app.py

import gradio as gr
import requests
from typing import List, Dict, Tuple
from flask import Flask, request, jsonify, send_from_directory
from transformers import AutoTokenizer, AutoModelForCausalLM
import threading
import torch
import os
import re

# Define the API URL to use the internal server
API_URL = "http://localhost:5000/chat"

History = List[Tuple[str, str]]
Messages = List[Dict[str, str]]

app = Flask(__name__)

# Load the model and tokenizer
model_name = "dicta-il/dictalm2.0-instruct"
model = AutoModelForCausalLM.from_pretrained(model_name, torch_dtype=torch.bfloat16)
tokenizer = AutoTokenizer.from_pretrained(model_name)

# Set the pad_token to eos_token if not already set
if tokenizer.pad_token is None:
    tokenizer.pad_token = tokenizer.eos_token

# Route to serve static files (e.g., images)
@app.route('/static/<path:path>')
def send_static(path):
    return send_from_directory('static', path)

@app.route('/chat', methods=['POST'])
def chat():
    data = request.json
    messages = data.get('messages', [])
    
    if not messages:
        return jsonify({"response": "No messages provided"}), 400
    
    # Concatenate all user inputs into a single string
    user_input = " ".join([msg['content'] for msg in messages if msg['role'] == 'user'])
    
    inputs = tokenizer(user_input, return_tensors='pt', padding=True, truncation=True)
    input_ids = inputs['input_ids']
    attention_mask = inputs['attention_mask']
    outputs = model.generate(input_ids, attention_mask=attention_mask, max_new_tokens=1000, pad_token_id=tokenizer.eos_token_id)
    response_text = tokenizer.decode(outputs[0], skip_special_tokens=True).replace(user_input, '').strip()
    
    return jsonify({"response": response_text})

# Function to run the Flask app
def run_flask():
    app.run(host='0.0.0.0', port=5000)

# Start the Flask app in a separate thread
threading.Thread(target=run_flask).start()

# Gradio interface functions
def clear_session() -> History:
    return []

def history_to_messages(history: History) -> Messages:
    messages = []
    for h in history:
        messages.append({'role': 'user', 'content': h[0].strip()})
        messages.append({'role': 'assistant', 'content': h[1].strip()})
    return messages

def messages_to_history(messages: Messages) -> History:
    history = []
    for q, r in zip(messages[0::2], messages[1::2]):
        history.append((q['content'], r['content']))
    return history

def is_hebrew(text: str) -> bool:
    return bool(re.search(r'[\u0590-\u05FF]', text))

def model_chat(query: str, history: History) -> Tuple[str, History]:
    if not query.strip():
        return '', history
    
    messages = history_to_messages(history)
    messages.append({'role': 'user', 'content': query.strip()})

    try:
        response = requests.post(API_URL, json={"messages": messages})
        response.raise_for_status()  # This will raise an HTTPError if the HTTP request returned an unsuccessful status code
        response_json = response.json()
        response_text = response_json.get("response", "Error: Response format is incorrect")
    except requests.exceptions.HTTPError as e:
        response_text = f"HTTPError: {str(e)}"
        print(f"HTTPError: {e.response.text}")  # Detailed error message
    except requests.exceptions.RequestException as e:
        response_text = f"RequestException: {str(e)}"
        print(f"RequestException: {e}")  # Debug print statement
    except ValueError as e:
        response_text = "ValueError: Invalid JSON response"
        print(f"ValueError: {e}")  # Debug print statement
    except Exception as e:
        response_text = f"Exception: {str(e)}"
        print(f"General Exception: {e}")  # Debug print statement
    
    history.append((query.strip(), response_text.strip()))
    return response_text.strip(), history

# Gradio interface setup
with gr.Blocks(css='''
    .gr-group {direction: rtl;}
    .chatbot{text-align:right;}
    .dicta-header {
        background-color: var(--input-background-fill);
        border-radius: 10px;
        padding: 20px;
        text-align: center;
        display: flex;
        flex-direction: row;
        align-items: center;
        box-shadow: var(--block-shadow);
        border-color: var(--block-border-color);
        border-width: 1px;
    }
    @media (max-width: 768px) {
        .dicta-header {
            flex-direction: column;
        }
    }
    .chatbot.prose {
        font-size: 1.2em;
    }
    .dicta-logo {
        width: 150px;
        height: auto;
        margin-bottom: 20px;
    }
    .dicta-intro-text {
        margin-bottom: 20px;
        text-align: center;
        display: flex;
        flex-direction: column;
        align-items: center;
        width: 100%;
        font-size: 1.1em;
    }
    textarea {
        font-size: 1.2em;
    }
''', js=None) as demo:
    gr.Markdown("""
<div class="dicta-header">
  <a href="/static/logo111.png">
    <img src="/static/logo111.png" alt="Logo" class="dicta-logo">
  </a>  
  <div class="dicta-intro-text">
    <h1>צ'אט מערכי - הדגמה ראשונית</h1>
    <span dir='rtl'>ברוכים הבאים לדמו האינטראקטיבי הראשון. חקרו את יכולות המודל וראו כיצד הוא יכול לסייע לכם במשימותיכם</span><br/>
    <span dir='rtl'>הדמו נכתב על ידי סרן רועי רתם תוך שימוש במודל שפה דיקטה שפותח על ידי מפא"ת</span><br/>
  </div>
</div>
""")
    
    chatbot = gr.Chatbot(height=600)
    query = gr.Textbox(placeholder="הכנס שאלה בעברית (או באנגלית!)", rtl=True)
    clear_btn = gr.Button("נקה שיחה")

    def respond(query, history):
        print(f"Query: {query}")  # Debug print statement
        response, history = model_chat(query, history)
        print(f"Response: {response}")  # Debug print statement
        if is_hebrew(response):
            return history, gr.update(value="", interactive=True, lines=2, rtl=True), history
        else:
            return history, gr.update(value="", interactive=True, lines=2, rtl=False), history

    demo_state = gr.State([])

    query.submit(respond, [query, demo_state], [chatbot, query, demo_state])
    clear_btn.click(clear_session, [], [chatbot, demo_state])

demo.launch(share=True)