Update app.py

app.py

@@ -1,5 +1,5 @@
 from transformers import AutoTokenizer, AutoModelForSeq2SeqLM
-import torch, gradio as gr, re, random
+import torch, gradio as gr, re
 
 # --- Load Model ---
 model_name = "prithivida/parrot_paraphraser_on_T5"
@@ -14,64 +14,52 @@ 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
 
-def clean_sentence(sent):
-    sent = re.sub(r'\s+', ' ', sent).strip()
-    if not sent.endswith(('.', '!', '?')):
-        sent += "."
-    return sent
-
-FILLERS = ["in fact", "notably", "interestingly", "remarkably", "as a matter of fact"]
-
-def inject_filler(paragraph):
-    sentences = split_sentences(paragraph)
-    if len(sentences) > 2:
-        idx = random.randint(1, len(sentences)-1)
-        sentences[idx] = FILLERS[random.randint(0, len(FILLERS)-1)].capitalize() + ", " + sentences[idx][0].lower() + sentences[idx][1:]
-    return " ".join(sentences)
-
-# --- Main function ---
-def humanize_text(text):
+# --- Main Humanizer ---
+def humanize_text(text, variants=1, temperature=1.0, top_p=0.92):
     if not text.strip():
         return "⚠️ Please enter some text"
 
     paragraphs = split_paragraphs(text)
-    final_paragraphs = []
+    final_output = []
 
     for para in paragraphs:
-        sentences = split_sentences(para)
-        out_sentences = []
-
-        for sent in sentences:
-            input_text = "paraphrase: " + sent + " </s>"
-            inputs = tokenizer([input_text], return_tensors="pt", truncation=True, padding=True).to(device)
+        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=64,
-                num_beams=3,          # Faster & more stable than sampling
-                do_sample=False       # Deterministic, no temp needed
-            )
+        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
+        )
 
-            decoded = tokenizer.decode(outputs[0], skip_special_tokens=True)
-            out_sentences.append(clean_sentence(decoded))
+        decoded = tokenizer.batch_decode(outputs, skip_special_tokens=True)
+        cleaned = [clean_text(d) for d in decoded]
 
-        # Rebuild paragraph + add filler once
-        new_para = " ".join(out_sentences)
-        new_para = inject_filler(new_para)
-        final_paragraphs.append(new_para)
+        # Pick the first variant for simplicity
+        final_output.append(cleaned[0])
 
-    return "\n\n".join(final_paragraphs)
+    return "\n\n".join(final_output)
 
 # --- Gradio Interface ---
 iface = gr.Interface(
     fn=humanize_text,
-    inputs=gr.Textbox(lines=10, placeholder="Paste text here..."),
+    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"),
+    ],
     outputs=gr.Textbox(label="Humanized Output"),
-    title="⚡ Writenix Fast Humanizer",
-    description="Stable & fast humanizer: deterministic paraphrasing + light filler once per paragraph."
+    title="✨ Writenix Humanizer Pro",
+    description="Paragraph-level paraphrasing with better flow and context. Combines beam search with sampling for more natural results."
 )
 
 iface.launch()