Update app.py

app.py

@@ -1,16 +1,20 @@
+"Single Thread"
+
+import os
+import multiprocessing
 import concurrent.futures
-import threading
+from langchain.document_loaders import TextLoader, DirectoryLoader
+from langchain.text_splitter import RecursiveCharacterTextSplitter
+from langchain.vectorstores import FAISS
+from sentence_transformers import SentenceTransformer
+import faiss
 import torch
+import numpy as np
+from transformers import AutoModelForCausalLM, AutoTokenizer, pipeline, BitsAndBytesConfig
 from datetime import datetime
 import json
 import gradio as gr
-import re
-import faiss
-import numpy as np
-from sentence_transformers import SentenceTransformer
-from transformers import AutoModelForCausalLM, AutoTokenizer, pipeline, BitsAndBytesConfig
-from langchain.document_loaders import DirectoryLoader, TextLoader
-from langchain.text_splitter import RecursiveCharacterTextSplitter
+import re 
 
 class DocumentRetrievalAndGeneration:
     def __init__(self, embedding_model_name, lm_model_id, data_folder):
@@ -18,7 +22,6 @@ class DocumentRetrievalAndGeneration:
         self.embeddings = SentenceTransformer(embedding_model_name)
         self.gpu_index = self.create_faiss_index()
         self.llm = self.initialize_llm(lm_model_id)
-        self.cancel_flag = threading.Event()
 
     def load_documents(self, folder_path):
         loader = DirectoryLoader(folder_path, loader_cls=TextLoader)
@@ -47,6 +50,7 @@ class DocumentRetrievalAndGeneration:
             bnb_4bit_quant_type="nf4",
             bnb_4bit_compute_dtype=torch.bfloat16
         )
+        device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
         model = AutoModelForCausalLM.from_pretrained(model_id, quantization_config=bnb_config)
         tokenizer = AutoTokenizer.from_pretrained(model_id)
         generate_text = pipeline(
@@ -60,85 +64,84 @@ class DocumentRetrievalAndGeneration:
         return generate_text
 
     def generate_response_with_timeout(self, model_inputs):
-        def target(future):
-            if self.cancel_flag.is_set():
-                return
-            generated_ids = self.llm.model.generate(model_inputs, max_new_tokens=1000, do_sample=True)
-            if not self.cancel_flag.is_set():
-                future.set_result(generated_ids)
-            else:
-                future.set_exception(TimeoutError("Text generation process was canceled"))
-
-        future = concurrent.futures.Future()
-        thread = threading.Thread(target=target, args=(future,))
-        thread.start()
-
         try:
-            generated_ids = future.result(timeout=60)  # Timeout set to 60 seconds
+            with concurrent.futures.ThreadPoolExecutor() as executor:
+                future = executor.submit(self.llm.model.generate, model_inputs, max_new_tokens=1000, do_sample=True)
+                generated_ids = future.result(timeout=60)  # Timeout set to 60 seconds
             return generated_ids
         except concurrent.futures.TimeoutError:
-            self.cancel_flag.set()
             raise TimeoutError("Text generation process timed out")
+    
+    def query_and_generate_response(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=5)
 
