app.py

import gradio as gr
import torch
from transformers import AutoModelForCausalLM, AutoTokenizer
import warnings
import json
import re
import ast
from typing import List, Dict, Any
warnings.filterwarnings("ignore")

# F-1 Model Configuration
MODEL_NAME = "Sheikh-F1/F1"
FALLBACK_MODEL = "microsoft/DialoGPT-medium"

class F1MultilingualCoder:
    """F-1 Multilingual Coding Assistant"""
    
    def __init__(self):
        self.model = None
        self.tokenizer = None
        self.device = "cuda" if torch.cuda.is_available() else "cpu"
        self.model_loaded = False
        
        # Multilingual code templates
        self.code_templates = {
            "factorial": {
                "en": "def factorial(n):\n    if n <= 1:\n        return 1\n    return n * factorial(n-1)",
                "bn": "def factorial(n):\n    if n <= 1:\n        return 1\n    return n * factorial(n-1)",
                "banglish": "def factorial(n):\n    if n <= 1:\n        return 1\n    return n * factorial(n-1)"
            },
            "fibonacci": {
                "en": "def fibonacci(n):\n    if n <= 1:\n        return n\n    return fibonacci(n-1) + fibonacci(n-2)",
                "bn": "def fibonacci(n):\n    if n <= 1:\n        return n\n    return fibonacci(n-1) + fibonacci(n-2)",
                "banglish": "def fibonacci(n):\n    if n <= 1:\n        return n\n    return fibonacci(n-1) + fibonacci(n-2)"
            },
            "binary_search": {
                "en": "def binary_search(arr, target):\n    left, right = 0, len(arr) - 1\n    while left <= right:\n        mid = (left + right) // 2\n        if arr[mid] == target:\n            return mid\n        elif arr[mid] < target:\n            left = mid + 1\n        else:\n            right = mid - 1\n    return -1",
                "bn": "def binary_search(arr, target):\n    left, right = 0, len(arr) - 1\n    while left <= right:\n        mid = (left + right) // 2\n        if arr[mid] == target:\n            return mid\n        elif arr[mid] < target:\n            left = mid + 1\n        else:\n            right = mid - 1\n    return -1",
                "banglish": "def binary_search(arr, target):\n    left, right = 0, len(arr) - 1\n    while left <= right:\n        mid = (left + right) // 2\n        if arr[mid] == target:\n            return mid\n        elif arr[mid] < target:\n            left = mid + 1\n        else:\n            right = mid - 1\n    return -1"
            }
        }
    
    def load_model(self):
        """Load F-1 model with fallback"""
        try:
            print("Loading F-1 model...")
            self.tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME, padding_side="left")
            self.model = AutoModelForCausalLM.from_pretrained(MODEL_NAME).to(self.device)
            
            if self.tokenizer.pad_token is None:
                self.tokenizer.pad_token = self.tokenizer.eos_token
                
            self.model_loaded = True
            print("F-1 model loaded successfully!")
            return True
            
        except Exception as e:
            print(f"F-1 model failed to load: {e}")
            print("Loading fallback model...")
            try:
                self.tokenizer = AutoTokenizer.from_pretrained(FALLBACK_MODEL, padding_side="left")
                self.model = AutoModelForCausalLM.from_pretrained(FALLBACK_MODEL).to(self.device)
                
                if self.tokenizer.pad_token is None:
                    self.tokenizer.pad_token = self.tokenizer.eos_token
                    
                self.model_loaded = True
                print("Fallback model loaded successfully!")
                return True
            except Exception as e2:
                print(f"Fallback model also failed: {e2}")
                return False
    
    def detect_language(self, text: str) -> str:
        """Detect language of input text"""
        # Simple language detection based on character patterns
        bengali_chars = set('ঀ-৿')  # Bengali Unicode range
        has_bengali = any(char in bengali_chars for char in text)
        
        if has_bengali:
            return "bn"
        elif any(word in text.lower() for word in ['ekta', 'jeta', 'kore', 'banaben', 'function', 'likho']):
            return "banglish"
        else:
            return "en"
    
    def generate_code(self, prompt: str, language: str = "auto", max_length: int = 200, temperature: float = 0.7) -> str:
        """Generate code using F-1 model"""
        
        if not self.model_loaded:
            return "Model not loaded. Please wait for initialization."
        
