app.py

"""
MM Coder Agent v1 - Gradio Space
A coding assistant AI based on Qwen2.5-1.5B
"""
import gradio as gr
from transformers import AutoModelForCausalLM, AutoTokenizer
import torch
from peft import PeftModel, PeftConfig
import os

# Configuration
MODEL_ID = "amkyawdev/mm-coder-agent-v1-combined"

# System prompt for better responses
SYSTEM_PROMPT = """You are a Python coding assistant. 

Output requirements:
1. Output ONLY the code, nothing else
2. Use proper Python syntax
3. Include necessary imports
4. Start with code directly in a python code block
5. Do not add explanations unless asked

Examples:
Input: print hello world
Output:
```python
print("Hello, World!")
```"""

# Global variables to store model and tokenizer
model = None
tokenizer = None

def load_model():
    """Load the model and tokenizer"""
    global model, tokenizer
    
    print("Loading model...")
    peft_config = PeftConfig.from_pretrained(MODEL_ID)
    
    base_model = AutoModelForCausalLM.from_pretrained(
        peft_config.base_model_name_or_path,
        torch_dtype=torch.float16,
        device_map="auto",
        trust_remote_code=True
    ).eval()
    
    tokenizer = AutoTokenizer.from_pretrained(
        peft_config.base_model_name_or_path,
        trust_remote_code=True
    )
    
    model = PeftModel.from_pretrained(base_model, MODEL_ID)
    print("Model loaded successfully!")

def generate_code(prompt, max_tokens=512, temperature=0.7):
    """Generate code based on the prompt"""
    if model is None:
        load_model()
    
    # Format prompt with system instruction
    formatted_prompt = f"{SYSTEM_PROMPT}\n\nUser: {prompt}\nAssistant:"
    
    inputs = tokenizer(formatted_prompt, return_tensors="pt").to(model.device)
    
    # Get pad token id safely
    pad_token_id = tokenizer.pad_token_id if tokenizer.pad_token_id is not None else tokenizer.eos_token_id
    
    with torch.no_grad():
        outputs = model.generate(
            **inputs, 
            max_new_tokens=int(max_tokens), 
            temperature=temperature,
            do_sample=True,
            top_p=0.95,
            pad_token_id=pad_token_id,
            eos_token_id=pad_token_id
        )
    
    response = tokenizer.decode(outputs[0], skip_special_tokens=True)
    
    # Remove the prompt and system prompt from response
    prefix_to_remove = f"{SYSTEM_PROMPT}\n\nUser: {prompt}\nAssistant:"
    if response.startswith(prefix_to_remove):
        response = response[len(prefix_to_remove):].strip()
    elif response.startswith(prompt):
        response = response[len(prompt):].strip()
    
    return response

# Create Gradio Interface
with gr.Blocks(title="🤖 MM Coder Agent v1") as demo:
    gr.Markdown("""
    # 🤖 MM Coder Agent v1
    
    A professional AI coding assistant fine-tuned from Qwen2.5-1.5B-Instruct.
    
    ## Features
    - Code Generation in multiple languages
    - Bug Detection and Fixing
    - Algorithm Implementation
    - Code Review Assistance
    """)
    
    with gr.Row():
        with gr.Column():
            prompt = gr.Textbox(
                label="Your Prompt", 
                placeholder="Write a Python function to calculate fibonacci numbers...",
                lines=5
            )
            with gr.Row():
                max_tokens = gr.Slider(64, 1024, value=512, step=64, label="Max Tokens")
                temperature = gr.Slider(0.1, 1.5, value=0.7, step=0.1, label="Temperature")
            
            generate_btn = gr.Button("Generate Code", variant="primary")
        
        with gr.Column():
            output = gr.Code(label="Generated Code", language="python")
    
    generate_btn.click(
        generate_code, 
        [prompt, max_tokens, temperature], 
        output
    )
    
    gr.Markdown("""
    ---
    ### Examples
    - "Write a Python function to reverse a string"
    - "Create a React component for a login form"
    - "How to fix CORS error in Express.js?"
    - "Explain bubble sort algorithm"
    """)

# Launch the demo
demo.launch(server_name="0.0.0.0", server_port=7860)