Update app.py

app.py

@@ -8,19 +8,18 @@ subprocess.check_call([sys.executable, "-m", "spacy", "download", "en_core_web_s
 nlp = spacy.load("en_core_web_sm")
 spell = SpellChecker(language='en')
 
-# load model once
+# 🔧 1) raise max_new_tokens so each chunk can be LONGER
 para = pipeline(
     "text2text-generation",
     model="Vamsi/T5_Paraphrase_Paws",
     device=-1,
-    max_new_tokens=1024  # ← allow longer outputs per chunk
+    max_new_tokens=2000  # 2000 tokens ≈ 1 500 words
 )
 
 def simple_correct(text):
     return " ".join(spell.correction(w) or w for w in text.split())
 
-def chunk_text(text, max_words=380):
-    """Split into exact word-length chunks."""
+def chunk_text(text, max_words=350):
     words = text.split()
     for i in range(0, len(words), max_words):
         yield " ".join(words[i:i + max_words])
@@ -29,33 +28,30 @@ def humanize(text, tone, strength, freeze):
     locked = [(ent.text, ent.label_) for ent in nlp(text).ents] if freeze else []
     out_chunks = []
 
-    for chunk in chunk_text(text, 380):
+    for chunk in chunk_text(text, 350):
+        # 🔧 2) tell the model *not to summarize*
+        prompt = f"Rewrite the following text naturally without shortening it:\n{chunk}"
         paraphrased = para(
-            chunk,
-            max_new_tokens=1024,
+            prompt,
+            max_new_tokens=2000,
             do_sample=True,
             temperature=0.7 * strength / 10
         )[0]['generated_text']
 
         paraphrased = simple_correct(paraphrased)
-
-        # restore locked entities
         for ent, label in locked:
             paraphrased = re.sub(re.escape(ent), ent, paraphrased, flags=re.IGNORECASE)
-
         out_chunks.append(paraphrased)
 
     full_text = " ".join(out_chunks)
 
-    # downloadable file
     file_path = "/tmp/full_humanized.txt"
     with open(file_path, "w", encoding="utf-8") as f:
         f.write(full_text)
-
     return full_text, file_path
 
 with gr.Blocks(title="AI Humanizer – Exact Length") as demo:
-    gr.Markdown("## AI Humanizer – Exact Length, No Compression")
+    gr.Markdown("## AI Humanizer – Exact Length, No Shortening")
     with gr.Row():
         with gr.Column():
             text_in = gr.Textbox(label="Paste your text (any length)", lines=15, max_lines=None)
@@ -65,15 +61,14 @@ with gr.Blocks(title="AI Humanizer – Exact Length") as demo:
             submit_btn = gr.Button("Humanize", variant="primary")
 
         with gr.Column():
-            text_out = gr.Textbox(label="Full humanized text (same word-count)", lines=25, interactive=True)
+            # 🔧 3) make the output box big enough
+            text_out = gr.Textbox(label="Humanized text (same length)", lines=30, interactive=True)
             with gr.Row():
                 copy_btn = gr.Button("📋 Copy")
                 file_out = gr.File(label="Download .txt")
 
     submit_btn.click(humanize, inputs=[text_in, tone_dd, strength_sl, lock_cb],
                      outputs=[text_out, file_out])
-
-    copy_btn.click(None, text_out, None,
-                   js="(txt) => navigator.clipboard.writeText(txt)")
+    copy_btn.click(None, text_out, None, js="(txt) => navigator.clipboard.writeText(txt)")
 
 demo.launch()