Spaces:
Running
Running
| import gradio as gr | |
| from huggingface_hub import InferenceClient | |
| import os | |
| import re | |
| # import traceback # Optional: for more detailed error logging if needed | |
| # --- Configuration --- | |
| API_TOKEN = os.getenv("HF_TOKEN", None) | |
| MODEL = "Qwen/Qwen2.5-Coder-32B-Instruct" | |
| # --- Initialize Inference Client --- | |
| try: | |
| print(f"Initializing Inference Client for model: {MODEL}") | |
| client = InferenceClient(model=MODEL, token=API_TOKEN) if API_TOKEN else InferenceClient(model=MODEL) | |
| except Exception as e: | |
| raise gr.Error(f"Failed to initialize model client for {MODEL}. Error: {e}. Check HF_TOKEN and model availability.") | |
| # --- Helper Function to Parse Code into Files --- | |
| def parse_code_into_files(raw_response: str) -> dict: | |
| """ | |
| Parses raw AI output containing .TAB separators | |
| into a dictionary where keys are filenames and values are code blocks. | |
| Returns keys like 'index.html', 'backend_file', 'backend_filename', 'backend_language'. | |
| """ | |
| files = {} | |
| # Default filename for the first block if no TAB is present or before the first TAB | |
| default_first_filename = "index.html" | |
| separator_pattern = r'\.TAB\[NAME=([^\]]+)\]\n?' # Capture filename | |
| # Find all separators and their positions | |
| matches = list(re.finditer(separator_pattern, raw_response)) | |
| start_index = 0 | |
| # Handle the first file (always assume index.html for now) | |
| first_separator_pos = matches[0].start() if matches else len(raw_response) | |
| first_block = raw_response[start_index:first_separator_pos].strip() | |
| if first_block: | |
| files[default_first_filename] = first_block | |
| # Handle the second file (if separator exists) | |
| if matches: | |
| backend_filename = matches[0].group(1).strip() # Get filename from first match | |
| start_index = matches[0].end() # Start after the first separator | |
| # Find the position of the *next* separator, or end of string | |
| second_separator_pos = matches[1].start() if len(matches) > 1 else len(raw_response) | |
| backend_code = raw_response[start_index:second_separator_pos].strip() | |
| if backend_code: | |
| files['backend_file'] = backend_code | |
| files['backend_filename'] = backend_filename | |
| # Determine language from filename extension | |
| if backend_filename.endswith(".py"): | |
| files['backend_language'] = 'python' | |
| elif backend_filename.endswith(".js"): | |
| files['backend_language'] = 'javascript' | |
| elif backend_filename.endswith(".css"): | |
| files['backend_language'] = 'css' | |
| else: | |
| files['backend_language'] = None # Default to plain text | |
| # If more files were generated (more separators), they are currently ignored by this simple parser. | |
| return files | |
| # --- Core Code Generation Function --- | |
| def generate_code( | |
| prompt: str, | |
| backend_choice: str, | |
| max_tokens: int, | |
| temperature: float, | |
| top_p: float, | |
| progress=gr.Progress(track_ τότε=True) # Add progress tracker | |
| ): | |
| print(f"Generating code for: {prompt[:100]}... | Backend: {backend_choice}") | |
| progress(0, desc="Initializing Request...") | |
| # System message remains the same - instructing the AI on format | |
| system_message = ( | |
| "You are an AI that generates website code. You MUST ONLY output the raw code, without any conversational text like 'Here is the code' or explanations before or after the code blocks. " | |
| "You MUST NOT wrap the code in markdown fences like ```html, ```python, or ```js. " | |
| "If the user requests 'Static' or the prompt clearly implies only frontend code, generate ONLY the content for the `index.html` file. " | |
| "If the user requests 'Flask' or 'Node.js' and the prompt requires backend logic, you MUST generate both the `index.html` content AND the corresponding main backend file content (e.g., `app.py` for Flask, `server.js` or `app.js` for Node.js). " | |
