Date: {date}
Location: {location}
Failed to fetch local events
" # 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""" #}.png){kind=icon}
# Temperature: {temp_fahrenheit}°F
#Condition: {condition}
#Humidity: {humidity}%
#Failed to fetch local weather: {e}
" # 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") # template1 = """You are an expert concierge who is helpful and a renowned guide for Birmingham,Alabama. Based on weather being a sunny bright day and the today's date is 1st july 2024, use the following pieces of context, # memory, and message history, along with your knowledge of perennial events in Birmingham,Alabama, to answer the question at the end. If you don't know the answer, just say "Homie, I need to get more data for this," and don't try to make up an answer. # Use fifteen sentences maximum. Keep the answer as detailed as possible. Always include the address, time, date, and # event type and description. Always say "It was my pleasure!" at the end of the answer. # {context} # Question: {question} # Helpful Answer:""" # template2 = """You are an expert concierge who is helpful and a renowned guide for Birmingham,Alabama. Based on today's weather being a sunny bright day and today's date is 1st july 2024, take the location or address but don't show the location or address on the output prompts. Use the following pieces of context, # memory, and message history, along with your knowledge of perennial events in Birmingham,Alabama, to answer the question at the end. If you don't know the answer, just say "Homie, I need to get more data for this," and don't try to make up an answer. # Keep the answer short and sweet and crisp. Always say "It was my pleasure!" at the end of the answer. # {context} # Question: {question} # Helpful Answer:""" # QA_CHAIN_PROMPT_1 = PromptTemplate(input_variables=["context", "question"], template=template1) # QA_CHAIN_PROMPT_2 = PromptTemplate(input_variables=["context", "question"], template=template2) # def build_qa_chain(prompt_template): # qa_chain = RetrievalQA.from_chain_type( # llm=chat_model, # chain_type="stuff", # retriever=retriever, # chain_type_kwargs={"prompt": prompt_template} # ) # tools = [ # Tool( # name='Knowledge Base', # func=qa_chain, # description='Use this tool when answering general knowledge queries to get more information about the topic' # ) # ] # return qa_chain, tools # def initialize_agent_with_prompt(prompt_template): # qa_chain, tools = build_qa_chain(prompt_template) # agent = initialize_agent( # agent='chat-conversational-react-description', # tools=tools, # llm=chat_model, # verbose=False, # max_iteration=5, # early_stopping_method='generate', # memory=conversational_memory # ) # return agent # def generate_answer(message, choice): # logging.debug(f"generate_answer called with prompt_choice: {choice}") # if choice == "Details": # agent = initialize_agent_with_prompt(QA_CHAIN_PROMPT_1) # elif choice == "Conversational": # agent = initialize_agent_with_prompt(QA_CHAIN_PROMPT_2) # else: # logging.error(f"Invalid prompt_choice: {choice}. Defaulting to 'Conversational'") # agent = initialize_agent_with_prompt(QA_CHAIN_PROMPT_2) # response = agent(message) # addresses = extract_addresses(response['output']) # return response['output'], addresses # def bot(history, choice, tts_choice): # if not history: # return history # response, addresses = generate_answer(history[-1][0], choice) # history[-1][1] = "" # with concurrent.futures.ThreadPoolExecutor() as executor: # if tts_choice == "Eleven Labs": # audio_future = executor.submit(generate_audio_elevenlabs, response) # elif tts_choice == "Parler-TTS": # audio_future = executor.submit(generate_audio_parler_tts, response) # elif tts_choice == "MARS5": # 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 # 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)' # ] # 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=16) # 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'], '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 = """ #{snippet}
#Failed to fetch local news
" # 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: # with tempfile.NamedTemporaryFile(delete=False, suffix=".mp3") as f: # for chunk in response.iter_content(chunk_size=1024): # f.write(chunk) # temp_audio_path = f.name # logging.debug(f"Audio saved to {temp_audio_path}") # return temp_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 = 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': 0.8, # '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 Bentley 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 # 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") # gr.Markdown("Date: {date}
Location: {location}
Failed to fetch local events
" 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"""
Temperature: {temp_fahrenheit}°F
Condition: {condition}
Humidity: {humidity}%
Failed to fetch local weather: {e}
