solution_text = standard_solution_text.strip()
+
+        # Agentic RAG
+        agentic_solution_text = self.run_agentic_rag(query)
+
+        combined_solution = f"Standard RAG Solution:\n{standard_solution_text}\n\nAgentic RAG Solution:\n{agentic_solution_text}"
+        return combined_solution, content
 
     def qa_infer_gradio(self, query):
         response = self.query_and_generate_response(query)
@@ -186,47 +202,40 @@ if __name__ == "__main__":
     lm_model_id = "meta-llama/Meta-Llama-3.1-8B-Instruct"
     data_folder = 'sample_embedding_folder2'
 
-    # Set your HuggingFace token here
-    os.environ["HUGGINGFACE_TOKEN"] = "your_huggingface_token_here"
+    doc_retrieval_gen = DocumentRetrievalAndGeneration(embedding_model_name, lm_model_id, data_folder)
 
-    try:
-        doc_retrieval_gen = DocumentRetrievalAndGeneration(embedding_model_name, lm_model_id, data_folder)
+    def launch_interface():
+        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;
+            }
+        """
+        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?"
+        ]
 
-        def launch_interface():
-            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;
-                }
-            """
-            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?"
-            ]
-
-            interface = gr.Interface(
-                fn=doc_retrieval_gen.qa_infer_gradio,
-                inputs=[gr.Textbox(label="QUERY", placeholder="Enter your query here")],
-                allow_flagging='never',
-                examples=EXAMPLES,
-                cache_examples=False,
-                outputs=[gr.Textbox(label="RESPONSE"), gr.Textbox(label="RELATED QUERIES")],
-                css=css_code,
-                title="TI E2E FORUM Multi-Agent RAG"
-            )
+        interface = gr.Interface(
+            fn=doc_retrieval_gen.qa_infer_gradio,
+            inputs=[gr.Textbox(label="QUERY", placeholder="Enter your query here")],
+            allow_flagging='never',
+            examples=EXAMPLES,
+            cache_examples=False,
+            outputs=[gr.Textbox(label="RESPONSE"), gr.Textbox(label="RELATED QUERIES")],
+            css=css_code,
+            title="TI E2E FORUM"
+        )
 
-            interface.launch(debug=True)
+        interface.launch(debug=True)
 
-        launch_interface()
-    except Exception as e:
-        logger.error(f"An error occurred: {str(e)}")
-        logger.info("Please check your environment setup and try again.")
+    launch_interface()