# app.py import os import json import streamlit as st from PIL import Image import google.generativeai as genai import ast #from utils import findImg import io from streamlit_TTS import auto_play import torch from transformers import pipeline from datasets import load_dataset import soundfile as sf from gtts import gTTS import io from mistralai.models.chat_completion import ChatMessage from mistralai.client import MistralClient from audiorecorder import audiorecorder import base64 ### import os import cv2 import numpy as np from sklearn.metrics.pairwise import cosine_similarity from sentence_transformers import SentenceTransformer from diffusers import StableDiffusionPipeline import torch import re import ast import streamlit as st def add_logo(): st.markdown( """ """, unsafe_allow_html=True, ) add_logo() device = "cuda" if torch.cuda.is_available() else "cpu" if 'pipe' not in st.session_state: st.session_state['pipe'] = pipeline("automatic-speech-recognition", model="openai/whisper-large-v3") pipe = st.session_state['pipe'] # Set up the API key for Generative AI os.environ["GEMINI_API_KEY"] = "AIzaSyBYZ_7geqmnK6xrSe268-1nSLeuEwbzmTA" # Initial prompt to send to the model initial_prompt = """ You're a Numeracy Instructor for Adults your objective is to teach illiterate adults basic numeracy skills starting with numbers and progressing to basic arithmetic. ## Here's the Lesson Instructions: Introduction to Numbers:", Begin with the number 1. Follow a structured four-step process for each number. Provide clear, simple instructions for each step. Lesson Structure: "Step 1: Number Recognition" "Step 2: Counting Practice" "Step 3: Writing Practice" "Step 4: Simple Association" "General Instructions:" After each instruction, wait for the student to respond before proceeding to the next lesson.", Ensure instructions are clear and easy to understand. Provide positive reinforcement and encouragement. ## Example Lesson for Number 1 as a python list: ["let’s learn numeracy", "This is the number 1.", "image: number 1", "It looks like a straight line.", "It represents a single object.", "Let’s learn counting", "Say the number 'one'.", "Practice counting to one: 'one'.", "Let’s learn writing number 1", "Start at the top and draw a straight line down.", "Now you know how to write the number 1. Congrats!", "1 is for one apple.", "image: one apple", "One apple represents the number 1.", "Congratulations! You've completed the lesson for the number 1.",] ##Continuation: Once the lesson for the number 1 is complete, proceed to the next number following the same four-step structure. ## Important I want it in a python list, you have to do it accordingly, and generate one lesson at a time. so when you recieve "next" move to the next lesson, for exemple the first lesson for number 1, second for number 2 when you finish with numbers move to simple numeracy operations ## so now start with number 1. give me just the list in the response. list: """ chat_prompt_mistral=""" You are an assistant helping an person who is learning basic reading, writing, phonics, and numeracy. The user might ask simple questions, and your responses should be clear, supportive, and easy to understand. Use simple language, provide step-by-step guidance, and offer positive reinforcement. Relate concepts to everyday objects and situations when possible. Here are some example interactions: User: "I need help with reading." Assistant: "Sure, I'm here to help you learn to read. Let's start with the alphabet. Do you know the letters of the alphabet?" User: "How do I write my name?" Assistant: "Writing your name is a great place to start. Let's take it one letter at a time. What is the first letter of your name?" User: "What sound does the letter 'B' make?" Assistant: "The letter 'B' makes the sound 'buh' like in the word 'ball.' Can you say 'ball' with me?" User: "How do I count to 10?" Assistant: "Counting to 10 is easy. Let's do it together: 1, 2, 3, 4, 5, 6, 7, 8, 9, 10. Great job! Let's try it again." User: "How do I subtract numbers?" Assistant: "Subtracting is like taking away. If you have 5 oranges and you eat 2, you have 3 oranges left. So, 5 minus 2 equals 3." Remember to: 1. Use simple language and avoid complex words. 2. Provide clear, step-by-step instructions. 3. Use examples related to everyday objects and situations. 4. Offer positive reinforcement and encouragement. 