" 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") template1 = """You are an expert concierge who is helpful and a renowned guide for Birmingham,Alabama. Based on weather being a sunny bright day and the today's date is 1st july 2024, use the following pieces of context, memory, and message history, along with your knowledge of perennial events in Birmingham,Alabama, to answer the question at the end. If you don't know the answer, just say "Homie, I need to get more data for this," and don't try to make up an answer. Use fifteen sentences maximum. Keep the answer as detailed as possible. Always include the address, time, date, and event type and description. Always say "It was my pleasure!" at the end of the answer. {context} Question: {question} Helpful Answer:""" template2 = """You are an expert concierge who is helpful and a renowned guide for Birmingham,Alabama. Based on today's weather being a sunny bright day and today's date is 1st july 2024, take the location or address but don't show the location or address on the output prompts. Use the following pieces of context, memory, and message history, along with your knowledge of perennial events in Birmingham,Alabama, to answer the question at the end. If you don't know the answer, just say "Homie, I need to get more data for this," and don't try to make up an answer. Keep the answer short and sweet and crisp. Always say "It was my pleasure!" at the end of the answer. {context} Question: {question} Helpful Answer:""" QA_CHAIN_PROMPT_1 = PromptTemplate(input_variables=["context", "question"], template=template1) QA_CHAIN_PROMPT_2 = PromptTemplate(input_variables=["context", "question"], template=template2) def build_qa_chain(prompt_template): qa_chain = RetrievalQA.from_chain_type( llm=chat_model, chain_type="stuff", retriever=retriever, chain_type_kwargs={"prompt": prompt_template} ) tools = [ Tool( name='Knowledge Base', func=qa_chain, description='Use this tool when answering general knowledge queries to get more information about the topic' ) ] return qa_chain, tools def initialize_agent_with_prompt(prompt_template): qa_chain, tools = build_qa_chain(prompt_template) agent = initialize_agent( agent='chat-conversational-react-description', tools=tools, llm=chat_model, verbose=False, max_iteration=5, early_stopping_method='generate', memory=conversational_memory ) return agent def generate_answer(message, choice): logging.debug(f"generate_answer called with prompt_choice: {choice}") if choice == "Details": agent = initialize_agent_with_prompt(QA_CHAIN_PROMPT_1) elif choice == "Conversational": agent = initialize_agent_with_prompt(QA_CHAIN_PROMPT_2) else: logging.error(f"Invalid prompt_choice: {choice}. Defaulting to 'Conversational'") agent = initialize_agent_with_prompt(QA_CHAIN_PROMPT_2) response = agent(message) addresses = extract_addresses(response['output']) return response['output'], addresses def bot(history, choice, tts_choice): if not history: return history response, addresses = generate_answer(history[-1][0], choice) history[-1][1] = "" with concurrent.futures.ThreadPoolExecutor() as executor: if tts_choice == "Eleven Labs": audio_future = executor.submit(generate_audio_elevenlabs, response) elif tts_choice == "Parler-TTS": audio_future = executor.submit(generate_audio_parler_tts, response) elif tts_choice == "MARS5": audio_future = executor.submit(generate_audio_mars5, response) elif tts_choice == "Coqui": audio_future = executor.submit(generate_audio_coqui, 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 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)' ] 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=16) 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 = """{snippet}
Failed to fetch local news
" 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: with tempfile.NamedTemporaryFile(delete=False, suffix=".mp3") as f: for chunk in response.iter_content(chunk_size=1024): f.write(chunk) temp_audio_path = f.name logging.debug(f"Audio saved to {temp_audio_path}") return temp_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 = 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': 0.8, '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 # Initialize Coqui XTTS os.system('python -m unidic download') os.environ["COQUI_TOS_AGREED"] = "1" from TTS.api import TTS from TTS.tts.configs.xtts_config import XttsConfig from TTS.tts.models.xtts import Xtts from TTS.utils.generic_utils import get_user_data_dir from huggingface_hub import HfApi api = HfApi(token=hf_token) repo_id = "coqui/xtts" model_name = "tts_models/multilingual/multi-dataset/xtts_v2" ModelManager().download_model(model_name) model_path = os.path.join(get_user_data_dir("tts"), model_name.replace("/", "--")) config = XttsConfig() config.load_json(os.path.join(model_path, "config.json")) model = Xtts.init_from_config(config) model.load_checkpoint( config, checkpoint_path=os.path.join(model_path, "model.pth"), vocab_path=os.path.join(model_path, "vocab.json"), eval=True, use_deepspeed=True, ) model.cuda() def generate_audio_coqui(text): language = "en" use_mic = False voice_cleanup = False no_lang_auto_detect = False agree = True mic_file_path = None audio_file_pth = None if agree: if language not in config.languages: raise Exception("Language not supported") if use_mic: if mic_file_path is not None: speaker_wav = mic_file_path else: raise Exception("Microphone input required") else: speaker_wav = audio_file_pth lowpass_highpass = "lowpass=8000,highpass=75," if voice_cleanup