import torch import json from torch import cuda, bfloat16 from transformers import AutoTokenizer, AutoModelForCausalLM, pipeline, BitsAndBytesConfig, StoppingCriteria, StoppingCriteriaList from langchain.llms import HuggingFacePipeline import gradio as gr import os import faiss import numpy as np from langchain.embeddings import HuggingFaceEmbeddings class Chatbot: def __init__(self): self.HF_TOKEN = os.environ.get("HF_TOKEN", None) self.model_id = "mistralai/Mistral-7B-Instruct-v0.2" self.device = f'cuda:{cuda.current_device()}' if cuda.is_available() else 'cpu' self.bnb_config = BitsAndBytesConfig( load_in_4bit=True, bnb_4bit_quant_type='nf4', bnb_4bit_use_double_quant=True, bnb_4bit_compute_dtype=bfloat16 ) self.tokenizer = AutoTokenizer.from_pretrained(self.model_id, token=self.HF_TOKEN) self.model = AutoModelForCausalLM.from_pretrained(self.model_id, device_map="auto", token=self.HF_TOKEN, quantization_config=self.bnb_config) self.stop_list = ['\nHuman:', '\n```\n'] self.stop_token_ids = [self.tokenizer(x)['input_ids'] for x in self.stop_list] self.stop_token_ids = [torch.LongTensor(x).to(self.device) for x in self.stop_token_ids] self.stopping_criteria = StoppingCriteriaList([self.StopOnTokens()]) self.generate_text = pipeline( model=self.model, tokenizer=self.tokenizer, return_full_text=True, task='text-generation', temperature=0.1, max_new_tokens=2048, ) self.llm = HuggingFacePipeline(pipeline=self.generate_text) # Initialize the embedding model self.embeddings = HuggingFaceEmbeddings(model_name="sentence-transformers/all-mpnet-base-v2", model_kwargs={"device": "cuda"}) try: cpu_index = faiss.read_index('faiss_index_new_model3.index') gpu_resource = faiss.StandardGpuResources() self.vectorstore = faiss.index_cpu_to_gpu(gpu_resource, 0, cpu_index) print("Loaded embedding successfully") except Exception as e: print("FAISS could not be imported or index could not be loaded.") raise e self.chain = ConversationalRetrievalChain.from_llm(self.llm, self.vectorstore.as_retriever(), return_source_documents=True) self.chat_history = [] class StopOnTokens(StoppingCriteria): def __call__(self, input_ids: torch.LongTensor, scores: torch.FloatTensor, **kwargs) -> bool: for stop_ids in self.stop_token_ids: if torch.eq(input_ids[0][-len(stop_ids):], stop_ids).all(): return True return False def format_prompt(self, query): prompt=f""" 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 four relatable json files , 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. Here's my question: Query:{query} Solution==> 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.", """ return prompt def qa_infer(self, query): content = "" formatted_prompt = self.format_prompt(query) # Embed the query query_embedding = self.embeddings.embed_query(formatted_prompt) # Perform the search distances, indices = self.vectorstore.search(np.array([query_embedding]), k=5) # Retrieve the top documents for idx in indices[0]: doc = self.vectorstore.get_document(idx) content += "-" * 50 + "\n" content += doc.page_content + "\n" result = self.chain({"question": formatted_prompt, "chat_history": self.chat_history}) print(content) print("#" * 100) print(result['answer']) output_file = "output.txt" with open(output_file, "w") as f: f.write("Query:\n") f.write(query + "\n\n") f.write("Answer:\n") f.write(result['answer'] + "\n\n") f.write("Source Documents:\n") f.write(content + "\n") download_link = f'Download Output File' return result['answer'], content, download_link def launch_interface(self): 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 = ["TDA4 product planning and datasheet release progress? ", "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?", "Master core in TDA2XX is a15 and in TDA3XX it is m4,so we have to shift all modules that are being used by a15 in TDA2XX to m4 in TDA3xx."] 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) tab1 = gr.Interface(fn=self.qa_infer, 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"), gr.HTML()], css=css_code) tab2 = gr.Interface(fn=self.qa_infer, inputs=[dropdown], allow_flagging='never', outputs=[gr.Textbox(label="SOLUTION"), gr.Textbox(label="RELATED QUERIES"), gr.HTML()], css=css_code)#, title="FAQs") # # Add dummy outputs to each interface # tab1.outputs = dummy_outputs # tab2.outputs = dummy_outputs gr.TabbedInterface([tab1, tab2],["Textbox Input", "FAQs"],title="TI E2E FORUM",css=css_code).launch(debug=True) # Instantiate and launch the chatbot chatbot = Chatbot() chatbot.launch_interface()