app.py

import numpy as np
import streamlit as st
import torch
from transformers import AutoModelForCausalLM, AutoTokenizer
import os

# Random dog images for error messages
random_dog = [
    "0f476473-2d8b-415e-b944-483768418a95.jpg",
    "1bd75c81-f1d7-4e55-9310-a27595fa8762.jpg",
]

def reset_conversation():
    '''Resets conversation'''
    st.session_state.conversation = []
    st.session_state.messages = []
    return None

# Sidebar controls
temp_values = st.sidebar.slider('Select a temperature value', 0.0, 1.0, 0.5)
max_token_value = st.sidebar.slider('Select a max_token value', 1000, 9000, 5000)
st.sidebar.button('Reset Chat', on_click=reset_conversation)

# Initialize chat history
if "messages" not in st.session_state:
    st.session_state.messages = []

# Display chat messages from history on app rerun
for message in st.session_state.messages:
    with st.chat_message(message["role"]):
        st.markdown(message["content"])

# Set cache directory path to /data
cache_dir = "/data"

# Load model and tokenizer on-demand to save memory
if prompt := st.chat_input(f"مرحبا انا سبيدي , كيف استطيع مساعدتك ؟"):
    with st.chat_message("user"):
        st.markdown(prompt)
    st.session_state.messages.append({"role": "user", "content": prompt})

    try:
        # Load the tokenizer and model with specific configuration
        tokenizer = AutoTokenizer.from_pretrained(
            "joermd/llma-speedy",
            cache_dir=cache_dir,
            local_files_only=False  # السماح بتحميل الملفات المتوفرة فقط
        )
        
        model = AutoModelForCausalLM.from_pretrained(
            "joermd/llma-speedy",
            cache_dir=cache_dir,
            local_files_only=False,  # السماح بتحميل الملفات المتوفرة فقط
            ignore_missing_weights=True  # تجاهل الأوزان المفقودة
        )
        
        # Generate response
        inputs = tokenizer(prompt, return_tensors="pt")
        outputs = model.generate(
            inputs.input_ids,
            max_new_tokens=max_token_value,
            temperature=temp_values,
            do_sample=True
        )
        assistant_response = tokenizer.decode(outputs[0], skip_special_tokens=True)
        
        # Clear memory
        if torch.cuda.is_available():
            torch.cuda.empty_cache()
        del model
    
    except Exception as e:
        assistant_response = "😵‍💫 Connection issue! Try again later. Here's a 🐶:"
        st.image(f'https://random.dog/{random_dog[np.random.randint(len(random_dog))]}')
        st.write("Error message:")
        st.write(e)

    # Display assistant response
    with st.chat_message("assistant"):
        st.markdown(assistant_response)
    st.session_state.messages.append({"role": "assistant", "content": assistant_response})