IT2091024v2 / app.py
Pijush2023's picture
Update app.py
800db56 verified
raw
history blame
79.7 kB
# #Main code header Library
# import gradio as gr
# import requests
# import os
# import time
# import re
# import logging
# import tempfile
# import folium
# import concurrent.futures
# import torch
# from PIL import Image
# from datetime import datetime
# from transformers import pipeline, AutoModelForSpeechSeq2Seq, AutoProcessor
# from googlemaps import Client as GoogleMapsClient
# from gtts import gTTS
# from diffusers import StableDiffusionPipeline
# from langchain_openai import OpenAIEmbeddings, ChatOpenAI
# from langchain_pinecone import PineconeVectorStore
# from langchain.prompts import PromptTemplate
# from langchain.chains import RetrievalQA
# from langchain.chains.conversation.memory import ConversationBufferWindowMemory
# from huggingface_hub import login
# from transformers.models.speecht5.number_normalizer import EnglishNumberNormalizer
# from parler_tts import ParlerTTSForConditionalGeneration
# from transformers import AutoTokenizer, AutoFeatureExtractor, set_seed
# from scipy.io.wavfile import write as write_wav
# from pydub import AudioSegment
# from string import punctuation
# import librosa
# from pathlib import Path
# import torchaudio
# import numpy as np
# # Neo4j imports
# from langchain.chains import GraphCypherQAChain
# from langchain_community.graphs import Neo4jGraph
# from langchain_community.document_loaders import HuggingFaceDatasetLoader
# from langchain_text_splitters import CharacterTextSplitter
# from langchain_experimental.graph_transformers import LLMGraphTransformer
# from langchain_core.prompts import ChatPromptTemplate
# from langchain_core.pydantic_v1 import BaseModel, Field
# from langchain_core.messages import AIMessage, HumanMessage
# from langchain_core.output_parsers import StrOutputParser
# from langchain_core.runnables import RunnableBranch, RunnableLambda, RunnableParallel, RunnablePassthrough
# # Set environment variables for Torch- CUDA
# os.environ['PYTORCH_USE_CUDA_DSA'] = '1'
# os.environ['CUDA_LAUNCH_BLOCKING'] = '1'
# #Hugging face token Initilization
# hf_token = os.getenv("HF_TOKEN")
# if hf_token is None:
# print("Please set your Hugging Face token in the environment variables.")
# else:
# login(token=hf_token)
# logging.basicConfig(level=logging.DEBUG)
# #Embedding the vector with openai
# embeddings = OpenAIEmbeddings(api_key=os.environ['OPENAI_API_KEY'])
# # Pinecone setup
# from pinecone import Pinecone
# pc = Pinecone(api_key=os.environ['PINECONE_API_KEY'])
# index_name = "radardata07242024"
# vectorstore = PineconeVectorStore(index_name=index_name, embedding=embeddings)
# retriever = vectorstore.as_retriever(search_kwargs={'k': 5})
# chat_model = ChatOpenAI(api_key=os.environ['OPENAI_API_KEY'], temperature=0, model='gpt-4o')
# conversational_memory = ConversationBufferWindowMemory(
# memory_key='chat_history',
# k=10,
# return_messages=True
# )
# # Prompt templates
# def get_current_date():
# return datetime.now().strftime("%B %d, %Y")
# current_date = get_current_date()
# template1 = f"""As an expert concierge in Birmingham, Alabama, known for being a helpful and renowned guide, I am here to assist you on this sunny bright day of {current_date}. Given the current weather conditions and date, I have access to a plethora of information regarding events, places, and activities in Birmingham that can enhance your experience.
# If you have any questions or need recommendations, feel free to ask. I have a wealth of knowledge of perennial events in Birmingham and can provide detailed information to ensure you make the most of your time here. Remember, I am here to assist you in any way possible.
# Now, let me guide you through some of the exciting events happening today in Birmingham, Alabama:
# Address: >>, Birmingham, AL
# Time: >>__
# Date: >>__
# Description: >>__
# Address: >>, Birmingham, AL
# Time: >>__
# Date: >>__
# Description: >>__
# Address: >>, Birmingham, AL
# Time: >>__
# Date: >>__
# Description: >>__
# Address: >>, Birmingham, AL
# Time: >>__
# Date: >>__
# Description: >>__
# Address: >>, Birmingham, AL
# Time: >>__
# Date: >>__
# Description: >>__
# If you have any specific preferences or questions about these events or any other inquiries, please feel free to ask. Remember, I am here to ensure you have a memorable and enjoyable experience in Birmingham, AL.
# It was my pleasure!
# {{context}}
# Question: {{question}}
# Helpful Answer:"""
# template2 = f"""As an expert concierge known for being helpful and a renowned guide for Birmingham, Alabama, I assist visitors in discovering the best that the city has to offer. Given today's sunny and bright weather on {current_date}, I am well-equipped to provide valuable insights and recommendations without revealing specific locations. I draw upon my extensive knowledge of the area, including perennial events and historical context.
# In light of this, how can I assist you today? Feel free to ask any questions or seek recommendations for your day in Birmingham. If there's anything specific you'd like to know or experience, please share, and I'll be glad to help. Remember, keep the question concise for a quick and accurate response.
# "It was my pleasure!"
# {{context}}
# Question: {{question}}
# Helpful Answer:"""
# #QA_Chain_templates
# QA_CHAIN_PROMPT_1 = PromptTemplate(input_variables=["context", "question"], template=template1)
# QA_CHAIN_PROMPT_2 = PromptTemplate(input_variables=["context", "question"], template=template2)
# # Neo4j setup
# graph = Neo4jGraph(
# url="neo4j+s://98f45cc0.databases.neo4j.io",
# username="neo4j",
# password="B_sZbapCTZoQDWj1JrhwqElsNa-jm5Zq1m_mAnyPYog"
# )
# # Avoid pushing the graph documents to Neo4j every time
# # Only push the documents once and comment the code below after the initial push
# # dataset_name = "Pijush2023/birmindata07312024"
# # page_content_column = 'events_description'
# # loader = HuggingFaceDatasetLoader(dataset_name, page_content_column)
# # data = loader.load()
# # text_splitter = CharacterTextSplitter(chunk_size=100, chunk_overlap=50)
# # documents = text_splitter.split_documents(data)
# # llm_transformer = LLMGraphTransformer(llm=chat_model)
# # graph_documents = llm_transformer.convert_to_graph_documents(documents)
# # graph.add_graph_documents(graph_documents, baseEntityLabel=True, include_source=True)
# #Neo4j Setup
# class Entities(BaseModel):
# names: list[str] = Field(..., description="All the person, organization, or business entities that appear in the text")
# entity_prompt = ChatPromptTemplate.from_messages([
# ("system", "You are extracting organization and person entities from the text."),
# ("human", "Use the given format to extract information from the following input: {question}"),
# ])
# entity_chain = entity_prompt | chat_model.with_structured_output(Entities)
# #Remove Lucene Characther
# def remove_lucene_chars(input: str) -> str:
# return input.translate(str.maketrans({"\\": r"\\", "+": r"\+", "-": r"\-", "&": r"\&", "|": r"\|", "!": r"\!",
# "(": r"\(", ")": r"\)", "{": r"\{", "}": r"\}", "[": r"\[", "]": r"\]",
# "^": r"\^", "~": r"\~", "*": r"\*", "?": r"\?", ":": r"\:", '"': r'\"',
# ";": r"\;", " ": r"\ "}))
# #Full Text query Generator
# def generate_full_text_query(input: str) -> str:
# full_text_query = ""
# words = [el for el in remove_lucene_chars(input).split() if el]
# for word in words[:-1]:
# full_text_query += f" {word}~2 AND"
# full_text_query += f" {words[-1]}~2"
# return full_text_query.strip()
# # Neo4j Retrieval connection
# def structured_retriever(question: str) -> str:
# result = ""
# entities = entity_chain.invoke({"question": question})
# for entity in entities.names:
# response = graph.query(
# """CALL db.index.fulltext.queryNodes('entity', $query, {limit:2})
# YIELD node,score
# CALL {
# WITH node
# MATCH (node)-[r:!MENTIONS]->(neighbor)
# RETURN node.id + ' - ' + type(r) + ' -> ' + neighbor.id AS output
# UNION ALL
# WITH node
# MATCH (node)<-[r:!MENTIONS]-(neighbor)
# RETURN neighbor.id + ' - ' + type(r) + ' -> ' + node.id AS output
# }
# RETURN output LIMIT 50
# """,
# {"query": generate_full_text_query(entity)},
# )
# result += "\n".join([el['output'] for el in response])
# return result
# def retriever_neo4j(question: str):
# structured_data = structured_retriever(question)
# logging.debug(f"Structured data: {structured_data}")
# return structured_data
# _template = """Given the following conversation and a follow-up question, rephrase the follow-up question to be a standalone question,
# in its original language.
# Chat History:
# {chat_history}
# Follow Up Input: {question}
# Standalone question:"""
# CONDENSE_QUESTION_PROMPT = PromptTemplate.from_template(_template)
# def _format_chat_history(chat_history: list[tuple[str, str]]) -> list:
# buffer = []
# for human, ai in chat_history:
# buffer.append(HumanMessage(content=human))
# buffer.append(AIMessage(content=ai))
# return buffer
# _search_query = RunnableBranch(
# (
# RunnableLambda(lambda x: bool(x.get("chat_history"))).with_config(
# run_name="HasChatHistoryCheck"
# ),
# RunnablePassthrough.assign(
# chat_history=lambda x: _format_chat_history(x["chat_history"])
# )
# | CONDENSE_QUESTION_PROMPT
# | ChatOpenAI(temperature=0, api_key=os.environ['OPENAI_API_KEY'])
# | StrOutputParser(),
# ),
# RunnableLambda(lambda x : x["question"]),
# )
# # # template = """Answer the question based only on the following context:
# # {context}
# # Question: {question}
# # Use natural language and be concise.
# # Answer:"""
# # Define conversational and detailed prompt templates for Neo4j responses
# neo4j_conversational_template = f"""As an expert concierge known for being helpful and a renowned guide for Birmingham, Alabama, I assist visitors in discovering the best that the city has to offer. Given today's sunny and bright weather on {current_date}, I am well-equipped to provide valuable insights and recommendations without revealing specific locations. I draw upon my extensive knowledge of the area, including perennial events and historical context.
