import openai
import gradio as gr
from gradio.components import Audio, Textbox
import os
import re
import tiktoken
from transformers import GPT2Tokenizer
import whisper
import pandas as pd
from datetime import datetime, timezone, timedelta
import notion_df
import concurrent.futures
import nltk
from nltk.tokenize import sent_tokenize
nltk.download('punkt')
import spacy
from spacy import displacy
from gradio import Markdown
import threading
# Define the tokenizer and model
tokenizer = GPT2Tokenizer.from_pretrained('gpt2-medium')
model = openai.api_key = os.environ["OAPI_KEY"]
# Define the initial message and messages list
initialt = 'You are a USMLE Tutor. Respond with ALWAYS layered "bullet points" (listing rather than sentences) \
to all input with a fun mneumonics to memorize that list. But you can answer up to 1200 words if the user requests longer response. \
You are going to keep answer and also challenge the student to learn USMLE anatomy, phsysiology, and pathology.'
initial_message = {"role": "system", "content": initialt}
messages = [initial_message]
messages_rev = [initial_message]
# Define the answer counter
answer_count = 0
# Define the Notion API key
API_KEY = os.environ["NAPI_KEY"]
nlp = spacy.load("en_core_web_sm")
def process_nlp(system_message):
# Colorize the system message text
colorized_text = colorize_text(system_message['content'])
return colorized_text
# # define color combinations for different parts of speech
# COLORS = {
# "NOUN": "#5e5e5e", # Dark gray
# "VERB": "#ff6936", # Orange
# "ADJ": "#4363d8", # Blue
# "ADV": "#228b22", # Green
# "digit": "#9a45d6", # Purple
# "punct": "#ffcc00", # Yellow
# "quote": "#b300b3" # Magenta
# }
# # define color combinations for individuals with dyslexia
# DYSLEXIA_COLORS = {
# "NOUN": "#5e5e5e",
# "VERB": "#ff6936",
# "ADJ": "#4363d8",
# "ADV": "#228b22",
# "digit": "#9a45d6",
# "punct": "#ffcc00",
# "quote": "#b300b3"
# }
# # define a muted background color
# BACKGROUND_COLOR = "#f5f5f5" # Light gray
# # define font and size
# FONT = "Arial"
# FONT_SIZE = "14px"
# # load the English language model
# nlp = spacy.load('en_core_web_sm')
# def colorize_text(text, colors=DYSLEXIA_COLORS, background_color=None):
# if colors is None:
# colors = COLORS
# colorized_text = ""
# lines = text.split("\n")
# # set background color
# if background_color is None:
# background_color = BACKGROUND_COLOR
# # iterate over the lines in the text
# for line in lines:
# # parse the line with the language model
# doc = nlp(line)
# # iterate over the tokens in the line
# for token in doc:
# # check if the token is an entity
# if token.ent_type_:
# # use dyslexia colors for entity if available
# if colors == COLORS:
# color = DYSLEXIA_COLORS.get(token.pos_, None)
# else:
# color = colors.get(token.pos_, None)
# # check if a color is available for the token
# if color is not None:
# colorized_text += (
# f'<span style="color: {color}; '
# f'background-color: {background_color}; '
# f'font-family: {FONT}; '
# f'font-size: {FONT_SIZE}; '
# f'font-weight: bold; '
# f'text-decoration: none; '
# f'padding-right: 0.5em;">' # Add space between tokens
# f"{token.text}</span>"
# )
# else:
# colorized_text += (
# f'<span style="font-family: {FONT}; '
# f'font-size: {FONT_SIZE}; '
# f'font-weight: bold; '
# f'text-decoration: none; '
# f'padding-right: 0.5em;">' # Add space between tokens
# f"{token.text}</span>"
# )
# else:
# # check if a color is available for the token
# color = colors.get(token.pos_, None)
# if color is not None:
# colorized_text += (
# f'<span style="color: {color}; '
# f'background-color: {background_color}; '
# f'font-family: {FONT}; '
# f'font-size: {FONT_SIZE}; '