else "" trim_silence = "areverse,silenceremove=start_periods=1:start_silence=0:start_threshold=0.02,areverse,silenceremove=start_periods=1:start_silence=0:start_threshold=0.02," if voice_cleanup else "" out_filename = speaker_wav + str(uuid.uuid4()) + ".wav" if voice_cleanup else speaker_wav if voice_cleanup: shell_command = f"./ffmpeg -y -i {speaker_wav} -af {lowpass_highpass}{trim_silence} {out_filename}".split(" ") subprocess.run([item for item in shell_command], capture_output=False, text=True, check=True) speaker_wav = out_filename t_latent = time.time() gpt_cond_latent, speaker_embedding = model.get_conditioning_latents(audio_path=speaker_wav, gpt_cond_len=30, gpt_cond_chunk_len=4, max_ref_length=60) latent_calculation_time = time.time() - t_latent prompt = re.sub("([^\x00-\x7F]|\w)(\.|\。|\?)", r"\1 \2\2", text) out = model.inference(prompt, language, gpt_cond_latent, speaker_embedding, repetition_penalty=5.0, temperature=0.75) torchaudio.save("output.wav", torch.tensor(out["wav"]).unsqueeze(0), 24000) return "output.wav" else: raise Exception("Terms & Conditions not accepted") 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 Bentley 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 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") gr.Markdown("Date: {date}
Location: {location}
Failed to fetch local events
" # 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""" #
# Temperature: {temp_fahrenheit}°F
#Condition: {condition}
#Humidity: {humidity}%
#Failed to fetch local weather: {e}
" # 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") # template1 = """You are an expert concierge who is helpful and a renowned guide for Birmingham,Alabama. Based on weather being a sunny bright day and the today's date is 1st july 2024, use the following pieces of context, # memory, and message history, along with your knowledge of perennial events in Birmingham,Alabama, to answer the question at the end. If you don't know the answer, just say "Homie, I need to get more data for this," and don't try to make up an answer. # Use fifteen sentences maximum. Keep the answer as detailed as possible. Always include the address, time, date, and # event type and description. Always say "It was my pleasure!" at the end of the answer. # {context} # Question: {question} # Helpful Answer:""" # template2 = """You are an expert concierge who is helpful and a renowned guide for Birmingham,Alabama. Based on today's weather being a sunny bright day and today's date is 1st july 2024, take the location or address but don't show the location or address on the output prompts. Use the following pieces of context, # memory, and message history, along with your knowledge of perennial events in Birmingham,Alabama, to answer the question at the end. If you don't know the answer, just say "Homie, I need to get more data for this," and don't try to make up an answer. # Keep the answer short and sweet and crisp. Always say "It was my pleasure!" at the end of the answer. # {context} # Question: {question} # Helpful Answer:""" # QA_CHAIN_PROMPT_1 = PromptTemplate(input_variables=["context", "question"], template=template1) # QA_CHAIN_PROMPT_2 = PromptTemplate(input_variables=["context", "question"], template=template2) # def build_qa_chain(prompt_template): # qa_chain = RetrievalQA.from_chain_type( # llm=chat_model, # chain_type="stuff", # retriever=retriever, # chain_type_kwargs={"prompt": prompt_template} # ) # tools = [ # Tool( # name='Knowledge Base', # func=qa_chain, # description='Use this tool when answering general knowledge queries to get more information about the topic' # ) # ] # return qa_chain, tools # def initialize_agent_with_prompt(prompt_template): # qa_chain, tools = build_qa_chain(prompt_template) # agent = initialize_agent( # agent='chat-conversational-react-description', # tools=tools, # llm=chat_model, # verbose=False, # max_iteration=5, # early_stopping_method='generate', # memory=conversational_memory # ) # return agent # def generate_answer(message, choice): # logging.debug(f"generate_answer called with prompt_choice: {choice}") # if choice == "Details": # agent = initialize_agent_with_prompt(QA_CHAIN_PROMPT_1) # elif choice == "Conversational": # agent = initialize_agent_with_prompt(QA_CHAIN_PROMPT_2) # else: # logging.error(f"Invalid prompt_choice: {choice}. Defaulting to 'Conversational'") # agent = initialize_agent_with_prompt(QA_CHAIN_PROMPT_2) # response = agent(message) # addresses = extract_addresses(response['output']) # return response['output'], addresses # def bot(history, choice, tts_choice): # if not history: # return history # response, addresses = generate_answer(history[-1][0], choice) # history[-1][1] = "" # with concurrent.futures.ThreadPoolExecutor() as executor: # if tts_choice == "Eleven Labs": # audio_future = executor.submit(generate_audio_elevenlabs, response) # elif tts_choice == "Parler-TTS": # audio_future = executor.submit(generate_audio_parler_tts, response) # elif tts_choice == "MARS5": # 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 # 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)' # ] # 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=16) # 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 = """ #{snippet}