# In light of this, how can I assist you today? Feel free to ask any questions or seek recommendations for your day in Birmingham. If there's anything specific you'd like to know or experience, please share, and I'll be glad to help. Remember, keep the question concise for a quick and accurate response.
# "It was my pleasure!"
# {{context}}
# Question: {{question}}
# Helpful Answer:"""
# neo4j_details_template = f"""As an expert concierge in Birmingham, Alabama, known for being a helpful and renowned guide, I am here to assist you on this sunny bright day of {current_date}. Given the current weather conditions and date, I have access to a plethora of information regarding events, places, and activities in Birmingham that can enhance your experience.
# If you have any questions or need recommendations, feel free to ask. I have a wealth of knowledge of perennial events in Birmingham and can provide detailed information to ensure you make the most of your time here. Remember, I am here to assist you in any way possible.
# Now, let me guide you through some of the exciting events happening today in Birmingham, Alabama:
# Address: >>, Birmingham, AL
# Time: >>__
# Date: >>__
# Description: >>__
# Address: >>, Birmingham, AL
# Time: >>__
# Date: >>__
# Description: >>__
# Address: >>, Birmingham, AL
# Time: >>__
# Date: >>__
# Description: >>__
# Address: >>, Birmingham, AL
# Time: >>__
# Date: >>__
# Description: >>__
# Address: >>, Birmingham, AL
# Time: >>__
# Date: >>__
# Description: >>__
# If you have any specific preferences or questions about these events or any other inquiries, please feel free to ask. Remember, I am here to ensure you have a memorable and enjoyable experience in Birmingham, AL.
# It was my pleasure!
# {{context}}
# Question: {{question}}
# Helpful Answer:"""
# # Create prompt templates
# QA_CHAIN_PROMPT_NEO4J_CONVERSATIONAL = PromptTemplate(input_variables=["context", "question"], template=neo4j_conversational_template)
# QA_CHAIN_PROMPT_NEO4J_DETAILS = PromptTemplate(input_variables=["context", "question"], template=neo4j_details_template)
# # Define Neo4j retrieval chain for conversational mode
# def neo4j_retrieval_conversational(question: str):
# structured_data = structured_retriever(question)
# logging.debug(f"Structured data (Conversational): {structured_data}")
# prompt = QA_CHAIN_PROMPT_NEO4J_CONVERSATIONAL.format(context=structured_data, question=question)
# response = chat_model({"query": prompt})
# return response, []
# # Define Neo4j retrieval chain for detailed mode
# def neo4j_retrieval_details(question: str):
# structured_data = structured_retriever(question)
# logging.debug(f"Structured data (Details): {structured_data}")
# prompt = QA_CHAIN_PROMPT_NEO4J_DETAILS.format(context=structured_data, question=question)
# response = chat_model({"query": prompt})
# return response, extract_addresses(response)
# # qa_prompt = ChatPromptTemplate.from_template(template)
# chain_neo4j = (
# RunnableParallel(
# {
# "context": _search_query | retriever_neo4j,
# "question": RunnablePassthrough(),
# }
# )
# | qa_prompt
# | chat_model
# | StrOutputParser()
# )
# # Define a function to select between Pinecone and Neo4j
# # def generate_answer(message, choice, retrieval_mode):
# # logging.debug(f"generate_answer called with choice: {choice} and retrieval_mode: {retrieval_mode}")
# # prompt_template = QA_CHAIN_PROMPT_1 if choice == "Details" else QA_CHAIN_PROMPT_2
# # if retrieval_mode == "Vector":
# # qa_chain = RetrievalQA.from_chain_type(
# # llm=chat_model,
# # chain_type="stuff",
# # retriever=retriever,
# # chain_type_kwargs={"prompt": prompt_template}
# # )
# # response = qa_chain({"query": message})
# # logging.debug(f"Vector response: {response}")
# # return response['result'], extract_addresses(response['result'])
# # elif retrieval_mode == "Knowledge-Graph":
# # response = chain_neo4j.invoke({"question": message})
# # logging.debug(f"Knowledge-Graph response: {response}")
# # return response, extract_addresses(response)
# # else:
# # return "Invalid retrieval mode selected.", []
# def generate_answer(message, choice, retrieval_mode):
# logging.debug(f"generate_answer called with choice: {choice} and retrieval_mode: {retrieval_mode}")
# prompt_template = QA_CHAIN_PROMPT_1 if choice == "Details" else QA_CHAIN_PROMPT_2
# if retrieval_mode == "Vector":
# qa_chain = RetrievalQA.from_chain_type(
# llm=chat_model,
# chain_type="stuff",
# retriever=retriever,
# chain_type_kwargs={"prompt": prompt_template}
# )
# response = qa_chain({"query": message})
# logging.debug(f"Vector response: {response}")
# return response['result'], extract_addresses(response['result'])
# elif retrieval_mode == "Knowledge-Graph":
# if choice == "Details":
# response, addresses = neo4j_retrieval_details(message)
# else:
# response, addresses = neo4j_retrieval_conversational(message)
# logging.debug(f"Knowledge-Graph response: {response}")
# return response, addresses
# else:
# return "Invalid retrieval mode selected.", []
# def bot(history, choice, tts_choice, retrieval_mode):
# if not history:
# return history
# response, addresses = generate_answer(history[-1][0], choice, retrieval_mode)
# history[-1][1] = ""
# with concurrent.futures.ThreadPoolExecutor() as executor:
# if tts_choice == "Alpha":
# audio_future = executor.submit(generate_audio_elevenlabs, response)
# elif tts_choice == "Beta":
# audio_future = executor.submit(generate_audio_parler_tts, response)
# elif tts_choice == "Gamma":
# audio_future = executor.submit(generate_audio_mars5, response)
# for character in response:
# history[-1][1] += character
# time.sleep(0.05)
# yield history, None
# audio_path = audio_future.result()
# yield history, audio_path
# history.append([response, None]) # Ensure the response is added in the correct format
# def add_message(history, message):
# history.append((message, None))
# return history, gr.Textbox(value="", interactive=True, placeholder="Enter message or upload file...", show_label=False)
# def print_like_dislike(x: gr.LikeData):
# print(x.index, x.value, x.liked)
# def extract_addresses(response):
# if not isinstance(response, str):
# response = str(response)
# address_patterns = [
# r'([A-Z].*,\sBirmingham,\sAL\s\d{5})',
# r'(\d{4}\s.*,\sBirmingham,\sAL\s\d{5})',
# r'([A-Z].*,\sAL\s\d{5})',
# r'([A-Z].*,.*\sSt,\sBirmingham,\sAL\s\d{5})',
# r'([A-Z].*,.*\sStreets,\sBirmingham,\sAL\s\d{5})',
# r'(\d{2}.*\sStreets)',
# r'([A-Z].*\s\d{2},\sBirmingham,\sAL\s\d{5})',
# r'([a-zA-Z]\s Birmingham)',
# r'([a-zA-Z].*,\sBirmingham,\sAL)',
# r'(^Birmingham,AL$)'
# ]
# addresses = []
# for pattern in address_patterns:
# addresses.extend(re.findall(pattern, response))
# return addresses
# all_addresses = []
# def generate_map(location_names):
# global all_addresses
# all_addresses.extend(location_names)
# api_key = os.environ['GOOGLEMAPS_API_KEY']
# gmaps = GoogleMapsClient(key=api_key)
# m = folium.Map(location=[33.5175, -86.809444], zoom_start=12)
# for location_name in all_addresses:
# geocode_result = gmaps.geocode(location_name)
# if geocode_result:
# location = geocode_result[0]['geometry']['location']
# folium.Marker(
# [location['lat'], location['lng']],
# tooltip=f"{geocode_result[0]['formatted_address']}"
# ).add_to(m)
# map_html = m._repr_html_()
# return map_html
# def fetch_local_news():
# api_key = os.environ['SERP_API']
# url = f'https://serpapi.com/search.json?engine=google_news&q=birmingham headline&api_key={api_key}'
# response = requests.get(url)
# if response.status_code == 200:
# results = response.json().get("news_results", [])
# news_html = """
# <h2 style="font-family: 'Georgia', serif; color: #ff0000; background-color: #f8f8f8; padding: 10px; border-radius: 10px;">Birmingham Today</h2>
# <style>
# .news-item {
# font-family: 'Verdana', sans-serif;
# color: #333;
# background-color: #f0f8ff;
# margin-bottom: 15px;
# padding: 10px;
# border-radius: 5px;
# transition: box-shadow 0.3s ease, background-color 0.3s ease;
# font-weight: bold;
# }
# .news-item:hover {
# box-shadow: 0 4px 8px rgba(0, 0, 0, 0.1);
# background-color: #e6f7ff;
# }
# .news-item a {
# color: #1E90FF;
# text-decoration: none;
# font-weight: bold;
# }
# .news-item a:hover {
# text-decoration: underline;
# }
# .news-preview {
# position: absolute;
# display: none;
# border: 1px solid #ccc;
# border-radius: 5px;
# box-shadow: 0 2px 4px rgba(0, 0, 0, 0.2);
# background-color: white;
# z-index: 1000;
# max-width: 300px;
# padding: 10px;
# font-family: 'Verdana', sans-serif;
# color: #333;
# }
# </style>
# <script>
# function showPreview(event, previewContent) {
# var previewBox = document.getElementById('news-preview');
# previewBox.innerHTML = previewContent;
# previewBox.style.left = event.pageX + 'px';
# previewBox.style.top = event.pageY + 'px';
# previewBox.style.display = 'block';
# }
# function hidePreview() {
# var previewBox = document.getElementById('news-preview');
# previewBox.style.display = 'none';
# }
# </script>
# <div id="news-preview" class="news-preview"></div>
# """
# for index, result in enumerate(results[:7]):
# title = result.get("title", "No title")
# link = result.get("link", "#")
# snippet = result.get("snippet", "")
# news_html += f"""
# <div class="news-item" onmouseover="showPreview(event, '{snippet}')" onmouseout="hidePreview()">
# <a href='{link}' target='_blank'>{index + 1}. {title}</a>
# <p>{snippet}</p>
# </div>
# """
# return news_html
# else:
# return "<p>Failed to fetch local news</p>"
# import numpy as np
# import torch
# from transformers import pipeline, AutoModelForSpeechSeq2Seq, AutoProcessor
# model_id = 'openai/whisper-large-v3'
# device = "cuda:0" if torch.cuda.is_available() else "cpu"
# torch_dtype = torch.float16 if torch.cuda.is_available() else torch.float32
# model = AutoModelForSpeechSeq2Seq.from_pretrained(model_id, torch_dtype=torch_dtype).to(device)
# processor = AutoProcessor.from_pretrained(model_id)
# pipe_asr = pipeline("automatic-speech-recognition", model=model, tokenizer=processor.tokenizer, feature_extractor=processor.feature_extractor, max_new_tokens=128, chunk_length_s=15, batch_size=16, torch_dtype=torch_dtype, device=device, return_timestamps=True)
