File size: 5,931 Bytes
b8d8517 8e15425 b8d8517 a048ee4 b8d8517 2c8a05a 00212f4 b8d8517 00212f4 b8d8517 8e15425 b8d8517 ad390da b8d8517 4694b1a b8d8517 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 |
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
# Define the tokenizer and model
tokenizer = GPT2Tokenizer.from_pretrained('gpt2-medium')
model = openai.api_key = os.environ["OPENAI_API_KEY"]
# Define the initial message and messages list
initial_message = {"role": "system", "content": '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.'}
messages = [initial_message]
messages_rev = [initial_message]
# Define the answer counter
answer_count = 0
# Define the Notion API key
API_KEY = os.environ["API_KEY"]
def transcribe(audio, text):
global messages
global answer_count
transcript = {'text': ''}
input_text = []
# 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=-5)))
# 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/US-62e861a0b35f43da8ef9a7789512b8c2?pvs=4', title=str(published_date+'FULL'), api_key=API_KEY)
# Reset the messages list and answer counter
messages = [initial_message]
answer_count = 0
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",
messages=messages,
max_tokens=2000
)["choices"][0]["message"]
# Wait for the completion of the OpenAI API call
# 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"]})
# 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"
# Save the chat transcript to a file
with open("conversation_history.txt", "a") as f:
f.write(chat_transcript)
# Upload the chat transcript to Notion
now_et = datetime.now(timezone(timedelta(hours=-5)))
published_date = now_et.strftime('%m-%d-%y %H:%M')
df = pd.DataFrame([chat_transcript])
notion_df.upload(df, 'https://www.notion.so/US-62e861a0b35f43da8ef9a7789512b8c2?pvs=4', title=str(published_date), api_key=API_KEY)
# Return the chat transcript
return 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 = gr.outputs.Textbox(label="Response")
output_audio = Audio()
# Define the Gradio interface
iface = gr.Interface(
fn=transcribe,
inputs=[audio_input, text_input],
outputs=[output_text],
title="Hold On, Pain Ends (HOPE)",
description="Talk to Your Nephrology Tutor HOPE",
theme="compact",
layout="vertical",
allow_flagging=False
)
# Run the Gradio interface
iface.launch() |