app.py

import sys
import toml
from omegaconf import OmegaConf
from query import VectaraQuery
import os
from transformers import pipeline
import numpy as np
import tempfile

import streamlit as st
from PIL import Image

master_prompt = """
As a Natural Farming Fertilizers Assistant, you will assist the user with any farming related question, always willing to answer any question and provide useful organic farming advice in the following format.
' ' '
** Format is: **
[Short Introduction]
[Nutritional Needs of the user's crops]
[List of plants available locally with the needed nutrients (using the chunks provided.) At least 5 different plants.]
[List of ingredients, quantities of those ingredients needed to fertilize the crop stated, and steps for multiple fertilizer Recipes  (using the chunks provided as Bioaccumulators List, you will match plants on the Bioaccumulators List with plants locally growing in the user's area)]
[Give three different sets of recipes using ingredients locally available for free to the user]
[Tables with bioaccumulators data and crop needs data, showing wildcrafted plant nutrient levels and crop nutritional needs, in text table format (not visual)]
[Instructions on using the fertilizers (SOPs)]
[Fertilizer application schedule (step by step in fundamental details) and crop rotation reccomendations]
[Brief Philosophical encouragement related to Natural Farming]
[Alternative set of recipes using localized free ingredients]
[Words of encouragement]
' ' '
User prompt:
"""

#temporary file system created: used to text-to-speech
fp = tempfile.TemporaryFile()

def launch_bot():
    def generate_response(question):
        response = vq.submit_query(question)
        return response

    if 'cfg' not in st.session_state:
        corpus_ids = str(os.environ['corpus_ids']).split(',')
        questions = list(eval(os.environ['examples']))
        cfg = OmegaConf.create({
            'customer_id': str(os.environ['customer_id']),
            'corpus_ids': corpus_ids,
            'api_key': str(os.environ['api_key']),
            'title': os.environ['title'],
            'description': os.environ['description'],
            'examples': questions,
            'source_data_desc': os.environ['source_data_desc']
        })
        st.session_state.cfg = cfg
        st.session_state.vq = VectaraQuery(cfg.api_key, cfg.customer_id, cfg.corpus_ids)

    cfg = st.session_state.cfg
    vq = st.session_state.vq
    st.set_page_config(page_title=cfg.title, layout="wide")

    # left side content
    with st.sidebar:
        image = Image.open('Vectara-logo.png')
        st.markdown(f"## Welcome to {cfg.title}\n\n"
                    f"This demo uses an AI organic farming expert and carefully currated library system to achieve greater accuracy in agronomics and agricultural methodology. Created by Copyleft Cultivars, a nonprofit, we hope you enjoy this beta-test early access version.\n\n")

        st.markdown("---")
        st.markdown(
            "## Democratizing access to farming knowledge.\n"
            "This app was built with the support of our Patreon subscribers. Thank you! [Click here to join our patreon or upgrade your membership.](https://www.patreon.com/CopyleftCultivarsNonprofit). \n"
        )
        st.markdown("---")
        st.image(image, width=250)

    st.markdown(f"<center> <h2> Copyleft Cultivars AI Agriculture Assistant demo: {cfg.title} </h2> </center>", unsafe_allow_html=True)
    st.markdown(f"<center> <h4> {cfg.description} <h4> </center>", unsafe_allow_html=True)

    if "messages" not in st.session_state.keys():
        st.session_state.messages = [{"role": "assistant", "content": "How may I help you?"}]

    # Display chat messages
    for message in st.session_state.messages:
        with st.chat_message(message["role"]):
            st.write(message["content"])

    # User-provided prompt
    if prompt := st.chat_input():
        st.session_state.messages.append({"role": "user", "content": prompt})
        with st.chat_message("user"):
            st.write(prompt)
    
    # Generate a new response if last message is not from assistant
    if st.session_state.messages[-1]["role"] != "assistant":
        with st.chat_message("assistant"):
            with st.spinner("Thinking..."):
                prompt2 = prompt + master_prompt
                response = generate_response(prompt2) 
              #  if response == 'The returned results did not contain sufficient information to be summarized into a useful answer for your query. Please try a different search or restate your query differently.':
                    #st.write("reroute to LLM")
                    #call in Mistral
                prompt3 = master_prompt + prompt2 + "context:" + response
                print("Here's where we would call in Mistral")
                    # ADD IN LLM
                # st.write("Mistral:" ) #Needs finishing
              #  else:
                st.write(response) 
        message = {"role": "assistant", "content": response}
        st.session_state.messages.append(message)

    # If assistant has most recently reaponded create audio of response 
    if st.session_state.messages[-1]["role"] == "assistant":
        audio_result = st.button("Convert to Audio 🔊") 
        if audio_result:
            st.session_state.messages.append({"role": "assistant", "content": prompt})
            with st.chat_message("text-to-speech (may take a few minutes"):
            #text-to-speech
                print("Calling in Text-to-speech via suno/bark-small")
                    
                synthesiser = pipeline("text-to-speech", "suno/bark-small")
            
                speech = synthesiser(response, forward_params={"do_sample": True})

                # write to temp file: the scipy wav write bark_out.wav
                fp.write(scipy.io.wavfile.write("bark_out.wav", rate=speech["sampling_rate"], data=speech["audio"]))
                    
            # ST interface for audio
                print("Now we try to display the audio file in the app")
                
                # Temp file access
                fp.seek(0)
                audio_file = fp.read('bark_out.wav', 'rb')
                    
                audio_bytes = audio_file.read()
                
                st.audio(audio_bytes, format='audio/wav')
                
                #audio_convert = st.audio(audio_bytes, format='audio/wav')
                #st.session_state.messages.append(audio_convert)
                #sample_rate = 44100  # 44100 samples per second
                #seconds = 2  # Note duration of 2 seconds
                    
                #frequency_la = 440  # Our played note will be 440 Hz
                    
                # Generate array with seconds*sample_rate steps, ranging between 0 and seconds
                #t = np.linspace(0, seconds, seconds * sample_rate, False)
                    
                # Generate a 440 Hz sine wave
                #note_la = np.sin(frequency_la * t * 2 * np.pi)
                #st.audio(note_la, sample_rate=sample_rate)

                # Close and delete temp file
                fp.close()
                
if __name__ == "__main__":
    launch_bot()