Spaces:
Sleeping
Sleeping
| import gradio as gr | |
| from transformers import GPT2Tokenizer, GPTNeoForCausalLM | |
| import torch | |
| import traceback | |
| import gc | |
| import psutil | |
| import time | |
| # Enable garbage collection | |
| gc.enable() | |
| # Properly handle CUDA settings | |
| if torch.cuda.is_available(): | |
| torch.backends.cudnn.benchmark = False | |
| torch.backends.cudnn.deterministic = True | |
| # Global variables | |
| model = None | |
| tokenizer = None | |
| def log_memory_usage(): | |
| process = psutil.Process() | |
| memory_info = process.memory_info() | |
| print(f"Memory usage: {memory_info.rss / 1024 / 1024:.2f} MB") | |
| def load_model(): | |
| global model, tokenizer | |
| try: | |
| print("Starting model loading...") | |
| log_memory_usage() | |
| tokenizer = GPT2Tokenizer.from_pretrained( | |
| "harshagnihotri14/SOP_Generator", | |
| local_files_only=False | |
| ) | |
| model = GPTNeoForCausalLM.from_pretrained( | |
| "harshagnihotri14/SOP_Generator", | |
| low_cpu_mem_usage=True, | |
| torch_dtype=torch.float32, | |
| device_map='auto' | |
| ) | |
| print("Model loaded successfully") | |
| log_memory_usage() | |
| return True | |
| except Exception as e: | |
| print(f"Error loading model: {str(e)}") | |
| return False | |
| def cleanup(): | |
| gc.collect() | |
| if torch.cuda.is_available(): | |
| torch.cuda.empty_cache() | |
| log_memory_usage() | |
| def generate_sop_with_retry(name, course_name, university, university_location, academic_background, | |
| marks, subjects, work_experience, extracurricular_activities, | |
| interests, short_term_goals, long_term_goals, max_retries=2): | |
| """Attempt to generate SOP multiple times if output is too short""" | |
| for attempt in range(max_retries): | |
| sop = generate_sop_attempt(name, course_name, university, university_location, | |
| academic_background, marks, subjects, work_experience, | |
| extracurricular_activities, interests, short_term_goals, | |
| long_term_goals) | |
| if not isinstance(sop, str) or "Error:" in sop: | |
| continue | |
| word_count = len(sop.split()) | |
| if word_count >= 800: # Reduced from 1000 for better performance | |
| return sop | |
| return "Error: Unable to generate a sufficiently detailed SOP after multiple attempts. Please try again with more detailed input information." | |
| def generate_sop_attempt(name, course_name, university, university_location, academic_background, | |
| marks, subjects, work_experience, extracurricular_activities, | |
| interests, short_term_goals, long_term_goals, progress=gr.Progress()): | |
| try: | |
| if model is None or tokenizer is None: | |
| progress(0.1, desc="Loading model...") | |
| success = load_model() | |
| if not success: | |
| return "Error: Failed to load the model. Please try again." | |
| progress(0.3, desc="Preparing prompt...") | |
| # Simplified prompt structure for better performance | |
| prompt = ( | |
| f"Write a Statement of Purpose (SOP) for {name} applying to {course_name} " | |
| f"at {university}, {university_location}.\n\n" | |
| f"Academic Background: {academic_background}\n" | |
| f"Academic Performance: {marks}\n" | |
| f"Key Subjects: {subjects}\n" | |
| f"Work Experience: {work_experience}\n" | |
| f"Extracurricular Activities: {extracurricular_activities}\n" | |
| f"Interests: {interests}\n" | |
| f"Short-term Goals: {short_term_goals}\n" | |
| f"Long-term Goals: {long_term_goals}\n\n" | |
| "Write a comprehensive SOP (800-1000 words) that includes:\n" | |
| "1. Strong introduction with personal motivation\n" | |
| "2. Academic achievements and relevant coursework\n" | |
| "3. Professional experience and projects\n" | |
| "4. Research interests and activities\n" | |
| "5. Extracurricular activities and leadership\n" | |
| "6. Career goals and program fit\n" | |
| "7. Conclusion\n\n" | |
| "BEGIN SOP:\n" | |
| ) | |
| progress(0.5, desc="Tokenizing input...") | |
| inputs = tokenizer(prompt, return_tensors="pt", truncation=True, max_length=1024) | |
