demo-crunchybot / backend_functions.py
Richie-O3's picture
add custom prompts
4c87232 verified
raw
history blame
14 kB
import gradio as gr
import requests
import os
import time
from datetime import timedelta
from openai import OpenAI
from pinecone import Pinecone
import uuid
import re
import pandas as pd
from google.cloud import storage
from elevenlabs.client import ElevenLabs, AsyncElevenLabs
from elevenlabs import play, save, Voice, stream
from pymongo.mongo_client import MongoClient
from utils import create_folders
from gcp import download_credentials
from csv import writer
import asyncio
import httpx
from dotenv import load_dotenv
load_dotenv()
OPENAI_API_KEY = os.getenv("OPENAI_API_KEY")
MODEL_OPENAI = os.getenv("MODEL_OPENAI")
PINECONE_API_TOKEN = os.getenv("PINECONE_API_TOKEN")
PINECONE_ENVIRONMENT = os.getenv("PINECONE_ENV")
PINECONE_HOST = os.getenv("PINECONE_HOST")
DB_USER_NAME = os.getenv("DB_USER_NAME")
DB_PASSWORD = os.getenv("DB_PASSWORD")
API_KEY_ELEVENLABS = os.getenv("API_KEY_ELEVENLABS")
D_ID_KEY = os.getenv("D_ID_KEY")
IMG_XAVY = os.getenv("IMG_XAVY")
CREDENTIALS_GCP = os.getenv("GOOGLE_APPLICATION_CREDENTIALS")
NAME_BUCKET = os.getenv("NAME_BUCKET")
URL_AUDIO = os.getenv("URL_AUDIO")
# Chat
openai_client = OpenAI(api_key=OPENAI_API_KEY)
# Vector store
pc = Pinecone(api_key=PINECONE_API_TOKEN)
index = pc.Index(host=PINECONE_HOST)
# Database
uri = f"mongodb+srv://{DB_USER_NAME}:{DB_PASSWORD}@cluster-rob01.3fpztfw.mongodb.net/?retryWrites=true&w=majority&appName=cluster-rob01"
client = MongoClient(uri)
db = client["ChatCrunchyroll"]
collection = db["history_msg"]
def _save_history_msg():
return None
def _add_question_vectorstore(question: str, response: str):
vector_id = str(uuid.uuid4())
vector_embedding = _call_embedding(question)
vector_metadata = {
'question': question,
'text': response
}
index.upsert([(vector_id, vector_embedding, vector_metadata)])
def _update_elements(question, chatbot, output, history_messages, url_audio, url_video, df_table_times):
chatbot.append([question, output])
new_comp_audio = gr.Audio(value=str(url_audio), autoplay=False, label="Audio")
new_comp_video = gr.Video(value=str(url_video), autoplay=True, height=400, label="Video")
history_messages.append({'role': 'user', 'content': question})
history_messages.append({'role': 'assistant', 'content': output})
return chatbot, new_comp_audio, new_comp_video, df_table_times
def _query_pinecone(embedding):
results = index.query(
vector=embedding,
top_k=10,
include_metadata=True,
)
final_results = """"""
for result in results['matches']:
final_results += f"{result['metadata']['text']}\n"
return final_results
def _general_prompt(context, option_prompt, general_prompt):
if option_prompt == "Default":
with open("prompt_general.txt", "r") as file:
file_prompt = file.read().replace("\n", "")
elif option_prompt == "Custom":
file_prompt = general_prompt
context_prompt = file_prompt.replace('CONTEXT', context)
print(context_prompt)
print("--------------------")
return context_prompt
def _call_embedding(text: str):
response = openai_client.embeddings.create(
input=text,
model='text-embedding-ada-002'
)
return response.data[0].embedding
def _call_gpt(prompt: str, message: str):
response = openai_client.chat.completions.create(
model=MODEL_OPENAI,
temperature=0.2,
messages=[
{'role': 'system', 'content': prompt},
{'role': 'user', 'content': message}
]
)
return response.choices[0].message.content
def _call_gpt_standalone(prompt: str):
response = openai_client.chat.completions.create(
model=MODEL_OPENAI,
temperature=0.2,
messages=[
{'role': 'system', 'content': prompt},
]
)
return response.choices[0].message.content
def _get_standalone_question(question, history_messages, option_prompt, standalone_prompt):
if option_prompt == "Default":
with open("prompt_standalone_message.txt", "r") as file:
file_prompt_standalone = file.read().replace("\n", "")
elif option_prompt == "Custom":
file_prompt_standalone = standalone_prompt
history = ''
for i, msg in enumerate(history_messages):
try:
if i == 0:
continue # Omit the prompt
if i % 2 == 0:
history += f'user: {msg["content"]}\n'
else:
history += f'assistant: {msg["content"]}\n'
except Exception as e:
print(e)
prompt_standalone = file_prompt_standalone.replace('HISTORY', history).replace('QUESTION', question)
