import streamlit as st
import streamlit.components.v1 as components
import os
import json
import random
import base64
import glob
import math
import openai
import pytz
import re
import requests
import textract
import time
import zipfile
import huggingface_hub
import dotenv
from audio_recorder_streamlit import audio_recorder
from bs4 import BeautifulSoup
from collections import deque
from datetime import datetime
from dotenv import load_dotenv
from huggingface_hub import InferenceClient
from io import BytesIO
from openai import ChatCompletion
from PyPDF2 import PdfReader
from templates import bot_template, css, user_template
from xml.etree import ElementTree as ET
from PIL import Image
from urllib.parse import quote # Ensure this import is included
# Set initial page and app customization and configuration -------------------------
st.set_page_config(
page_title="๐๐GraphicNovelAI",
page_icon="๐๐",
layout="wide",
initial_sidebar_state="expanded",
menu_items={
'Get Help': 'https://huggingface.co/awacke1',
'Report a bug': "https://huggingface.co/spaces/awacke1/GraphicAINovel",
'About': "# Midjourney: https://discord.com/channels/@me/997514686608191558"
}
)
# Title, Base Content and Help/About
st.markdown('''### ๐โจ๐ GraphicNovelAI ''')
# Base Content Prompts for App, for App Product, and App Product Code
PromptPrefix = 'Create a graphic novel story with streamlit markdown outlines and tables with appropriate emojis for graphic novel rules defining the method steps of play. Use story structure architect rules using plan, structure and top three dramatic situations matching the theme for topic of '
PromptPrefix2 = 'Create a streamlit python user app with full code listing to create a UI implementing the plans, structure, situations and tables as python functions creating a game which operates like choose your own adventure graphic novel rules and creates a compelling fun story using streamlit to create user interface elements like emoji buttons, sliders, drop downs, and data interfaces like dataframes to show tables, session_state to track inventory, character advancement and experience, locations, file_uploader to allow the user to add images which are saved and referenced shown in gallery, camera_input to take character picture, on_change = function callbacks with continual running plots that change when you change data or click a button, randomness and dice rolls using emojis and st.markdown, st.expander for groupings and clusters of things, st.columns and other UI controls in streamlit as a game. Create inline data tables and list dictionaries for entities implemented as variables for the game rule entities and stats. Design it as a fun data driven game app and show full python code listing for this ruleset and thematic story plot line: '
PromptPrefix3 = 'Create a HTML5 aframe and javascript app using appropriate libraries to create a simulation and use more advanced libraries like aframe to render 3d scenes creating moving entities that stay within a bounding box but show text and animation in 3d for inventory, components and story entities. Show full code listing. Add a list of new random entities say 3 of a few different types to any list appropriately and use emojis to make things easier and fun to read. Use appropriate emojis in labels. Create the UI to implement storytelling in the style of a dungeon master, with features using three emoji appropriate text plot twists and recurring interesting funny fascinating and complex almost poetic named characters with genius traits and file IO, randomness, ten point choice lists, math distribution tradeoffs, witty humorous dilemnas with emoji , rewards, variables, reusable functions with parameters, and data driven app with python libraries and streamlit components for Javascript and HTML5. Use appropriate emojis for labels to summarize and list parts, function, conditions for topic:'
with st.expander("Help / About ๐", expanded=False):
st.markdown('''
- ๐ **Unlock Plots:** Elevate your vocabulary with AI. Turns plots into thrilling experiences.
- ๐ **Features:** Creates extensive glossaries & exciting challenges.
- ๐งโโ๏ธ **Experience:** Become a graphic novel plot wizard, boost your language skills.
- ๐ **Query Use:** Input `?q=Palindrome` or `?query=Anagram` in URL for new challenges.
''')
# Aaron's Intelligent Style Guide for AI Graphic Novel Writers
parts_of_speech = [
{"type": "Noun", "description": "Person, place, thing, or idea", "example": "Hero, city, spaceship, justice"},
{"type": "Verb", "description": "Action or state of being", "example": "Fight, transform, is, become"},
{"type": "Adjective", "description": "Describes a noun", "example": "Mysterious, ancient, powerful, dark"},
{"type": "Adverb", "description": "Modifies verbs, adjectives, or other adverbs", "example": "Mysteriously, very, suddenly, heroically"},
{"type": "Conjunction", "description": "Connects clauses, sentences, or words", "example": "And, but, or, yet"},
{"type": "Interjection", "description": "Expresses emotion", "example": "Wow!, Ouch!, Haha!, Shhh!"},
{"type": "Idiom", "description": "Phrase with a figurative meaning", "example": "Break a leg, Spill the beans, Hit the road"},
{"type": "Symbolism", "description": "Objects, figures, or colors used to represent ideas or concepts", "example": "A rose for love, a storm for chaos"},
{"type": "Theme", "description": "Underlying message or main idea", "example": "The quest for identity, the battle between good and evil"},
{"type": "Motif", "description": "Recurring element that has symbolic significance", "example": "Repeated imagery of masks to signify identity"}
]
language_structures = [
{"type": "Glossary", "description": "Vocabulary Reference: List of terms and their definitions", "example": "Villain: The antagonist of the story"},
{"type": "Dialogue", "description": "Conversational Text: Characters' spoken words", "example": "We must act now! exclaimed the hero"},