# f'font-weight: bold; '
# f'text-decoration: none; '
# f'padding-right: 0.5em;">' # Add space between tokens
# f"{token.text}</span>"
# )
# elif token.is_digit:
# colorized_text += (
# f'<span style="color: {colors["digit"]}; '
# f'background-color: {background_color}; '
# f'font-family: {FONT}; '
# f'font-size: {FONT_SIZE}; '
# f'font-weight: bold; '
# f'text-decoration: none; '
# f'padding-right: 0.5em;">' # Add space between tokens
# f"{token.text}</span>"
# )
# elif token.is_punct:
# colorized_text += (
# f'<span style="color: {colors["punct"]}; '
# f'background-color: {background_color}; '
# f'font-family: {FONT}; '
# f'font-size: {FONT_SIZE}; '
# f'font-weight: bold; '
# f'text-decoration: none; '
# f'padding-right: 0.5em;">' # Add space between tokens
# f"{token.text}</span>"
# )
# elif token.is_quote:
# colorized_text += (
# f'<span style="color: {colors["quote"]}; '
# f'background-color: {background_color}; '
# f'font-family: {FONT}; '
# f'font-size: {FONT_SIZE}; '
# f'text-decoration: none; '
# f'padding-right: 0.5em;">' # Add space between tokens
# f"{token.text}</span>"
# )
# else:
# colorized_text += (
# f'<span style="font-family: {FONT}; '
# f'font-size: {FONT_SIZE}; '
# f'font-weight: bold; '
# f'text-decoration: none; '
# f'padding-right: 0.5em;">' # Add space between tokens
# f"{token.text}</span>"
# )
# colorized_text += "<br>"
# return colorized_text
# define color combinations for different parts of speech
COLORS = {
"NOUN": "#FF3300",
"VERB": "#008000",
"ADJ": "#1E90FF",
"ADV": "#FF8C00",
"digit": "#FF1493",
"punct": "#8B0000",
"quote": "#800080",
}
# define color combinations for individuals with dyslexia
DYSLEXIA_COLORS = {
"NOUN": "#1E90FF",
"VERB": "#006400",
"ADJ": "#00CED1",
"ADV": "#FF8C00",
"digit": "#FF1493",
"punct": "#A0522D",
"quote": "#800080",
}
# define a muted background color
BACKGROUND_COLOR = "#EAEAEA"
# define font and size
FONT = "Georgia"
FONT_SIZE = "18px"
def colorize_text(text, colors=None, background_color=None):
if colors is None:
colors = COLORS
colorized_text = ""
lines = text.split("\n")
# set background color
if background_color is None:
background_color = BACKGROUND_COLOR
for line in lines:
doc = nlp(line)
for token in doc:
if token.ent_type_:
# use dyslexia colors for entity if available
if colors == COLORS:
color = DYSLEXIA_COLORS.get(token.pos_, None)
else:
color = colors.get(token.pos_, None)
if color is not None:
colorized_text += (
f'<span style="color: {color}; '
f'background-color: {background_color}; '
f'font-family: {FONT}; '
f'font-size: {FONT_SIZE}; '
f'text-decoration: underline;">'
f"{token.text}</span>"
)
else:
colorized_text += (
f'<span style="font-family: {FONT}; '
f'font-size: {FONT_SIZE}; '
f'text-decoration: underline;">'
f"{token.text}</span>"
)
else:
color = colors.get(token.pos_, None)
if color is not None:
colorized_text += (
f'<span style="color: {color}; '
f'background-color: {background_color}; '
f'font-family: {FONT}; '
f'font-size: {FONT_SIZE}; '
f'text-decoration: underline;">'
f"{token.text}</span>"
)
elif token.is_digit:
colorized_text += (
f'<span style="color: {colors["digit"]}; '
f'background-color: {background_color}; '
f'font-family: {FONT}; '
f'font-size: {FONT_SIZE}; '
f'text-decoration: underline;">'
f"{token.text}</span>"
)
elif token.is_punct:
colorized_text += (
f'<span style="color: {colors["punct"]}; '
f'background-color: {background_color}; '
f'font-family: {FONT}; '
f'font-size: {FONT_SIZE}; '
f'text-decoration: underline;">'
f"{token.text}</span>"
)
elif token.is_quote:
colorized_text += (
f'<span style="color: {colors["quote"]}; '
f'background-color: {background_color}; '
f'font-family: {FONT}; '
