Update app.py

app.py

@@ -14,52 +14,53 @@ model.eval()
 def split_paragraphs(text):
     return [p.strip() for p in text.split("\n") if p.strip()]
 
+def split_sentences(text):
+    return re.split(r'(?<=[.!?])\s+', text.strip())
+
 def clean_text(text):
-    text = re.sub(r'\s+', ' ', text).strip()
-    return text
+    return re.sub(r'\s+', ' ', text).strip()
+
+def paraphrase_chunk(text_chunk):
+    inputs = tokenizer([text_chunk], return_tensors="pt", truncation=True, padding=True).to(device)
+    outputs = model.generate(
+        **inputs,
+        max_new_tokens=100,  # small chunks only
+        num_beams=4,
+        do_sample=False
+    )
+    return clean_text(tokenizer.decode(outputs[0], skip_special_tokens=True))
 
 # --- Main Humanizer ---
-def humanize_text(text, variants=1, temperature=1.0, top_p=0.92):
+def humanize_text(text):
     if not text.strip():
         return "⚠️ Please enter some text"
 
     paragraphs = split_paragraphs(text)
-    final_output = []
+    humanized_paragraphs = []
 
     for para in paragraphs:
-        input_text = "paraphrase: " + para + " </s>"
-        inputs = tokenizer([input_text], return_tensors="pt", truncation=True, padding=True).to(device)
-
-        outputs = model.generate(
-            **inputs,
-            max_new_tokens=256,          # More room per paragraph
-            num_return_sequences=variants,
-            do_sample=True,
-            temperature=float(temperature),
-            top_p=float(top_p),
-            num_beams=4,                 # balance between stable & creative
-        )
+        sentences = split_sentences(para)
+        paraphrased_sentences = []
 
-        decoded = tokenizer.batch_decode(outputs, skip_special_tokens=True)
-        cleaned = [clean_text(d) for d in decoded]
+        # Paraphrase each sentence separately for accuracy
+        for sent in sentences:
+            paraphrased_sentences.append(paraphrase_chunk("paraphrase: " + sent))
 
-        # Pick the first variant for simplicity
-        final_output.append(cleaned[0])
+        # Rebuild paragraph and optionally add light filler once
+        new_para = " ".join(paraphrased_sentences)
+        if len(paraphrased_sentences) > 2:
+            new_para = "In fact, " + new_para[0].lower() + new_para[1:]
+        humanized_paragraphs.append(new_para)
 
-    return "\n\n".join(final_output)
+    return "\n\n".join(humanized_paragraphs)
 
 # --- Gradio Interface ---
 iface = gr.Interface(
     fn=humanize_text,
-    inputs=[
-        gr.Textbox(lines=10, placeholder="Paste text here..."),
-        gr.Slider(1, 3, step=1, value=1, label="Variants"),
-        gr.Slider(0.5, 2.0, step=0.1, value=1.0, label="Temperature"),
-        gr.Slider(0.6, 1.0, step=0.01, value=0.92, label="Top-p"),
-    ],
+    inputs=gr.Textbox(lines=15, placeholder="Paste text here..."),
     outputs=gr.Textbox(label="Humanized Output"),
-    title="✨ Writenix Humanizer Pro",
-    description="Paragraph-level paraphrasing with better flow and context. Combines beam search with sampling for more natural results."
+    title="✨ Writenix Humanizer v3",
+    description="Paraphrases large text paragraph-by-paragraph for natural flow, keeps full content, adds light filler per paragraph."
 )
 
 iface.launch()