{"type": "Narration", "description": "Storytelling Text: Text that tells the story", "example": "The city had never seen such despair"},
{"type": "Captions", "description": "Descriptive Text: Describes scene, setting, or action", "example": "New York, 2050. A city in turmoil"},
{"type": "Sound Effects", "description": "Auditory Text: Words that mimic sounds", "example": "BOOM! The spaceship landed"},
{"type": "Thought Bubbles", "description": "Internal Monologue Text: Characters' thoughts", "example": "I wonder if they know my secret"},
{"type": "Panel Transitions", "description": "Visual Storytelling Technique: Movement between scenes or ideas", "example": "Meanwhile, across the galaxy..."},
{"type": "Character Development", "description": "Evolution of characters throughout the story", "example": "From a timid schoolgirl to a fearless warrior"},
{"type": "Plot Twists", "description": "Unexpected changes in the story direction", "example": "The hero discovers their enemy is their sibling"},
{"type": "Backstory", "description": "Historical or background context of characters or setting", "example": "Once a celebrated hero, now a forgotten legend"}
]
# Assuming 'parts_of_speech' and 'language_structures' are defined as above
def display_elements(elements, title):
st.markdown(f"## {title}")
for element in elements:
st.markdown(f"""
- **Type**: {element['type']}
- **Description**: {element['description']}
- **Example**: {element['example']}
""")
# process sets:
st.title("Graphic Novel Creation Toolkit")
display_elements(parts_of_speech, "Parts of Speech for Dramatic Situations")
display_elements(language_structures, "Language Structures for Dramatic Situations")
# MoE Context Glossary
roleplaying_glossary = {
"๐จโ๐ฉโ๐งโ๐ฆ Top Graphic Novel Plot Themes": {
"Epic Fantasy": [
"Ancient prophecies and mystical artifacts",
"Epic battles between good and evil",
"Complex world-building with diverse cultures",
"Journey of a reluctant hero",
"Alliance of unlikely companions",
"Betrayal and redemption arcs",
"Magic systems and mythical creatures",
"Climactic confrontation with a dark lord"
],
"Superhero Sagas": [
"Origin stories of heroes and villains",
"Struggle with personal identity and responsibility",
"Formation of superhero teams",
"Epic battles to save the city/world",
"Moral dilemmas and ethical questions",
"Interdimensional threats and cosmic wars",
"Evolution of powers and discovery of new abilities",
"Legacy heroes and passing of the mantle"
],
"Post-Apocalyptic Survival": [
"Survival in a world after a global catastrophe",
"Rebuilding society from the ashes",
"Conflict between surviving factions",
"Quests for scarce resources",
"Encounters with mutated creatures",
"Moral ambiguity and survival ethics",
"Exploration of human resilience",
"Discovery of a safe haven or cure"
],
"Science Fiction and Space Opera": [
"Exploration of distant galaxies",
"Conflict between alien species",
"Advanced technology and space travel",
"Utopian and dystopian societies",
"Time travel and alternate realities",
"Artificial intelligence and robotics",
"Quests for knowledge and discovery",
"Rebellion against oppressive regimes"
],
"Horror and Supernatural": [
"Haunted locations and ghost stories",
"Battles against demonic forces",
"Survival horror and psychological terror",
"Folklore and urban legends",
"Vampires, werewolves, and other monsters",
"Occult practices and dark magic",
"Apocalyptic and Lovecraftian themes",
"Investigations into the unknown"
],
"Romance and Relationship Dramas": [
"Complex romantic entanglements",
"Struggles with identity and societal expectations",
"Heartbreak, healing, and growth",
"Forbidden love and star-crossed lovers",
"Contemporary relationship dynamics",
"Cultural and social differences",
"Self-discovery and personal fulfillment",
"Romantic comedies and tragedies"
]
}
}
# Set initial page and app configs ------------------------------------------
# Function to display the entire glossary in a grid format with links
def display_glossary_grid(roleplaying_glossary):
search_urls = {
"๐": lambda k: f"https://en.wikipedia.org/wiki/{quote(k)}",
"๐": lambda k: f"https://www.google.com/search?q={quote(k)}",
"โถ๏ธ": lambda k: f"https://www.youtube.com/results?search_query={quote(k)}",
"๐": lambda k: f"https://www.bing.com/search?q={quote(k)}",
"๐ฆ": lambda k: f"https://twitter.com/search?q={quote(k)}",
"๐ฒ": lambda k: f"https://huggingface.co/spaces/awacke1/GraphicAINovel?q={quote(k)}", # this url plus query!
"๐": lambda k: f"https://huggingface.co/spaces/awacke1/GraphicAINovel?q={quote(PromptPrefix)}{quote(k)}", # this url plus query!
"๐": lambda k: f"https://huggingface.co/spaces/awacke1/GraphicAINovel?q={quote(PromptPrefix2)}{quote(k)}", # this url plus query!
"๐ฌ": lambda k: f"https://huggingface.co/spaces/awacke1/GraphicAINovel?q={quote(PromptPrefix3)}{quote(k)}", # this url plus query!
}
for category, details in roleplaying_glossary.items():
st.write(f"### {category}")
cols = st.columns(len(details)) # Create dynamic columns based on the number of games
for idx, (game, terms) in enumerate(details.items()):
with cols[idx]:
st.markdown(f"#### {game}")
for term in terms:
links_md = ' '.join([f"[{emoji}]({url(term)})" for emoji, url in search_urls.items()])
st.markdown(f"{term} {links_md}", unsafe_allow_html=True)
def display_glossary_entity(k):
search_urls = {
"๐": lambda k: f"https://en.wikipedia.org/wiki/{quote(k)}",
"๐": lambda k: f"https://www.google.com/search?q={quote(k)}",
"โถ๏ธ": lambda k: f"https://www.youtube.com/results?search_query={quote(k)}",
"๐": lambda k: f"https://www.bing.com/search?q={quote(k)}",
"๐ฆ": lambda k: f"https://twitter.com/search?q={quote(k)}",
"๐ฒ": lambda k: f"https://huggingface.co/spaces/awacke1/GraphicAINovel?q={quote(k)}", # this url plus query!
"๐": lambda k: f"https://huggingface.co/spaces/awacke1/GraphicAINovel?q={quote(PromptPrefix)}{quote(k)}", # this url plus query!