# base_audio_drive = "/data/audio"
# def transcribe_function(stream, new_chunk):
# try:
# sr, y = new_chunk[0], new_chunk[1]
# except TypeError:
# print(f"Error chunk structure: {type(new_chunk)}, content: {new_chunk}")
# return stream, "", None
# y = y.astype(np.float32) / np.max(np.abs(y))
# if stream is not None:
# stream = np.concatenate([stream, y])
# else:
# stream = y
# result = pipe_asr({"array": stream, "sampling_rate": sr}, return_timestamps=False)
# full_text = result.get("text","")
# return stream, full_text, result
# def update_map_with_response(history):
# if not history:
# return ""
# response = history[-1][1]
# addresses = extract_addresses(response)
# return generate_map(addresses)
# def clear_textbox():
# return ""
# def show_map_if_details(history, choice):
# if choice in ["Details", "Conversational"]:
# return gr.update(visible=True), update_map_with_response(history)
# else:
# return gr.update(visible=False), ""
# def generate_audio_elevenlabs(text):
# XI_API_KEY = os.environ['ELEVENLABS_API']
# VOICE_ID = 'd9MIrwLnvDeH7aZb61E9'
# tts_url = f"https://api.elevenlabs.io/v1/text-to-speech/{VOICE_ID}/stream"
# headers = {
# "Accept": "application/json",
# "xi-api-key": XI_API_KEY
# }
# data = {
# "text": str(text),
# "model_id": "eleven_multilingual_v2",
# "voice_settings": {
# "stability": 1.0,
# "similarity_boost": 0.0,
# "style": 0.60,
# "use_speaker_boost": False
# }
# }
# response = requests.post(tts_url, headers=headers, json=data, stream=True)
# if response.ok:
# audio_segments = []
# with tempfile.NamedTemporaryFile(delete=False, suffix=".mp3") as f:
# for chunk in response.iter_content(chunk_size=1024):
# if chunk:
# f.write(chunk)
# audio_segments.append(chunk)
# temp_audio_path = f.name
# # Combine all audio chunks into a single file
# combined_audio = AudioSegment.from_file(temp_audio_path, format="mp3")
# combined_audio_path = os.path.join(tempfile.gettempdir(), "elevenlabs_combined_audio.mp3")
# combined_audio.export(combined_audio_path, format="mp3")
# logging.debug(f"Audio saved to {combined_audio_path}")
# return combined_audio_path
# else:
# logging.error(f"Error generating audio: {response.text}")
# return None
# repo_id = "parler-tts/parler-tts-mini-expresso"
# parler_model = ParlerTTSForConditionalGeneration.from_pretrained(repo_id).to(device)
# parler_tokenizer = AutoTokenizer.from_pretrained(repo_id)
# parler_feature_extractor = AutoFeatureExtractor.from_pretrained(repo_id)
# SAMPLE_RATE = parler_feature_extractor.sampling_rate
# SEED = 42
# def preprocess(text):
# number_normalizer = EnglishNumberNormalizer()
# text = number_normalizer(text).strip()
# if text[-1] not in punctuation:
# text = f"{text}."
# abbreviations_pattern = r'\b[A-Z][A-Z\.]+\b'
# def separate_abb(chunk):
# chunk = chunk.replace(".", "")
# return " ".join(chunk)
# abbreviations = re.findall(abbreviations_pattern, text)
# for abv in abbreviations:
# if abv in text:
# text is text.replace(abv, separate_abb(abv))
# return text
# def chunk_text(text, max_length=250):
# words = text.split()
# chunks = []
# current_chunk = []
# current_length = 0
# for word in words:
# if current_length + len(word) + 1 <= max_length:
# current_chunk.append(word)
# current_length += len(word) + 1
# else:
# chunks.append(' '.join(current_chunk))
# current_chunk = [word]
# current_length = len(word) + 1
# if current_chunk:
# chunks.append(' '.join(current_chunk))
# return chunks
# def generate_audio_parler_tts(text):
# description = "Thomas speaks with emphasis and excitement at a moderate pace with high quality."
# chunks = chunk_text(preprocess(text))
# audio_segments = []
# for chunk in chunks:
# inputs = parler_tokenizer(description, return_tensors="pt").to(device)
# prompt = parler_tokenizer(chunk, return_tensors="pt").to(device)
# set_seed(SEED)
# generation = parler_model.generate(input_ids=inputs.input_ids, prompt_input_ids=prompt.input_ids)
# audio_arr = generation.cpu().numpy().squeeze()
# temp_audio_path = os.path.join(tempfile.gettempdir(), f"parler_tts_audio_{len(audio_segments)}.wav")
# write_wav(temp_audio_path, SAMPLE_RATE, audio_arr)
# audio_segments.append(AudioSegment.from_wav(temp_audio_path))
# combined_audio = sum(audio_segments)
# combined_audio_path = os.path.join(tempfile.gettempdir(), "parler_tts_combined_audio.wav")
# combined_audio.export(combined_audio_path, format="wav")
# logging.debug(f"Audio saved to {combined_audio_path}")
# return combined_audio_path
# # Load the MARS5 model
# mars5, config_class = torch.hub.load('Camb-ai/mars5-tts', 'mars5_english', trust_repo=True)
# def generate_audio_mars5(text):
# description = "Thomas speaks with emphasis and excitement at a moderate pace with high quality."
# kwargs_dict = {
# 'temperature': 0.2,
# 'top_k': -1,
# 'top_p': 0.2,
# 'typical_p': 1.0,
# 'freq_penalty': 2.6,
# 'presence_penalty': 0.4,
# 'rep_penalty_window': 100,
# 'max_prompt_phones': 360,
# 'deep_clone': True,
# 'nar_guidance_w': 3
# }
# chunks = chunk_text(preprocess(text))
# audio_segments = []
# for chunk in chunks:
# wav = torch.zeros(1, mars5.sr) # Use a placeholder silent audio for the reference
# cfg = config_class(**{k: kwargs_dict[k] for k in kwargs_dict if k in config_class.__dataclass_fields__})
# ar_codes, wav_out = mars5.tts(chunk, wav, "", cfg=cfg)
# temp_audio_path = os.path.join(tempfile.gettempdir(), f"mars5_audio_{len(audio_segments)}.wav")
# torchaudio.save(temp_audio_path, wav_out.unsqueeze(0), mars5.sr)
# audio_segments.append(AudioSegment.from_wav(temp_audio_path))
# combined_audio = sum(audio_segments)
# combined_audio_path = os.path.join(tempfile.gettempdir(), "mars5_combined_audio.wav")
# combined_audio.export(combined_audio_path, format="wav")
# logging.debug(f"Audio saved to {combined_audio_path}")
# return combined_audio_path
# pipe = StableDiffusionPipeline.from_pretrained("stabilityai/stable-diffusion-2", torch_dtype=torch.float16)
# pipe.to(device)
# def generate_image(prompt):
# with torch.cuda.amp.autocast():
# image = pipe(
# prompt,
# num_inference_steps=28,
# guidance_scale=3.0,
# ).images[0]
# return image
# hardcoded_prompt_1 = "Give a high quality photograph of a great looking red 2026 Toyota coupe against a skyline setting in the night, michael mann style in omaha enticing the consumer to buy this product"
# hardcoded_prompt_2 = "A vibrant and dynamic football game scene in the style of Peter Paul Rubens, showcasing the intense match between Alabama and Nebraska. The players are depicted with the dramatic, muscular physiques and expressive faces typical of Rubens' style. The Alabama team is wearing their iconic crimson and white uniforms, while the Nebraska team is in their classic red and white attire. The scene is filled with action, with players in mid-motion, tackling, running, and catching the ball. The background features a grand stadium filled with cheering fans, banners, and the natural landscape in the distance. The colors are rich and vibrant, with a strong use of light and shadow to create depth and drama. The overall atmosphere captures the intensity and excitement of the game, infused with the grandeur and dynamism characteristic of Rubens' work."
# hardcoded_prompt_3 = "Create a high-energy scene of a DJ performing on a large stage with vibrant lights, colorful lasers, a lively dancing crowd, and various electronic equipment in the background."
