Update app.py

app.py

@@ -1,53 +1,53 @@
 import gradio as gr
-from transformers import T5Tokenizer, T5ForConditionalGeneration
+import re
+import torch
+from transformers import AutoTokenizer, AutoModelForSeq2SeqLM, pipeline
 
-# Load T5-small model and tokenizer
+# Load T5-small paraphrase model
 model_name = "t5-small"
-tokenizer = T5Tokenizer.from_pretrained(model_name)
-model = T5ForConditionalGeneration.from_pretrained(model_name)
+tokenizer = AutoTokenizer.from_pretrained(model_name)
+model = AutoModelForSeq2SeqLM.from_pretrained(model_name, torch_dtype=torch.float16)  # Use fp16 for speed
+
+# Move model to CPU (remove if using GPU)
+model.to("cpu")
+
+# Initialize paraphrase pipeline with optimized settings
+paraphrase_pipeline = pipeline(
+    "text2text-generation",
+    model=model,
+    tokenizer=tokenizer,
+    truncation=True
+)
+
+def split_sentences(text):
+    """Split text into sentences using regex (faster than nltk)."""
+    return re.split(r'(?<=[.!?])\s+', text.strip())
 
-def generate_paraphrase(text, max_length, temperature, num_outputs):
-    """Generate paraphrased versions of the input text using T5-small."""
+def paraphrase_text(text):
+    """Paraphrases input text while maintaining sentence structure."""
     if not text.strip():
-        return ["⚠️ Please enter some text to paraphrase."]
-
-    input_text = f"paraphrase: {text} </s>"
-    input_ids = tokenizer.encode(input_text, return_tensors="pt")
-
-    outputs = model.generate(
-        input_ids,
-        max_length=max_length,
-        top_k=50,
-        top_p=0.95,
-        num_return_sequences=num_outputs,
-        do_sample=True
+        return "⚠️ Please enter some text to paraphrase."
+
+    sentences = split_sentences(text)
+
+    # Apply T5 paraphrasing to each sentence
+    paraphrased_results = paraphrase_pipeline(
+        [f"paraphrase: {sentence} </s>" for sentence in sentences if sentence], 
+        max_length=50, do_sample=True, batch_size=8, num_return_sequences=1  # Faster settings
     )
 
-    paraphrased_texts = [tokenizer.decode(output, skip_special_tokens=True) for output in outputs]
-    return paraphrased_texts  # Returns a list of paraphrases
+    paraphrased_sentences = [result['generated_text'] for result in paraphrased_results]
+    return " ".join(paraphrased_sentences)
 
 # Define Gradio Interface
-description = """
-## ✨ AI Paraphrasing Tool
-Enter a sentence and let AI generate multiple paraphrased versions!
-- Adjust **max length** for longer outputs.
-- Tune **temperature** for more creative results.
-- Choose **number of outputs** to generate multiple variations.
-"""
-
 demo = gr.Interface(
-    fn=generate_paraphrase,
-    inputs=[
-        gr.Textbox(label="Enter text", placeholder="Type a sentence to paraphrase..."),
-        gr.Slider(20, 100, value=50, step=5, label="Max Output Length"),
-        gr.Slider(0.5, 1.5, value=1.0, step=0.1, label="Creativity (Temperature)"),
-        gr.Dropdown(choices=[1, 2, 3, 4, 5], value=1, label="Number of Outputs")
-    ],
-    outputs=gr.Textbox(label="Paraphrased Text", lines=5),  # Allows multiple outputs
-    title="📝 AI Paraphraser",
-    description=description,
-    theme="huggingface",
-    live=True,
+    fn=paraphrase_text,
+    inputs=gr.Textbox(label="Enter text", placeholder="Type your text to paraphrase...", lines=10),
+    outputs=gr.Textbox(label="Paraphrased Text", lines=10),
+    title="🚀 Fast & Clean T5-Small Paraphraser",
+    description="Enter text and let AI generate a paraphrased version using an optimized T5-small model!",
+    theme="huggingface"
 )
 
-demo.launch()
+if __name__ == "__main__":
+    demo.launch()