"๐": lambda k: f"https://huggingface.co/spaces/awacke1/GraphicAINovel?q={quote(PromptPrefix2)}{quote(k)}", # this url plus query!
"๐ฌ": lambda k: f"https://huggingface.co/spaces/awacke1/GraphicAINovel?q={quote(PromptPrefix3)}{quote(k)}", # this url plus query!
}
links_md = ' '.join([f"[{emoji}]({url(k)})" for emoji, url in search_urls.items()])
st.markdown(f"{k} {links_md}", unsafe_allow_html=True)
# HTML5 based Speech Synthesis (Text to Speech in Browser)
@st.cache_resource
def SpeechSynthesis(result):
documentHTML5='''
Read It Aloud
๐ Read It Aloud
'''
components.html(documentHTML5, width=1280, height=300)
# 9. Chat History File Sidebar
@st.cache_resource
def get_table_download_link(file_path):
with open(file_path, 'r') as file:
data = file.read()
b64 = base64.b64encode(data.encode()).decode()
file_name = os.path.basename(file_path)
ext = os.path.splitext(file_name)[1] # get the file extension
if ext == '.txt':
mime_type = 'text/plain'
elif ext == '.py':
mime_type = 'text/plain'
elif ext == '.xlsx':
mime_type = 'text/plain'
elif ext == '.csv':
mime_type = 'text/plain'
elif ext == '.htm':
mime_type = 'text/html'
elif ext == '.md':
mime_type = 'text/markdown'
elif ext == '.wav':
mime_type = 'audio/wav'
else:
mime_type = 'application/octet-stream' # general binary data type
href = f'{file_name}'
return href
@st.cache_resource
def create_zip_of_files(files):
#zip_name = "all_files.zip"
zip_name = "GraphicAINovels.zip"
with zipfile.ZipFile(zip_name, 'w') as zipf:
for file in files:
zipf.write(file)
return zip_name
@st.cache_resource
def get_zip_download_link(zip_file):
with open(zip_file, 'rb') as f:
data = f.read()
b64 = base64.b64encode(data).decode()
href = f'Download All'
return href
def FileSidebar():
# ----------------------------------------------------- File Sidebar for Jump Gates ------------------------------------------
# Compose a file sidebar of markdown md files:
all_files = glob.glob("*.md")
all_files = [file for file in all_files if len(os.path.splitext(file)[0]) >= 10] # exclude files with short names
all_files.sort(key=lambda x: (os.path.splitext(x)[1], x), reverse=True) # sort by file type and file name in descending order
if st.sidebar.button("๐ Delete All Text"):
for file in all_files:
os.remove(file)
st.experimental_rerun()
if st.sidebar.button("โฌ๏ธ Download All"):
zip_file = create_zip_of_files(all_files)
st.sidebar.markdown(get_zip_download_link(zip_file), unsafe_allow_html=True)
file_contents=''
next_action=''
for file in all_files:
col1, col2, col3, col4, col5 = st.sidebar.columns([1,6,1,1,1]) # adjust the ratio as needed
with col1:
if st.button("๐", key="md_"+file): # md emoji button
with open(file, 'r') as f:
file_contents = f.read()
next_action='md'
with col2:
st.markdown(get_table_download_link(file), unsafe_allow_html=True)
with col3:
if st.button("๐", key="open_"+file): # open emoji button
with open(file, 'r') as f:
file_contents = f.read()
next_action='open'
with col4:
if st.button("๐", key="read_"+file): # search emoji button
with open(file, 'r') as f:
file_contents = f.read()
next_action='search'
with col5:
if st.button("๐", key="delete_"+file):
os.remove(file)
st.experimental_rerun()
if len(file_contents) > 0:
if next_action=='open':
file_content_area = st.text_area("File Contents:", file_contents, height=500)
try:
if st.button("๐", key="filecontentssearch"):
search_glossary(file_content_area)
except:
st.markdown('GPT is sleeping. Restart ETA 30 seconds.')
if next_action=='md':
st.markdown(file_contents)
buttonlabel = '๐Run'
if st.button(key='RunWithLlamaandGPT', label = buttonlabel):
user_prompt = file_contents
try:
search_glossary(file_contents)
except:
st.markdown('GPT is sleeping. Restart ETA 30 seconds.')
if next_action=='search':
file_content_area = st.text_area("File Contents:", file_contents, height=500)
user_prompt = file_contents
try:
search_glossary(file_contents)
except:
st.markdown('GPT is sleeping. Restart ETA 30 seconds.')
# ----------------------------------------------------- File Sidebar for Jump Gates ------------------------------------------
FileSidebar()
# ---- Art Card Sidebar with Random Selection of image:
@st.cache_resource
def get_image_as_base64(url):
response = requests.get(url)
if response.status_code == 200:
# Convert the image to base64
return base64.b64encode(response.content).decode("utf-8")
else:
return None
@st.cache_resource
def create_download_link(filename, base64_str):
href = f'Download Image'
return href
# List of image URLs
image_urls = [
"https://cdn-uploads.huggingface.co/production/uploads/620630b603825909dcbeba35/W1omJItftG3OkW9sj-Ckb.png",
"https://cdn-uploads.huggingface.co/production/uploads/620630b603825909dcbeba35/Djx-k4WOxzlXEQPzllP3r.png"
]
UseSidebarArtCard=True
if UseSidebarArtCard:
# Select a random URL from the list
selected_image_url = random.choice(image_urls)
# Get the base64 encoded string of the selected image
# st.write(selected_image_url)
try:
selected_image_base64 = get_image_as_base64(selected_image_url)
if selected_image_base64 is not None:
with st.sidebar:
#st.markdown("""### Graphic Novel AI""")
# Display the image
st.markdown(f"")
# Create and display the download link
#download_link = create_download_link("downloaded_image.png", selected_image_base64)
#st.markdown(download_link, unsafe_allow_html=True)
else:
st.sidebar.write("Failed to load the image.")
