app.py

import spaces
import gradio as gr
from transformers import pipeline, TextIteratorStreamer
import torch
import threading
import os

# Load model and tokenizer
model_name = os.getenv("MODEL_ID")
pipe = pipeline("text-generation", model=model_name, device=0)
tokenizer = pipe.tokenizer
model = pipe.model

# Fixed generation config
MAX_TOKENS = 3000
TEMPERATURE = 0.1
TOP_P = 0.9

@spaces.GPU
def respond_stream(population, intervention, comparison, outcome, study_design, summary, title, abstract):
    # Validate required fields
    if not title.strip() or not abstract.strip():
        return "❌ Error: Title and Abstract are required."

    criteria_parts = []
    if population.strip():
        criteria_parts.append(f"Population of interest = {population.strip()}")
    if intervention.strip():
        criteria_parts.append(f"Intervention/exposure of interest = {intervention.strip()}")
    if comparison.strip():
        criteria_parts.append(f"Comparison of interest = {comparison.strip()}")
    if outcome.strip():
        criteria_parts.append(f"Outcome of interest = {outcome.strip()}")
    if study_design.strip():
        criteria_parts.append(f"Study design of interest = {study_design.strip()}")

    if not criteria_parts:
        return "❌ Error: At least one of the five PICOS criteria must be filled."

    # Build instruction section
    instruction = "Instruction: " + "\n".join(criteria_parts)

    # Construct full prompt
    prompt = instruction
    if summary.strip():
        prompt += f"\n\nPICOS Summary: {summary.strip()}"
    prompt += f"\n\nTitle: {title.strip()}\nAbstract: {abstract.strip()}"

    # Wrap into message for chat template
    messages = [{"role": "user", "content": prompt}]
    prompt_text = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)

    # Tokenize and prepare streamer
    inputs = tokenizer(prompt_text, return_tensors="pt").to("cuda")
    streamer = TextIteratorStreamer(tokenizer, skip_prompt=True, skip_special_tokens=True)

    generation_kwargs = dict(
        input_ids=inputs["input_ids"],
        streamer=streamer,
        max_new_tokens=MAX_TOKENS,
        temperature=TEMPERATURE,
        top_p=TOP_P,
        do_sample=True,
        pad_token_id=tokenizer.eos_token_id,
    )

    thread = threading.Thread(target=model.generate, kwargs=generation_kwargs)
    thread.start()

    partial_text = ""
    for token in streamer:
        partial_text += token
        yield partial_text

# Build Gradio interface
with gr.Blocks() as demo:
    gr.Markdown("## PICO screener")

    with gr.Column():
        population = gr.Textbox(label="Population of interest", lines=1)
        intervention = gr.Textbox(label="Intervention/exposure of interest", lines=1)
        comparison = gr.Textbox(label="Comparison of interest", lines=1)
        outcome = gr.Textbox(label="Outcome of interest", lines=1)
        study_design = gr.Textbox(label="Study design of interest", lines=1)

    with gr.Column():
        summary = gr.Textbox(label="PICOS Summary (optional)", lines=4)
        title = gr.Textbox(label="Title", lines=2, placeholder="Required")
        abstract = gr.Textbox(label="Abstract", lines=10, placeholder="Required")

    output_box = gr.Textbox(label="Model Response", lines=15, interactive=False)
    generate_btn = gr.Button("Generate")

    generate_btn.click(
        fn=respond_stream,
        inputs=[population, intervention, comparison, outcome, study_design, summary, title, abstract],
        outputs=[output_box]
    )

# Launch the app
if __name__ == "__main__":
    demo.launch()