| progress(0.7, desc="Generating SOP...") | |
| # Optimized generation parameters for CPU | |
| with torch.no_grad(): | |
| outputs = model.generate( | |
| **inputs, | |
| max_length=2048, # Reduced for better performance | |
| min_length=500, | |
| temperature=0.7, | |
| repetition_penalty=1.5, | |
| num_beams=4, # Reduced for better performance | |
| length_penalty=2.0, | |
| top_p=0.85, | |
| top_k=50, | |
| do_sample=True, | |
| pad_token_id=tokenizer.eos_token_id | |
| ) | |
| progress(0.9, desc="Processing output...") | |
| generated_text = tokenizer.decode(outputs[0], skip_special_tokens=True) | |
| # Extract the SOP content | |
| sop_start = generated_text.find("BEGIN SOP:") | |
| if sop_start != -1: | |
| sop_text = generated_text[sop_start + len("BEGIN SOP:"):].strip() | |
| else: | |
| sop_text = generated_text.split("Now, write a detailed")[-1].strip() | |
| # Format the final output | |
| formatted_sop = ( | |
| "STATEMENT OF PURPOSE\n\n" | |
| f"Name: {name}\n" | |
| f"Program: {course_name}\n" | |
| f"University: {university}, {university_location}\n\n" | |
| f"{sop_text}" | |
| ) | |
| cleanup() | |
| progress(1.0, desc="Done!") | |
| return formatted_sop | |
| except Exception as e: | |
| cleanup() | |
| error_msg = f"Error during SOP generation: {str(e)}\n{traceback.format_exc()}" | |
| print(error_msg) | |
| return f"An error occurred while generating the SOP. Please try again. Error: {str(e)}" | |
| def generate_sop(*args): | |
| """Wrapper function to handle the retry logic""" | |
| return generate_sop_with_retry(*args) | |
| # Gradio interface with optimized settings | |
| with gr.Blocks(theme=gr.themes.Soft()) as demo: | |
| gr.Markdown(""" | |
| # Professional SOP Generator | |
| Generate a comprehensive Statement of Purpose for your graduate school application. | |
| **Guidelines:** | |
| - Provide detailed information in all fields | |
| - Include specific examples and achievements | |
| - The generated SOP will be approximately 800-1000 words | |
| """) | |
| with gr.Row(): | |
| with gr.Column(): | |
| name = gr.Textbox(label="Full Name", placeholder="Enter your full name") | |
| course_name = gr.Textbox(label="Course Name", placeholder="Enter the program name") | |
| university = gr.Textbox(label="University", placeholder="Enter university name") | |
| university_location = gr.Textbox(label="University Location", placeholder="City, Country") | |
| academic_background = gr.Textbox(label="Academic Background", placeholder="Your current/previous degree") | |
| marks = gr.Textbox(label="Academic Marks/Grades", placeholder="Your GPA or percentage") | |
| with gr.Column(): | |
| subjects = gr.Textbox(label="Major Subjects", placeholder="List your major subjects") | |
| work_experience = gr.Textbox(label="Work Experience", placeholder="Relevant work experience") | |
| extracurricular_activities = gr.Textbox(label="Extracurricular Activities", placeholder="List your activities") | |
| interests = gr.Textbox(label="Personal Interests", placeholder="Your interests and hobbies") | |
| short_term_goals = gr.Textbox(label="Short Term Goals", placeholder="Your immediate career goals") | |
| long_term_goals = gr.Textbox(label="Long Term Goals", placeholder="Your long-term career aspirations") | |
| submit = gr.Button("Generate SOP", variant="primary") | |
| output = gr.Textbox(label="Generated SOP", lines=30) | |
| # Fixed submit.click() without invalid parameters | |
| submit.click( | |
| fn=generate_sop, | |
| inputs=[ | |
| name, course_name, university, university_location, academic_background, | |
| marks, subjects, work_experience, extracurricular_activities, | |
| interests, short_term_goals, long_term_goals | |
| ], | |
| outputs=output | |
| ) | |
| if __name__ == "__main__": | |
| # Configure Gradio queue for better resource management | |
| demo.queue(max_size=5) # Limit concurrent requests | |
| demo.launch( | |
| server_name="0.0.0.0", | |
| server_port=7860, | |
| share=True, | |
| show_error=True, | |
| debug=True, | |
| ) |