except:
st.write('Sidebar Fail - Check your Images')
# ---- Art Card Sidebar with random selection of image.
# Ensure the directory for storing scores exists
score_dir = "scores"
os.makedirs(score_dir, exist_ok=True)
# Function to generate a unique key for each button, including an emoji
def generate_key(label, header, idx):
return f"{header}_{label}_{idx}_key"
# Function to increment and save score
def update_score(key, increment=1):
score_file = os.path.join(score_dir, f"{key}.json")
if os.path.exists(score_file):
with open(score_file, "r") as file:
score_data = json.load(file)
else:
score_data = {"clicks": 0, "score": 0}
score_data["clicks"] += 1
score_data["score"] += increment
with open(score_file, "w") as file:
json.dump(score_data, file)
return score_data["score"]
# Function to load score
def load_score(key):
score_file = os.path.join(score_dir, f"{key}.json")
if os.path.exists(score_file):
with open(score_file, "r") as file:
score_data = json.load(file)
return score_data["score"]
return 0
@st.cache_resource
def search_glossary(query):
for category, terms in roleplaying_glossary.items():
if query.lower() in (term.lower() for term in terms):
st.markdown(f"#### {category}")
st.write(f"- {query}")
all=""
#query2 = PromptPrefix + query
query2 = query
response = chat_with_model(query2)
all = query + ' ' + response
filename = generate_filename(response, "md")
create_file(filename, query, response, should_save)
#query3 = PromptPrefix2 + query + ' for story outline of method steps: ' + response # Add prompt preface for coding task behavior
#response2 = chat_with_model(query3)
#query4 = PromptPrefix3 + query + ' using this streamlit python programspecification to define features. Create entities for each variable and generate UI with HTML5 and JS that matches the streamlit program: ' + response2 # Add prompt preface for coding task behavior
#response3 = chat_with_model(query4)
#all = query + ' ' + response + ' ' + response2 + ' ' + response3
#filename = generate_filename(all, "md")
#create_file(filename, query, all, should_save)
SpeechSynthesis(all)
return all
# Function to display the glossary in a structured format
def display_glossary(glossary, area):
if area in glossary:
st.subheader(f"๐ Glossary for {area}")
for game, terms in glossary[area].items():
st.markdown(f"### {game}")
for idx, term in enumerate(terms, start=1):
st.write(f"{idx}. {term}")
game_emojis = {
"Dungeons and Dragons": "๐",
"Call of Cthulhu": "๐",
"GURPS": "๐ฒ",
"Pathfinder": "๐บ๏ธ",
"Kindred of the East": "๐ ",
"Changeling": "๐",
}
topic_emojis = {
"Core Rulebooks": "๐",
"Maps & Settings": "๐บ๏ธ",
"Game Mechanics & Tools": "โ๏ธ",
"Monsters & Adversaries": "๐น",
"Campaigns & Adventures": "๐",
"Creatives & Assets": "๐จ",
"Game Master Resources": "๐ ๏ธ",
"Lore & Background": "๐",
"Character Development": "๐ง",
"Homebrew Content": "๐ง",
"General Topics": "๐",
}
# Adjusted display_buttons_with_scores function
def display_buttons_with_scores():
for category, games in roleplaying_glossary.items():
category_emoji = topic_emojis.get(category, "๐") # Default to search icon if no match
st.markdown(f"## {category_emoji} {category}")
for game, terms in games.items():
game_emoji = game_emojis.get(game, "๐ฎ") # Default to generic game controller if no match
for term in terms:
key = f"{category}_{game}_{term}".replace(' ', '_').lower()
score = load_score(key)
if st.button(f"{game_emoji} {term} {score}", key=key):
update_score(key)
# Create a dynamic query incorporating emojis and formatting for clarity
query_prefix = f"{category_emoji} {game_emoji} **{game} - {category}:**"
# ----------------------------------------------------------------------------------------------
#query_body = f"Create a detailed outline for **{term}** with subpoints highlighting key aspects, using emojis for visual engagement. Include step-by-step rules and boldface important entities and ruleset elements."
query_body = f"Create a streamlit python app.py that produces a detailed markdown outline and emoji laden user interface with labels with the entity name and emojis in all labels with a set of streamlit UI components with drop down lists and dataframes and buttons with expander and sidebar for the app to run the data as default values mostly in text boxes. Feature a 3 point outline sith 3 subpoints each where each line has about six words describing this and also contain appropriate emoji for creating sumamry of all aspeccts of this topic. an outline for **{term}** with subpoints highlighting key aspects, using emojis for visual engagement. Include step-by-step rules and boldface important entities and ruleset elements."
response = search_glossary(query_prefix + query_body)
def fetch_wikipedia_summary(keyword):
# Placeholder function for fetching Wikipedia summaries
# In a real app, you might use requests to fetch from the Wikipedia API
return f"Summary for {keyword}. For more information, visit Wikipedia."
def create_search_url_youtube(keyword):
base_url = "https://www.youtube.com/results?search_query="
return base_url + keyword.replace(' ', '+')
def create_search_url_bing(keyword):
base_url = "https://www.bing.com/search?q="
return base_url + keyword.replace(' ', '+')
def create_search_url_wikipedia(keyword):
base_url = "https://www.wikipedia.org/search-redirect.php?family=wikipedia&language=en&search="
return base_url + keyword.replace(' ', '+')
def create_search_url_google(keyword):
base_url = "https://www.google.com/search?q="
return base_url + keyword.replace(' ', '+')
def create_search_url_ai(keyword):
base_url = "https://huggingface.co/spaces/awacke1/GraphicAINovel?q="
return base_url + keyword.replace(' ', '+')
@st.cache_resource
def display_videos_and_links():
video_files = [f for f in os.listdir('.') if f.endswith('.mp4')]
if not video_files:
st.write("No MP4 videos found in the current directory.")
