Update app.py

app.py

@@ -7,14 +7,13 @@ import os
 import torch
 from transformers import AutoTokenizer, AutoModelForSeq2SeqLM, pipeline
 
-# 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)
+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")
 
@@ -43,22 +42,19 @@ Answer: {answer}
 Score:
 """
     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.
+    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
-    # 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)
@@ -69,12 +65,10 @@ def judge_ecolinguistics_from_csv(csv_file):
     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
 
@@ -83,20 +77,19 @@ def judge_ecolinguistics_from_csv(csv_file):
             "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()
+    # 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_file.name
+    return out_path
 
 demo = gr.Interface(
     fn=judge_ecolinguistics_from_csv,