Update app.py

app.py

@@ -20,23 +20,27 @@ def paraphrase_t5(text, temperature=0.9, top_p=0.92):
     paraphrased_paragraphs = []
 
     for p in paragraphs:
-        inputs = tokenizer([f"paraphrase: {p}"], return_tensors="pt", truncation=True, padding=True).to(device)
+        # Stronger paraphrasing prompt
+        prompt = f"Paraphrase this in a more natural, human style while keeping meaning:\n{p}"
+
+        inputs = tokenizer([prompt], return_tensors="pt", truncation=True, padding=True).to(device)
 
         outputs = model.generate(
             **inputs,
             max_new_tokens=256,
             do_sample=True,
-            top_p=float(top_p),       # nucleus sampling
-            temperature=float(temperature),  # creativity
-            num_return_sequences=1
+            top_p=float(top_p),
+            temperature=float(temperature),
+            num_return_sequences=1,
+            no_repeat_ngram_size=3  # avoid repeats like 'inequality, inequality'
         )
 
         paraphrased = tokenizer.decode(outputs[0], skip_special_tokens=True)
         paraphrased_paragraphs.append(paraphrased)
 
-    # Join paraphrased paragraphs with newlines
     return "\n\n".join(paraphrased_paragraphs)
 
+
 # Gradio UI
 iface = gr.Interface(
     fn=paraphrase_t5,