return
video_files_sorted = sorted(video_files, key=lambda x: len(x.split('.')[0]))
cols = st.columns(2) # Define 2 columns outside the loop
col_index = 0 # Initialize column index
for video_file in video_files_sorted:
with cols[col_index % 2]: # Use modulo 2 to alternate between the first and second column
# Embedding video with autoplay and loop using HTML
#video_html = ("""""")
#st.markdown(video_html, unsafe_allow_html=True)
k = video_file.split('.')[0] # Assumes keyword is the file name without extension
st.video(video_file, format='video/mp4', start_time=0)
display_glossary_entity(k)
col_index += 1 # Increment column index to place the next video in the next column
@st.cache_resource
def display_images_and_wikipedia_summaries():
image_files = [f for f in os.listdir('.') if f.endswith('.png')]
if not image_files:
st.write("No PNG images found in the current directory.")
return
image_files_sorted = sorted(image_files, key=lambda x: len(x.split('.')[0]))
grid_sizes = [len(f.split('.')[0]) for f in image_files_sorted]
col_sizes = ['small' if size <= 4 else 'medium' if size <= 8 else 'large' for size in grid_sizes]
num_columns_map = {"small": 4, "medium": 3, "large": 2}
current_grid_size = 0
for image_file, col_size in zip(image_files_sorted, col_sizes):
if current_grid_size != num_columns_map[col_size]:
cols = st.columns(num_columns_map[col_size])
current_grid_size = num_columns_map[col_size]
col_index = 0
with cols[col_index % current_grid_size]:
image = Image.open(image_file)
st.image(image, caption=image_file, use_column_width=True)
k = image_file.split('.')[0] # Assumes keyword is the file name without extension
display_glossary_entity(k)
def get_all_query_params(key):
return st.query_params().get(key, [])
def clear_query_params():
st.query_params()
# Function to display content or image based on a query
@st.cache_resource
def display_content_or_image(query):
for category, terms in transhuman_glossary.items():
for term in terms:
if query.lower() in term.lower():
st.subheader(f"Found in {category}:")
st.write(term)
return True # Return after finding and displaying the first match
image_dir = "images" # Example directory where images are stored
image_path = f"{image_dir}/{query}.png" # Construct image path with query
if os.path.exists(image_path):
st.image(image_path, caption=f"Image for {query}")
return True
st.warning("No matching content or image found.")
return False
# 1. Constants and Top Level UI Variables
# My Inference API Copy
API_URL = 'https://qe55p8afio98s0u3.us-east-1.aws.endpoints.huggingface.cloud' # Dr Llama
# Meta's Original - Chat HF Free Version:
#API_URL = "https://api-inference.huggingface.co/models/meta-llama/Llama-2-7b-chat-hf"
API_KEY = os.getenv('API_KEY')
MODEL1="meta-llama/Llama-2-7b-chat-hf"
MODEL1URL="https://huggingface.co/meta-llama/Llama-2-7b-chat-hf"
HF_KEY = os.getenv('HF_KEY')
headers = {
"Authorization": f"Bearer {HF_KEY}",
"Content-Type": "application/json"
}
key = os.getenv('OPENAI_API_KEY')
prompt = f"..."
should_save = st.sidebar.checkbox("๐พ Save", value=True, help="Save your session data.")
# 3. Stream Llama Response
# @st.cache_resource
def StreamLLMChatResponse(prompt):
try:
endpoint_url = API_URL
hf_token = API_KEY
#st.write('Running client ' + endpoint_url)
client = InferenceClient(endpoint_url, token=hf_token)
gen_kwargs = dict(
max_new_tokens=512,
top_k=30,
top_p=0.9,
temperature=0.2,
repetition_penalty=1.02,
stop_sequences=["\nUser:", "<|endoftext|>", ""],
)
stream = client.text_generation(prompt, stream=True, details=True, **gen_kwargs)
report=[]
res_box = st.empty()
collected_chunks=[]
collected_messages=[]
allresults=''
for r in stream:
if r.token.special:
continue
if r.token.text in gen_kwargs["stop_sequences"]:
break
collected_chunks.append(r.token.text)
chunk_message = r.token.text
collected_messages.append(chunk_message)
try:
report.append(r.token.text)
if len(r.token.text) > 0:
result="".join(report).strip()
res_box.markdown(f'*{result}*')
except:
st.write('Stream llm issue')
SpeechSynthesis(result)
return result
except:
st.write('Llama model is asleep. Starting up now on A10 - please give 5 minutes then retry as KEDA scales up from zero to activate running container(s).')
# 4. Run query with payload
@st.cache_resource
def query(payload):
response = requests.post(API_URL, headers=headers, json=payload)
st.markdown(response.json())
return response.json()
def get_output(prompt):
return query({"inputs": prompt})
# 5. Auto name generated output files from time and content
def generate_filename(prompt, file_type):
central = pytz.timezone('US/Central')
safe_date_time = datetime.now(central).strftime("%m%d_%H%M")
replaced_prompt = prompt.replace(" ", "_").replace("\n", "_")
safe_prompt = "".join(x for x in replaced_prompt if x.isalnum() or x == "_")[:255] # 255 is linux max, 260 is windows max
return f"{safe_date_time}_{safe_prompt}.{file_type}"
# 6. Speech transcription via OpenAI service
def transcribe_audio(openai_key, file_path, model):
openai.api_key = openai_key
OPENAI_API_URL = "https://api.openai.com/v1/audio/transcriptions"
headers = {
"Authorization": f"Bearer {openai_key}",
}
with open(file_path, 'rb') as f:
data = {'file': f}
st.write('STT transcript ' + OPENAI_API_URL)
response = requests.post(OPENAI_API_URL, headers=headers, files=data, data={'model': model})
if response.status_code == 200:
st.write(response.json())
chatResponse = chat_with_model(response.json().get('text'), '') # *************************************
transcript = response.json().get('text')
filename = generate_filename(transcript, 'txt')
response = chatResponse
user_prompt = transcript
create_file(filename, user_prompt, response, should_save)
return transcript
else:
st.write(response.json())
st.error("Error in API call.")