print(prompt_standalone)
print("------------------")
standalone_msg_q = _call_gpt_standalone(prompt_standalone)
print(standalone_msg_q)
print("------------------")
return standalone_msg_q
def _create_clean_message(text: str):
clean_answer = re.sub(r'http[s]?://\S+', 'el siguiente link', text)
return clean_answer
async def _create_audio(clean_text: str, option_audio: str):
download_credentials()
create_folders()
STORAGE_CLIENT = storage.Client.from_service_account_json(CREDENTIALS_GCP)
unique_id = str(uuid.uuid4())
signed_url_audio = "None"
if option_audio == "Elevenlabs":
# Create audio file with elevenlabs
client_elevenlabs = ElevenLabs(api_key=API_KEY_ELEVENLABS)
voice_custom = Voice(voice_id = "ZQe5CZNOzWyzPSCn5a3c")
audio = client_elevenlabs.generate(
text=clean_text,
voice=voice_custom,
model="eleven_multilingual_v2"
)
source_audio_file_name = f'./audios/file_audio_{unique_id}.wav'
try:
save(audio, source_audio_file_name)
except Exception as e:
print(e)
# Save audio and get url of gcp
destination_blob_name_audio = unique_id + '.wav'
bucket = STORAGE_CLIENT.bucket(NAME_BUCKET)
blob = bucket.blob(destination_blob_name_audio)
try:
blob.upload_from_filename(source_audio_file_name)
except Exception as e:
print(e)
try:
url_expiration = timedelta(minutes=15)
signed_url_audio = blob.generate_signed_url(expiration=url_expiration)
except Exception as e:
print(e)
elif option_audio == "XTTS":
params = {'text': clean_text, 'language': 'es'}
headers = {'accept': 'application/json'}
# Makes a request to the instance with the audio api
async with httpx.AsyncClient() as client:
try:
response = await client.get(URL_AUDIO, params=params, headers=headers, timeout=120)
except Exception as e:
print(f'There is a problem with the audio. Check that instance. ERROR: {e}')
# Check if everything was successful
if response.status_code == 200:
r = response.json()
signed_url_audio = r['link_audio']
else:
print(f'There is a problem with the audio. Check that instance. ERROR: {response.status_code}')
return signed_url_audio, unique_id
def _create_video(link_audio: str, unique_id: str):
download_credentials()
create_folders()
STORAGE_CLIENT = storage.Client.from_service_account_json(CREDENTIALS_GCP)
bucket = STORAGE_CLIENT.bucket(NAME_BUCKET)
# Create video talk with file audio created by elevenlabs api
url_did = "https://api.d-id.com/talks"
payload = {
"script": {
"type": "audio",
"provider": {
"type": "microsoft",
"voice_id": "en-US-JennyNeural"
},
"ssml": "false",
"audio_url": link_audio
},
"config": {
"fluent": "false",
"pad_audio": "0.0",
"stitch": True
},
"source_url": IMG_XAVY
}
headers = {
"accept": "application/json",
"content-type": "application/json",
"authorization": f"Basic {D_ID_KEY}"
}
request_create_talk = requests.post(url_did, json=payload, headers=headers)
resp_create_talk = request_create_talk.json()
talk_id = "None"
try:
talk_id = resp_create_talk['id']
except Exception as e:
print(e)
# Get url of video file
url_get_talk_id = f"https://api.d-id.com/talks/{talk_id}"
while True:
request_video_url = requests.get(url_get_talk_id, headers=headers)
resp_video_url = request_video_url.json()
if resp_video_url['status'] == 'done':
break
# Sleep until the video is ready
time.sleep(0.5)
result_url_video = resp_video_url['result_url']
# Saves the video into a file to later upload it to the GCP
source_video_file_name = f'./videos/video_final_{unique_id}.mp4'
request_video = requests.get(result_url_video)
if request_video.status_code == 200:
with open(source_video_file_name, 'wb') as outfile:
outfile.write(request_video.content)
# Save video file to the GCP
destination_blob_name_video = unique_id + '.mp4'
# Configure bucket
blob = bucket.blob(destination_blob_name_video)
try:
blob.upload_from_filename(source_video_file_name)
except Exception as e:
print(e)
signed_url_video = "None"
try:
url_expiration_video = timedelta(minutes=15)
signed_url_video = blob.generate_signed_url(expiration=url_expiration_video)
except Exception as e:
print(e)
return signed_url_video
def get_answer(question: str, chatbot: list[tuple[str, str]], history_messages, comp_audio, comp_video, df_table, option_audio, option_prompt, general_prompt, standalone_prompt):
"""
Gets the answer of the chatbot
"""
if len(chatbot) == 8:
message_output = 'Un placer haberte ayudado, hasta luego!'