-    def qa_infer_gradio(self, query):
-        # Set the cancel flag to false for the new query
-        self.cancel_flag.clear()
-
-        try:
-            query_embedding = self.embeddings.encode(query, convert_to_tensor=True).cpu().numpy()
-            distances, indices = self.gpu_index.search(np.array([query_embedding]), k=5)
-
-            content = ""
-            for idx in indices[0]:
-                content += "-" * 50 + "\n"
-                content += self.all_splits[idx].page_content + "\n"
-
-            prompt = f"""<s>
-            You are a knowledgeable assistant with access to a comprehensive database. 
-            I need you to answer my question and provide related information in a specific format.
-            I have provided five relatable json files {content}, choose the most suitable chunks for answering the query
-            Here's what I need:
-            Include a final answer without additional comments, sign-offs, or extra phrases. Be direct and to the point.
-            content
-            Here's my question:
-            Query:{query}
-            Solution==>
-            RETURN ONLY SOLUTION . IF THEIR IS NO ANSWER RELATABLE IN RETRIEVED CHUNKS , RETURN " NO SOLUTION AVAILABLE"
-            Example1
-            Query: "How to use IPU1_0 instead of A15_0 to process NDK in TDA2x-EVM",
-            Solution: "To use IPU1_0 instead of A15_0 to process NDK in TDA2x-EVM, you need to modify the configuration file of the NDK application. Specifically, change the processor reference from 'A15_0' to 'IPU1_0'.",
-            
-            Example2
-            Query: "Can BQ25896 support I2C interface?",
-            Solution: "Yes, the BQ25896 charger supports the I2C interface for communication."
-            </s>
-            """
-            
-            messages = [{"role": "user", "content": prompt}]
-            encodeds = self.llm.tokenizer.apply_chat_template(messages, return_tensors="pt")
-            model_inputs = encodeds.to(self.llm.model.device)
-
-            start_time = datetime.now()
-            generated_ids = self.generate_response_with_timeout(model_inputs)
-            elapsed_time = datetime.now() - start_time
-
-            decoded = self.llm.tokenizer.batch_decode(generated_ids)
-            generated_response = decoded[0]
+        content = ""
+        for idx in indices[0]:
+            content += "-" * 50 + "\n"
+            content += self.all_splits[idx].page_content + "\n"
+            print("CHUNK", idx)
+            print(self.all_splits[idx].page_content)
+            print("############################")
+        prompt = f"""<s>
+        You are a knowledgeable assistant with access to a comprehensive database. 
+        I need you to answer my question and provide related information in a specific format.
+        I have provided five relatable json files {content}, choose the most suitable chunks for answering the query
+        Here's what I need:
+        Include a final answer without additional comments, sign-offs, or extra phrases. Be direct and to the point.
+        content
+        Here's my question:
+        Query:{query}
+        Solution==>
+        RETURN ONLY SOLUTION . IF THEIR IS NO ANSWER RELATABLE IN RETRIEVED CHUNKS , RETURN " NO SOLUTION AVAILABLE"
+        Example1
+        Query: "How to use IPU1_0 instead of A15_0 to process NDK in TDA2x-EVM",
+        Solution: "To use IPU1_0 instead of A15_0 to process NDK in TDA2x-EVM, you need to modify the configuration file of the NDK application. Specifically, change the processor reference from 'A15_0' to 'IPU1_0'.",
+        
+        Example2
+        Query: "Can BQ25896 support I2C interface?",
+        Solution: "Yes, the BQ25896 charger supports the I2C interface for communication."
+        </s>
+        """
+        # prompt = f"Query: {query}\nSolution: {content}\n"
 
-            match = re.search(r'Solution:(.*?)</s>', generated_response, re.DOTALL | re.IGNORECASE)
-            if match:
-                solution_text = match.group(1).strip()
-            else:
-                solution_text = "NO SOLUTION AVAILABLE"
+        # Encode and prepare inputs
+        messages = [{"role": "user", "content": prompt}]
+        encodeds = self.llm.tokenizer.apply_chat_template(messages, return_tensors="pt")
+        model_inputs = encodeds.to(self.llm.device)
+        
+        # Perform inference and measure time
+        start_time = datetime.now()
+        generated_ids = self.generate_response_with_timeout(model_inputs)
+        # generated_ids = self.llm.model.generate(model_inputs, max_new_tokens=1000, do_sample=True)
+        elapsed_time = datetime.now() - start_time
 