# def update_images():
# image_1 = generate_image(hardcoded_prompt_1)
# image_2 = generate_image(hardcoded_prompt_2)
# image_3 = generate_image(hardcoded_prompt_3)
# return image_1, image_2, image_3
# def fetch_local_events():
# api_key = os.environ['SERP_API']
# url = f'https://serpapi.com/search.json?engine=google_events&q=Events+in+Birmingham&hl=en&gl=us&api_key={api_key}'
# response = requests.get(url)
# if response.status_code == 200:
# events_results = response.json().get("events_results", [])
# events_html = """
# <h2 style="font-family: 'Georgia', serif; color: #ff0000; background-color: #f8f8f8; padding: 10px; border-radius: 10px;">Local Events</h2>
# <style>
# table {
# font-family: 'Verdana', sans-serif;
# color: #333;
# border-collapse: collapse;
# width: 100%;
# }
# th, td {
# border: 1px solid #fff !important;
# padding: 8px;
# }
# th {
# background-color: #f2f2f2;
# color: #333;
# text-align: left;
# }
# tr:hover {
# background-color: #f5f5f5;
# }
# .event-link {
# color: #1E90FF;
# text-decoration: none;
# }
# .event-link:hover {
# text-decoration: underline;
# }
# </style>
# <table>
# <tr>
# <th>Title</th>
# <th>Date and Time</th>
# <th>Location</th>
# </tr>
# """
# for event in events_results:
# title = event.get("title", "No title")
# date_info = event.get("date", {})
# date = f"{date_info.get('start_date', '')} {date_info.get('when', '')}".replace("{", "").replace("}", "")
# location = event.get("address", "No location")
# if isinstance(location, list):
# location = " ".join(location)
# location = location.replace("[", "").replace("]", "")
# link = event.get("link", "#")
# events_html += f"""
# <tr>
# <td><a class='event-link' href='{link}' target='_blank'>{title}</a></td>
# <td>{date}</td>
# <td>{location}</td>
# </tr>
# """
# events_html += "</table>"
# return events_html
# else:
# return "<p>Failed to fetch local events</p>"
# def get_weather_icon(condition):
# condition_map = {
# "Clear": "c01d",
# "Partly Cloudy": "c02d",
# "Cloudy": "c03d",
# "Overcast": "c04d",
# "Mist": "a01d",
# "Patchy rain possible": "r01d",
# "Light rain": "r02d",
# "Moderate rain": "r03d",
# "Heavy rain": "r04d",
# "Snow": "s01d",
# "Thunderstorm": "t01d",
# "Fog": "a05d",
# }
# return condition_map.get(condition, "c04d")
# def fetch_local_weather():
# try:
# api_key = os.environ['WEATHER_API']
# url = f'https://weather.visualcrossing.com/VisualCrossingWebServices/rest/services/timeline/birmingham?unitGroup=metric&include=events%2Calerts%2Chours%2Cdays%2Ccurrent&key={api_key}'
# response = requests.get(url)
# response.raise_for_status()
# jsonData = response.json()
# current_conditions = jsonData.get("currentConditions", {})
# temp_celsius = current_conditions.get("temp", "N/A")
# if temp_celsius != "N/A":
# temp_fahrenheit = int((temp_celsius * 9/5) + 32)
# else:
# temp_fahrenheit = "N/A"
# condition = current_conditions.get("conditions", "N/A")
# humidity = current_conditions.get("humidity", "N/A")
# weather_html = f"""
# <div class="weather-theme">
# <h2 style="font-family: 'Georgia', serif; color: #ff0000; background-color: #f8f8f8; padding: 10px; border-radius: 10px;">Local Weather</h2>
# <div class="weather-content">
# <div class="weather-icon">
# <img src="https://www.weatherbit.io/static/img/icons/{get_weather_icon(condition)}.png" alt="{condition}" style="width: 100px; height: 100px;">
# </div>
# <div class="weather-details">
# <p style="font-family: 'Verdana', sans-serif; color: #333; font-size: 1.2em;">Temperature: {temp_fahrenheit}°F</p>
# <p style="font-family: 'Verdana', sans-serif; color: #333; font-size: 1.2em;">Condition: {condition}</p>
# <p style="font-family: 'Verdana', sans-serif; color: #333; font-size: 1.2em;">Humidity: {humidity}%</p>
# </div>
# </div>
# </div>
# <style>
# .weather-theme {{
# animation: backgroundAnimation 10s infinite alternate;
# border-radius: 10px;
# padding: 10px;
# margin-bottom: 15px;
# background: linear-gradient(45deg, #ffcc33, #ff6666, #ffcc33, #ff6666);
# background-size: 400% 400%;
# box-shadow: 0 4px 8px rgba(0, 0, 0, 0.1);
# transition: box-shadow 0.3s ease, background-color 0.3s ease;
# }}
# .weather-theme:hover {{
# box-shadow: 0 8px 16px rgba(0, 0, 0, 0.2);
# background-position: 100% 100%;
# }}
# @keyframes backgroundAnimation {{
# 0% {{ background-position: 0% 50%; }}
# 100% {{ background-position: 100% 50%; }}
# }}
# .weather-content {{
# display: flex;
# align-items: center;
# }}
# .weather-icon {{
# flex: 1;
# }}
# .weather-details {{
# flex 3;
# }}
# </style>
# """
# return weather_html
# except requests.exceptions.RequestException as e:
# return f"<p>Failed to fetch local weather: {e}</p>"
# with gr.Blocks(theme='Pijush2023/scikit-learn-pijush') as demo:
# with gr.Row():
# with gr.Column():
# state = gr.State()
# chatbot = gr.Chatbot([], elem_id="RADAR:Channel 94.1", bubble_full_width=False)
# choice = gr.Radio(label="Select Style", choices=["Details", "Conversational"], value="Conversational")
# retrieval_mode = gr.Radio(label="Retrieval Mode", choices=["Vector", "Knowledge-Graph"], value="Vector")
# gr.Markdown("<h1 style='color: red;'>Talk to RADAR</h1>", elem_id="voice-markdown")
# chat_input = gr.Textbox(show_copy_button=True, interactive=True, show_label=False, label="ASK Radar !!!", placeholder="After Prompt,click Retriever Only")
# tts_choice = gr.Radio(label="Select TTS System", choices=["Alpha", "Beta", "Gamma"], value="Alpha")
# retriever_button = gr.Button("Retriever")
# clear_button = gr.Button("Clear")
# clear_button.click(lambda:[None,None] ,outputs=[chat_input, state])
# gr.Markdown("<h1 style='color: red;'>Radar Map</h1>", elem_id="Map-Radar")
# location_output = gr.HTML()
# # Define a single audio component
# audio_output = gr.Audio(interactive=False, autoplay=True)
# def stop_audio():
# audio_output.stop()
# return None
# # Define the sequence of actions for the "Retriever" button
# retriever_sequence = (
# retriever_button.click(fn=stop_audio, inputs=[], outputs=[audio_output], api_name="Ask_Retriever")
# .then(fn=add_message, inputs=[chatbot, chat_input], outputs=[chatbot, chat_input], api_name="voice_query")
# .then(fn=bot, inputs=[chatbot, choice, tts_choice, retrieval_mode], outputs=[chatbot, audio_output], api_name="generate_voice_response")
# .then(fn=show_map_if_details, inputs=[chatbot, choice], outputs=[location_output, location_output], api_name="map_finder")
# .then(fn=clear_textbox, inputs=[], outputs=[chat_input])
# )
# # Link the "Enter" key (submit event) to the same sequence of actions
# chat_input.submit(fn=stop_audio, inputs=[], outputs=[audio_output])
# chat_input.submit(fn=add_message, inputs=[chatbot, chat_input], outputs=[chatbot, chat_input], api_name="voice_query").then(
# fn=bot, inputs=[chatbot, choice, tts_choice, retrieval_mode], outputs=[chatbot, audio_output], api_name="generate_voice_response"
# ).then(
# fn=show_map_if_details, inputs=[chatbot, choice], outputs=[location_output, location_output], api_name="map_finder"
# ).then(
# fn=clear_textbox, inputs=[], outputs=[chat_input]
# )
# audio_input = gr.Audio(sources=["microphone"], streaming=True, type='numpy', every=0.1)
# audio_input.stream(transcribe_function, inputs=[state, audio_input], outputs=[state, chat_input], api_name="voice_query_to_text")
# #Api Integration to gradio call function
# # with gr.Column():
# # weather_output = gr.HTML(value=fetch_local_weather())
# # news_output = gr.HTML(value=fetch_local_news())
# # events_output = gr.HTML(value=fetch_local_events())
# with gr.Column():
# image_output_1 = gr.Image(value=generate_image(hardcoded_prompt_1), width=400, height=400)
# image_output_2 = gr.Image(value=generate_image(hardcoded_prompt_2), width=400, height=400)
# image_output_3 = gr.Image(value=generate_image(hardcoded_prompt_3), width=400, height=400)
# refresh_button = gr.Button("Refresh Images")
# refresh_button.click(fn=update_images, inputs=None, outputs=[image_output_1, image_output_2, image_output_3], api_name="update_image")
# demo.queue()
# demo.launch(share=True)
# Main code header Library
import gradio as gr
import requests
import os
import time
import re
import logging
import tempfile
import folium
import concurrent.futures
import torch
from PIL import Image
from datetime import datetime
from transformers import pipeline, AutoModelForSpeechSeq2Seq, AutoProcessor