#Failed to fetch local news
" # 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: # with tempfile.NamedTemporaryFile(delete=False, suffix=".mp3") as f: # for chunk in response.iter_content(chunk_size=1024): # f.write(chunk) # temp_audio_path = f.name # logging.debug(f"Audio saved to {temp_audio_path}") # return temp_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 = 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': 0.8, # '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 Bentley 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 # 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") # gr.Markdown("Date: {date}
Location: {location}
Failed to fetch local events
" # 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""" #
# Temperature: {temp_fahrenheit}°F
#Condition: {condition}
#Humidity: {humidity}%
#Failed to fetch local weather: {e}
" # 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") # template1 = """You are an expert concierge who is helpful and a renowned guide for Birmingham,Alabama. Based on weather being a sunny bright day and the today's date is 1st july 2024, use the following pieces of context, # memory, and message history, along with your knowledge of perennial events in Birmingham,Alabama, to answer the question at the end. If you don't know the answer, just say "Homie, I need to get more data for this," and don't try to make up an answer. # Use fifteen sentences maximum. Keep the answer as detailed as possible. Always include the address, time, date, and # event type and description. Always say "It was my pleasure!" at the end of the answer. # {context} # Question: {question} # Helpful Answer:""" # template2 = """You are an expert concierge who is helpful and a renowned guide for Birmingham,Alabama. Based on today's weather being a sunny bright day and today's date is 1st july 2024, take the location or address but don't show the location or address on the output prompts. Use the following pieces of context, # memory, and message history, along with your knowledge of perennial events in Birmingham,Alabama, to answer the question at the end. If you don't know the answer, just say "Homie, I need to get more data for this," and don't try to make up an answer. # Keep the answer short and sweet and crisp. Always say "It was my pleasure!" at the end of the answer. # {context} # Question: {question} # Helpful Answer:""" # QA_CHAIN_PROMPT_1 = PromptTemplate(input_variables=["context", "question"], template=template1) # QA_CHAIN_PROMPT_2 = PromptTemplate(input_variables=["context", "question"], template=template2) # def build_qa_chain(prompt_template): # qa_chain = RetrievalQA.from_chain_type( # llm=chat_model, # chain_type="stuff", # retriever=retriever, # chain_type_kwargs={"prompt": prompt_template} # ) # tools = [ # Tool( # name='Knowledge Base', # func=qa_chain, # description='Use this tool when answering general knowledge queries to get more information about the topic' # ) # ] # return qa_chain, tools # def initialize_agent_with_prompt(prompt_template): # qa_chain, tools = build_qa_chain(prompt_template) # agent = initialize_agent( # agent='chat-conversational-react-description', # tools=tools, # llm=chat_model, # verbose=False, # max_iteration=5, # early_stopping_method='generate', # memory=conversational_memory # ) # return agent # def generate_answer(message, choice): # logging.debug(f"generate_answer called with prompt_choice: {choice}") # if choice == "Details": # agent = initialize_agent_with_prompt(QA_CHAIN_PROMPT_1) # elif choice == "Conversational": # agent = initialize_agent_with_prompt(QA_CHAIN_PROMPT_2) # else: # logging.error(f"Invalid prompt_choice: {choice}. Defaulting to 'Conversational'") # agent = initialize_agent_with_prompt(QA_CHAIN_PROMPT_2) # response = agent(message) # addresses = extract_addresses(response['output']) # return response['output'], addresses # def bot(history, choice, tts_choice): # if not history: # return history # response, addresses = generate_answer(history[-1][0], choice) # history[-1][1] = "" # with concurrent.futures.ThreadPoolExecutor() as executor: # if tts_choice == "Eleven Labs": # audio_future = executor.submit(generate_audio_elevenlabs, response) # elif tts_choice == "Parler-TTS": # audio_future = executor.submit(generate_audio_parler_tts, 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 # 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)' # ] # 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=16) # 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 = """ #{snippet}
#Failed to fetch local news
" # 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: # with tempfile.NamedTemporaryFile(delete=False, suffix=".mp3") as f: # for chunk in response.iter_content(chunk_size=1024): # f.write(chunk) # temp_audio_path = f.name # logging.debug(f"Audio saved to {temp_audio_path}") # return temp_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 = 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 # 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 Bentley 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 # 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") # gr.Markdown("