import gradio as gr
from transformers import T5Tokenizer, T5ForConditionalGeneration

# Load the base T5 model and tokenizer
model = T5ForConditionalGeneration.from_pretrained('t5-small')
tokenizer = T5Tokenizer.from_pretrained('t5-small')

def generate_clinical_report(input_text):
    """
    Generate a clinical report from the input text using the T5 model.
    """
    try:
        # Prepare input text
        input_ids = tokenizer.encode("summarize: " + input_text, return_tensors="pt", max_length=512, truncation=True)
        
        # Generate report
        outputs = model.generate(
            input_ids,
            max_length=256,
            num_beams=4,
            no_repeat_ngram_size=3,
            length_penalty=2.0,
            early_stopping=True,
            bad_words_ids=[[tokenizer.encode(word, add_special_tokens=False)[0]] 
                          for word in ['http', 'www', '.com', '.org']]
        )
        
        # Decode and return the generated report
        return tokenizer.decode(outputs[0], skip_special_tokens=True)
    except Exception as e:
        print(f"Error generating report: {str(e)}")
        return f"Error: {str(e)}"

# Create Gradio interface
demo = gr.Interface(
    fn=generate_clinical_report,
    inputs=gr.Textbox(
        lines=8,
        placeholder="Enter clinical notes here...",
        label="Clinical Notes"
    ),
    outputs=gr.Textbox(
        lines=8,
        label="Generated Clinical Report"
    ),
    title="Clinical Report Generator",
    description="Generate professional clinical reports from clinical notes using a T5 model.",
    allow_flagging="never"
)

# Launch the app
if __name__ == "__main__":
    demo.queue(concurrency_count=1)  # Single request at a time for stability
    demo.launch(
        server_name="0.0.0.0",
        server_port=7860,
        share=False,
        root_path="",
        show_api=True,
        allowed_paths=None,  # Allow all paths
        quiet=True  # Reduce logging noise
    )