from googlemaps import Client as GoogleMapsClient
from gtts import gTTS
from diffusers import StableDiffusionPipeline
from langchain_openai import OpenAIEmbeddings, ChatOpenAI
from langchain_pinecone import PineconeVectorStore
from langchain.prompts import PromptTemplate
from langchain.chains import RetrievalQA
from langchain.chains.conversation.memory import ConversationBufferWindowMemory
from huggingface_hub import login
from transformers.models.speecht5.number_normalizer import EnglishNumberNormalizer
from parler_tts import ParlerTTSForConditionalGeneration
from transformers import AutoTokenizer, AutoFeatureExtractor, set_seed
from scipy.io.wavfile import write as write_wav
from pydub import AudioSegment
from string import punctuation
import librosa
from pathlib import Path
import torchaudio
import numpy as np
# Neo4j imports
from langchain.chains import GraphCypherQAChain
from langchain_community.graphs import Neo4jGraph
from langchain_community.document_loaders import HuggingFaceDatasetLoader
from langchain_text_splitters import CharacterTextSplitter
from langchain_experimental.graph_transformers import LLMGraphTransformer
from langchain_core.prompts import ChatPromptTemplate
from langchain_core.pydantic_v1 import BaseModel, Field
from langchain_core.messages import AIMessage, HumanMessage
from langchain_core.output_parsers import StrOutputParser
from langchain_core.runnables import RunnableBranch, RunnableLambda, RunnableParallel, RunnablePassthrough
# Set environment variables for Torch- CUDA
os.environ['PYTORCH_USE_CUDA_DSA'] = '1'
os.environ['CUDA_LAUNCH_BLOCKING'] = '1'
# Hugging face token Initialization
hf_token = os.getenv("HF_TOKEN")
if hf_token is None:
print("Please set your Hugging Face token in the environment variables.")
else:
login(token=hf_token)
logging.basicConfig(level=logging.DEBUG)
# Embedding the vector with OpenAI
embeddings = OpenAIEmbeddings(api_key=os.environ['OPENAI_API_KEY'])
# Pinecone setup
from pinecone import Pinecone
pc = Pinecone(api_key=os.environ['PINECONE_API_KEY'])
index_name = "radardata07242024"
vectorstore = PineconeVectorStore(index_name=index_name, embedding=embeddings)
retriever = vectorstore.as_retriever(search_kwargs={'k': 5})
chat_model = ChatOpenAI(api_key=os.environ['OPENAI_API_KEY'], temperature=0, model='gpt-4o')
conversational_memory = ConversationBufferWindowMemory(
memory_key='chat_history',
k=10,
return_messages=True
)
# Prompt templates
def get_current_date():
return datetime.now().strftime("%B %d, %Y")
current_date = get_current_date()
template1 = f"""As an expert concierge in Birmingham, Alabama, known for being a helpful and renowned guide, I am here to assist you on this sunny bright day of {current_date}. Given the current weather conditions and date, I have access to a plethora of information regarding events, places, and activities in Birmingham that can enhance your experience.
If you have any questions or need recommendations, feel free to ask. I have a wealth of knowledge of perennial events in Birmingham and can provide detailed information to ensure you make the most of your time here. Remember, I am here to assist you in any way possible.
Now, let me guide you through some of the exciting events happening today in Birmingham, Alabama:
Address: >>, Birmingham, AL
Time: >>__
Date: >>__
Description: >>__
Address: >>, Birmingham, AL
Time: >>__
Date: >>__
Description: >>__
Address: >>, Birmingham, AL
Time: >>__
Date: >>__
Description: >>__
Address: >>, Birmingham, AL
Time: >>__
Date: >>__
Description: >>__
Address: >>, Birmingham, AL
Time: >>__
Date: >>__
Description: >>__
If you have any specific preferences or questions about these events or any other inquiries, please feel free to ask. Remember, I am here to ensure you have a memorable and enjoyable experience in Birmingham, AL.
It was my pleasure!
{{context}}
Question: {{question}}
Helpful Answer:"""
template2 = f"""As an expert concierge known for being helpful and a renowned guide for Birmingham, Alabama, I assist visitors in discovering the best that the city has to offer. Given today's sunny and bright weather on {current_date}, I am well-equipped to provide valuable insights and recommendations without revealing specific locations. I draw upon my extensive knowledge of the area, including perennial events and historical context.
In light of this, how can I assist you today? Feel free to ask any questions or seek recommendations for your day in Birmingham. If there's anything specific you'd like to know or experience, please share, and I'll be glad to help. Remember, keep the question concise for a quick and accurate response.
"It was my pleasure!"
{{context}}
Question: {{question}}
Helpful Answer:"""
# QA_Chain_templates
QA_CHAIN_PROMPT_1 = PromptTemplate(input_variables=["context", "question"], template=template1)
QA_CHAIN_PROMPT_2 = PromptTemplate(input_variables=["context", "question"], template=template2)
# Neo4j setup
graph = Neo4jGraph(
url="neo4j+s://98f45cc0.databases.neo4j.io",
username="neo4j",
password="B_sZbapCTZoQDWj1JrhwqElsNa-jm5Zq1m_mAnyPYog"
)
# Avoid pushing the graph documents to Neo4j every time
# Only push the documents once and comment the code below after the initial push
# dataset_name = "Pijush2023/birmindata07312024"
# page_content_column = 'events_description'
# loader = HuggingFaceDatasetLoader(dataset_name, page_content_column)
# data = loader.load()
# text_splitter = CharacterTextSplitter(chunk_size=100, chunk_overlap=50)
# documents = text_splitter.split_documents(data)
# llm_transformer = LLMGraphTransformer(llm=chat_model)
# graph_documents = llm_transformer.convert_to_graph_documents(documents)
# graph.add_graph_documents(graph_documents, baseEntityLabel=True, include_source=True)
class Entities(BaseModel):
names: list[str] = Field(..., description="All the person, organization, or business entities that appear in the text")
entity_prompt = ChatPromptTemplate.from_messages([
("system", "You are extracting organization and person entities from the text."),
("human", "Use the given format to extract information from the following input: {question}"),
])
entity_chain = entity_prompt | chat_model.with_structured_output(Entities)
# Remove Lucene Character
def remove_lucene_chars(input: str) -> str:
return input.translate(str.maketrans({"\\": r"\\", "+": r"\+", "-": r"\-", "&": r"\&", "|": r"\|", "!": r"\!",
"(": r"\(", ")": r"\)", "{": r"\{", "}": r"\}", "[": r"\[", "]": r"\]",
"^": r"\^", "~": r"\~", "*": r"\*", "?": r"\?", ":": r"\:", '"': r'\"',
";": r"\;", " ": r"\ "}))
# Full Text query Generator
def generate_full_text_query(input: str) -> str:
full_text_query = ""
words = [el for el in remove_lucene_chars(input).split() if el]
for word in words[:-1]:
full_text_query += f" {word}~2 AND"
full_text_query += f" {words[-1]}~2"
return full_text_query.strip()
# Neo4j Retrieval connection
def structured_retriever(question: str) -> str:
result = ""
entities = entity_chain.invoke({"question": question})
for entity in entities.names:
response = graph.query(
"""CALL db.index.fulltext.queryNodes('entity', $query, {limit:2})
YIELD node,score
CALL {
WITH node
MATCH (node)-[r:!MENTIONS]->(neighbor)
RETURN node.id + ' - ' + type(r) + ' -> ' + neighbor.id AS output
UNION ALL
WITH node
MATCH (node)<-[r:!MENTIONS]-(neighbor)
RETURN neighbor.id + ' - ' + type(r) + ' -> ' + node.id AS output
}
RETURN output LIMIT 50
""",
{"query": generate_full_text_query(entity)},
)
result += "\n".join([el['output'] for el in response])
return result
def retriever_neo4j(question: str):
structured_data = structured_retriever(question)
logging.debug(f"Structured data: {structured_data}")
return structured_data
_template = """Given the following conversation and a follow-up question, rephrase the follow-up question to be a standalone question,
in its original language.
