Update app.py

app.py

@@ -1,5 +1,6 @@
 from transformers import AutoTokenizer, AutoModelForSeq2SeqLM
 import torch, gradio as gr
+import re
 
 # Load Model (lighter + faster)
 model_name = "humarin/chatgpt_paraphraser_on_T5_base"
@@ -10,38 +11,48 @@ device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
 model = model.to(device)
 model.eval()
 
-# Paraphrasing Function with Paragraph Splitting
+# --- Helpers ---
+def split_sentences(paragraph):
+    # Split into sentences based on punctuation + space
+    sentences = re.split(r'(?<=[.!?])\s+', paragraph.strip())
+    return [s for s in sentences if s]
+
+# --- Main Paraphrasing Function ---
 def paraphrase_t5(text, temperature=0.9, top_p=0.92):
     if not text.strip():
         return "⚠️ Please enter some text"
 
-    # Split by paragraphs
     paragraphs = [p.strip() for p in text.split("\n") if p.strip()]
     paraphrased_paragraphs = []
 
     for p in paragraphs:
-        # Stronger paraphrasing prompt
-        prompt = f"Paraphrase this in a more natural, human style while keeping meaning:\n{p}"
+        sentences = split_sentences(p)
+        paraphrased_sentences = []
 
-        inputs = tokenizer([prompt], return_tensors="pt", truncation=True, padding=True).to(device)
+        for s in sentences:
+            prompt = f"Paraphrase this in a more natural, human style while keeping meaning:\n{s}"
 
-        outputs = model.generate(
-            **inputs,
-            max_new_tokens=256,
-            do_sample=True,
-            top_p=float(top_p),
-            temperature=float(temperature),
-            num_return_sequences=1,
-            no_repeat_ngram_size=3  # avoid repeats like 'inequality, inequality'
-        )
+            inputs = tokenizer([prompt], return_tensors="pt", truncation=True, padding=True).to(device)
 
-        paraphrased = tokenizer.decode(outputs[0], skip_special_tokens=True)
-        paraphrased_paragraphs.append(paraphrased)
+            outputs = model.generate(
+                **inputs,
+                max_new_tokens=512,   # allow long outputs
+                do_sample=True,
+                top_p=float(top_p),
+                temperature=float(temperature),
+                num_return_sequences=1,
+                no_repeat_ngram_size=3
+            )
 
-    return "\n\n".join(paraphrased_paragraphs)
+            paraphrased = tokenizer.decode(outputs[0], skip_special_tokens=True).strip()
+            paraphrased_sentences.append(paraphrased)
 
+        # Rejoin sentences into a paragraph
+        paraphrased_paragraphs.append(" ".join(paraphrased_sentences))
+
+    return "\n\n".join(paraphrased_paragraphs)
 
-# Gradio UI
+# --- Gradio UI ---
 iface = gr.Interface(
     fn=paraphrase_t5,
     inputs=[