| "When generating multiple files, you MUST separate them EXACTLY as follows: " | |
| "1. Output the complete code for the first file (e.g., `index.html`). " | |
| "2. On a new line immediately after the first file's code, add the separator '.TAB[NAME=filename.ext]' (e.g., '.TAB[NAME=app.py]' or '.TAB[NAME=server.js]'). " | |
| "3. On the next line, immediately start the code for the second file. " | |
| "Generate only the necessary files (usually index.html and potentially one backend file). " | |
| "The generated website code must be SFW and have minimal errors. " | |
| "Only include comments where user modification is strictly required. Avoid explanatory comments. " | |
| "If the user asks you to create code that is NOT for a website, you MUST respond ONLY with the exact phrase: " | |
| "'hey there! am here to create websites for you unfortunately am programmed to not create codes! otherwise I would go on the naughty list :-('" | |
| ) | |
| user_prompt = f"USER_PROMPT = {prompt}\nUSER_BACKEND = {backend_choice}" | |
| messages = [ | |
| {"role": "system", "content": system_message}, | |
| {"role": "user", "content": user_prompt} | |
| ] | |
| full_response = "" | |
| token_count = 0 | |
| est_total_tokens = max_tokens # Rough estimate for progress | |
| try: | |
| progress(0.1, desc="Sending Request to Model...") | |
| stream = client.chat_completion( | |
| messages=messages, | |
| max_tokens=max_tokens, | |
| stream=True, | |
| temperature=temperature, | |
| top_p=top_p, | |
| ) | |
| progress(0.2, desc="Receiving Stream...") | |
| for message in stream: | |
| token = message.choices[0].delta.content | |
| if isinstance(token, str): | |
| full_response += token | |
| token_count += 1 | |
| # Update progress based on tokens received vs max_tokens | |
| # Adjust the scaling factor (e.g., 0.7) as needed | |
| prog = min(0.2 + (token_count / est_total_tokens) * 0.7, 0.9) | |
| progress(prog, desc="Generating Code...") | |
| progress(0.9, desc="Processing Response...") | |
| # --- Post-processing --- | |
| cleaned_response = full_response.strip() | |
| # Fallback fence removal | |
| cleaned_response = re.sub(r"^\s*```[a-z]*\s*\n?", "", cleaned_response) | |
| cleaned_response = re.sub(r"\n?\s*```\s*$", "", cleaned_response) | |
| # Remove potential chat markers | |
| cleaned_response = re.sub(r"<\s*\|?\s*(user|system|assistant)\s*\|?\s*>", "", cleaned_response, flags=re.IGNORECASE).strip() | |
| # Remove common conversational phrases (if they slip through) | |
| common_phrases = [ # Simplified list as prompt should handle most | |
| "Here is the code:", "Okay, here is the code:", "Here's the code:", | |
| "Sure, here is the code you requested:", | |
| ] | |
| temp_response_lower = cleaned_response.lower() | |
| for phrase in common_phrases: | |
| if temp_response_lower.startswith(phrase.lower()): | |
| cleaned_response = cleaned_response[len(phrase):].lstrip() | |
| temp_response_lower = cleaned_response.lower() | |
| # Check for refusal message | |
| refusal_message = "hey there! am here to create websites for you unfortunately am programmed to not create codes! otherwise I would go on the naughty list :-(" | |
| if refusal_message in full_response: | |
| # Return updates to clear both code blocks and show refusal in the first | |
| progress(1, desc="Refusal Message Generated") | |
| return gr.update(value=refusal_message, language=None, visible=True), gr.update(value="", visible=False) | |
| # --- PARSE the final cleaned response into files --- | |
| parsed_files = parse_code_into_files(cleaned_response) | |
| html_code = parsed_files.get("index.html", "") | |
| backend_code = parsed_files.get("backend_file", "") | |
| backend_filename = parsed_files.get("backend_filename", "Backend") | |
| backend_language = parsed_files.get("backend_language", None) | |
| # --- Prepare Gradio Updates --- | |