else:
start_get_standalone_question = time.time()
standalone_msg_q = _get_standalone_question(question, history_messages, option_prompt, standalone_prompt) # create standalone question or message
end_get_standalone_question = time.time()
time_get_standalone_question = end_get_standalone_question - start_get_standalone_question
start_call_embedding = time.time()
output_embedding = _call_embedding(standalone_msg_q) # create embedding of standalone question or message
end_call_embedding = time.time()
time_call_embedding = end_call_embedding - start_call_embedding
start_query_pinecone = time.time()
best_results = _query_pinecone(output_embedding) # get nearest embeddings
end_query_pinecone = time.time()
time_query_pinecone = end_query_pinecone - start_query_pinecone
start_general_prompt = time.time()
final_context_prompt = _general_prompt(best_results, option_prompt, general_prompt) # create context/general prompt
end_general_prompt = time.time()
time_general_prompt = end_general_prompt - start_general_prompt
start_call_gpt = time.time()
message_output = _call_gpt(final_context_prompt, question) # final response (to user)
end_call_gpt = time.time()
time_call_gpt = end_call_gpt - start_call_gpt
if "Respuesta:" in message_output:
message_output.replace("Respuesta:", "")
start_create_clean_message = time.time()
processed_message = _create_clean_message(message_output) # clean message output
end_create_clean_message = time.time()
time_create_clean_message = end_create_clean_message - start_create_clean_message
start_create_audio = time.time()
url_audio, unique_id = asyncio.run(_create_audio(processed_message, option_audio)) # create audio
end_create_audio = time.time()
time_create_audio = end_create_audio - start_create_audio
start_create_video = time.time()
url_video = _create_video(url_audio, unique_id) # create video with d-id no streaming
end_create_video = time.time()
time_create_video = end_create_video - start_create_video
final_time = time_get_standalone_question + time_call_embedding + time_query_pinecone + time_general_prompt
final_time += (time_call_gpt + time_create_clean_message + time_create_audio + time_create_video)
df_table = pd.DataFrame(df_table)
df_table.loc[len(df_table.index)] = [question,
message_output,
time_get_standalone_question,
time_call_embedding,
time_query_pinecone,
time_general_prompt,
time_call_gpt,
time_create_clean_message,
time_create_audio,
time_create_video,
final_time]
new_df_table = gr.DataFrame(df_table, interactive=False, visible=True)
print(history_messages)
return _update_elements(question, chatbot, message_output, history_messages, url_audio, url_video, new_df_table)
def init_greeting(chatbot, history_messages):
if len(chatbot) == 0:
greeting = ('Hola 👋, soy tu asistente de recomendación de series y películas animadas en Crunchyroll. ¿En qué puedo ayudarte hoy?')
history_messages.append({'role': 'assistant', 'content': greeting})
chatbot.append([None, greeting])
return chatbot, history_messages
def export_dataframe(df):
final_df = pd.DataFrame(df)
final_df = final_df.iloc[1:]
final_df.to_csv("./csv_times/csv_times.csv", index=False, encoding='utf-8')
return gr.File(value="./csv_times/csv_times.csv", visible=True)