f'font-size: {FONT_SIZE}; '
f'text-decoration: underline;">'
f"{token.text}</span>"
)
else:
colorized_text += (
f'<span style="font-family: {FONT}; '
f'font-size: {FONT_SIZE}; '
f'text-decoration: underline;">'
f"{token.text}</span>"
)
colorized_text += " "
colorized_text += "<br>"
return colorized_text
# def colorize_text(text):
# colorized_text = ""
# lines = text.split("\n")
# for line in lines:
# doc = nlp(line)
# for token in doc:
# if token.ent_type_:
# colorized_text += f'**{token.text_with_ws}**'
# elif token.pos_ == 'NOUN':
# colorized_text += f'<span style="color: #FF3300; background-color: transparent;">{token.text_with_ws}</span>'
# elif token.pos_ == 'VERB':
# colorized_text += f'<span style="color: #FFFF00; background-color: transparent;">{token.text_with_ws}</span>'
# elif token.pos_ == 'ADJ':
# colorized_text += f'<span style="color: #00CC00; background-color: transparent;">{token.text_with_ws}</span>'
# elif token.pos_ == 'ADV':
# colorized_text += f'<span style="color: #FF6600; background-color: transparent;">{token.text_with_ws}</span>'
# elif token.is_digit:
# colorized_text += f'<span style="color: #9900CC; background-color: transparent;">{token.text_with_ws}</span>'
# elif token.is_punct:
# colorized_text += f'<span style="color: #8B4513; background-color: transparent;">{token.text_with_ws}</span>'
# elif token.is_quote:
# colorized_text += f'<span style="color: #008080; background-color: transparent;">{token.text_with_ws}</span>'
# else:
# colorized_text += token.text_with_ws
# colorized_text += "<br>"
# return colorized_text
def colorize_and_update(system_message, submit_update):
colorized_system_message = colorize_text(system_message['content'])
submit_update(None, colorized_system_message) # Pass the colorized_system_message as the second output
def update_text_output(system_message, submit_update):
submit_update(system_message['content'], None)
def train(text):
now_et = datetime.now(timezone(timedelta(hours=-4)))
published_date = now_et.strftime('%m-%d-%y %H:%M')
df = pd.DataFrame([text])
notion_df.upload(df, 'https://www.notion.so/US-62e861a0b35f43da8ef9a7789512b8c2?pvs=4', title=str(published_date), api_key=API_KEY)
def transcribe(audio, text, submit_update=None):
global messages
global answer_count
transcript = {'text': ''}
input_text = []
# Check if the first word of the first line is "COLORIZE"
if text and text.split("\n")[0].split(" ")[0].strip().upper() == "COLORIZE":
train(text)
colorized_input = colorize_text(text)
return text, colorized_input
# Transcribe the audio if provided
if audio is not None:
audio_file = open(audio, "rb")
transcript = openai.Audio.transcribe("whisper-1", audio_file, language="en")
# Tokenize the text input
if text is not None:
# Split the input text into sentences
sentences = re.split("(?<=[.!?]) +", text)
# Initialize a list to store the tokens
input_tokens = []
# Add each sentence to the input_tokens list
for sentence in sentences:
# Tokenize the sentence using the GPT-2 tokenizer
sentence_tokens = tokenizer.encode(sentence)
# Check if adding the sentence would exceed the token limit
if len(input_tokens) + len(sentence_tokens) < 1440:
# Add the sentence tokens to the input_tokens list
input_tokens.extend(sentence_tokens)
else:
# If adding the sentence would exceed the token limit, truncate it
sentence_tokens = sentence_tokens[:1440-len(input_tokens)]
input_tokens.extend(sentence_tokens)
break
# Decode the input tokens into text
input_text = tokenizer.decode(input_tokens)
# Add the input text to the messages list
messages.append({"role": "user", "content": transcript["text"]+input_text})