Chat History:
{chat_history}
Follow Up Input: {question}
Standalone question:"""
CONDENSE_QUESTION_PROMPT = PromptTemplate.from_template(_template)
def _format_chat_history(chat_history: list[tuple[str, str]]) -> list:
buffer = []
for human, ai in chat_history:
buffer.append(HumanMessage(content=human))
buffer.append(AIMessage(content=ai))
return buffer
_search_query = RunnableBranch(
(
RunnableLambda(lambda x: bool(x.get("chat_history"))).with_config(
run_name="HasChatHistoryCheck"
),
RunnablePassthrough.assign(
chat_history=lambda x: _format_chat_history(x["chat_history"])
)
| CONDENSE_QUESTION_PROMPT
| ChatOpenAI(temperature=0, api_key=os.environ['OPENAI_API_KEY'])
| StrOutputParser(),
),
RunnableLambda(lambda x : x["question"]),
)
# Define conversational and detailed prompt templates for Neo4j responses
neo4j_conversational_template = f"""As an expert concierge known for being helpful and a renowned guide for Birmingham, Alabama, I assist visitors in discovering the best that the city has to offer. Given today's sunny and bright weather on {current_date}, I am well-equipped to provide valuable insights and recommendations without revealing specific locations. I draw upon my extensive knowledge of the area, including perennial events and historical context.
In light of this, how can I assist you today? Feel free to ask any questions or seek recommendations for your day in Birmingham. If there's anything specific you'd like to know or experience, please share, and I'll be glad to help. Remember, keep the question concise for a quick and accurate response.
"It was my pleasure!"
{{context}}
Question: {{question}}
Helpful Answer:"""
neo4j_details_template = f"""As an expert concierge in Birmingham, Alabama, known for being a helpful and renowned guide, I am here to assist you on this sunny bright day of {current_date}. Given the current weather conditions and date, I have access to a plethora of information regarding events, places, and activities in Birmingham that can enhance your experience.
If you have any questions or need recommendations, feel free to ask. I have a wealth of knowledge of perennial events in Birmingham and can provide detailed information to ensure you make the most of your time here. Remember, I am here to assist you in any way possible.
Now, let me guide you through some of the exciting events happening today in Birmingham, Alabama:
Address: >>, Birmingham, AL
Time: >>__
Date: >>__
Description: >>__
Address: >>, Birmingham, AL
Time: >>__
Date: >>__
Description: >>__
Address: >>, Birmingham, AL
Time: >>__
Date: >>__
Description: >>__
Address: >>, Birmingham, AL
Time: >>__
Date: >>__
Description: >>__
Address: >>, Birmingham, AL
Time: >>__
Date: >>__
Description: >>__
If you have any specific preferences or questions about these events or any other inquiries, please feel free to ask. Remember, I am here to ensure you have a memorable and enjoyable experience in Birmingham, AL.
It was my pleasure!
{{context}}
Question: {{question}}
Helpful Answer:"""
# Create prompt templates
QA_CHAIN_PROMPT_NEO4J_CONVERSATIONAL = PromptTemplate(input_variables=["context", "question"], template=neo4j_conversational_template)
QA_CHAIN_PROMPT_NEO4J_DETAILS = PromptTemplate(input_variables=["context", "question"], template=neo4j_details_template)
# Neo4j Retrieval chain for conversational mode
def neo4j_retrieval_conversational(question: str):
structured_data = structured_retriever(question)
logging.debug(f"Structured data (Conversational): {structured_data}")
prompt = QA_CHAIN_PROMPT_NEO4J_CONVERSATIONAL.format(context=structured_data, question=question)
response = chat_model({"query": prompt})
return response, []
# Neo4j Retrieval chain for detailed mode
def neo4j_retrieval_details(question: str):
structured_data = structured_retriever(question)
logging.debug(f"Structured data (Details): {structured_data}")
prompt = QA_CHAIN_PROMPT_NEO4J_DETAILS.format(context=structured_data, question=question)
response = chat_model({"query": prompt})
return response, extract_addresses(response)
# Update the generate_answer function to include Neo4j retrieval modes
def generate_answer(message, choice, retrieval_mode):
logging.debug(f"generate_answer called with choice: {choice} and retrieval_mode: {retrieval_mode}")
prompt_template = QA_CHAIN_PROMPT_1 if choice == "Details" else QA_CHAIN_PROMPT_2
if retrieval_mode == "Vector":
qa_chain = RetrievalQA.from_chain_type(
llm=chat_model,
chain_type="stuff",
retriever=retriever,
chain_type_kwargs={"prompt": prompt_template}
)
response = qa_chain({"query": message})
logging.debug(f"Vector response: {response}")
return response['result'], extract_addresses(response['result'])
elif retrieval_mode == "Knowledge-Graph":
if choice == "Details":
response, addresses = neo4j_retrieval_details(message)
else:
response, addresses = neo4j_retrieval_conversational(message)
logging.debug(f"Knowledge-Graph response: {response}")
return response, addresses
else:
return "Invalid retrieval mode selected.", []
# Full Text query Generator
def generate_full_text_query(input: str) -> str:
full_text_query = ""
words = [el for el in remove_lucene_chars(input).split() if el]
for word in words[:-1]:
full_text_query += f" {word}~2 AND"
full_text_query += f" {words[-1]}~2"
return full_text_query.strip()
# Rest of the code remains the same
def bot(history, choice, tts_choice, retrieval_mode):
if not history:
return history
response, addresses = generate_answer(history[-1][0], choice, retrieval_mode)
history[-1][1] = ""
with concurrent.futures.ThreadPoolExecutor() as executor:
if tts_choice == "Alpha":
audio_future = executor.submit(generate_audio_elevenlabs, response)
elif tts_choice == "Beta":
audio_future = executor.submit(generate_audio_parler_tts, response)
elif tts_choice == "Gamma":
audio_future = executor.submit(generate_audio_mars5, response)
for character in response:
history[-1][1] += character
time.sleep(0.05)
yield history, None
audio_path = audio_future.result()
yield history, audio_path
history.append([response, None]) # Ensure the response is added in the correct format
def add_message(history, message):
history.append((message, None))
return history, gr.Textbox(value="", interactive=True, placeholder="Enter message or upload file...", show_label=False)
def print_like_dislike(x: gr.LikeData):
print(x.index, x.value, x.liked)
def extract_addresses(response):
if not isinstance(response, str):
response = str(response)
address_patterns = [
r'([A-Z].*,\sBirmingham,\sAL\s\d{5})',
r'(\d{4}\s.*,\sBirmingham,\sAL\s\d{5})',
r'([A-Z].*,\sAL\s\d{5})',
r'([A-Z].*,.*\sSt,\sBirmingham,\sAL\s\d{5})',
r'([A-Z].*,.*\sStreets,\sBirmingham,\sAL\s\d{5})',
r'(\d{2}.*\sStreets)',
r'([A-Z].*\s\d{2},\sBirmingham,\sAL\s\d{5})',
r'([a-zA-Z]\s Birmingham)',
r'([a-zA-Z].*,\sBirmingham,\sAL)',
r'(^Birmingham,AL$)'
]
addresses = []
for pattern in address_patterns:
addresses.extend(re.findall(pattern, response))
return addresses
all_addresses = []
def generate_map(location_names):
global all_addresses
all_addresses.extend(location_names)
api_key = os.environ['GOOGLEMAPS_API_KEY']
gmaps = GoogleMapsClient(key=api_key)
m = folium.Map(location=[33.5175, -86.809444], zoom_start=12)
for location_name in all_addresses:
geocode_result = gmaps.geocode(location_name)
if geocode_result:
location = geocode_result[0]['geometry']['location']
folium.Marker(
[location['lat'], location['lng']],
tooltip=f"{geocode_result[0]['formatted_address']}"
).add_to(m)
map_html = m._repr_html_()
return map_html
def fetch_local_news():
api_key = os.environ['SERP_API']
url = f'https://serpapi.com/search.json?engine=google_news&q=birmingham headline&api_key={api_key}'
response = requests.get(url)
if response.status_code == 200:
results = response.json().get("news_results", [])
news_html = """
<h2 style="font-family: 'Georgia', serif; color: #ff0000; background-color: #f8f8f8; padding: 10px; border-radius: 10px;">Birmingham Today</h2>
<style>
.news-item {
font-family: 'Verdana', sans-serif;
color: #333;
background-color: #f0f8ff;
margin-bottom: 15px;
padding: 10px;
border-radius: 5px;
transition: box-shadow 0.3s ease, background-color 0.3s ease;
font-weight: bold;
}
.news-item:hover {
box-shadow: 0 4px 8px rgba(0, 0, 0, 0.1);
background-color: #e6f7ff;
}
.news-item a {
color: #1E90FF;
text-decoration: none;
font-weight: bold;
}
.news-item a:hover {
text-decoration: underline;
}
.news-preview {
position: absolute;
display: none;
border: 1px solid #ccc;
border-radius: 5px;
box-shadow: 0 2px 4px rgba(0, 0, 0, 0.2);