return None
# 7. Auto stop on silence audio control for recording WAV files
def save_and_play_audio(audio_recorder):
audio_bytes = audio_recorder(key='audio_recorder')
if audio_bytes:
filename = generate_filename("Recording", "wav")
with open(filename, 'wb') as f:
f.write(audio_bytes)
st.audio(audio_bytes, format="audio/wav")
return filename
return None
# 8. File creator that interprets type and creates output file for text, markdown and code
@st.cache_resource
def create_file(filename, prompt, response, should_save=True):
if not should_save:
return
base_filename, ext = os.path.splitext(filename)
if ext in ['.txt', '.htm', '.md']:
with open(f"{base_filename}.md", 'w') as file:
try:
content = prompt.strip() + '\r\n' + response
file.write(content)
except:
st.write('.')
#has_python_code = re.search(r"```python([\s\S]*?)```", prompt.strip() + '\r\n' + response)
#has_python_code = bool(re.search(r"```python([\s\S]*?)```", prompt.strip() + '\r\n' + response))
#if has_python_code:
# python_code = re.findall(r"```python([\s\S]*?)```", response)[0].strip()
# with open(f"{base_filename}-Code.py", 'w') as file:
# file.write(python_code)
# with open(f"{base_filename}.md", 'w') as file:
# content = prompt.strip() + '\r\n' + response
# file.write(content)
def truncate_document(document, length):
return document[:length]
def divide_document(document, max_length):
return [document[i:i+max_length] for i in range(0, len(document), max_length)]
def CompressXML(xml_text):
root = ET.fromstring(xml_text)
for elem in list(root.iter()):
if isinstance(elem.tag, str) and 'Comment' in elem.tag:
elem.parent.remove(elem)
return ET.tostring(root, encoding='unicode', method="xml")
# 10. Read in and provide UI for past files
@st.cache_resource
def read_file_content(file,max_length):
if file.type == "application/json":
content = json.load(file)
return str(content)
elif file.type == "text/html" or file.type == "text/htm":
content = BeautifulSoup(file, "html.parser")
return content.text
elif file.type == "application/xml" or file.type == "text/xml":
tree = ET.parse(file)
root = tree.getroot()
xml = CompressXML(ET.tostring(root, encoding='unicode'))
return xml
elif file.type == "text/markdown" or file.type == "text/md":
md = mistune.create_markdown()
content = md(file.read().decode())
return content
elif file.type == "text/plain":
return file.getvalue().decode()
else:
return ""
# 11. Chat with GPT - Caution on quota - now favoring fastest AI pipeline STT Whisper->LLM Llama->TTS
@st.cache_resource
def chat_with_model(prompt, document_section='', model_choice='gpt-3.5-turbo'): # gpt-4-0125-preview gpt-3.5-turbo
#def chat_with_model(prompt, document_section='', model_choice='gpt-4-0125-preview'): # gpt-4-0125-preview gpt-3.5-turbo
model = model_choice
conversation = [{'role': 'system', 'content': 'You are a helpful assistant.'}]
conversation.append({'role': 'user', 'content': prompt})
if len(document_section)>0:
conversation.append({'role': 'assistant', 'content': document_section})
start_time = time.time()
report = []
res_box = st.empty()
collected_chunks = []
collected_messages = []
for chunk in openai.ChatCompletion.create(model=model_choice, messages=conversation, temperature=0.5, stream=True):
collected_chunks.append(chunk)
chunk_message = chunk['choices'][0]['delta']
collected_messages.append(chunk_message)
content=chunk["choices"][0].get("delta",{}).get("content")
try:
report.append(content)
if len(content) > 0:
result = "".join(report).strip()
res_box.markdown(f'*{result}*')
except:
st.write(' ')
full_reply_content = ''.join([m.get('content', '') for m in collected_messages])
st.write("Elapsed time:")
st.write(time.time() - start_time)
return full_reply_content
@st.cache_resource
def chat_with_file_contents(prompt, file_content, model_choice='gpt-3.5-turbo'): # gpt-4-0125-preview gpt-3.5-turbo
#def chat_with_file_contents(prompt, file_content, model_choice='gpt-4-0125-preview'): # gpt-4-0125-preview gpt-3.5-turbo
conversation = [{'role': 'system', 'content': 'You are a helpful assistant.'}]
conversation.append({'role': 'user', 'content': prompt})
if len(file_content)>0:
conversation.append({'role': 'assistant', 'content': file_content})
response = openai.ChatCompletion.create(model=model_choice, messages=conversation)
return response['choices'][0]['message']['content']
def extract_mime_type(file):
if isinstance(file, str):
pattern = r"type='(.*?)'"