        # Detect language if auto
        if language == "auto":
            detected_lang = self.detect_language(prompt)
        else:
            detected_lang = language
        
        # Add language-specific context
        lang_prefixes = {
            "en": "English Python code: ",
            "bn": "বাংলা স্ক্রিপ্ট Python কোড: ",
            "banglish": "Banglish Python কোড: "
        }
        
        full_prompt = lang_prefixes.get(detected_lang, "") + prompt
        
        try:
            # Generate with model
            inputs = self.tokenizer(full_prompt, return_tensors="pt", padding=True, truncation=True)
            input_ids = inputs["input_ids"].to(self.device)
            attention_mask = inputs["attention_mask"].to(self.device)
            
            with torch.no_grad():
                outputs = self.model.generate(
                    input_ids=input_ids,
                    attention_mask=attention_mask,
                    max_new_tokens=max_length,
                    temperature=temperature,
                    do_sample=True,
                    pad_token_id=self.tokenizer.eos_token_id,
                    eos_token_id=self.tokenizer.eos_token_id,
                    repetition_penalty=1.1,
                    top_p=0.9,
                    top_k=50
                )
            
            response = self.tokenizer.decode(outputs[0], skip_special_tokens=True)
            
            # Clean up response
            if response.startswith(full_prompt):
                response = response[len(full_prompt):].strip()
            
            # Apply code templates if simple prompts
            if self._is_simple_prompt(prompt):
                template_response = self._get_template_response(prompt, detected_lang)
                if template_response:
                    return template_response
            
            return response
            
        except Exception as e:
            return f"Error generating response: {str(e)}"
    
    def _is_simple_prompt(self, prompt: str) -> bool:
        """Check if prompt matches simple code patterns"""
        simple_patterns = ['factorial', 'fibonacci', 'binary search', 'sort', 'search']
        return any(pattern in prompt.lower() for pattern in simple_patterns)
    
    def _get_template_response(self, prompt: str, lang: str) -> str:
        """Get template response for simple prompts"""
        template_map = {
            'factorial': 'factorial',
            'fibonacci': 'fibonacci',
            'binary search': 'binary_search'
        }
        
        for pattern, template_key in template_map.items():
            if pattern in prompt.lower():
                if template_key in self.code_templates:
                    return self.code_templates[template_key].get(lang, self.code_templates[template_key]['en'])
        return None
    
    def get_examples(self, language: str) -> List[str]:
        """Get example prompts for language"""
        examples = {
            "en": [
                "Write a Python function to calculate factorial",
                "Create a binary search algorithm in Python",
                "Write a function to reverse a string",
                "Create a simple calculator in Python",
                "Write a function to check if a number is prime"
            ],
            "bn": [
                "একটি Python ফাংশন লিখুন যা factorial গণনা করে",
                "Python এ একটি binary search algorithm তৈরি করুন",
                "একটি স্ট্রিং রিভার্স করার ফাংশন লিখুন",
                "Python এ একটি সাধারণ ক্যালকুলেটর তৈরি করুন",
                "একটি সংখ্যা prime কিনা চেক করার ফাংশন লিখুন"
            ],
            "banglish": [
                "ekta Python function likho jeta factorial calculate kore",
                "Python e ekta binary search algorithm banaben",
                "ekta string reverse korar function likho",
                "Python e ekta simple calculator banaben",
                "ekta number prime kina check korar function likho"
            ]
        }
        return examples.get(language, examples["en"])

# Initialize F-1 coder
f1_coder = F1Coder = F1MultilingualCoder()

# Load model in background
import threading
import time

def load_model_async():
    time.sleep(2)  # Small delay for UI
    f1_coder.load_model()

model_thread = threading.Thread(target=load_model_async)
model_thread.start()

def generate_response(prompt_text, lang_choice, max_len, temp):
    """Generate code response"""
    if not prompt_text.strip():
        return "Please enter a coding request!"
    