background-color: white;
z-index: 1000;
max-width: 300px;
padding: 10px;
font-family: 'Verdana', sans-serif;
color: #333;
}
</style>
<script>
function showPreview(event, previewContent) {
var previewBox = document.getElementById('news-preview');
previewBox.innerHTML = previewContent;
previewBox.style.left = event.pageX + 'px';
previewBox.style.top = event.pageY + 'px';
previewBox.style.display = 'block';
}
function hidePreview() {
var previewBox = document.getElementById('news-preview');
previewBox.style.display = 'none';
}
</script>
<div id="news-preview" class="news-preview"></div>
"""
for index, result in enumerate(results[:7]):
title = result.get("title", "No title")
link = result.get("link", "#")
snippet = result.get("snippet", "")
news_html += f"""
<div class="news-item" onmouseover="showPreview(event, '{snippet}')" onmouseout="hidePreview()">
<a href='{link}' target='_blank'>{index + 1}. {title}</a>
<p>{snippet}</p>
</div>
"""
return news_html
else:
return "<p>Failed to fetch local news</p>"
import numpy as np
import torch
from transformers import pipeline, AutoModelForSpeechSeq2Seq, AutoProcessor
model_id = 'openai/whisper-large-v3'
device = "cuda:0" if torch.cuda.is_available() else "cpu"
torch_dtype = torch.float16 if torch.cuda.is_available() else torch.float32
model = AutoModelForSpeechSeq2Seq.from_pretrained(model_id, torch_dtype=torch_dtype).to(device)
processor = AutoProcessor.from_pretrained(model_id)
pipe_asr = pipeline("automatic-speech-recognition", model=model, tokenizer=processor.tokenizer, feature_extractor=processor.feature_extractor, max_new_tokens=128, chunk_length_s=15, batch_size=16, torch_dtype=torch_dtype, device=device, return_timestamps=True)
base_audio_drive = "/data/audio"
def transcribe_function(stream, new_chunk):
try:
sr, y = new_chunk[0], new_chunk[1]
except TypeError:
print(f"Error chunk structure: {type(new_chunk)}, content: {new_chunk}")
return stream, "", None
y = y.astype(np.float32) / np.max(np.abs(y))
if stream is not None:
stream = np.concatenate([stream, y])
else:
stream = y
result = pipe_asr({"array": stream, "sampling_rate": sr}, return_timestamps=False)
full_text = result.get("text","")
return stream, full_text, result
def update_map_with_response(history):
if not history:
return ""
response = history[-1][1]
addresses = extract_addresses(response)
return generate_map(addresses)
def clear_textbox():
return ""
def show_map_if_details(history, choice):
if choice in ["Details", "Conversational"]:
return gr.update(visible=True), update_map_with_response(history)
else:
return gr.update(visible=False), ""
def generate_audio_elevenlabs(text):
XI_API_KEY = os.environ['ELEVENLABS_API']
VOICE_ID = 'd9MIrwLnvDeH7aZb61E9'
tts_url = f"https://api.elevenlabs.io/v1/text-to-speech/{VOICE_ID}/stream"
headers = {
"Accept": "application/json",
"xi-api-key": XI_API_KEY
}
data = {
"text": str(text),
"model_id": "eleven_multilingual_v2",
"voice_settings": {
"stability": 1.0,
"similarity_boost": 0.0,
"style": 0.60,
"use_speaker_boost": False
}
}
response = requests.post(tts_url, headers=headers, json=data, stream=True)
if response.ok:
audio_segments = []
with tempfile.NamedTemporaryFile(delete=False, suffix=".mp3") as f:
for chunk in response.iter_content(chunk_size=1024):
if chunk:
f.write(chunk)
audio_segments.append(chunk)
temp_audio_path = f.name
# Combine all audio chunks into a single file
combined_audio = AudioSegment.from_file(temp_audio_path, format="mp3")
combined_audio_path = os.path.join(tempfile.gettempdir(), "elevenlabs_combined_audio.mp3")
combined_audio.export(combined_audio_path, format="mp3")
logging.debug(f"Audio saved to {combined_audio_path}")
return combined_audio_path
else:
logging.error(f"Error generating audio: {response.text}")
return None
repo_id = "parler-tts/parler-tts-mini-expresso"
parler_model = ParlerTTSForConditionalGeneration.from_pretrained(repo_id).to(device)
parler_tokenizer = AutoTokenizer.from_pretrained(repo_id)
parler_feature_extractor = AutoFeatureExtractor.from_pretrained(repo_id)
SAMPLE_RATE = parler_feature_extractor.sampling_rate
SEED = 42
def preprocess(text):
number_normalizer = EnglishNumberNormalizer()
text = number_normalizer(text).strip()
if text[-1] not in punctuation:
text = f"{text}."
abbreviations_pattern = r'\b[A-Z][A-Z\.]+\b'
def separate_abb(chunk):
chunk = chunk.replace(".", "")
return " ".join(chunk)
abbreviations = re.findall(abbreviations_pattern, text)
for abv in abbreviations:
if abv in text:
text is text.replace(abv, separate_abb(abv))
return text
def chunk_text(text, max_length=250):
words = text.split()
chunks = []
current_chunk = []
current_length = 0
for word in words:
if current_length + len(word) + 1 <= max_length:
current_chunk.append(word)
current_length += len(word) + 1
else:
chunks.append(' '.join(current_chunk))
current_chunk = [word]
current_length = len(word) + 1
if current_chunk:
chunks.append(' '.join(current_chunk))
return chunks
def generate_audio_parler_tts(text):
description = "Thomas speaks with emphasis and excitement at a moderate pace with high quality."
chunks = chunk_text(preprocess(text))
audio_segments = []
for chunk in chunks:
inputs = parler_tokenizer(description, return_tensors="pt").to(device)
prompt = parler_tokenizer(chunk, return_tensors="pt").to(device)
set_seed(SEED)
generation = parler_model.generate(input_ids=inputs.input_ids, prompt_input_ids=prompt.input_ids)
audio_arr = generation.cpu().numpy().squeeze()
temp_audio_path = os.path.join(tempfile.gettempdir(), f"parler_tts_audio_{len(audio_segments)}.wav")
write_wav(temp_audio_path, SAMPLE_RATE, audio_arr)
audio_segments.append(AudioSegment.from_wav(temp_audio_path))
combined_audio = sum(audio_segments)
combined_audio_path = os.path.join(tempfile.gettempdir(), "parler_tts_combined_audio.wav")
combined_audio.export(combined_audio_path, format="wav")
logging.debug(f"Audio saved to {combined_audio_path}")
return combined_audio_path
# Load the MARS5 model
mars5, config_class = torch.hub.load('Camb-ai/mars5-tts', 'mars5_english', trust_repo=True)
def generate_audio_mars5(text):
description = "Thomas speaks with emphasis and excitement at a moderate pace with high quality."
kwargs_dict = {
'temperature': 0.2,
'top_k': -1,
'top_p': 0.2,
'typical_p': 1.0,
'freq_penalty': 2.6,
'presence_penalty': 0.4,
'rep_penalty_window': 100,
'max_prompt_phones': 360,
'deep_clone': True,
'nar_guidance_w': 3
}
chunks = chunk_text(preprocess(text))
audio_segments = []
for chunk in chunks:
wav = torch.zeros(1, mars5.sr) # Use a placeholder silent audio for the reference
cfg = config_class(**{k: kwargs_dict[k] for k in kwargs_dict if k in config_class.__dataclass_fields__})
ar_codes, wav_out = mars5.tts(chunk, wav, "", cfg=cfg)
temp_audio_path = os.path.join(tempfile.gettempdir(), f"mars5_audio_{len(audio_segments)}.wav")
torchaudio.save(temp_audio_path, wav_out.unsqueeze(0), mars5.sr)
audio_segments.append(AudioSegment.from_wav(temp_audio_path))
combined_audio = sum(audio_segments)
combined_audio_path = os.path.join(tempfile.gettempdir(), "mars5_combined_audio.wav")
combined_audio.export(combined_audio_path, format="wav")
logging.debug(f"Audio saved to {combined_audio_path}")
return combined_audio_path
pipe = StableDiffusionPipeline.from_pretrained("stabilityai/stable-diffusion-2", torch_dtype=torch.float16)
pipe.to(device)
def generate_image(prompt):
with torch.cuda.amp.autocast():
image = pipe(
prompt,
num_inference_steps=28,
guidance_scale=3.0,
).images[0]
return image
hardcoded_prompt_1 = "Give a high quality photograph of a great looking red 2026 Toyota coupe against a skyline setting in the night, michael mann style in omaha enticing the consumer to buy this product"
hardcoded_prompt_2 = "A vibrant and dynamic football game scene in the style of Peter Paul Rubens, showcasing the intense match between Alabama and Nebraska. The players are depicted with the dramatic, muscular physiques and expressive faces typical of Rubens' style. The Alabama team is wearing their iconic crimson and white uniforms, while the Nebraska team is in their classic red and white attire. The scene is filled with action, with players in mid-motion, tackling, running, and catching the ball. The background features a grand stadium filled with cheering fans, banners, and the natural landscape in the distance. The colors are rich and vibrant, with a strong use of light and shadow to create depth and drama. The overall atmosphere captures the intensity and excitement of the game, infused with the grandeur and dynamism characteristic of Rubens' work."