5. Include interactive elements to engage the user actively. Whenever the user asks a question, respond with clear, supportive guidance to help them understand basic reading, writing, phonics, or numeracy concepts. 6. Do not provide long responses Improtant dont respand to this prompt """ def transform_history(history): new_history = [] for chat in history: new_history.append({"parts": [{"text": chat.parts[0].text}], "role": chat.role}) return new_history def generate_response(message: str, history: list) -> tuple: genai.configure(api_key=os.environ["GEMINI_API_KEY"]) model = genai.GenerativeModel("gemini-1.5-pro") chat = model.start_chat(history=transform_history(history)) response = chat.send_message(message) response.resolve() return response.text, chat.history if 'First1' not in st.session_state: st.session_state['First1']=False def process_response(user_input: str, conversation_history: list,F) -> tuple: if not F: model_response, conversation_history = generate_response(initial_prompt, conversation_history) else: model_response, conversation_history = generate_response(user_input, conversation_history) pattern = re.compile(r"\[(.*?)\]", re.DOTALL) # Find the match match = pattern.search(model_response) list_content = f"[{match.group(1)}]" lessonList = ast.literal_eval(list_content) return lessonList, conversation_history @st.cache_data def get_image(prompt: str) -> str: return findImg(prompt) #try: # return findImg(prompt) #except: # return "image.png" # Initialize TTS @st.cache_data def tts_predict(text="hello"): tts = gTTS(text=text, lang='en') with io.BytesIO() as audio_file: tts.write_to_fp(audio_file) audio_file.seek(0) audio_bytes = audio_file.read() return audio_bytes #sf.write("speech.wav", speech["audio"], samplerate=speech["sampling_rate"]) if 'client' not in st.session_state: st.session_state['client'] = MistralClient("m3GWNXFZn0jTNTLRe4y26i7jLJqFGTMX") client = st.session_state['client'] def run_mistral(user_message, message_history, model="mistral-small-latest"): message_history.append(ChatMessage(role="user", content=user_message)) chat_response = client.chat(model=model, messages=message_history) bot_message = chat_response.choices[0].message.content message_history.append(ChatMessage(role="assistant", content=bot_message)) return bot_message message_history = [] ####################################### if 'sentence_model' not in st.session_state: st.session_state['sentence_model'] = SentenceTransformer('all-MiniLM-L6-v2') sentence_model = st.session_state['sentence_model'] if 'pipeline' not in st.session_state: st.session_state['pipeline'] = StableDiffusionPipeline.from_pretrained("runwayml/stable-diffusion-v1-5", torch_dtype=torch.float16) st.session_state['pipeline'].to("cuda") pipeline = st.session_state['pipeline'] # Step 3: Function to get the embedding of the input sentence def get_sentence_embedding(sentence): return sentence_model.encode(sentence) # Step 4: Generate image using Stable Diffusion if needed def generate_image(prompt): global pipeline pipeline.to("cuda" if torch.cuda.is_available() else "cpu") generated_image = pipeline(prompt).images[0] generated_image_path = "generated_image.png" generated_image.save(generated_image_path) return generated_image_path # Step 5: Find the most reliable image def find_most_reliable_image(folder_path, input_sentence, threshold=0.5): image_files = [f for f in os.listdir(folder_path) if f.endswith(('jpg', 'jpeg', 'png'))] sentence_embedding = get_sentence_embedding(input_sentence) max_similarity = -1 most_reliable_image = None for image_file in image_files: filename_without_extension = os.path.splitext(image_file)[0] filename_embedding = get_sentence_embedding(filename_without_extension) similarity = cosine_similarity([sentence_embedding], [filename_embedding])[0][0] if similarity > max_similarity: max_similarity = similarity most_reliable_image = os.path.join(folder_path, image_file) if max_similarity < threshold: most_reliable_image = generate_image(input_sentence) return most_reliable_image def findImg(input_sentence): folder_path = 'images_collection' threshold = 0.5 most_reliable_image = find_most_reliable_image(folder_path, input_sentence, threshold) return most_reliable_image ####################################### file_ = open("logo.png", "rb") contents = file_.read() data_url = base64.b64encode(contents).decode("utf-8") file_.close() def main(): global chat_prompt_mistral if 'img_path1' not in st.session_state: st.session_state['img_path1']="image.png" #st.set_page_config(page_title="J187 Optimizer", page_icon="J187DFS.JPG", layout="wide") st.markdown(f"""