Update app.py

app.py

@@ -1,168 +1,261 @@
+"Single Thread"
+
 import os
-import time
-import spaces
+import multiprocessing
+import concurrent.futures
+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
-from transformers import AutoModelForCausalLM, AutoTokenizer, TextIteratorStreamer, BitsAndBytesConfig
+import numpy as np
+from transformers import AutoModelForCausalLM, AutoTokenizer, pipeline, BitsAndBytesConfig
+from datetime import datetime
+import json
 import gradio as gr
-from threading import Thread
-
-
-HF_TOKEN = os.environ.get("HF_TOKEN", None)
-MODEL = "meta-llama/Meta-Llama-3.1-8B-Instruct"
-
-TITLE = "<h1><center>Meta-Llama3.1-8B</center></h1>"
-
-PLACEHOLDER = """
-<center>
-<p>Hi! How can I help you today?</p>
-</center>
-"""
-
-
-CSS = """
-.duplicate-button {
-    margin: auto !important;
-    color: white !important;
-    background: black !important;
-    border-radius: 100vh !important;
-}
-h3 {
-    text-align: center;
-}
-"""
-
-device = "cuda" # for GPU usage or "cpu" for CPU usage
-
-quantization_config = BitsAndBytesConfig(
-    load_in_4bit=True,
-    bnb_4bit_compute_dtype=torch.bfloat16,
-    bnb_4bit_use_double_quant=True,
-    bnb_4bit_quant_type= "nf4")
-
-tokenizer = AutoTokenizer.from_pretrained(MODEL)
-model = AutoModelForCausalLM.from_pretrained(
-    MODEL,
-    torch_dtype=torch.bfloat16,
-    device_map="auto",
-    quantization_config=quantization_config)
-
-@spaces.GPU()
-def stream_chat(
-    message: str, 
-    history: list,
-    system_prompt: str,
-    temperature: float = 0.8, 
-    max_new_tokens: int = 1024, 
-    top_p: float = 1.0, 
-    top_k: int = 20, 
-    penalty: float = 1.2,
-):
-    print(f'message: {message}')
-    print(f'history: {history}')
-
-    conversation = [
-        {"role": "system", "content": system_prompt}
-    ]
-    for prompt, answer in history:
-        conversation.extend([
-            {"role": "user", "content": prompt}, 
-            {"role": "assistant", "content": answer},
-        ])
-
-    conversation.append({"role": "user", "content": message})
-
-    input_ids = tokenizer.apply_chat_template(conversation, add_generation_prompt=True, return_tensors="pt").to(model.device)
-    
-    streamer = TextIteratorStreamer(tokenizer, timeout=60.0, skip_prompt=True, skip_special_tokens=True)
+import re 
+
+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.llm = self.initialize_llm(lm_model_id)
+
+    def load_documents(self, folder_path):
+        loader = DirectoryLoader(folder_path, loader_cls=TextLoader)
+        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(5):
+            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.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
+
+    def initialize_llm(self, model_id):
+        bnb_config = BitsAndBytesConfig(
+            load_in_4bit=True,
+            bnb_4bit_use_double_quant=True,
+            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(
+            model=model,
+            tokenizer=tokenizer,
+            return_full_text=True,
+            task='text-generation',
+            temperature=0.6,
+            max_new_tokens=256,
+        )
+        return generate_text
+
+    def generate_response_with_timeout(self, model_inputs):
+        try:
+            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:
+            raise TimeoutError("Text generation process timed out")
     
-    generate_kwargs = dict(
-        input_ids=input_ids, 
-        max_new_tokens = max_new_tokens,
-        do_sample = False if temperature == 0 else True,
-        top_p = top_p,
-        top_k = top_k,
-        temperature = temperature,
-        repetition_penalty=penalty,
-        eos_token_id=[128001,128008,128009],
-        streamer=streamer,
-    )
-
-    with torch.no_grad():
-        thread = Thread(target=model.generate, kwargs=generate_kwargs)
-        thread.start()
+    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)
+
+        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"
+
+        # 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)
         
-    buffer = ""
-    for new_text in streamer:
-        buffer += new_text
-        yield buffer
+        # 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)
+        
+        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)
             
-chatbot = gr.Chatbot(height=600, placeholder=PLACEHOLDER)
-
-with gr.Blocks(css=CSS, theme="soft") as demo:
-    gr.HTML(TITLE)
-    gr.DuplicateButton(value="Duplicate Space for private use", elem_classes="duplicate-button")
-    gr.ChatInterface(
-        fn=stream_chat,
-        chatbot=chatbot,
-        fill_height=True,
-        additional_inputs_accordion=gr.Accordion(label="⚙️ Parameters", open=False, render=False),
-        additional_inputs=[
-            gr.Textbox(
-                value="You are a helpful assistant",
-                label="System Prompt",
-                render=False,
-            ),
-            gr.Slider(
-                minimum=0,
-                maximum=1,
-                step=0.1,
-                value=0.8,
-                label="Temperature",
-                render=False,
-            ),
-            gr.Slider(
-                minimum=128,
-                maximum=8192,
-                step=1,
-                value=1024,
-                label="Max new tokens",
-                render=False,
-            ),
-            gr.Slider(
-                minimum=0.0,
-                maximum=1.0,
-                step=0.1,
-                value=1.0,
-                label="top_p",
-                render=False,
-            ),
-            gr.Slider(
-                minimum=1,
-                maximum=20,
-                step=1,
-                value=20,
-                label="top_k",
-                render=False,
-            ),
-            gr.Slider(
-                minimum=0.0,
-                maximum=2.0,
-                step=0.1,
-                value=1.2,
-                label="Repetition penalty",
-                render=False,
-            ),
-        ],
-        examples=[
-            ["Help me study vocabulary: write a sentence for me to fill in the blank, and I'll try to pick the correct option."],
-            ["What are 5 creative things I could do with my kids' art? I don't want to throw them away, but it's also so much clutter."],
-            ["Tell me a random fun fact about the Roman Empire."],
-            ["Show me a code snippet of a website's sticky header in CSS and JavaScript."],
-        ],
-        cache_examples=False,
-    )
+        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
+
+    def qa_infer_gradio(self, query):
+        response = self.query_and_generate_response(query)
+        return response
 
 if __name__ == "__main__":
-    demo.launch()
+    # Example usage
+    embedding_model_name = 'flax-sentence-embeddings/all_datasets_v3_MiniLM-L12'
+    lm_model_id = "meta-llama/Meta-Llama-3.1-8B-Instruct"
+    data_folder = 'sample_embedding_folder2'
+
+    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 {
+    #             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()