match = re.search(pattern, file)
if match:
return match.group(1)
else:
raise ValueError(f"Unable to extract MIME type from {file}")
elif isinstance(file, streamlit.UploadedFile):
return file.type
else:
raise TypeError("Input should be a string or a streamlit.UploadedFile object")
def extract_file_extension(file):
# get the file name directly from the UploadedFile object
file_name = file.name
pattern = r".*?\.(.*?)$"
match = re.search(pattern, file_name)
if match:
return match.group(1)
else:
raise ValueError(f"Unable to extract file extension from {file_name}")
# Normalize input as text from PDF and other formats
@st.cache_resource
def pdf2txt(docs):
text = ""
for file in docs:
file_extension = extract_file_extension(file)
st.write(f"File type extension: {file_extension}")
if file_extension.lower() in ['py', 'txt', 'html', 'htm', 'xml', 'json']:
text += file.getvalue().decode('utf-8')
elif file_extension.lower() == 'pdf':
from PyPDF2 import PdfReader
pdf = PdfReader(BytesIO(file.getvalue()))
for page in range(len(pdf.pages)):
text += pdf.pages[page].extract_text() # new PyPDF2 syntax
return text
def txt2chunks(text):
text_splitter = CharacterTextSplitter(separator="\n", chunk_size=1000, chunk_overlap=200, length_function=len)
return text_splitter.split_text(text)
# Vector Store using FAISS
@st.cache_resource
def vector_store(text_chunks):
embeddings = OpenAIEmbeddings(openai_api_key=key)
return FAISS.from_texts(texts=text_chunks, embedding=embeddings)
# Memory and Retrieval chains
@st.cache_resource
def get_chain(vectorstore):
llm = ChatOpenAI()
memory = ConversationBufferMemory(memory_key='chat_history', return_messages=True)
return ConversationalRetrievalChain.from_llm(llm=llm, retriever=vectorstore.as_retriever(), memory=memory)
def process_user_input(user_question):
response = st.session_state.conversation({'question': user_question})
st.session_state.chat_history = response['chat_history']
for i, message in enumerate(st.session_state.chat_history):
template = user_template if i % 2 == 0 else bot_template
st.write(template.replace("{{MSG}}", message.content), unsafe_allow_html=True)
filename = generate_filename(user_question, 'txt')
response = message.content
user_prompt = user_question
create_file(filename, user_prompt, response, should_save)
def divide_prompt(prompt, max_length):
words = prompt.split()
chunks = []
current_chunk = []
current_length = 0
for word in words:
if len(word) + current_length <= max_length:
current_length += len(word) + 1
current_chunk.append(word)
else:
chunks.append(' '.join(current_chunk))
current_chunk = [word]
current_length = len(word)
chunks.append(' '.join(current_chunk))
return chunks
# 14. Inference Endpoints for Whisper (best fastest STT) on NVIDIA T4 and Llama (best fastest AGI LLM) on NVIDIA A10
API_URL_IE = f'https://tonpixzfvq3791u9.us-east-1.aws.endpoints.huggingface.cloud'
API_URL_IE = "https://api-inference.huggingface.co/models/openai/whisper-small.en"
MODEL2 = "openai/whisper-small.en"
MODEL2_URL = "https://huggingface.co/openai/whisper-small.en"
HF_KEY = st.secrets['HF_KEY']
headers = {
"Authorization": f"Bearer {HF_KEY}",
"Content-Type": "audio/wav"
}
def query(filename):
with open(filename, "rb") as f:
data = f.read()
response = requests.post(API_URL_IE, headers=headers, data=data)
return response.json()
def generate_filename(prompt, file_type):
central = pytz.timezone('US/Central')
safe_date_time = datetime.now(central).strftime("%m%d_%H%M")
replaced_prompt = prompt.replace(" ", "_").replace("\n", "_")
safe_prompt = "".join(x for x in replaced_prompt if x.isalnum() or x == "_")[:90]
return f"{safe_date_time}_{safe_prompt}.{file_type}"
# 15. Audio recorder to Wav file
def save_and_play_audio(audio_recorder):
audio_bytes = audio_recorder()
if audio_bytes:
filename = generate_filename("Recording", "wav")
with open(filename, 'wb') as f:
f.write(audio_bytes)
st.audio(audio_bytes, format="audio/wav")
return filename
# 16. Speech transcription to file output
def transcribe_audio(filename):
output = query(filename)
return output
def whisper_main():
filename = save_and_play_audio(audio_recorder)
if filename is not None:
transcription = transcribe_audio(filename)
try:
transcript = transcription['text']
st.write(transcript)
except:
transcript=''
st.write(transcript)
st.write('Reasoning with your inputs..')
response = chat_with_model(transcript)
st.write('Response:')
st.write(response)
filename = generate_filename(response, "txt")
create_file(filename, transcript, response, should_save)
# Whisper to Llama:
response = StreamLLMChatResponse(transcript)
filename_txt = generate_filename(transcript, "md")
create_file(filename_txt, transcript, response, should_save)
filename_wav = filename_txt.replace('.txt', '.wav')
import shutil
try:
if os.path.exists(filename):
shutil.copyfile(filename, filename_wav)
except:
st.write('.')
if os.path.exists(filename):
os.remove(filename)
# 17. Main
def main():
prompt = PromptPrefix2
with st.expander("Prompts ๐", expanded=False):
example_input = st.text_input("Enter your prompt text:", value=prompt, help="Enter text to get a response.")
if st.button("Run Prompt", help="Click to run."):
try:
response=StreamLLMChatResponse(example_input)
create_file(filename, example_input, response, should_save)
except:
st.write('model is asleep. Starting now on A10 GPU. Please wait one minute then retry. KEDA triggered.')