| # Update for the HTML code block (always visible) | |
| html_update = gr.update(value=html_code, language='html', visible=True) | |
| # Update for the Backend code block (visible only if backend code exists) | |
| if backend_code: | |
| backend_update = gr.update(value=backend_code, language=backend_language, label=backend_filename, visible=True) | |
| else: | |
| backend_update = gr.update(value="", visible=False) # Hide if no backend code | |
| progress(1, desc="Done") | |
| # Return tuple of updates for the outputs list | |
| return html_update, backend_update | |
| except Exception as e: | |
| print(f"ERROR during code generation: {e}") # Log detailed error | |
| # traceback.print_exc() # Uncomment for full traceback | |
| progress(1, desc="Error Occurred") | |
| error_message = f"## Error\n\nFailed to generate or process code.\n**Reason:** {e}" | |
| # Return updates to show error in the first block and hide the second | |
| return gr.update(value=error_message, language=None, visible=True), gr.update(value="", visible=False) | |
| # --- Build Gradio Interface --- | |
| with gr.Blocks(css=".gradio-container { max-width: 90% !important; }") as demo: | |
| gr.Markdown("# ✨ Website Code Generator ✨") | |
| gr.Markdown( | |
| "Describe the website you want. The AI will generate the necessary code.\n" | |
| "If multiple files are generated (e.g., for Flask/Node.js), they will appear in separate tabs below." # Updated description | |
| ) | |
| with gr.Row(): | |
| with gr.Column(scale=2): | |
| prompt_input = gr.Textbox( | |
| label="Website Description", | |
| placeholder="e.g., A Flask app with a simple chat using Socket.IO", | |
| lines=6, | |
| ) | |
| backend_radio = gr.Radio( | |
| ["Static", "Flask", "Node.js"], | |
| label="Backend Context", | |
| value="Static", | |
| info="Guides AI if backend code (like Python/JS) is needed alongside HTML." | |
| ) | |
| generate_button = gr.Button("✨ Generate Website Code", variant="primary") | |
| with gr.Column(scale=3): | |
| # Define Tabs to hold the code outputs | |
| with gr.Tabs(elem_id="code-tabs") as code_tabs: | |
| # Tab 1: Always present for HTML | |
| with gr.Tab("index.html", elem_id="html-tab") as html_tab: | |
| html_code_output = gr.Code( | |
| label="index.html", # Label for the code block itself | |
| language="html", | |
| lines=25, # Adjusted lines slightly | |
| interactive=False, | |
| elem_id="html_code", # Unique ID for targeting | |
| ) | |
| # Tab 2: For Backend code, initially hidden | |
| with gr.Tab("Backend", elem_id="backend-tab", visible=False) as backend_tab: | |
| backend_code_output = gr.Code( | |
| label="Backend Code", # Label will be updated dynamically | |
| language=None, # Language updated dynamically | |
| lines=25, | |
| interactive=False, | |
| elem_id="backend_code", # Unique ID for targeting | |
| visible=False # Component also starts hidden | |
| ) | |
| # Add more tabs here if needed (e.g., for CSS) following the same pattern | |
| with gr.Accordion("Advanced Settings", open=False): | |
| max_tokens_slider = gr.Slider( | |
| minimum=512, maximum=4096, value=3072, step=256, label="Max New Tokens" | |
| ) | |
| temperature_slider = gr.Slider( | |
| minimum=0.1, maximum=1.2, value=0.7, step=0.1, label="Temperature" | |
| ) | |
| top_p_slider = gr.Slider( | |
| minimum=0.1, maximum=1.0, value=0.9, step=0.05, label="Top-P" | |
| ) | |
| # The click function now targets the specific code blocks within the tabs | |
| generate_button.click( | |
| fn=generate_code, | |
| inputs=[prompt_input, backend_radio, max_tokens_slider, temperature_slider, top_p_slider], | |
| # The outputs list MUST match the order and number of code blocks we want to update | |
| outputs=[html_code_output, backend_code_output], | |
| ) | |
| if __name__ == "__main__": | |
| if not API_TOKEN: | |
| print("Warning: HF_TOKEN environment variable not set. Using anonymous access.") | |
| demo.queue(max_size=10).launch() # Allow queueing |