# Check if the accumulated tokens have exceeded 2096
num_tokens = sum(len(tokenizer.encode(message["content"])) for message in messages)
if num_tokens > 2096:
# Concatenate the chat history
chat_transcript = "\n\n".join([f"[ANSWER {answer_count}]{message['role']}: {message['content']}" for message in messages if message['role'] != 'system'])
# Append the number of tokens used to the end of the chat transcript
chat_transcript += f"\n\nNumber of tokens used: {num_tokens}\n\n"
# Get the current time in Eastern Time (ET)
now_et = datetime.now(timezone(timedelta(hours=-4)))
# Format the time as string (YY-MM-DD HH:MM)
published_date = now_et.strftime('%m-%d-%y %H:%M')
# Upload the chat transcript to Notion
df = pd.DataFrame([chat_transcript])
notion_df.upload(df, 'https://www.notion.so/YENA-be569d0a40c940e7b6e0679318215790?pvs=4', title=str(published_date+'back_up'), api_key=API_KEY)
# Reset the messages list and answer counter
messages = [initial_message]
messages.append({"role": "user", "content": initialt})
answer_count = 0
# Add the input text to the messages list
messages.append({"role": "user", "content": input_text})
else:
# Increment the answer counter
answer_count += 1
# Generate the system message using the OpenAI API
with concurrent.futures.ThreadPoolExecutor() as executor:
prompt = [{"text": f"{message['role']}: {message['content']}\n\n"} for message in messages]
system_message = openai.ChatCompletion.create(
# model="gpt-3.5-turbo",
model="gpt-4",
messages=messages,
max_tokens=2000
)["choices"][0]["message"]
# Wait for the completion of the OpenAI API call
if submit_update: # Check if submit_update is not None
update_text_output(system_message, submit_update)
# Add the system message to the messages list
messages.append(system_message)
# Add the system message to the beginning of the messages list
messages_rev.insert(0, system_message)
# Add the input text to the messages list
messages_rev.insert(0, {"role": "user", "content": input_text + transcript["text"]})
# Start a separate thread to process the colorization and update the Gradio interface
if submit_update: # Check if submit_update is not None
colorize_thread = threading.Thread(target=colorize_and_update, args=(system_message, submit_update))
colorize_thread.start()
# Return the system message immediately
chat_transcript = system_message['content']
# Concatenate the chat history
chat_transcript = "\n\n".join([f"[ANSWER {answer_count}]{message['role']}: {message['content']}" for message in messages_rev if message['role'] != 'system'])
# Append the number of tokens used to the end of the chat transcript
chat_transcript += f"\n\nNumber of tokens used: {num_tokens}\n\n"
now_et = datetime.now(timezone(timedelta(hours=-4)))
published_date = now_et.strftime('%m-%d-%y %H:%M')
df = pd.DataFrame([chat_transcript])
notion_df.upload(df, 'https://www.notion.so/YENA-be569d0a40c940e7b6e0679318215790?pvs=4', title=str(published_date), api_key=API_KEY)
# Return the chat transcript
return system_message['content'], colorize_text(system_message['content'])
# Define the input and output components for Gradio
audio_input = Audio(source="microphone", type="filepath", label="Record your message")
text_input = Textbox(label="Type your message", max_length=4096)
output_text = Textbox(label="Text Output")
output_html = Markdown()
# Define the Gradio interface
iface = gr.Interface(
fn=transcribe,
inputs=[audio_input, text_input],
outputs=[output_text, output_html],
title="Hold On, Pain Ends (HOPE)",
description="Talk to Your USMLE Tutor HOPE. \n If you want to colorize your note, type COLORIZE in the first line of your input.",
theme="compact",
layout="vertical",
allow_flagging=False
)
# Run the Gradio interface
iface.launch()