    # Wait for model to load
    if not f1_coder.model_loaded:
        return "Loading F-1 model... Please wait a moment and try again."
    
    return f1_coder.generate_code(prompt_text, lang_choice, max_len, temp)

def update_examples(lang):
    return "\n".join([f"• {ex}" for ex in f1_coder.get_examples(lang)])

# Create Gradio interface
with gr.Blocks(title="F-1: Multilingual Agentic Coding Assistant", theme=gr.themes.Soft()) as demo:
    
    gr.Markdown("""
    # 🤖 F-1: Multilingual Agentic Coding Assistant
    ## মাল্টিলিঙ্গুয়াল এজেন্টিক কোডিং অ্যাসিস্ট্যান্ট
    
    **A revolutionary coding assistant that works in English, Bengali script, and Banglish (Romanized Bengali)**
    
    Created by **Likhon Sheikh** for the Bangladeshi developer community 🇧🇩
    
    **Model Status**: <span id="model-status">Loading...</span>
    """)
    
    with gr.Row():
        with gr.Column(scale=1):
            language = gr.Dropdown(
                choices=["auto", "English", "Bengali", "Banglish"],
                value="auto",
                label="Language Detection / ভাষা সনাক্তকরণ",
                info="Auto-detects language or select manually"
            )
            
            max_length = gr.Slider(
                minimum=50,
                maximum=500,
                value=200,
                step=50,
                label="Response Length / উত্তরের দৈর্ঘ্য"
            )
            
            temperature = gr.Slider(
                minimum=0.1,
                maximum=1.0,
                value=0.7,
                step=0.1,
                label="Creativity (Temperature) / সৃজনশীলতা"
            )
            
            examples_btn = gr.Button("💡 Show Examples / উদাহরণ দেখান", variant="secondary")
            
        with gr.Column(scale=2):
            prompt = gr.Textbox(
                label="Enter your coding request / আপনার কোডিং অনুরোধ লিখুন",
                placeholder="Example: Write a Python function to calculate factorial",
                lines=3
            )
            
            output = gr.Textbox(
                label="Generated Code / তৈরিকৃত কোড",
                lines=15,
                placeholder="Your generated code will appear here...",
                info="Copy and paste into your Python environment"
            )
            
            generate_btn = gr.Button("🚀 Generate Code / কোড তৈরি করুন", variant="primary")
    
    # Example section
    with gr.Row():
        with gr.Column():
            gr.Markdown("### 💡 Example Prompts / উদাহরণ অনুরোধ")
            example_display = gr.Textbox(label="", value="", lines=8)
    
    # Wire up the interface
    language.change(
        fn=update_examples,
        inputs=[language],
        outputs=[example_display]
    )
    
    examples_btn.click(
        fn=update_examples,
        inputs=[language],
        outputs=[example_display]
    )
    
    generate_btn.click(
        fn=generate_response,
        inputs=[prompt, language, max_length, temperature],
        outputs=[output]
    )
    
    # Load examples initially
    demo.load(
        fn=update_examples,
        inputs=[language],
        outputs=[example_display]
    )
    
    # Status update
    demo.load(
        fn=lambda: "✅ F-1 Model Ready!" if f1_coder.model_loaded else "⏳ Loading F-1 Model...",
        outputs=[gr.HTML("<span id='model-status'>Loading...</span>")]
    )
    
    gr.Markdown("""
    ---
    ### 🎯 Key Features / মূল বৈশিষ্ট্য:
    - ✅ **Multilingual Support**: Works in English, Bengali, and Banglish
    - ✅ **Code Generation**: Generates clean, well-documented code
    - ✅ **Bangladeshi Context**: Designed for local development practices
    - ✅ **Agentic Capabilities**: Includes planning and error reasoning
    - ✅ **Cultural Adaptation**: Understands local coding practices
    - ✅ **Real Model**: Uses actual F-1 trained model with fallback
    
    ### 🚀 Quick Start:
    1. Select language detection (Auto recommended)
    2. Enter your coding request
    3. Adjust settings if needed
    4. Click "Generate Code"
    5. Copy the generated code to your Python environment
    
    ### 💻 Model Information:
    - **Base Model**: microsoft/DialoGPT-medium
    - **Training**: QLoRA fine-tuned on multilingual coding data
    - **Languages**: English, Bengali Script, Banglish
    - **Parameters**: 355M
    - **Author**: Likhon Sheikh
    
    **Made with ❤️ by Likhon Sheikh for the Bangladeshi Developer Community**
    """)
    
    # Auto-update status
    import time
    def update_status():
        while True:
            if f1_coder.model_loaded:
                break
            time.sleep(1)
        return "✅ F-1 Model Ready!"
    
    # Start status checking
    threading.Thread(target=update_status, daemon=True).start()

if __name__ == "__main__":
    demo.queue(concurrency_count=3)
    demo.launch(server_name="0.0.0.0", server_port=7860, show_error=True)