Update app.py

app.py

@@ -15,10 +15,11 @@ model.eval()
 # ------------------------
 # Helpers
 # ------------------------
-def split_sentences(text):
-    # Split by punctuation
-    sentences = re.split(r'(?<=[.!?])\s+', text.strip())
-    return [s for s in sentences if s]
+def split_paragraphs(text):
+    return [p.strip() for p in text.split("\n") if p.strip()]
+
+def split_sentences(paragraph):
+    return re.split(r'(?<=[.!?])\s+', paragraph.strip())
 
 def clean_sentence(sent):
     sent = re.sub(r'\s+', ' ', sent).strip()
@@ -26,12 +27,15 @@ def clean_sentence(sent):
         sent += "."
     return sent
 
-FILLERS = ["actually", "indeed", "quite", "essentially", "additionally", "remarkably"]
+FILLERS = ["actually", "indeed", "essentially", "remarkably", "interestingly", "notably"]
 
-def add_fillers(sentence):
+def maybe_add_filler(sentence, add=False):
+    """Insert a filler only if flagged."""
+    if not add:
+        return sentence
     words = sentence.split()
-    if len(words) > 6:  # only add if long enough
-        insert_pos = random.randint(2, min(len(words)-2, 8))
+    if len(words) > 6:
+        insert_pos = random.randint(2, min(len(words) - 2, 8))
         filler = random.choice(FILLERS)
         words.insert(insert_pos, filler)
     return " ".join(words)
@@ -43,29 +47,41 @@ def humanize_text(text, temperature=1.0, top_p=0.92):
     if not text.strip():
         return "⚠️ Please enter some text"
 
-    sentences = split_sentences(text)
-    paraphrased_sentences = []
+    paragraphs = split_paragraphs(text)
+    final_output = []
+
+    for paragraph in paragraphs:
+        sentences = split_sentences(paragraph)
+        paraphrased_sentences = []
+
+        for i, sent in enumerate(sentences):
+            if not sent.strip():
+                continue
+
+            # Run through model
+            input_text = "paraphrase: " + sent + " </s>"
+            inputs = tokenizer([input_text], return_tensors="pt", truncation=True, padding=True).to(device)
+
+            outputs = model.generate(
+                **inputs,
+                max_new_tokens=80,
+                do_sample=True,
+                top_p=float(top_p),
+                temperature=float(temperature)
+            )
 
-    for sent in sentences:
-        input_text = "paraphrase: " + sent + " </s>"
-        inputs = tokenizer([input_text], return_tensors="pt", truncation=True, padding=True).to(device)
+            decoded = tokenizer.decode(outputs[0], skip_special_tokens=True)
+            decoded = clean_sentence(decoded)
 
-        outputs = model.generate(
-            **inputs,
-            max_new_tokens=80,
-            do_sample=True,
-            top_p=float(top_p),
-            temperature=float(temperature)
-        )
+            # Add filler only to one sentence per paragraph
+            add_filler_here = (i == random.randint(0, max(0, len(sentences)-1)))
+            final_sentence = maybe_add_filler(decoded, add=add_filler_here)
 
-        decoded = tokenizer.decode(outputs[0], skip_special_tokens=True)
-        decoded = clean_sentence(decoded)
+            paraphrased_sentences.append(final_sentence)
 
-        # Add filler word for naturalness
-        final_sentence = add_fillers(decoded)
-        paraphrased_sentences.append(final_sentence)
+        final_output.append(" ".join(paraphrased_sentences))
 
-    return " ".join(paraphrased_sentences)
+    return "\n\n".join(final_output)
 
 # ------------------------
 # Gradio Interface
@@ -78,8 +94,8 @@ iface = gr.Interface(
         gr.Slider(0.6, 1.0, step=0.01, value=0.92, label="Top-p"),
     ],
     outputs=gr.Textbox(label="Final Humanized Text"),
-    title="⚡ Writenix Fast Humanizer",
-    description="Fast pipeline: Parrot paraphraser + smart filler injection. Keeps full text, avoids truncation, adds subtle human touch."
+    title="⚡ Writenix Humanizer (Balanced)",
+    description="Parrot paraphraser + subtle filler words. Injects fillers only once per paragraph for natural variation."
 )
 
 iface.launch()