app.py

import gradio as gr
import csv
import re
import tempfile
import os

import torch
from transformers import AutoTokenizer, AutoModelForSeq2SeqLM, pipeline

MODEL_NAME = "google/flan-t5-large"

tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME)
model = AutoModelForSeq2SeqLM.from_pretrained(
    MODEL_NAME,
    torch_dtype=torch.float16 if torch.cuda.is_available() else None
)
if torch.cuda.is_available():
    model = model.to("cuda")

pipe = pipeline(
    "text2text-generation",
    model=model,
    tokenizer=tokenizer,
    device=0 if torch.cuda.is_available() else -1
)

def score_qa(question, answer):
    """
    Prompt Flan-T5 to return just an integer score.
    Fallback to 5 if we fail to parse a number.
    """
    prompt = f"""
You are an ecolinguistics judge. You will be given a question and an answer.
Please return a single integer score from 0 to 5 based on how well the answer
addresses ecological concerns, clarity, and factual correctness.

Only return the number, nothing else.

Question: {question}
Answer: {answer}

Score:
"""
    result = pipe(prompt, max_new_tokens=3, do_sample=False)[0]["generated_text"]
    match = re.search(r"\d+", result)
    if match:
        return int(match.group(0))
    return 5

def judge_ecolinguistics_from_csv(csv_file):
    """
    1. Parse the CSV for each question–answer pair.
    2. Use the model to score each pair.
    3. Save a new CSV with (question_number, score) per row + a final Total row.
    4. Return a path to the new CSV for download.
    """
    # Read rows from the uploaded CSV
    rows = []
    with open(csv_file.name, "r", encoding="utf-8") as f:
        reader = csv.DictReader(f)
        for r in reader:
            rows.append(r)
    
    results = []
    total_score = 0

    for r in rows:
        question_number = r.get("question_number", "")
        question = r.get("question", "")
        answer = r.get("answer", "")

        sc = score_qa(question, answer)
        total_score += sc

        results.append({
            "question_number": question_number,
            "score": sc
        })
    
    # Write results to a new CSV in text mode
    with tempfile.NamedTemporaryFile(mode="w", delete=False, suffix=".csv", encoding="utf-8") as out_file:
        fieldnames = ["question_number", "score"]
        writer = csv.DictWriter(out_file, fieldnames=fieldnames)
        writer.writeheader()
        for row in results:
            writer.writerow(row)
        # Add a final row showing the total
        writer.writerow({"question_number": "Total", "score": total_score})
        out_path = out_file.name

    # Return the path to the CSV so Gradio can serve it as a downloadable file
    return out_path

demo = gr.Interface(
    fn=judge_ecolinguistics_from_csv,
    inputs=gr.File(label="Upload a CSV with question_number, question, answer columns"),
    outputs=gr.File(label="Download the scored CSV"),
    title="Ecolinguistics Q&A Scoring",
    description=(
        "Upload a CSV containing columns: question_number, question, and answer. "
        "The model assigns each answer a score (0-5) based on ecolinguistic criteria. "
        "A final row shows the total score across all questions."
    )
)

if __name__ == "__main__":
    demo.launch()