-            print("Generated response:", generated_response)
-            print("Time elapsed:", elapsed_time)
-            print("Device in use:", self.llm.model.device)
+        # Decode and return output
+        decoded = self.llm.tokenizer.batch_decode(generated_ids)
+        generated_response = decoded[0]
+        match1 = re.search(r'\[/INST\](.*?)</s>', generated_response, re.DOTALL)
+        
+        match2 = re.search(r'Solution:(.*?)</s>', generated_response, re.DOTALL | re.IGNORECASE)
+        if match1:
+            solution_text = match1.group(1).strip()
+            print(solution_text)
+            if "Solution:" in solution_text:
+                solution_text = solution_text.split("Solution:", 1)[1].strip()
+        elif match2:
+            solution_text = match2.group(1).strip()
+            print(solution_text)
+            
+        else:
+            solution_text=generated_response
+        print("Generated response:", generated_response)
+        print("Time elapsed:", elapsed_time)
+        print("Device in use:", self.llm.device)
 
-            return solution_text, content
+        return solution_text, content
 
-        except TimeoutError:
-            return "timeout", content
+    def qa_infer_gradio(self, query):
+        response = self.query_and_generate_response(query)
+        return response
 
 if __name__ == "__main__":
     # Example usage
@@ -148,7 +151,8 @@ if __name__ == "__main__":
 
     doc_retrieval_gen = DocumentRetrievalAndGeneration(embedding_model_name, lm_model_id, data_folder)
 
-    # Define Gradio interface function
+    """Dual Interface"""
+    
     def launch_interface():
         css_code = """
             .gradio-container {
@@ -164,21 +168,23 @@ if __name__ == "__main__":
                 font-size: 16px; /* Increase font size */
                 font-weight: bold; /* Make text bold */
             }
-            """
-        EXAMPLES = ["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?"]
-        
+        """
+        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?"
+        ]
+    
         file_path = "ticketNames.txt"
-
+    
         # Read the file content
         with open(file_path, "r") as file:
             content = file.read()
         ticket_names = json.loads(content)
         dropdown = gr.Dropdown(label="Sample queries", choices=ticket_names)
-        
-        # Define Gradio interface
-        interface = gr.Interface(
+    
+        # Define Gradio interfaces
+        tab1 = gr.Interface(
             fn=doc_retrieval_gen.qa_infer_gradio,
             inputs=[gr.Textbox(label="QUERY", placeholder="Enter your query here")],
             allow_flagging='never',
@@ -187,9 +193,69 @@ if __name__ == "__main__":
             outputs=[gr.Textbox(label="SOLUTION"), gr.Textbox(label="RELATED QUERIES")],
             css=css_code
         )
-
-        # Launch Gradio interface
-        interface.launch(debug=True)
-
+        tab2 = gr.Interface(
+            fn=doc_retrieval_gen.qa_infer_gradio,
+            inputs=[dropdown],
+            allow_flagging='never',
+            outputs=[gr.Textbox(label="SOLUTION"), gr.Textbox(label="RELATED QUERIES")],
+            css=css_code
+        )
+    
+        # Combine interfaces into a tabbed interface
+        gr.TabbedInterface(
+            [tab1, tab2],
+            ["Textbox Input", "FAQs"],
+            title="TI E2E FORUM",
+            css=css_code
+        ).launch(debug=True)
+    
     # Launch the interface
     launch_interface()
+
+
+
+    """Single Interface"""
+    # def launch_interface():
+    #     css_code = """
+    #         .gradio-container {
+    #             background-color: #daccdb;
+    #         }
+    #         /* Button styling for all buttons */
+    #         button {
+    #             background-color: #927fc7; /* Default color for all other buttons */
+    #             color: black;
+    #             border: 1px solid black;
+    #             padding: 10px;
+    #             margin-right: 10px;
+    #             font-size: 16px; /* Increase font size */
+    #             font-weight: bold; /* Make text 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?"]
+        
+    #     file_path = "ticketNames.txt"
+
+    #     # Read the file content
+    #     with open(file_path, "r") as file:
+    #         content = file.read()
+    #     ticket_names = json.loads(content)
+    #     dropdown = gr.Dropdown(label="Sample queries", choices=ticket_names)
+        
+    #     # Define Gradio interface
+    #     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="SOLUTION"), gr.Textbox(label="RELATED QUERIES")],
+    #         css=css_code
+    #     )
+
+    #     # Launch Gradio interface
+    #     interface.launch(debug=True)
+
+    # # Launch the interface
+    # launch_interface()