Update app.py

app.py

@@ -1,5 +1,3 @@
-"Single Thread"
-
 import os
 import multiprocessing
 import concurrent.futures
@@ -20,7 +18,7 @@ class DocumentRetrievalAndGeneration:
     def __init__(self, embedding_model_name, lm_model_id, data_folder):
         self.all_splits = self.load_documents(data_folder)
         self.embeddings = SentenceTransformer(embedding_model_name)
-        self.gpu_index = self.create_faiss_index()
+        self.cpu_index = self.create_faiss_index()
         self.llm = self.initialize_llm(lm_model_id)
 
     def load_documents(self, folder_path):
@@ -30,8 +28,6 @@ class DocumentRetrievalAndGeneration:
         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(5):
-        #     print(all_splits[i].page_content)
         return all_splits
 
     def create_faiss_index(self):
@@ -39,9 +35,7 @@ class DocumentRetrievalAndGeneration:
         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
+        return index
 
     def initialize_llm(self, model_id):
         bnb_config = BitsAndBytesConfig(
@@ -75,7 +69,7 @@ class DocumentRetrievalAndGeneration:
     
     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)
+        distances, indices = self.cpu_index.search(np.array([query_embedding]), k=5)
 
         content = ""
         for idx in indices[0]:
@@ -113,20 +107,15 @@ class DocumentRetrievalAndGeneration:
         Solution:"NO SOLUTION AVAILABLE"
         </s>
         """
-        # prompt = f"Query: {query}\nSolution: {content}\n"
 
-        # 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
 
-        # 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)
@@ -134,13 +123,10 @@ class DocumentRetrievalAndGeneration:
         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)
@@ -154,77 +140,12 @@ class DocumentRetrievalAndGeneration:
         return response
 
 if __name__ == "__main__":
-    # Example usage
     embedding_model_name = 'flax-sentence-embeddings/all_datasets_v3_MiniLM-L12'
     lm_model_id = "mistralai/Mistral-7B-Instruct-v0.2"
     data_folder = 'text_files'
 
     doc_retrieval_gen = DocumentRetrievalAndGeneration(embedding_model_name, lm_model_id, data_folder)
 
-    # """Dual 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 interfaces
-    #     tab1 = 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
-    #     )
-    #     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 {
@@ -245,15 +166,6 @@ if __name__ == "__main__":
                     "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")],
@@ -264,8 +176,6 @@ if __name__ == "__main__":
             css=css_code
         )
 
-        # Launch Gradio interface
         interface.launch(debug=True)
 
-    # Launch the interface
     launch_interface()