hardcoded_prompt_3 = "Create a high-energy scene of a DJ performing on a large stage with vibrant lights, colorful lasers, a lively dancing crowd, and various electronic equipment in the background."
def update_images():
image_1 = generate_image(hardcoded_prompt_1)
image_2 = generate_image(hardcoded_prompt_2)
image_3 = generate_image(hardcoded_prompt_3)
return image_1, image_2, image_3
def fetch_local_events():
api_key = os.environ['SERP_API']
url = f'https://serpapi.com/search.json?engine=google_events&q=Events+in+Birmingham&hl=en&gl=us&api_key={api_key}'
response = requests.get(url)
if response.status_code == 200:
events_results = response.json().get("events_results", [])
events_html = """
<h2 style="font-family: 'Georgia', serif; color: #ff0000; background-color: #f8f8f8; padding: 10px; border-radius: 10px;">Local Events</h2>
<style>
table {
font-family: 'Verdana', sans-serif;
color: #333;
border-collapse: collapse;
width: 100%;
}
th, td {
border: 1px solid #fff !important;
padding: 8px;
}
th {
background-color: #f2f2f2;
color: #333;
text-align: left;
}
tr:hover {
background-color: #f5f5f5;
}
.event-link {
color: #1E90FF;
text-decoration: none;
}
.event-link:hover {
text-decoration: underline;
}
</style>
<table>
<tr>
<th>Title</th>
<th>Date and Time</th>
<th>Location</th>
</tr>
"""
for event in events_results:
title = event.get("title", "No title")
date_info = event.get("date", {})
date = f"{date_info.get('start_date', '')} {date_info.get('when', '')}".replace("{", "").replace("}", "")
location = event.get("address", "No location")
if isinstance(location, list):
location = " ".join(location)
location = location.replace("[", "").replace("]", "")
link = event.get("link", "#")
events_html += f"""
<tr>
<td><a class='event-link' href='{link}' target='_blank'>{title}</a></td>
<td>{date}</td>
<td>{location}</td>
</tr>
"""
events_html += "</table>"
return events_html
else:
return "<p>Failed to fetch local events</p>"
def get_weather_icon(condition):
condition_map = {
"Clear": "c01d",
"Partly Cloudy": "c02d",
"Cloudy": "c03d",
"Overcast": "c04d",
"Mist": "a01d",
"Patchy rain possible": "r01d",
"Light rain": "r02d",
"Moderate rain": "r03d",
"Heavy rain": "r04d",
"Snow": "s01d",
"Thunderstorm": "t01d",
"Fog": "a05d",
}
return condition_map.get(condition, "c04d")
def fetch_local_weather():
try:
api_key = os.environ['WEATHER_API']
url = f'https://weather.visualcrossing.com/VisualCrossingWebServices/rest/services/timeline/birmingham?unitGroup=metric&include=events%2Calerts%2Chours%2Cdays%2Ccurrent&key={api_key}'
response = requests.get(url)
response.raise_for_status()
jsonData = response.json()
current_conditions = jsonData.get("currentConditions", {})
temp_celsius = current_conditions.get("temp", "N/A")
if temp_celsius != "N/A":
temp_fahrenheit = int((temp_celsius * 9/5) + 32)
else:
temp_fahrenheit = "N/A"
condition = current_conditions.get("conditions", "N/A")
humidity = current_conditions.get("humidity", "N/A")
weather_html = f"""
<div class="weather-theme">
<h2 style="font-family: 'Georgia', serif; color: #ff0000; background-color: #f8f8f8; padding: 10px; border-radius: 10px;">Local Weather</h2>
<div class="weather-content">
<div class="weather-icon">
<img src="https://www.weatherbit.io/static/img/icons/{get_weather_icon(condition)}.png" alt="{condition}" style="width: 100px; height: 100px;">
</div>
<div class="weather-details">
<p style="font-family: 'Verdana', sans-serif; color: #333; font-size: 1.2em;">Temperature: {temp_fahrenheit}°F</p>
<p style="font-family: 'Verdana', sans-serif; color: #333; font-size: 1.2em;">Condition: {condition}</p>
<p style="font-family: 'Verdana', sans-serif; color: #333; font-size: 1.2em;">Humidity: {humidity}%</p>
</div>
</div>
</div>
<style>
.weather-theme {{
animation: backgroundAnimation 10s infinite alternate;
border-radius: 10px;
padding: 10px;
margin-bottom: 15px;
background: linear-gradient(45deg, #ffcc33, #ff6666, #ffcc33, #ff6666);
background-size: 400% 400%;
box-shadow: 0 4px 8px rgba(0, 0, 0, 0.1);
transition: box-shadow 0.3s ease, background-color 0.3s ease;
}}
.weather-theme:hover {{
box-shadow: 0 8px 16px rgba(0, 0, 0, 0.2);
background-position: 100% 100%;
}}
@keyframes backgroundAnimation {{
0% {{ background-position: 0% 50%; }}
100% {{ background-position: 100% 50%; }}
}}
.weather-content {{
display: flex;
align-items: center;
}}
.weather-icon {{
flex: 1;
}}
.weather-details {{
flex 3;
}}
</style>
"""
return weather_html
except requests.exceptions.RequestException as e:
return f"<p>Failed to fetch local weather: {e}</p>"
with gr.Blocks(theme='Pijush2023/scikit-learn-pijush') as demo:
with gr.Row():
with gr.Column():
state = gr.State()
chatbot = gr.Chatbot([], elem_id="RADAR:Channel 94.1", bubble_full_width=False)
choice = gr.Radio(label="Select Style", choices=["Details", "Conversational"], value="Conversational")
retrieval_mode = gr.Radio(label="Retrieval Mode", choices=["Vector", "Knowledge-Graph"], value="Vector")
gr.Markdown("<h1 style='color: red;'>Talk to RADAR</h1>", elem_id="voice-markdown")
chat_input = gr.Textbox(show_copy_button=True, interactive=True, show_label=False, label="ASK Radar !!!", placeholder="After Prompt,click Retriever Only")
tts_choice = gr.Radio(label="Select TTS System", choices=["Alpha", "Beta", "Gamma"], value="Alpha")
retriever_button = gr.Button("Retriever")
clear_button = gr.Button("Clear")
clear_button.click(lambda:[None,None] ,outputs=[chat_input, state])
gr.Markdown("<h1 style='color: red;'>Radar Map</h1>", elem_id="Map-Radar")
location_output = gr.HTML()
# Define a single audio component
audio_output = gr.Audio(interactive=False, autoplay=True)
def stop_audio():
audio_output.stop()
return None
# Define the sequence of actions for the "Retriever" button
retriever_sequence = (
retriever_button.click(fn=stop_audio, inputs=[], outputs=[audio_output], api_name="Ask_Retriever")
.then(fn=add_message, inputs=[chatbot, chat_input], outputs=[chatbot, chat_input], api_name="voice_query")
.then(fn=bot, inputs=[chatbot, choice, tts_choice, retrieval_mode], outputs=[chatbot, audio_output], api_name="generate_voice_response")
.then(fn=show_map_if_details, inputs=[chatbot, choice], outputs=[location_output, location_output], api_name="map_finder")
.then(fn=clear_textbox, inputs=[], outputs=[chat_input])
)
# Link the "Enter" key (submit event) to the same sequence of actions
chat_input.submit(fn=stop_audio, inputs=[], outputs=[audio_output])
chat_input.submit(fn=add_message, inputs=[chatbot, chat_input], outputs=[chatbot, chat_input], api_name="voice_query").then(
fn=bot, inputs=[chatbot, choice, tts_choice, retrieval_mode], outputs=[chatbot, audio_output], api_name="generate_voice_response"
).then(
fn=show_map_if_details, inputs=[chatbot, choice], outputs=[location_output, location_output], api_name="map_finder"
).then(
fn=clear_textbox, inputs=[], outputs=[chat_input]
)
audio_input = gr.Audio(sources=["microphone"], streaming=True, type='numpy', every=0.1)
audio_input.stream(transcribe_function, inputs=[state, audio_input], outputs=[state, chat_input], api_name="voice_query_to_text")
# Api Integration to gradio call function
# with gr.Column():
# weather_output = gr.HTML(value=fetch_local_weather())
# news_output = gr.HTML(value=fetch_local_news())
# events_output = gr.HTML(value=fetch_local_events())
with gr.Column():
image_output_1 = gr.Image(value=generate_image(hardcoded_prompt_1), width=400, height=400)
image_output_2 = gr.Image(value=generate_image(hardcoded_prompt_2), width=400, height=400)
image_output_3 = gr.Image(value=generate_image(hardcoded_prompt_3), width=400, height=400)
refresh_button = gr.Button("Refresh Images")
refresh_button.click(fn=update_images, inputs=None, outputs=[image_output_1, image_output_2, image_output_3], api_name="update_image")
demo.queue()
demo.launch(share=True)