openai.api_key = os.getenv('OPENAI_API_KEY')
if openai.api_key == None: openai.api_key = st.secrets['OPENAI_API_KEY']
menu = ["txt", "htm", "xlsx", "csv", "md", "py"]
choice = st.sidebar.selectbox("Output File Type:", menu)
model_choice = st.sidebar.radio("Select Model:", ('gpt-3.5-turbo', 'gpt-3.5-turbo-0301'))
user_prompt = st.text_area("Enter prompts, instructions & questions:", '', height=100)
collength, colupload = st.columns([2,3]) # adjust the ratio as needed
with collength:
max_length = st.slider("File section length for large files", min_value=1000, max_value=128000, value=12000, step=1000)
with colupload:
uploaded_file = st.file_uploader("Add a file for context:", type=["pdf", "xml", "json", "xlsx", "csv", "html", "htm", "md", "txt"])
document_sections = deque()
document_responses = {}
if uploaded_file is not None:
file_content = read_file_content(uploaded_file, max_length)
document_sections.extend(divide_document(file_content, max_length))
if len(document_sections) > 0:
if st.button("๐๏ธ View Upload"):
st.markdown("**Sections of the uploaded file:**")
for i, section in enumerate(list(document_sections)):
st.markdown(f"**Section {i+1}**\n{section}")
st.markdown("**Chat with the model:**")
for i, section in enumerate(list(document_sections)):
if i in document_responses:
st.markdown(f"**Section {i+1}**\n{document_responses[i]}")
else:
if st.button(f"Chat about Section {i+1}"):
st.write('Reasoning with your inputs...')
#response = chat_with_model(user_prompt, section, model_choice)
st.write('Response:')
st.write(response)
document_responses[i] = response
filename = generate_filename(f"{user_prompt}_section_{i+1}", choice)
create_file(filename, user_prompt, response, should_save)
st.sidebar.markdown(get_table_download_link(filename), unsafe_allow_html=True)
if st.button('๐ฌ Chat'):
st.write('Reasoning with your inputs...')
user_prompt_sections = divide_prompt(user_prompt, max_length)
full_response = ''
for prompt_section in user_prompt_sections:
response = chat_with_model(prompt_section, ''.join(list(document_sections)), model_choice)
full_response += response + '\n' # Combine the responses
response = full_response
st.write('Response:')
st.write(response)
filename = generate_filename(user_prompt, choice)
create_file(filename, user_prompt, response, should_save)
# Function to encode file to base64
def get_base64_encoded_file(file_path):
with open(file_path, "rb") as file:
return base64.b64encode(file.read()).decode()
# Function to create a download link
def get_audio_download_link(file_path):
base64_file = get_base64_encoded_file(file_path)
return f'โฌ๏ธ Download Audio'
# Compose a file sidebar of past encounters
all_files = glob.glob("*.wav")
all_files = [file for file in all_files if len(os.path.splitext(file)[0]) >= 10] # exclude files with short names
all_files.sort(key=lambda x: (os.path.splitext(x)[1], x), reverse=True) # sort by file type and file name in descending order
filekey = 'delall'
if st.sidebar.button("๐ Delete All Audio", key=filekey):
for file in all_files:
os.remove(file)
st.experimental_rerun()
for file in all_files:
col1, col2 = st.sidebar.columns([6, 1]) # adjust the ratio as needed
with col1:
st.markdown(file)
if st.button("๐ต", key="play_" + file): # play emoji button
audio_file = open(file, 'rb')
audio_bytes = audio_file.read()
st.audio(audio_bytes, format='audio/wav')
#st.markdown(get_audio_download_link(file), unsafe_allow_html=True)
#st.text_input(label="", value=file)
with col2:
if st.button("๐", key="delete_" + file):
os.remove(file)
st.experimental_rerun()
GiveFeedback=False
if GiveFeedback:
with st.expander("Give your feedback ๐", expanded=False):
feedback = st.radio("Step 8: Give your feedback", ("๐ Upvote", "๐ Downvote"))
if feedback == "๐ Upvote":
st.write("You upvoted ๐. Thank you for your feedback!")
else:
st.write("You downvoted ๐. Thank you for your feedback!")
load_dotenv()
st.write(css, unsafe_allow_html=True)
st.header("Chat with documents :books:")
user_question = st.text_input("Ask a question about your documents:")
if user_question:
process_user_input(user_question)
with st.sidebar:
st.subheader("Your documents")
docs = st.file_uploader("import documents", accept_multiple_files=True)
with st.spinner("Processing"):
raw = pdf2txt(docs)
if len(raw) > 0:
length = str(len(raw))
text_chunks = txt2chunks(raw)
vectorstore = vector_store(text_chunks)
st.session_state.conversation = get_chain(vectorstore)
st.markdown('# AI Search Index of Length:' + length + ' Created.') # add timing
filename = generate_filename(raw, 'txt')
create_file(filename, raw, '', should_save)
try:
query_params = st.query_params
query = (query_params.get('q') or query_params.get('query') or [''])
if query: search_glossary(query)
except:
st.markdown(' ')
# Display the glossary grid
st.markdown("### ๐ฒ๐บ๏ธ Graphic Novel Gallery")
display_videos_and_links() # Video Jump Grid
display_images_and_wikipedia_summaries() # Image Jump Grid
display_glossary_grid(roleplaying_glossary) # Word Glossary Jump Grid
display_buttons_with_scores() # Feedback Jump Grid
if 'action' in st.query_params:
action = st.query_params()['action'][0] # Get the first (or only) 'action' parameter
if action == 'show_message':
st.success("Showing a message because 'action=show_message' was found in the URL.")
elif action == 'clear':
clear_query_params()
st.experimental_rerun()
# Handling repeated keys
if 'multi' in st.query_params:
multi_values = get_all_query_params('multi')
st.write("Values for 'multi':", multi_values)
# Manual entry for demonstration
st.write("Enter query parameters in the URL like this: ?action=show_message&multi=1&multi=2")
if 'query' in st.query_params:
query = st.query_params['query'][0] # Get the query parameter
# Display content or image based on the query
display_content_or_image(query)
# Add a clear query parameters button for convenience
if st.button("Clear Query Parameters", key='ClearQueryParams'):
# This will clear the browser URL's query parameters
st.experimental_set_query_params
st.experimental_rerun()
# 18. Run AI Pipeline
if __name__ == "__main__":
whisper_main()
main()