Update app.py

app.py

@@ -1,46 +1,112 @@
 import gradio as gr
-from transformers import pipeline
+import csv
+import re
+import tempfile
+import os
 
-# Load the Flan-T5-Large model and tokenizer via a pipeline.
-# If you're on a GPU Space, device=0 will place it on GPU.
-# If you're on CPU only, leave device=-1.
-model_name = "google/flan-t5-large"
-pipe = pipeline("text2text-generation", model=model_name, device=-1)
+import torch
+from transformers import AutoTokenizer, AutoModelForSeq2SeqLM, pipeline
 
-def judge_ecolinguistics(pairs_text):
+# MODEL_NAME can be "google/flan-t5-large", or switch to "google/flan-t5-base" if you need faster/less memory usage.
+MODEL_NAME = "google/flan-t5-large"
+
+# Load the tokenizer and model. 
+# If you're on a GPU Space, device=0 places the model on GPU. 
+# If CPU only, leave device=-1.
+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):
     """
-    Takes a multiline string of question–answer pairs and returns a model-generated
-    scoring from 1 to 10 for each pair, along with a brief explanation.
+    Prompt Flan-T5 to return just an integer score from 1–10 (no extra text).
+    Fallback to 5 if we fail to parse a number.
     """
-    # Construct a single prompt that instructs the model to score each Q&A pair.
     prompt = f"""
-You are an ecolinguistics judge. You evaluate answers based on how thoroughly
-they address ecological concerns, clarity of expression, and factual correctness.
+You are an ecolinguistics judge. You will be given a question and an answer.
+Please return a single integer score from 1 to 10 based on how well the answer
+addresses ecological concerns, clarity, and factual correctness.
 
-Below is a set of question–answer pairs:
+Only return the number, nothing else.
 
-{pairs_text}
+Question: {question}
+Answer: {answer}
 
-Please provide, for each pair, a single numerical score from 1 to 10 and a brief explanation.
+Score:
 """
-    # Use the pipeline to generate a response.
-    response = pipe(prompt, max_length=512, truncation=True)[0]["generated_text"]
-    return response
+    result = pipe(prompt, max_new_tokens=3, do_sample=False)[0]["generated_text"]
+    # Extract the first integer found in the output
+    match = re.search(r"\d+", result)
+    if match:
+        return int(match.group(0))
+    # If no number is found, return a default (e.g. 5)
+    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 plus a total row.
+    4. Return the path to the new CSV for download.
+    """
+    # Read rows from the uploaded CSV
+    # Adjust these column names to match your actual CSV headers
+    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:
+        # Extract needed columns; adapt to match your CSV
+        question_number = r.get("question_number", "")
+        question = r.get("question", "")
+        answer = r.get("answer", "")
+
+        # Score each Q&A with the model
+        sc = score_qa(question, answer)
+        total_score += sc
+
+        results.append({
+            "question_number": question_number,
+            "score": sc
+        })
+    
+    # Write results to a new CSV
+    # We'll place it in a temporary file so Gradio can return it
+    out_file = tempfile.NamedTemporaryFile(delete=False, suffix=".csv")
+    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_file.close()
+
+    # Return the path to the CSV so Gradio can serve it as a downloadable file
+    return out_file.name
 
-# Build the Gradio interface.
 demo = gr.Interface(
-    fn=judge_ecolinguistics,
-    inputs=gr.Textbox(
-        lines=10,
-        label="Enter Your Question–Answer Pairs",
-        placeholder="Example:\nQ1: What is an ecological niche?\nA1: It is the role a species plays in its environment.\n\nQ2: How does deforestation affect the climate?\nA2: It can reduce carbon sequestration and disrupt rainfall patterns.\n"
-    ),
-    outputs="text",
-    title="Ecolinguistics Q&A Scorer (Flan-T5-Large)",
+    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 (Flan‑T5‑Large)",
     description=(
-        "Paste multiple question–answer pairs. The model will assign a score from 1–10 "
-        "to each answer, considering ecological relevance and clarity. "
-        "It will also provide a brief rationale for its scoring."
+        "Upload a CSV containing columns: question_number, question, and answer. "
+        "The model assigns each answer a score (1–10) based on ecolinguistic criteria. "
+        "A final row shows the total score across all questions."
     )
 )